AI Revolution Hub

Own the Layer That Learns

Wilson CELY — Fri, 08 May 2026 02:17:00 +0000

How Commoditization Eats the Chassis and Software Captures the Margin

The car industry is not just going electric. It is being reclassified.

For more than a century, automakers won by mastering the physical machine: engines, transmissions, chassis, manufacturing quality, and brand trust. But the electric vehicle transition has quietly changed the basis of competition. The physical car still matters, but it is no longer where the most strategic value is accumulating.

The new battlefield is the intelligence layer: the operating system, the assisted-driving stack, the digital cockpit, the data feedback loop, and the software that decides how the product improves after it leaves the factory.

That shift is forcing automakers into a brutal choice.

Some are outsourcing their digital brains to move faster. Others are spending billions to keep their software proprietary. The difference between those two strategies may decide who remains a product company and who becomes a hardware wrapper.

Some layers are once difficult, expensive, and strategically powerful. Over time, they become standardized, accessible, and cheaper. That is commoditization.

When a layer commoditizes, the profit pool does not vanish. It migrates to the next layer that is still scarce, complex, and strategically important. That new layer becomes the differentiator.

The product leader’s job is to know which layer is losing power and which layer is gaining it.

In automotive, the internal combustion engine used to be the defining layer. It shaped brand identity, performance, maintenance, and customer loyalty. But in electric vehicles, batteries and motors are becoming more standardized. They are still hard to manufacture well, but they are less able to create lasting differentiation on their own.

The new differentiator is software intelligence.

The car is becoming a physical platform that hosts a continuously improving digital system. That means the most important question is no longer: “Who built the car?”

It is: “Who owns the layer that gets smarter every time the car is used?”

The most important automotive story is not simply that Chinese technology companies are gaining power. That is true, but it is not the deeper pattern.

The deeper pattern is that Western automakers are increasingly buying the future operating model of the car industry from external software and technology suppliers.

Toyota’s China-market bZ7 is a striking example. The Toyota badge remains on the car, but much of the experience is powered by external Chinese technology partners. Huawei contributes the cockpit and electric drive technology. Xiaomi connects the vehicle into a broader smart-home ecosystem. Momenta provides advanced assisted-driving capabilities.

From the customer’s perspective, the daily experience is increasingly shaped by the software stack, not by the badge on the hood.

That is the strategic danger.

If the partner owns the interface, the partner shapes the experience. If the partner owns the driving data, the partner improves the system. If the partner improves faster, the partner becomes the real control point in the ecosystem.

Volkswagen has taken a similarly pragmatic route. In China, it is partnering with XPeng to accelerate its local software-defined vehicle strategy. In the West, it has committed billions to Rivian to access zonal architecture and software capability. This may be rational. Volkswagen needs speed. Its internal software arm, CARIAD, struggled with delays and complexity. But the strategic implication is clear: the company is buying back relevance from players that move faster in software.

BMW is taking the opposite bet.

Instead of outsourcing the intelligence layer, BMW is trying to protect it. Its Neue Klasse architecture is built around centralized computing systems, including the “Heart of Joy,” an in-house control system designed to preserve BMW’s driving dynamics as a proprietary differentiator.

This is not just an engineering preference. It is a product strategy.

BMW is effectively saying: if driving feel becomes software-defined, then outsourcing that software means outsourcing the soul of the brand.

Strategic Camp	What They Do	Example	Main Risk

Wrappers

Outsource the intelligence layer to move faster

Toyota in China, parts of VW strategy

Lose control of the user experience

Ecosystem Providers

Own the software, data, and learning loop

Huawei, Xiaomi, XPeng, Rivian

Must prove reliability at massive scale

In-House Believer

Build and protect proprietary intelligence

BMW

High cost, slower iteration, execution risk

The most powerful company in a value chain is not always the one that sells the final product.

It is often the one that controls the layer everyone else depends on.

That is stack capture.

In software-defined vehicles, the capture point is the intelligence layer. This includes ADAS, autonomy, cockpit systems, operating systems, data pipelines, and over-the-air update infrastructure.

Huawei’s automotive strategy shows this clearly. The company does not need to manufacture cars to become powerful in automotive. It can supply the systems that make cars feel intelligent, collect driving data, improve assisted-driving models, and create dependence across multiple OEMs.

That is a more scalable position than building every car itself.

Every vehicle using the stack generates data. Every kilometer improves the model. Every improvement makes the stack more attractive to other automakers. Every new automaker adds more vehicles. The flywheel compounds.

The supplier becomes stronger every time the OEM’s customer uses the product.

The automotive example is dramatic, but the pattern applies far beyond cars.

Every company now faces a similar question: which parts of our product are becoming commodities, and which parts are becoming strategic control points?

This matters because outsourcing is not automatically bad. In fact, strong product companies should outsource commodity layers aggressively. They should not waste resources rebuilding infrastructure that no longer differentiates them.

But they must be extremely careful when outsourcing layers that shape the user experience, capture proprietary data, or improve through feedback loops.

That is where the danger lives.

The wrong build-versus-buy decision can quietly move the company out of the value pool.

The global car may be dying.

Software-defined vehicles are not just products. They are rolling sensor networks. They collect location data, driving behavior, camera inputs, infrastructure signals, and system telemetry. That turns the car from a mechanical export into a geopolitical object.

This creates a new product constraint: software localization.

The United States is moving to restrict connected vehicle technology linked to China and other foreign adversaries. China is enforcing strict data sovereignty rules that limit how vehicle data can leave the country. The result is a fragmented software world where automakers can no longer assume that one global architecture will work everywhere.

Volkswagen’s East-West software strategy reflects this reality. One stack for China. Another for the West. Different partners. Different rules. Different data environments.

That destroys one of the old advantages of global automotive scale.

For product leaders, this is the overlooked lesson: the more your product depends on intelligence, data, and connectivity, the more regulation becomes part of the architecture.

You do not add compliance at the end. You design around regulatory borders from the beginning.

Product Questions to Ask

Which layer of our product stack is actively commoditizing right now?
Which layer is becoming the new source of margin, data, and user control?
Are we treating a future differentiator as a vendor-supplied feature?
If we outsource our intelligence layer, what prevents the supplier from eventually replacing us?
What proprietary data are we surrendering to solve a short-term delivery bottleneck?
If cross-border data transfer became impossible tomorrow, how would our product architecture need to change?

The Reusable Principle

Never outsource the layer of your product that defines the user experience and captures the compounding data loop.

15 years navigating corporate complexity. An entrepreneur's instinct for what matters. The AI fluency to build it. Let's work together →

For more details and sources you can look at our Deep Dive.

Deep Dive Article

beehiiv — The newsletter platform built for growth

Access the best tools available in email, helping your newsletter scale and monetize like never before.

beehiiv.com

Est rerum sunt eum nisi dolores est velit voluptatem non exercitationem consequatur id Quis ipsa. Sed autem fugit eos consequuntur maxime est corrupti laboriosam ea totam amet.

Lorem Ipsum!

[Deep Dive] Chinese Tech in Western Cars

Wilson CELY — Thu, 07 May 2026 09:19:22 +0000

The Intelligence Layer Shift: Stack Capture and the Reconfiguration of the Global Automotive Value Chain

Article - AI Narrated

NotebookLM - Podcast style Analysis

The global automotive sector has entered a period of brutal, structural realignment. The foundational transition from internal combustion engines to battery-electric propulsion, which dominated industry discourse for the past decade, is increasingly recognized as merely a prerequisite to a far more profound disruption. The true battleground has migrated upward within the vehicle's architecture. The core differentiator of the modern automobile is no longer mechanical engineering, fluid dynamics, or transmission calibration; it is the software-defined intelligence layer. As advanced driving assistance systems, intelligent digital cockpits, interconnected smart-home ecosystems, and over-the-air update architectures become the primary drivers of consumer value, legacy Western automakers are facing an existential strategic dilemma.

A prevailing industry narrative suggests that Western automakers are rapidly outsourcing this critical intelligence layer to Chinese technology firms and manufacturing partners. However, categorizing this phenomenon simply as "China replacing the West" is an analytical oversimplification. The sharper, more critical reality is that Western automakers are not merely procuring components; they are increasingly purchasing the future operating model of the car industry itself. When an original equipment manufacturer outsources the highly complex layers that define the daily user experience, it may retain the intellectual property of the hood badge, but it fundamentally surrenders control over the systemic evolution of the vehicle.

Through the rigorous application of value chain analysis and the concept of "Stack Capture," this report exhaustively dissects current industry developments. It examines the deep integration of Chinese intelligence stacks into Western vehicles, the strategic capitulation of certain legacy brands, the defensive fortification efforts of others, the geopolitical fracturing of automotive software, and the long-term implications for the global industrial power balance.

The Commoditization of Hardware and the Mechanisms of Stack Capture

To accurately interpret the strategic maneuvers of global automakers, it is essential to apply a dual theoretical framework: Value Chain Commoditization and Stack Capture. These models elucidate why the outsourcing of certain vehicle components is financially prudent, while the outsourcing of the intelligence layer constitutes a fundamental surrender of market leverage and future viability.

Value Chain Commoditization versus Differentiation

For the vast majority of the twentieth and early twenty-first centuries, automotive original equipment manufacturers derived their high profit margins, competitive moats, and brand differentiation from mechanical excellence. The internal combustion engine, transmission tuning, chassis dynamics, supply-chain discipline, and sprawling dealership networks formed an unassailable barrier to entry. In the modern electric vehicle era, relentless competitive pressure has forced these traditional mechanical layers to become commoditized. Electric motors and lithium-ion battery cells, while undeniably complex to manufacture at scale, are increasingly standardized components that offer rapidly diminishing returns in terms of unique brand differentiation.

Simultaneously, the software and electronic architecture layers have emerged as the paramount core differentiators. These are the layers that dictate the user experience, command the highest margins, and ultimately drive consumer purchasing decisions. A rigorous product strategy dictates that an automaker should eagerly outsource the commodity layers to reduce costs, buy speed to market, and mitigate supply chain risks. Conversely, the automaker must fiercely protect, build, and own the differentiating layers. If an automaker outsources the core differentiator, it ceases to be a product company and is effectively reduced to a hardware distributor.

The financial reality of this value chain inversion is already visible in global market data. Analyses of automotive profitability through late 2025 and early 2026 indicate that original equipment manufacturer margins have continued to decline precipitously, reaching an average of 3.9%, representing a drop of nearly 60% from their peak in 2021.¹ In stark contrast, supplier margins have rebounded and stabilized, averaging 6.4%, driven predominantly by strong financial results in the semiconductor, battery, and advanced technology-component segments.¹ Automakers are bearing the massive, crushing capital costs of running parallel vehicle architectures—maintaining legacy combustion platforms while funding new electric platforms—while technology suppliers extract higher margins from the software and intelligence layers without bearing the heavy industrial capital expenditure of final vehicle assembly.¹

The Stack Capture Framework

The concept of Stack Capture posits that a corporate entity does not need to own or assemble the final physical product to capture the most valuable layer of an industry. Power naturally accumulates with the entity that controls the foundational layer upon which all other companies must build. In the realm of artificial intelligence and software-defined vehicles, the most valuable layer is the one that improves autonomously with continuous usage through massive data feedback loops.

When a vehicle's intelligence layer—comprising the advanced driving assistance systems, the digital cockpit, and the core electronic-electrical architecture—is outsourced, the supplier captures the learning loop. The supplier gains real-world edge-case data, vital system validation, operational credibility, and massive scale. This allows the supplier's algorithms to compound in efficacy at a rate far exceeding the buyer's internal capabilities. Over time, the supplier defines the industry standard for "good enough," while the automaker becomes a hollow brand wrapper, capturing only the lowest-margin elements of the physical assembly process.

This dynamic mirrors the artificial intelligence software industry, where wrapper companies own the user interface, but the foundational model providers own the intelligence, the infrastructure providers own the deployment, and the data platforms own the context. In the automotive sector, the critical question is no longer "Who sells the car?" but rather, "Who controls the layer that gets smarter every time the customer uses the product?"

Shift in Automotive Value Chain Dynamics

Strategic Element	Traditional Automotive Era	Software-Defined Vehicle (SDV) Era
Core Differentiator	Internal Combustion Engine, Transmission, Chassis	ADAS, Operating System, User Interface, OTA Updates
Commoditized Layer	Radios, Buttons, Basic Electronics	Battery Cells, Electric Motors, Basic Assembly
Value Capture Point	Dealership Sales, Mechanical Aftersales	Software Subscriptions, Ecosystem Integration, Data
Pace of Innovation	5 to 7-year traditional model cycles	Continuous, agile software delivery via OTA
Supplier Relationship	OEM dictates specifications to Tier 1 suppliers	Tech suppliers dictate architecture to OEMs

Empirical Validation: The Integration of the Chinese Stack

Recent product launches and strategic alliances in the Asian market provide concrete, empirical evidence validating the hypothesis that legacy automakers are increasingly integrating Chinese intelligence stacks to remain relevant in high-velocity, hyper-competitive markets.

The Ultimate Hardware Wrapper: Toyota's bZ7

The most stark and revealing illustration of this new paradigm is the Toyota bZ7. Unveiled as a flagship pure-electric sedan developed primarily by the GAC-Toyota joint venture's local Chinese engineering team, the bZ7 was launched to the market with an aggressive, limited-time subsidized starting price of approximately $21,380 (147,800 yuan).² The vehicle generated significant immediate traction, securing over 10,000 pre-sale orders and capturing an additional 3,000 firm orders within its first hour of official availability.²

While the vehicle prominently bears the Toyota badge and benefits from the Japanese automaker's historical reputation for safety and manufacturing quality, its foundational technologies are almost entirely sourced from an alliance of Chinese technology leaders. The bZ7 is, fundamentally, a meticulously assembled composite of external intelligence and powertrain layers. The vehicle utilizes Huawei's DriveONE electric drive system, an advanced integrated platform combining the motor, inverter, and transmission into a single unit delivering 207 kW of power and a top speed of 180 km/h.³ Energy storage is handled by lithium-iron-phosphate battery packs supplied by the Chinese firm CALB, offering 71 kWh and 88 kWh capacities that yield impressive ranges of 600 kilometers and 710 kilometers on the CLTC cycle, respectively.³

More critically, the intelligence layer is entirely outsourced. The intelligent cockpit operates natively on Huawei's HarmonyOS 5.0 (HarmonySpace 5).³ The vehicle is deeply integrated into Xiaomi's "Human x Car x Home" smart ecosystem, allowing users to seamlessly control domestic smart appliances directly from the vehicle's expansive 15.6-inch central touchscreen.³ Finally, the autonomous driving layer is provided by Chinese specialist Momenta, whose R6 advanced driving assistance system, augmented by a roof-mounted LiDAR unit, supports Navigation on Autopilot functionality for both highly congested urban environments and high-speed highways.³

The Toyota bZ7 serves as the quintessential case study of Stack Capture. The daily, tactile consumer experience is defined entirely by Huawei's interface logic, Xiaomi's connectivity ecosystem, and Momenta's algorithmic driving decisions. Toyota acts primarily as the hardware integrator, the distributor, and the brand guarantor. While this strategy successfully and rapidly delivered a highly competitive, affordable electric vehicle to the brutal Chinese market, it raises profound, existential questions about Toyota's long-term capability retention in the software domain. If every interaction the driver has with the vehicle is mediated by a Chinese technology firm, Toyota risks losing its direct relationship with the user, becoming entirely dependent on external suppliers for future feature velocity.

Mercedes-Benz and the Geely Architecture Conundrum

The immense pressure to drastically reduce engineering costs and aggressively shorten development cycles is driving legacy luxury brands toward similar architectural compromises. Extensive industry reports from highly informed Asian supply-chain portals indicate that Mercedes-Benz is developing a new global compact electric vehicle platform, internally codenamed "Phoenix," heavily utilizing Geely's electronic-electrical architecture.⁶

These reports claim that the Mercedes-Benz China Research and Development Center, which employs approximately 2,000 personnel, has been elevated to serve as the global headquarters for all compact vehicle development.⁶ In a historic shift for the 130-year-old German automaker, this would mark the first time that independent development rights for a new global vehicle platform have been granted to an engineering team outside of Germany.⁷ The Phoenix platform is allegedly based directly on Geely's GEEA 4.0 (Geely Electronic & Electrical Architecture), a highly integrated system that serves as the foundational backbone for intelligent cockpits, connectivity modules, and advanced driver assistance integration.⁷ Internal tear-down evaluations and feasibility studies reportedly demonstrated that Geely's underlying architecture possesses a massive advantage in cost control—a vital metric as Mercedes-Benz faces severe margin pressures and a 19% year-over-year decline in its Chinese sales volumes.⁷

Mercedes-Benz officially and vehemently denied these reports, issuing statements labeling the described cooperation as "fabricated and untrue".⁶ However, regardless of the veracity of the specific Phoenix rumors, the strategic logic behind the proposed move aligns perfectly with the broader, inescapable industry trend. Mercedes-Benz already maintains a deep, structural relationship with Geely; the Chinese automaker holds a significant financial stake in Mercedes-Benz's parent company, and the two entities already operate the Smart brand as a joint venture, explicitly utilizing Geely-developed electric vehicle platforms and engineering.¹⁰ The economic gravity pulling Western automakers toward mature, highly optimized Chinese supply chains remains incredibly potent, forcing executives to constantly weigh the preservation of in-house engineering pride against the reality of brutal cost disadvantages.

Audi's Advanced Digitized Platform and the Risk of Saturation

Volkswagen Group's premium subsidiary, Audi, has overtly embraced a localized stack strategy in an attempt to halt its slide in the Asian market. Deepening a decades-long industrial collaboration with SAIC Motor, Audi has established an "Advanced Digitized Platform" to produce a completely new line of vehicles exclusively for the Chinese market.¹²

This partnership is comprehensive, featuring the establishment of a dedicated AUDI Innovation & Technology Center in Shanghai, tasked with co-developing intelligent connected vehicles across the entire value chain.¹² To differentiate these highly localized vehicles from its global lineup, the company launched a new sub-brand, stylized simply as "AUDI" without the traditional four-ring logo. The first model under this new architecture, the all-electric E5 Sportback, entered series production in late 2025, with the E7X SUV scheduled for launch in the first half of 2026.¹³ By integrating local technological ecosystems explicitly tailored to Chinese consumer preferences, Audi aims to capture younger, highly tech-centric demographics that have abandoned legacy luxury brands.

However, early market data suggests severe headwinds. The E5 Sportback faced rapidly cooling demand shortly after its initial launch, forcing AUDI to enact significant price reductions by February 2026—a mere six months into its lifecycle—under the guise of a "limited-time promotion".¹³ This highlights a secondary, highly dangerous risk for Western automakers: simply adopting local Chinese technology and software stacks does not automatically guarantee market dominance. If the underlying brand identity is diluted, or if native Chinese competitors iterate on their software at a faster pace, the Western automaker is left with neither engineering supremacy nor pricing power, resulting in a rapid race to the bottom.

Huawei's Ascendancy: From Component Supplier to Strategic Control Point

To fully comprehend the magnitude of the intelligence layer shift, one must examine the entity currently capturing the most significant volume of automotive software dependency globally: Huawei. Huawei's executive leadership explicitly states that the company does not intend to manufacture vehicles itself; rather, its overarching objective is to become the fundamental "electronic nuts and bolts" of the intelligent connected vehicle era, empowering global automakers while capturing the highest-value data streams.¹⁴

The Unassailable Dominance of the Data Feedback Loop

Huawei's intelligent automotive solution unit is currently executing a textbook, flawless Stack Capture strategy. By establishing its proprietary hardware and software suites as the underlying operational infrastructure across a vast multitude of rival original equipment manufacturer brands, Huawei aggregates massive volumes of real-world driving telemetry.

The scale of this deployment is staggering. By early 2026, Huawei's Qiankun advanced driving system had been installed in over 1.7 million vehicles across 28 distinct models.¹⁵ The partner roster reads like a comprehensive catalog of the Asian automotive industry, including Avatr, Deepal, Voyah, M-Hero, Trumpchi, Fangchengbao, Dongfeng (under the Yijing brand), GAC (under the Qijing brand), and FAW-Audi.¹⁶ More critically, the cumulative assisted-driving mileage logged by the Qiankun system surpassed the 10.47 billion-kilometer threshold by April 2026, with expectations to reach 20 billion kilometers by the end of the year.¹⁵

This vast, exponentially growing data repository represents the ultimate, unbreachable competitive moat. In the autonomous driving industry, real-world mileage serves simultaneously as the primary training resource for algorithms and the ultimate consumer trust signal.¹⁷ Every single kilometer logged by a Huawei-equipped vehicle feeds back into Huawei's cloud infrastructure, refining the neural networks, mapping new road topologies, and improving edge-case handling logic. A superior system inherently attracts more automaker partnerships, which directly expands the physical deployment base, thereby generating even more data. As this cycle compounds, Huawei rapidly transitions from filling a technological gap to actively defining the industry standard for what constitutes "good enough" autonomy, leaving unaligned legacy automakers at a severe, potentially fatal capability deficit.

Technological Architecture: World Action vs. Vision-Language-Action

The depth and sophistication of Huawei's strategy are highly evident in its architectural divergence from current Western software norms. While many global automakers and autonomous startups are heavily pursuing Vision-Language-Action models—which operate by converting video inputs into language tokens that are subsequently trained by large language models to produce vehicle control actions—Huawei views this methodology as a fundamentally flawed shortcut that will never achieve true Level 4 autonomy.¹⁶

Instead, Huawei has aggressively developed a proprietary "World Action" framework.¹⁶ This architecture bypasses the language conversion step entirely, processing raw visual, auditory, and tactile sensor data directly into physical driving actions without the latency of language tokenization. This advanced system is underpinned by the WEWA (World Engine World Action) infrastructure:

Cloud World Engine: A massive cloud-based diffusion generation model utilized to synthesize extreme edge-case scenarios. This engine produces difficult, highly complex driving simulations at 1,000 times the density of real-world data collection, having already completed 600 million kilometers of pure highway Level 3 simulation.¹⁷
On-Vehicle World Action Model: An on-board system that utilizes a highly advanced Mixture of Experts architecture to achieve full-modal perception and scene management. This specific design reduces end-to-end processing latency by 50% compared to previous generations, vastly improving traffic efficiency and drastically lowering the rate of harsh, uncomfortable braking events.¹⁷

The safety metrics derived from this massive architectural advantage are striking. According to Huawei's April 2026 operational data, the deployment of Qiankun ADS resulted in a severe accident rate (defined as an event requiring airbag deployment or seatbelt pretension) of just one incident per 1.8 million kilometers for standard vehicles, but improved to one incident per 11 million kilometers when the latest version of the ADS was actively engaged.¹⁵ Furthermore, the system completed 300,000 automated parking maneuvers with only a single recorded collision, vastly outperforming the human average of one collision per 6,700 attempts.¹⁵

Huawei's financial commitment underscores its total dominance ambitions. The company explicitly allocated 18 billion yuan for autonomous driving research and development in 2026 alone, with a projected, massive five-year investment ranging from 70 to 80 billion yuan.¹⁵ This immense capitalization allows Huawei's automotive unit to endure incredibly long research cycles without the crushing pressure of immediate commercialization, creating a formidable barrier to entry for legacy automakers attempting to build competing software from scratch.¹⁶

Huawei's Qiankun ADS Deployment Tiers

Architecture Tier	Target Segment	Core Capabilities	Hardware Configuration
ADS SE	Entry-level (Sub-200k RMB)	Basic active safety, automatic parking, highway cruise control	Vision-centric, limited sensor suite
ADS Pro	Mid-market	Highway NOA, enhanced active safety (130 km/h AEB)	Addition of in-cabin laser vision
ADS Max	Premium	Urban NOA, point-to-point automated driving	1 to 4 advanced LiDAR units
ADS Ultra	Flagship/Luxury	Level 3 conditional autonomy, full WEWA integration	Maximum sensor redundancy, high compute

The Western Bifurcation: Outsourcing to Tech versus In-House Fortification

Faced with the terrifyingly fast, compounding lead of tech-native software providers, Western automakers are pursuing radically divergent strategies to solve their intelligence layer deficit. This response has broadly bifurcated into two distinct philosophical camps: total capitulation to technology partnerships (exemplified by Volkswagen Group) and aggressive, highly capitalized in-house fortification (exemplified by BMW).

Volkswagen's Strategic Pivot: SDV East and SDV West

Volkswagen Group represents the most prominent, globally visible example of a legacy automaker publically recognizing its inability to match the software velocity of native tech firms. Following years of severe developmental delays, massive budget overruns, and highly public delayed model launches (such as the highly anticipated Porsche Macan EV and Audi Q6 e-tron) caused directly by its internal software subsidiary, CARIAD, Volkswagen executed a sweeping, desperate strategic pivot.²⁰

CARIAD, originally intended to be the monolithic, central software developer for all Volkswagen Group vehicles globally, has been effectively demoted to a mere coordinator for legacy maintenance issues.²⁰ In its place, Volkswagen has adopted a highly complex, regionalized partnership strategy, breaking the concept of a unified global vehicle architecture:

SDV East: To remain competitive in the hyper-agile Asian market, Volkswagen established a deep partnership with Chinese electric vehicle maker XPeng.²⁰ Starting in 2026, the China Electronic Architecture, co-developed with XPeng, will serve as the central digital nervous system for all locally produced Volkswagen models based on the China Main Platform and the MEB architecture.²² This explicitly allows Volkswagen to utilize XPeng's advanced XNGP driver assistance systems, entirely bypassing the legacy internal systems that crippled their global operations.²³
SDV West: For Western markets, Volkswagen committed a staggering $5.8 billion to a joint venture with American tech-native automaker Rivian, establishing "Rivian and VW Group Technologies".²⁵ Through this highly capitalized venture, Volkswagen will abandon its piecemeal, supplier-driven electronic control unit strategy and adopt Rivian's advanced zonal electronic architecture.²⁶

Rivian's zonal architecture drastically reduces manufacturing and software complexity by eliminating 1.6 miles of physical wiring (as seen in the R1 Gen 2) and consolidating computing power into a few highly powerful, centralized zonal nodes.²⁶ By essentially purchasing Rivian's "vehicle nervous system," Volkswagen aims to deploy highly automated software-defined vehicles across its upcoming SSP platform by the end of the decade, with winter testing of reference vehicles already successfully completed in early 2026.²⁶ However, massive internal friction persists; premium brands like Porsche and Audi are reportedly pushing back heavily against these standardized platforms in favor of bespoke brand features, highlighting the deep cultural resistance legacy automakers face when attempting to adopt tech-industry standardization.²¹

BMW's Defensive Fortification: The "Heart of Joy"

In stark, diametric contrast to Volkswagen's capitulation, BMW has explicitly and aggressively rejected the outsourcing of its core software and dynamic control systems. Operating on the fundamental belief that a premium vehicle's dynamic feel is its ultimate, irreplaceable differentiator, BMW has built a massive, proprietary software-defined architecture for its upcoming "Neue Klasse" electric vehicles, debuting with the iX3.²⁸

BMW's strategy is anchored by an advanced electronic-electrical architecture that bundles all critical computing power into four centralized, highly integrated "superbrains".²⁹ This configuration features twenty times the processing capability of previous generations and utilizes a zonal wiring harness that reduces cable length by 600 meters, saving a crucial 30% in physical weight.²⁹ The four superbrains independently govern distinct domains: basic vehicle functions, infotainment via BMW Operating System X, automated driving, and most importantly, driving dynamics.³⁰

The cornerstone of this architecture is the "Heart of Joy" (officially designated as BMW Dynamic Performance Control), a super-computer developed entirely in-house.³⁰ The Heart of Joy consolidates the control of the electric drivetrain, energy recuperation, precise steering, and braking into a single, cohesive neural node.³¹ By processing inputs up to ten times faster than fragmented legacy ECU systems, the software dynamically manages torque distribution and traction with unparalleled precision, reducing latency to as low as 1 millisecond.³²

For BMW, this strategy is an existential imperative. As R&D head Joachim Post articulated, digital functions are now the absolute core of all vehicle development.²⁸ In an era entirely devoid of the acoustic and mechanical feedback of combustion engines, proprietary software is the only medium through which BMW can deliver its signature driving experience.²⁸ By ensuring that up to 200,000 daily software builds remain under strict, internal control, BMW is placing a massive bet that preserving its intelligence layer is the only possible way to avoid the commoditization trap currently ensnaring nearly all other global automakers.²⁸

Comparison of Western SDV Strategies

Automaker	Software Strategy	Primary Intelligence Partner	Core Strategic Philosophy
Volkswagen	Aggressive Outsourcing / Regionalization	Rivian (West), XPeng (East)	Speed to market; public acknowledgement of internal software failure (CARIAD demotion).
BMW	Strict In-House Full-Stack Control	Internal ("Heart of Joy" / Superbrains)	Retain brand differentiator; driving dynamics must not be commoditized by third parties.
Mercedes-Benz	Hybrid / Platform Sourcing (Compact)	Geely (Rumored GEEA 4.0 for Phoenix)	Leverage external scale for entry-level cost reduction while maintaining high-end brand prestige.

Distribution Colonization and the Re-education of Global Consumers

While the intelligence layer battle rages fiercely at the high end of the market, Chinese automakers are executing a parallel, highly effective strategy in emerging and budget-conscious segments: distribution colonization. This strategy involves inserting Chinese technological capabilities, battery systems, and software under the comforting guise of familiar, highly trusted Western badges.

General Motors and the SAIC-GM-Wuling Pipeline in Latin America

In highly contested emerging markets across Latin America, Africa, and the Middle East, General Motors has heavily leveraged its joint venture, SAIC-GM-Wuling, to rapidly electrify its product portfolio without bearing massive internal R&D costs. Two key vehicles beautifully illustrate this phenomenon: the Chevrolet Spark EUV and the Chevrolet Captiva EV.

Neither of these vehicles possesses a lineage connected to traditional American automotive engineering. The newly introduced Chevrolet Spark EUV is a direct, unapologetic rebadge of the Baojun Yep Plus, a highly affordable, boxy electric crossover originating entirely in China.³⁴ Manufactured initially in China and subsequently assembled in Brazil via semi-knock-down kits at the PACE plant in Ceará, the Spark EUV features a 75 kW front motor, a 41.9 kWh LFP battery offering roughly 401 km of CLTC range, and an advanced driver assistance suite originally developed by the Chinese drone manufacturer DJI.³⁵

Similarly, the upcoming Chevrolet Captiva EV is a rebadged Wuling Starlight S.³⁷ Beyond very minor front-fascia styling adjustments designed to seamlessly accommodate the Chevy bowtie grille, the vehicle's core architecture, 317-mile range capability, and 20-minute rapid-charging systems are entirely Chinese.³⁸

This specific maneuver represents a highly pragmatic, yet incredibly dangerous, short-term calculation by General Motors. GM obtains immediate access to highly affordable, technologically advanced EVs, allowing it to defend vital market share against the direct influx of Chinese brands like BYD in Latin America.³⁵ However, this strategy actively facilitates the systemic conditioning of the global consumer base. Drivers learn that the underlying EV technology is reliable, local regulatory bodies establish certification precedents for Chinese E/E architectures, and local dealership networks learn exactly how to service these software systems.³⁵ By utilizing the historic Chevrolet brand as a trojan horse, Chinese engineering achieves mass distribution and consumer validation globally without facing any direct brand resistance.

The Generational Shift in Brand Loyalty

This technological infiltration aligns perfectly with a profound, highly disruptive demographic shift regarding brand loyalty in the automotive sector. Recent consumer perception research conducted by Cox Automotive reveals a striking, unprecedented generational divide within the United States market.⁴¹

While older demographics stubbornly maintain their traditional loyalty to legacy domestic or established import brands, a staggering 69% of Generation Z vehicle shoppers indicated they were "more likely" to explicitly consider purchasing a Chinese automotive brand.⁴² The primary driver of this immense openness is affordability, coupled directly with high technological inclusion. With the average cost of a new vehicle in the U.S. approaching $50,000, younger consumers—who are heavily burdened by economic pressures and student debt—prioritize value, digital ecosystem experiences, and novel tech integration far above historical brand prestige.⁴²

Chinese EVs, which consistently price highly advanced digital cockpits and ADAS features at roughly half the cost of Western equivalents, precisely target this demographic demand.⁴² This data strongly suggests that if artificial trade barriers were removed, Chinese automakers would not need to engage in decades-long, expensive brand-building exercises; their perfect alignment with the software-centric priorities of young consumers would facilitate immediate, devastating market penetration.

The Splinternet of Mobility: Geopolitics and Regulatory Fencing

The rapid, undeniable technological dominance of the Chinese intelligence layer has triggered severe geopolitical blowback. Western governments, increasingly recognizing that modern software-defined vehicles are effectively highly advanced, roaming sensor networks, are hastily erecting massive legislative barriers. As a direct result, the global automotive supply chain is fracturing irrevocably into a "splinternet of mobility."

The United States: The Connected Vehicle Security Act

In direct response to both economic panic and genuine national security anxieties, the United States is advancing highly aggressive protectionist measures. Senators Bernie Moreno and Elissa Slotkin recently introduced the Connected Vehicle Security Act of 2026, a bipartisan initiative designed to permanently codify and drastically expand upon previous Department of Commerce regulations.⁴⁴

The legislation explicitly and comprehensively bans the importation, manufacture, integration, and sale of connected vehicles, software, and hardware linked to "foreign adversaries," with a primary, unwavering focus on China and Russia.⁴⁴ The justification is fundamentally rooted in intelligence risks. Modern connected vehicles are openly characterized by legislators as "surveillance packages on wheels," equipped with high-resolution cameras, LiDAR, highly sensitive microphones, and telematics systems capable of continuously mapping critical infrastructure, tracking civilian movements, and potentially serving as vectors for mass remote cyberattacks or coordinated vehicle immobilization during a geopolitical crisis.⁴⁴

To minimize immediate, catastrophic supply chain collapse, the regulations feature a strictly phased implementation timeline. The integration of prohibited "covered software"—defined incredibly broadly to include any application, middleware, and operating systems developed by entities organized under the laws of a covered country—will be completely banned starting with Model Year 2027.⁴⁴ Prohibitions targeting specific physical hardware modules (such as Vehicle Connectivity Systems) will take absolute effect for Model Year 2030.⁴⁴ Furthermore, these rules apply regardless of where the vehicle is physically assembled, meaning that vehicles built in Mexico by Chinese subsidiaries (or utilizing Chinese software subcomponents) will be permanently denied entry to the U.S. market.⁴⁹

Chinese Data Export Restrictions

Conversely, the Chinese government is ruthlessly enforcing its own stringent data sovereignty frameworks, ensuring that intelligence gathered within its borders never benefits foreign entities. The Ministry of Industry and Information Technology, operating in tandem with the Cyberspace Administration of China, recently published heavily updated "Guidelines for the Security of Cross-Border Transfer of Vehicle Data".⁵⁰

Under these incredibly strict regulations, automotive data controllers are subject to intense, continuous scrutiny regarding the export of "important data." This category includes geographic mapping near sensitive areas, high-volume traffic flow metrics, operational data from charging networks, and personal information involving more than 100,000 individuals.⁵² All such data must be stored locally within physically secure Chinese servers, and any potential export requires a rigorous, multi-step security assessment by federal regulatory authorities.⁵²

While the 2026 guidelines provide certain minor exemptions to ease basic operational burdens (such as allowing the export of standard security vulnerability data without specific approvals), the overarching structure mandates that automakers maintain completely isolated, localized data silos.⁵¹

The Eradication of the Global Car

The compounding, devastating effect of these dual regulatory walls is the absolute eradication of the "global car" paradigm. Automakers can no longer develop a single, unified E/E architecture and software stack for deployment across all global markets. As clearly evidenced by Volkswagen's panicked SDV East/West strategy, multinational original equipment manufacturers are being forced to duplicate immense R&D expenditures, maintaining distinct, totally isolated intelligence layers for the Western Hemisphere and the Asian sphere. This forced regionalization permanently destroys the massive economies of scale that historically defined automotive profitability, further compressing legacy margins and granting an insurmountable structural advantage to domestic Chinese champions operating within a single, massive, unified market.

Strategic Outlook and Market Restructuring

The exhaustive analysis of current technological deployments, strategic alliances, E/E architectures, and regulatory shifts yields several profound conclusions regarding the immediate future of the global automotive industry.

First, the intelligence layer has officially, permanently decoupled from the hardware layer. The automotive product is now hopelessly bifurcated. The physical hardware—the chassis, the battery cells, the electric motors—operates merely as a low-margin commodity platform designed strictly to host the high-margin, software-defined intelligence layer. The traditional moat of mechanical refinement has evaporated entirely.

Second, "Stack Capture" is undeniably the dominant industrial strategy of the 2020s. Technology entities, led most visibly and aggressively by Huawei, are capturing the global automotive industry not by building low-margin physical cars, but by monopolizing the data feedback loop. By supplying the operating systems, the AI agent frameworks, and the complex ADAS logic to a broad coalition of OEMs, these tech firms ensure that their algorithms improve exponentially with every single kilometer driven. The OEMs that outsource these layers achieve short-term cost efficiency but sacrifice their long-term capability, devolving into high-capital distributors of external software.

Finally, the market is rapidly structuring into three distinct, highly polarized archetypes. "Ecosystem Providers" (such as Huawei and Xiaomi) possess the capital and software velocity to build hyper-advanced intelligence layers and monetize data. "In-House Purists" (such as BMW) will endure massive capital burn to retain absolute control over their software stacks, surviving on the premise that the integration of hardware and proprietary software is the only true brand differentiator left. The final archetype, the "Wrappers," encompasses legacy OEMs that fail to bridge the software divide and are forced to partner with Ecosystem Providers to survive. These companies will compete purely on price, aesthetic design, and the lingering momentum of localized dealer networks.

The next decade of the automotive industry will not be defined by which company can engineer the most efficient electric motor or the most aerodynamic chassis. It will be defined entirely by a brutal, existential battle over who owns the E/E architecture that dictates the vehicle's intelligence. For legacy automakers, the strategic mandate is stark and unforgiving: they must either rigorously define and aggressively defend the boundaries of the software they build internally, or accept their permanent, irreversible relegation to the commodity layer of a technology-dominated ecosystem.

Thanks!

15 years navigating corporate complexity. An entrepreneur's instinct for what matters. The AI fluency to build it. Let's work together →

Works cited

Automotive Profitability: How OEM and Supplier Margins Are Faring | Bain & Company, accessed May 6, 2026, https://www.bain.com/insights/automotive-profitability-how-oem-and-supplier-margins-are-faring-interactive/
GAC Toyota launches bZ7 electric sedan with Huawei tech to boost China sales - CnEVPost, accessed May 6, 2026, https://cnevpost.com/2026/03/29/gac-toyota-launches-bz7-electric-sedan/
Toyota bZ7 electric sedan receives strong China launch - New Atlas, accessed May 6, 2026, https://newatlas.com/automotive/toyota-bz7/
Toyota launches China-built bZ7 electric saloon, accessed May 6, 2026, https://www.electrive.com/2026/04/08/toyota-launches-china-built-bz7-electric-saloon/
Mercedes-Benz to develop new EV platform in China using Geely architecture, report says, accessed May 6, 2026, https://carnewschina.com/2026/03/09/mercedes-benz-to-develop-new-ev-platform-in-china-using-geely-architecture-report-says/
Is Mercedes-Benz planning a new EV platform based on Geely technology? - electrive.com, accessed May 6, 2026, https://www.electrive.com/2026/03/06/is-mercedes-benz-planning-a-new-ev-platform-based-on-geely-technology/
Mercedes reportedly taps Geely architecture for new global EV platform - CnEVPost, accessed May 6, 2026, https://cnevpost.com/2026/03/05/mercedes-taps-geely-architecture-for-new-global-ev-platform/
Why Are Mercedes And Geely Partnering Up? - PakWheels Blog, accessed May 6, 2026, https://www.pakwheels.com/blog/mercedes-benz-geely-partnership-compact-ev-platform-phoenix/
Mercedes and Geely Join Forces for Global EV Platform in China - YouTube, accessed May 6, 2026, https://www.youtube.com/watch?v=_esAc8o7u4I
If you can't beat them... Mercedes-Benz deepening ties with China's Geely: report, accessed May 6, 2026, https://evcentral.com.au/if-you-cant-beat-them-mercedes-benz-deepening-ties-with-chinas-geely-report/
Mercedes-Benz Turns to China for Future EV Technology - Autoblog, accessed May 6, 2026, https://www.autoblog.com/news/mercedes-benz-turns-to-china-for-future-ev-technology
Vorsprung in China: Audi and SAIC deepen their partnership for future generation of AUDI models | Audi MediaCenter, accessed May 6, 2026, https://www.audi-mediacenter.com/en/press-releases/vorsprung-in-china-audi-and-saic-deepen-their-partnership-for-future-generation-of-audi-models-17089
Audi and SAIC plan four new AUDI models - electrive.com, accessed May 6, 2026, https://www.electrive.com/2026/04/20/audi-and-saic-plan-four-new-audi-models/
Huawei ADS Surpasses 10 Billion Kilometers Driven in China-Internet Info Agency - Car Reviews, Automotive News and Analysis, accessed May 6, 2026, https://english.news18a.com/news/english_248918.html
Huawei Unveils ADS Safety and Parking Performance Data; Plans L3/L4 Autonomous Driving Pilots from 2026, accessed May 6, 2026, https://english.news18a.com/news/english_251292.html
“Nothing is free”: Huawei's Richard Jin on assisted driving and the path to Level 4 autonomy, accessed May 6, 2026, https://kr-asia.com/nothing-is-free-huaweis-richard-jin-on-assisted-driving-and-the-path-to-level-4-autonomy
Huawei's Qiankun ADS autopilot system closes in on 10 billion km in China, accessed May 6, 2026, https://www.globalchinaev.com/post/huaweis-qiankun-ads-autopilot-system-closes-in-on-10-billion-km-in-china
Huawei autopilot system logs over 10 billion km as user adoption hits 94.8%, accessed May 6, 2026, https://www.globalchinaev.com/post/huawei-autopilot-system-logs-over-10-billion-km-as-user-adoption-hits-948
Huawei Invests 80 Billion Yuan to Expand Computing Power for Autonomous Driving, accessed May 6, 2026, https://www.indexbox.io/blog/huawei-invests-80-billion-yuan-to-expand-computing-power-for-autonomous-driving/
Volkswagen overtakes Amazon as Rivian's largest shareholder with 15.9% stake after $1B software milestone payment - TNW, accessed May 6, 2026, https://thenextweb.com/news/volkswagen-rivian-top-shareholder-amazon-software
Report: VW forms task force to deal with Rivian software JV partnership disagreements & delays, accessed May 6, 2026, https://www.rivianforums.com/report-vw-forms-task-force-to-deal-with-rivian-software-partnership-disagreements-delays/
One system for all: China Electronic Architecture becomes the central digital system for the Volkswagen brand's e-portfolio in China, accessed May 6, 2026, https://volkswagengroupchina.com.cn/en/news/Detail?ArticleID=74C1976E1920434D9399A951525F3F27
VW reassigns Cariad as coordinator of Rivian and Xpeng software - electrive.com, accessed May 6, 2026, https://www.electrive.com/2025/10/06/vw-reassigns-cariad-as-coordinator-of-rivian-and-xpeng-software/
Volkswagen Forms Task Force as Rivian Software Alliance Falls Short: Report - Reddit, accessed May 6, 2026, https://www.reddit.com/r/RIVNstock/comments/1qk7kss/volkswagen_forms_task_force_as_rivian_software/
Rivian and Volkswagen Group Announce the Launch of their Joint Venture, accessed May 6, 2026, https://rivian.com/newsroom/article/rivian-and-volkswagen-group-announce-the-launch-of-their-joint-venture
Rivian's Secret Weapon: Zonal Architecture (The Tech VW Is Paying Billions For), accessed May 6, 2026, https://www.rivianforums.com/rivians-secret-weapon-zonal-architecture-the-tech-vw-is-paying-billions-for/
One year after its founding: Joint venture between Volkswagen Group and Rivian shows strong progress, accessed May 6, 2026, https://www.volkswagen-group.com/en/press-releases/one-year-after-its-founding-joint-venture-between-volkswagen-group-and-rivian-shows-strong-progress-19980
BMW Highlights Growing Role of Software in Next-Gen Models, accessed May 6, 2026, https://www.bmwblog.com/2026/04/09/bmw-software-role-next-gen-models-neue-klasse/
Four “Superbrains“ for the Neue Klasse by BMW - BMW Group, accessed May 6, 2026, https://www.bmwgroup.com/en/news/general/2025/superbrains.html
Inside BMW's Next-Gen Car Tech: The Four “Superbrains” Explained - BMW Blog, accessed May 6, 2026, https://www.bmwblog.com/2025/03/10/bmw-neue-klasse-digital-architecture-superbrains/
Details about Heart of Joy - BMW Group, accessed May 6, 2026, https://www.bmwgroup.com/en/company/neue-klasse/details-zu-heart-of-joy.html
BMW VDX Vision Driving Experience: Neue Klasse drivetrain & driving dynamics test rig - Page 5 - BMW 3-Series and 4-Series Forum (F30 / F32) | F30POST - Bimmerpost, accessed May 6, 2026, https://f30.bimmerpost.com/forums/showthread.php?p=31873308
What is the BMW Heart of Joy? - BMW Blog, accessed May 6, 2026, https://www.bmwblog.com/2023/09/05/what-is-the-bmw-heart-of-joy/
Chevy Unveils New Spark EV For South America, Courtesy Of China - Reddit, accessed May 6, 2026, https://www.reddit.com/r/electricvehicles/comments/1iq5g00/chevy_unveils_new_spark_ev_for_south_america/
China's Baojun Yep Plus, South America's Chevrolet Spark EUV 2025 - YouTube, accessed May 6, 2026, https://www.youtube.com/watch?v=2--uEpgRxPQ
Baojun Yep - Wikipedia, accessed May 6, 2026, https://en.wikipedia.org/wiki/Baojun_Yep
Wuling Starlight S - Wikipedia, accessed May 6, 2026, https://en.wikipedia.org/wiki/Wuling_Starlight_S
Chevrolet gives sneak peek at next-gen EV based on popular Chinese model: 'Seems to be the trend' - The Cool Down, accessed May 6, 2026, https://www.thecooldown.com/green-business/captiva-ev-chevrolet-evs-expansion-china/
Chevrolet's Latest Electric SUV Has A Secret Chinese Twin | Carscoops, accessed May 6, 2026, https://www.carscoops.com/2025/07/gms-captiva-ev-is-based-on-a-chinese-suv/
The Baojun Yep Plus — I Mean, the Chevrolet Spark EUV — Lands in Mexico. Will This Turn the Tide for GM? - Reddit, accessed May 6, 2026, https://www.reddit.com/r/electricvehicles/comments/1ler7gx/the_baojun_yep_plus_i_mean_the_chevrolet_spark/
Chinese Auto Brands: What U.S. Consumers and Dealers Think - Cox Automotive Inc., accessed May 6, 2026, https://www.coxautoinc.com/insights/chinese-auto-brands-what-u-s-consumers-and-dealers-think/
Gen Z Buyers Open to Chinese EV Brands in the US - Electric Cars Report, accessed May 6, 2026, https://electriccarsreport.com/2026/03/gen-z-car-buyers-more-open-to-chinese-ev-brands-cox-study-finds/
Gen Z Consumers Are Warming Up To The Idea Of Chinese EVs - CleanTechnica, accessed May 6, 2026, https://cleantechnica.com/2026/03/06/gen-z-consumers-are-warming-up-to-the-idea-of-chinese-evs/
Moreno, Slotkin Bill to Ban Chinese Vehicles, Connected Components From U.S. Market, accessed May 6, 2026, https://www.moreno.senate.gov/press-releases/moreno-slotkin-bill-to-ban-chinese-vehicles-connected-components-from-u-s-market/
Senators Move to Ban Chinese Vehicles and Parts From US Market, accessed May 6, 2026, https://www.thomasnet.com/insights/senators-ban-chinese-cars-parts-entering-us/
Commerce Finalizes Rule to Secure Connected Vehicle Supply Chains from Foreign Adversary Threats, accessed May 6, 2026, https://www.bis.gov/press-release/commerce-finalizes-rule-secure-connected-vehicle-supply-chains-foreign-adversary-threats
Paulina Uznańska SMARTPHONES ON WHEELS CONNECTED CARS FROM CHINA AND THE EU'S CYBERSECURITY - OSW, accessed May 6, 2026, https://www.osw.waw.pl/sites/default/files/OSW%20REPORT%20-%20Smartphones%20on%20wheels.pdf
U.S. Department of Commerce Finalizes Connected Vehicles Supply Chain Restrictions, accessed May 6, 2026, https://datamatters.sidley.com/2025/01/22/u-s-department-of-commerce-finalizes-connected-vehicles-supply-chain-restrictions/
How Connected Vehicle Regulations May Impact US Automotive Imports - Baker Institute, accessed May 6, 2026, https://www.bakerinstitute.org/research/how-connected-vehicle-regulations-may-impact-us-automotive-imports
Vehicle Data Export in China: New Guidelines Clarify Rules for Automakers, accessed May 6, 2026, https://www.china-briefing.com/news/vehicle-data-export-rules-china/
China: New guidance on data transfer and identification of important data in the automotive sector | Privacy Matters, accessed May 6, 2026, https://privacymatters.dlapiper.com/2026/02/china-new-guidance-on-data-transfer-and-identification-of-important-data-in-the-automotive-sector/
Europe's vehicle-data gap: How under-regulation gives Chinese carmakers an edge, accessed May 6, 2026, https://merics.org/en/comment/europes-vehicle-data-gap-how-under-regulation-gives-chinese-carmakers-edgef

Thanks

beehiiv — The newsletter platform built for growth

Access the best tools available in email, helping your newsletter scale and monetize like never before.

beehiiv.com

Est rerum sunt eum nisi dolores est velit voluptatem non exercitationem consequatur id Quis ipsa. Sed autem fugit eos consequuntur maxime est corrupti laboriosam ea totam amet.

Lorem Ipsum!

When authorship stops being enough

Wilson CELY — Tue, 06 Jan 2026 03:53:05 +0000

What Karpathy's struggle tells us about leadership

Open thought about Karpathy's struggle

Andrej Karpathy admitted recently he feels "behind."

Twitter tweet

One of the world's best programmers (the guy who built Tesla's Autopilot, led AI at OpenAI) can't keep up.

I've been sitting with that for a while now.

Because I think what he's feeling isn't a personal problem. It might be a signal that something fundamental has shifted.

For years, we built our reputations on being the "authors" of our results. We prized the person who knew every line of code or every detail of a contract because they had direct control over the logic. It was deterministic.

But we've introduced this "alien component" into the stack now.

AI isn't a logic machine; it's a probabilistic one. It's brilliant but fallible, which means the old link between effort and leverage is effectively broken.

I'm starting to think the most important skill we need to develop (and frankly, hire for) isn't "technical" in the old sense at all.

It's the ability to preserve authority while delegating generation.

It's like moving from being a craftsman to a systems architect.

In the "Factorio" mindset, nobody cares if you personally hand-crafted a gear. What matters is that you've built a system where gears are produced reliably at scale, even when the individual parts are uncertain.

The ones who figure out how to orchestrate outcomes (rather than just do the work themselves) are the ones who will actually find that 10x leverage we keep hearing about.

It's a strange moment to be a leader.

Did you that NotebookLM make infographics?

LINK HERE=> https://notebooklm.google.com/

Stop treating AI as a "chatbot" and start treating it as a Production Studio.

Google’s NotebookLM just moved past simple summaries. It’s now generating high-fidelity slide decks and infographics directly from your raw data.

Most leaders use AI to "ask questions." Smart leaders are using it to "build assets."

The problem? Most AI-generated presentations look like generic templates from 2012. But NotebookLM is different.

Because it’s grounded in your specific sources—research papers, 20-page strategy docs, or raw data tables—it doesn't hallucinate "fluff." It builds logic.

I recently tested this with a dense research paper. Usually, reading the abstract and conclusion takes 15 minutes, but the "nuance" is buried in the middle. But instead of reading, I prompted it to "Create an infographic summarizing the core technical friction points."

The result? A visual map of the paper’s logic that I could digest in 60 seconds.

Therefore, the workflow for a C-suite executive changes:

Upload the "impossible to read" 50-page industry report.
Generate a Slide Deck (Presenter Mode) for your next board update.
Generate an Infographic for the internal Slack channel to align the team.
Listen to the Audio Overview on your drive home to prep for Q&A.

This isn't just "saving time." It’s increasing the velocity of internal clarity.

If you want to try this today, don't just "upload and click." Use a "Specific Lens" prompt.

Instead of: "Make an infographic." Try: "Create an infographic for a non-technical board of directors focusing on the ROI metrics of this implementation."

The tool is free (for now) via Google Labs. Move your research from "unread tabs" to "ready-to-share assets."

beehiiv — The newsletter platform built for growth

Access the best tools available in email, helping your newsletter scale and monetize like never before.

beehiiv.com

31 | AI: Can I Recognize You from a Picture? (1)

Wilson CELY — Fri, 11 Oct 2024 04:53:00 +0000

Our Menu :)

I-XRAY: The AI Glasses That Reveal Anyone’s Personal Details
Meta Ups the Ante with Movie Gen: Revolutionizing Text-to-Video Generation
Glean who?

Jobs and Training

Job Listings this week
Andrew Ng’s AI Python for Beginners.

Read time: 5 Minutes

WEEKLY DIGEST
I-XRAY: The AI Glasses That Reveal Anyone’s Personal Details

A few weeks ago, Meta unveiled an intriguing feature for its new Ray-Ban Meta glasses: the ability to gather information about your surroundings. This includes innocuous tasks like identifying flowers. However, two Harvard students recently demonstrated how easily these smart glasses can turn into a privacy nightmare.

Twitter tweet

Anhphu Nguyen and Caine Ardayfio developed an app called I-XRAY that transforms the Ray-Bans into a powerful doxxing machine. This app can reveal personal details such as names, addresses, and phone numbers just by looking at someone through the glasses.

How Does It Work?

The Ray-Bans can record up to three minutes of video, with a privacy light that's barely noticeable—akin to a firefly in broad daylight. This video is streamed to Instagram, where an AI monitors the feed. I-XRAY then uses PimEyes, a facial recognition tool, to match these faces to public images and employs AI to dig up personal details from public databases.

The duo's demo had strangers alarmed when they realized how easily identifiable they were from public online information. They tested the app on unsuspecting subway riders, convincingly pretending they knew them using details from I-XRAY. (More like I-SCARY.)

You can see the app in action here: Watch the Demo.

Fortunately, Nguyen and Ardayfio are not releasing the app. It's just a demonstration of how easy this is to do. But if two students can create this, you can bet someone else will, too.

Big Tech's Role

While big tech companies like Meta might not be the ones developing such invasive apps, they are certainly involved in other ways. Meta has been evasive about whether they're using your glasses' pics to train their AI. When asked, they simply said, “We're not publicly discussing that.”

Protecting Your Privacy

If you're concerned about your privacy, the Harvard duo shared tips on how to scrub yourself from facial recognition databases. You can find these tips in this Google Doc.

Main Services to Opt Out Of

Here are the main services they suggest you opt out of (with links to directly opt out):

PimEyes - Opt Out
FaceCheck.ID - Opt Out
FastPeopleSearch - Opt Out
Instant Checkmate - Opt Out

Stay informed and take steps to protect your privacy in this increasingly digital world.

Learn more

WEEKLY DIGEST
Meta Ups the Ante with Movie Gen: Revolutionizing Text-to-Video Generation

Meta has introduced Movie Gen, a groundbreaking series of four systems that generate videos with consistent characters, alter imagery, and add matching sound effects and music. This innovative technology will be available on Instagram in 2025, but you can already view and listen to examples here. The team has detailed their work in an extensive 92-page paper.

Key Features

Generated Videos: Movie Gen Video outputs 256 frames (up to 16 seconds) at 1920x1080 resolution. It uses flow matching to produce detailed imagery from text prompts.
Consistent Characters: Given an image of a face, Movie Gen Video generates a video depicting a person with that face, with varied poses and expressions.
Altered Clips: The system can alter videos by changing backgrounds, adding objects, or restoring altered clips to their original content.
Synthetic Soundtracks: Movie Gen Audio generates sound effects and music for video clips up to 30 seconds long, starting from pure noise and extending audio as needed.

Performance Highlights

Overall Video Quality: Movie Gen outperformed competitors like Runway Gen3, Sora, and Kling 1.5 in qualitative evaluations.
Character Generation: Achieved a net win rate of 64.74 percent versus ID-Animator.
Soundtrack Generation: Outperformed various video-to-audio models with a net win rate between 32 percent and 85 percent.
Video Alteration: Beat all competitors more than 70 percent of the time in the TGVE+ dataset.

Movie Gen

WEEKLY DIGEST
Glean who?

Rumors in valley suggest that Sam Altman, the CEO of OpenAI, has identified five AI startups that OpenAI's recent investors should avoid backing:

🛑 Anthropic
🛑 xAI (Elon Musk)
🛑 Safe Superintelligence (Ilya Sutskever)
🛑 Perplexity
🛑 Glean

While the first four are well-known, Glean AI stands out as a lesser-known but formidable competitor. Let's dive into what makes Glean a potential future winner in the enterprise AI space.

Glean AI: The Work AI Platform

Glean AI, founded by a team of former Google search engineers and industry veterans, is a Work AI platform connected to enterprise data. It allows users to find, create, and automate anything, making it a powerful tool for enhancing workplace productivity.

Revenue Growth: Glean's annualized revenue grew to $39 million in January 2024, marking a 4x growth from the $10 million recorded in 2023.
Funding: In September 2024, Glean raised over $260M in a Series E, valuing the company at $4.6 billion.
Founders: Arvind Jain (CEO), T.R. Vishwanath (CTO, formerly at Meta and Microsoft), Tony Gentilcore (Product Engineering, formerly at Google), and Piyush Prahladka (now CEO of his own stealth startup).

Glean's Suite of Products

Work AI Platform: A unified search tool that aggregates information from various applications, allowing users to access critical data efficiently.
AI-Powered Workspace Search: Enhances workplace productivity by providing enterprise-grade AI and search capabilities.
Customer Testimonials and Stories: Real-world examples of how Glean is transforming workplaces.

The Future of Enterprise AI Search

Glean specializes in enterprise-grade AI and search capabilities that have the potential to scale well. The startup focuses on enhancing workplace productivity by providing a unified search tool that aggregates information from various applications.

Arvind Jain, co-founder and CEO of Glean, has a vision to expand human potential to do extraordinary work. The trajectory of Glean AI follows an expected path, and I expect it to become one of the first winners in the pure-play RAG (Retrieval Augmented Generation) category.

Conclusion

As the enterprise AI search space becomes more consolidated, Glean AI is poised to emerge as a leader. Its innovative approach to enhancing workplace productivity and its impressive revenue growth trajectory make it a startup to watch.

More Info

GEN AI STARTUP

Waymo is seeking a Backend Software Engineer for their Simulation Infrastructure team.

Mistral AI is seeking a Machine Learning Engineer to join their team.

Databricks is hiring a Sr. Corporate Communications Manager to join their team.

xAI has an opening for an Office Manager to oversee daily operations.

JOBS & TRAININGS
Andrew Ng’s AI Python for Beginners.

Working With Your Own Data and Documents in Python

Learn how to handle both structured and unstructured data using Python, from reading text files to processing data in CSV format for tasks like vacation planning.
Use Python and AI to process large volumes of text, such as extracting relevant details from food critics' journals or other documents, and generate useful insights.
Turn repetitive code blocks into reusable Python functions to make your code more modular and efficient.
Combine data processing techniques with AI to generate detailed travel itineraries, integrating multiple data sources.

GOT TO COURSE

Thank you, see you next week!

30 | NoteBookLM: Create an Instant Podcast of your PDFs

Wilson CELY — Fri, 20 Sep 2024 07:15:00 +0000

Our Menu :)

Google's NotebookLM: Instant Podcast for YOU
AI Weekly News Digest

Actionable Tips:

ChatGPT Prompts for Business
Create a Synthetic Dataset Using Llama 3.1 to Fine-Tune Your LLM

Jobs and Training

Anthropic's LLM Bootcamp: Improve Your Prompts
Job Listings this week

Read time: 4 Minutes

GEN AI AT WORK
Google's NotebookLM: Instant Podcast for YOU 🎧

Buckle up, because Google just dropped a BOMB, that's about to turn your boring old documents into a binge-worthy podcast series. Say hello to NotebookLM, the tool that's putting the 'AI' in 'auditory learning'!

What Does NotebookLM Do?

It uses two AI voices to read your documents, making it feel more conversational. The goal is to make information retention as easy as listening to a podcast.

You read that right! It transforms any document you give it into an INSTANT PODCAST!

How Does It Work?

AI Model: NotebookLM uses Google's Gemini 1.5, a multimodal AI model, to process the text.
Text-to-Speech: The AI-powered text-to-speech technology ensures the voices sound natural.
Summarization: NotebookLM doesn't just read the text; it understands, summarizes, and highlights important information.
Citations: It includes citations and quotes to back up its content.

Additional Features

Study Guide: Generates instant summaries for quick review.
Briefing: Creates quick overviews of documents.
Privacy: Your personal data is not used for AI training.

Technical Details

Context window capacity: ~50 PDFs
Language support: Currently English-only
Input formats: PDFs, Google Docs, Slides, web URLs
Output: Downloadable audio file
NotebookLM's architecture manages large input volumes and cross-references information between multiple documents.
However, audio generation for extensive notebooks can take several minutes, indicating computational intensity.

How to

1/ Go to https://notebooklm.google/

2/ On the left panel, click on

3/ Drag and drop your document, in my case I have added a scientific paper about <What is the Role of Small Models in the LLM Era: A Survey>, scheduled reading

📝 Paper HERE

4/ Result .. I am dreaming!!!

What is the Role of Small Models in the LLM Era: A Survey

5/ Chat with your document, the interface is super awesome.

Try NoteBookLM

ACTIONABLE TIPS
ChatGPT Prompts for Business

1/ PEST Analysis Prompt:

Perform a comprehensive PEST analysis for a [industry] company operating in [location]. Examine the political, economic, social, and technological factors that could impact the business. Provide detailed insights and strategic recommendations based on your analysis."

2/ The Five Whys Prompt:

ChatGPT, use the Five Whys technique to identify the root cause of [Specific Problem] in my business."

3/ Customer Journey Mapping Prompt:

Assist me in mapping the customer journey for a [type of customer] purchasing [product/service]. Identify each touchpoint, the customer's goals and emotions at each stage, and potential pain points. Offer suggestions to enhance the customer experience throughout the journey

4/ Growth Hacking Strategies Prompt:

Suggest innovative growth hacking strategies for a [industry] startup aiming to rapidly increase user acquisition and engagement with a minimal budget. Include actionable tactics and real-world examples of successful implementations

5/ Lean Startup Methodology Prompt:

Explain how to apply the Lean Startup methodology to develop a new product in the [industry] sector. Outline key steps like hypothesis formulation, building a minimum viable product (MVP), testing, and iterating based on customer feedback. Provide tips to maximize efficiency and success during this process.

ACTIONABLE TIPS
Create a Synthetic Dataset Using Llama 3.1 to Fine-Tune Your LLM

This guide will show you how to use Llama 3.1 405B and Nvidia Nemotron 4 to create a synthetic dataset for instruction fine-tuning.

Key Steps

Get Access: Obtain an Nvidia NIM API key to access Llama 3.1 405B.
Generate Subtopics: Use Llama 3.1 to generate subtopics for your dataset.
Create Instructions: Use Llama 3.1 to create instructions for each subtopic.
Generate Responses: Use Llama 3.1 to generate responses for each instruction.
Score and Filter: Use Nvidia's Nemotron 4 to score and filter the responses.
Remove Low-Quality Responses: Remove responses based on helpfulness and verbosity scores.
Save Dataset: Save the filtered dataset as a JSON Lines file.
Upload Dataset: Log in to Hugging Face and upload your dataset.

This process helps you create a synthetic dataset for fine-tuning smaller LLMs. It uses Llama 3.1 405B to generate high-quality instruction-response pairs and Nemotron 4 to refine the dataset.

This method is cost-effective and gives you control over the dataset creation process.

See Tutorial here

JOBS AND TRAININGS
Anthropic's LLM Bootcamp: Improve Your Prompts

Anthropic has released a free course on LLM prompt evaluations to enhance your AI applications.

What's the Big Deal?

If your AI isn't performing well, LLM prompt evaluations can help. These evaluations include:

Systematic tests to assess your prompts
Performance measurements to improve AI capabilities
Methods to significantly boost AI performance

Course Overview

The course consists of 9 chapters, covering techniques used by Anthropic with enterprise clients. Topics include:

Test Dataset Mastery
Exact String Matching
Keyword Detective Work
Regex Wizardry
LLM Grading Guru
Custom Rubric Design
Prompt Iteration Station
Model Version Showdown
Quality Gatekeeper

Why Should You Care?

Improve AI performance
Free course

Check out the BoothCamp

🦾 Master AI & ChatGPT for FREE in just 3 hours 🤯

1 Million+ people have attended, and are RAVING about this AI Workshop.
Don’t believe us? Attend it for free and see it for yourself.

Save your spot here. (100 free spots only)

Highly Recommended: 🚀

Join this 3-hour Power-Packed Masterclass worth $399 for absolutely free and learn 20+ AI tools to become 10x better & faster at what you do

👉Save your seat now (FREE for First 100)

🗓️ Tomorrow | ⏱️ 10 AM EST

In this Masterclass, you’ll learn how to:

🚀 Do quick excel analysis & make AI-powered PPTs
🚀 Build your own personal AI assistant to save 10+ hours
🚀 Become an expert at prompting & learn 20+ AI tools
🚀 Research faster & make your life a lot simpler & more…

👉 Register here (Offer valid for First 100 people only)🎁

Thank you, see you next week!

29 | OpenAI launched yesterday o1 | Waymo is proving safety | Pixtral Multimodal

Wilson CELY — Fri, 13 Sep 2024 11:07:00 +0000

Our Menu :)

OpenAI Unveils New "01" Model Series
Waymo Leads the Pack in Autonomous Vehicle Safety, New Data Reveals
Mistral Challenges Tech Giants with Launch of Pixtral 12B, Its First Multimodal AI Model

Actionable Tips:

Listen to anything on the go with the highest-quality voices
Replit Agent: AI for development everybody is talking about

Jobs and Training

Next Week

Read time: 10 Minutes

WEEKLY DIGEST
OpenAI Unveils New "01" Model Series

ElevenLabs_2024-09-23T11_53_46_Adam_pre_s50_sb75_se0_b_m2(5)

Sep 10, 2024. OpenAI, the company behind ChatGPT, has just released a new AI system called O1 (also known as "Strawberry"). This new system represents a big change in how AI works and thinks. Here's what you need to know:

What's New About O1?

Thinking Before Speaking: Unlike older AI systems, O1 takes time to think before giving an answer. You can even see its thought process.
Solving Tough Problems: O1 is great at tasks that need careful thinking, like advanced math and complex coding. It sometimes does better than human experts.
Using Less to Do More: O1 doesn't need to be as big as other AI systems to be smart. It's about how it uses information, not how much it knows.
Learning While Working: Every time O1 solves a problem, it gets better at solving similar problems in the future. It's like it's always learning on the job.

Availability and Versions:

O-1 Mini: A faster, more efficient version
O-1 Preview: The more powerful version available to select users
A third, more advanced version is being used internally at OpenAI, to generate synthetic data potentially to train GPT-5

30 messages/week for o1-preview (the big berry).
50 messages/week for o1-mini (80% cheaper and 3-5x faster).

Train of Thought.

OpenAI doesn't share many details about how their model is trained, which is typical for them. They mention two things: first, the model is trained to "think before it answers," and second, this training uses a method called reinforcement learning.

"Think before you answer" is a technique where the model explains its reasoning step by step. You might have seen this if you've worked with advanced prompts. In this case, the model is trained to give detailed, thoughtful answers.

What you're seeing here is a summary of the model's output. For safety and competitive reasons, OpenAI doesn't show the exact answers from the model. Instead, another model summarizes the steps, though these summaries might not be perfect.

This use of secondary AI models is a common theme in the training process. OpenAI is getting better at using existing models to improve and speed up new training. For example, after collecting the initial training data, they used a special tool to remove harmful examples, like explicit content.

Similarly, another model called GPT-4o was used to check for safety and preparedness. It acted as both a security guard and a partner in tests. In two exercises, "MakeMeSay" and "MakeMePay," the model played the roles of a manipulator and a con artist, respectively.

This brings us to the topic of safety with this model.

Why Is This Important?

More Human-Like Reasoning The way O1 thinks through problems is more similar to how humans think, which could make it easier for people to work with and understand.
New Possibilities in Many Fields From education to scientific research, O1 could open up new ways of learning and discovering.

Technical aspects

49th percentile in the 2024 International Olympiad in Informatics
83.3% score on AIME math exam with consensus voting (GPT-4o: 13.4%)
Elo rating of 1673 on Codeforces, outperforming 89% of human competitors
362.14 points (above gold medal threshold) with relaxed submission constraints in International Olympiad in Informatics
01-preview is top-ranked in LiveBench, a contamination-free benchmark.

How Does Reasoning work?

The o1 models introduce reasoning tokens. The models use these reasoning tokens to "think", breaking down their understanding of the prompt and considering multiple approaches to generating a response.
The reinforcement learning approach trains the model to think through problems, similar to AlphaGo's Monte Carlo tree search.
This shifts computational focus from pre-training to inference time, allowing the model to explore multiple strategies and scenarios during problem-solving.

In conclusion, This release aligns more closely with the original GPT-3 than the ChatGPT release. It introduces a new approach that some may find incredibly valuable in ways even we at OpenAI can't predict.

However, it's not a mass-market product that works seamlessly and unlocks new value for everyone effortlessly.

I'm confident that we will have another significant moment like ChatGPT soon!

OpenAI’s Blog

There are already some interesting use cases

Favorite o1 Use-Cases: Standouts

An absolute must-watch: A researcher's reaction after o1 wrote his PhD code in just one hour, a task that originally took him ten months. The reaction video went so viral that he had to make a second response video.

YouTube video by Kyle Kabasares: "ChatGPT o1 Preview + Mini: Wrote My PhD Code in 1 Hour*—What Took Me ~1 Year"

Another notable example: A doctor used o1 to write a "major cancer treatment project" in under a minute, a task that would have taken him days. o1 even provided at least one creative idea he might not have thought of in "thirty years."

Aaron Levie’s take: o1 could potentially replace many enterprise SaaS use-cases by handling complex tasks, such as determining a contract's effective date through reasoning and self-validating. Levie (of Box fame) says this capability is "extremely useful for complex business processes" typically solved by SaaS products.

PRO(MPT) TIP: Ethan Mollick emphasized prompting your question in a way that an AI might find interesting, otherwise it might not think so hard.

More o1 Prompting Tips:

First, have a casual back-and-forth with 4o or Claude Sonnet about your ideas. Then, clearly summarize what you want done, along with any constraints.
Be specific about what you're asking. Instead of a broad question, narrow it down to a particular area or issue.
When the AI gives you an answer, ask it to explain why. This helps you spot any mistakes in its reasoning.

P.S.: If you want to learn more about how o1 works, this is a must-read. o1 can handle problems with up to 200 steps, but struggles approaching ~250 steps… so it’s perfect for difficult seating charts!

Twitter tweet

Update September 15th:
WILD o1 Demos We Discovered- YOU CAN YRY:

Generated an Animated Solar System: Created in just 5 prompts.
Wrote a Poem with Incredibly Strict Rules: Showcasing its ability to adhere to complex constraints.
Solved Complex Crossword Clues and Intelligence Tests: Demonstrating advanced problem-solving capabilities.
Created Fractal Art Using JavaScript and Adobe Firefly: Combining coding and artistic creativity.
Completed White-Collar Tasks: Such as estimating the number of Chinese people with an annual disposable income over 100K Yuan.
Achieved First Place on the Norway Mensa IQ Test: Outperforming other models tested, with an IQ score of approximately 120.

WEEKLY DIGEST
Waymo Leads the Pack in Autonomous Vehicle Safety, New Data Reveals

Recent data and analysis paint a picture of a technology that's not just futuristic, but increasingly safer than human drivers.

Widespread Deployment

Waymo's autonomous vehicles are no longer confined to controlled environments. They're now operational in major U.S. cities including San Francisco, Phoenix, Austin, Miami, Seattle, Los Angeles, and Atlanta. This widespread deployment is a testament to the company's confidence in its technology and its ability to navigate diverse urban environments.

Impressive Growth and Safety Record

The numbers speak volumes about Waymo's progress:

A fleet of approximately 700 vehicles
100,000 rides per week (up from 10,000 a year ago)
22 million miles driven over 2 million paid rides

But it's the safety statistics that are truly turning heads. Waymo reports that its autonomous vehicles cause fewer than one crash per million miles driven. To put this into perspective, in the 22 million miles driven, Waymo has reported:

20 crashes with injuries
5 crashes serious enough to deploy airbags
Only 1 crash involving serious injuries

Waymo claims that human drivers in San Francisco and Phoenix would have caused 64 crashes over the same distance, 31 of which would have been severe enough to trigger airbags.

Human Error: The Biggest Threat to Self-Driving Cars?

An in-depth analysis by Timothy B. Lee of Understanding AI reveals an interesting twist: out of the 23 most serious crashes involving Waymo vehicles, 19 were actually caused by human error. Specifically:

16 incidents involved humans rear-ending Waymo cars
3 cases where humans ran red lights and collided with Waymo vehicles

This leaves only 4 serious crashes that could be attributed to Waymo's technology itself.

Safety Comparison with Human Drivers

Lee's analysis suggests that Waymo's injury crash rates are approximately 60-70% lower than those of human drivers. This stark contrast highlights the potential of autonomous vehicles to significantly reduce road accidents and improve overall traffic safety.

The Road Ahead

As the technology continues to improve and gain public trust, the autonomous vehicle market is poised for significant growth. S&P Global projects that by 2034, there will be 230,000 driverless taxis sold in the U.S. alone.

However, the irony isn't lost on industry observers: the biggest challenge facing self-driving cars might not be the technology itself, but rather the unpredictable behavior of human drivers sharing the road.

The Bottom Line

Waymo's impressive safety record and rapid expansion across major U.S. cities signal a turning point in the autonomous vehicle industry. As public perception catches up with the reality of the technology's safety benefits, we may be witnessing the early stages of a transportation revolution.

The question now isn't whether self-driving cars will become mainstream, but how quickly society will adapt to this safer, more efficient mode of transportation. As one industry insider quipped, "The safest self-driving car tech will be the one everyone uses—we might not be the safest drivers, but us humans don't mess around when it comes to 'unsafe' new tech."

With Waymo leading the charge, the future of transportation looks not just autonomous, but demonstrably safer.

WEEKLY DIGEST
Mistral Challenges Tech Giants with Launch of Pixtral 12B, Its First Multimodal AI Model

In a bold move that signals its ambitions in the AI race, French startup Mistral AI has unveiled Pixtral 12B, its first-ever multimodal model capable of processing both language and visual data. This release marks Mistral's entry into the competitive field of vision-language models, putting it in direct competition with tech giants like Meta and OpenAI.

A New Contender in Multimodal AI

Pixtral 12B builds upon Mistral's text-based Nemo 12B model, incorporating a 400 million parameter vision adapter. This architectural choice allows the model to leverage Mistral's existing language prowess while adding robust image processing capabilities.

"With Pixtral 12B, we're pushing the boundaries of what's possible in AI," said Arthur Mensch, CEO and co-founder of Mistral AI. "This model represents a significant step forward in our mission to create more versatile and powerful AI systems."

Technical Specifications and Features

Pixtral 12B boasts impressive specifications:

Text backbone: Mistral Nemo 12B
Vision Adapter: 400M parameters
Vocabulary size: 131,072 tokens
Image processing: Up to 1024x1024 pixels, divided into 16x16 pixel patches
New special tokens: 'img', 'img_break', 'img_end'
Model weights in bfloat16 format
Total download size: 24GB

The model uses GeLU activation for the vision adapter and 2D Rotary Position Embedding (RoPE) for the vision encoder, techniques that have shown promise in improving model performance and efficiency.

Accessibility and Implementation

In line with Mistral's commitment to open-source AI, the company has made Pixtral 12B's weights available on the Hugging Face Hub. Developers can access the model through the mistral_common Python package, which now supports image input alongside text in user messages.

Installation is straightforward:

pip install --upgrade mistral_commonCopy

This integration allows for seamless incorporation of image data within the text processing pipeline, potentially enabling applications like visual question answering and image captioning.

Additionally, following Mistral's previous practice, the model is also distributed via a peer-to-peer torrent network, further enhancing accessibility for developers worldwide.

Performance and Future Prospects

While specific benchmarks are yet to be released, the AI community is eagerly anticipating performance comparisons with other multimodal models like OpenAI's GPT-4V or Meta's CLIP.

"The release of Pixtral 12B opens up exciting new possibilities in AI application development," noted AI researcher Dr. Emily Chen, unaffiliated with Mistral. "Its large scale and advanced architecture could enable more sophisticated AI applications across various industries, from content creation to data analysis."

As Mistral continues to push the boundaries of AI technology, all eyes will be on how Pixtral 12B performs in real-world applications and how it stacks up against offerings from more established players in the field.

With this release, Mistral has made it clear that it's not just participating in the AI race – it's aiming to lead it. As the company refines its multimodal capabilities, the broader implications for the AI industry and beyond remain to be seen.

Access the model in Hugging Face

ACTIONABLE TIPS
Listen to anything on the go with the highest quality voices

This week introduces an exciting free mobile app from Eleven Labs that transforms any text into high-quality audio using advanced AI voices.

Why It’s Cool and Useful:

Versatile Import Options: Easily import text by typing, pasting URLs, uploading files like PDFs, or using your phone’s camera for OCR to read text from images.
Customizable Playback: Adjust the speed and choose from a variety of voices based on language, gender, and style to suit your preferences.

Additional Tip: For complex documents like white papers that include charts and extensive formatting, use an LLM like ChatGPT to adapt the content for audio. This involves:

Uploading the Document: Import the PDF into the LLM platform.
Preparing a Prompt: Instruct the LLM to translate (if needed) and simplify the text, removing citations and complex formatting.
Generating a Clean Version: The LLM creates a streamlined version suitable for narration.
Listening on the App: Load the adapted document into the Eleven Labs app for a smooth listening experience.

Go to 11Labs Reader

ACTIONABLE TIPS
Replit Agent: AI for development everybody is talking about

Twitter tweet

Go to Replit

Thank you, see you next week!

28 | AI Scientists: LLMs Can Create New Knowledge for Humanity

Wilson CELY — Fri, 30 Aug 2024 11:00:00 +0000

Our Menu :)

The AI Scientist: When Machines Start Doing the Research
Eric Schmidt Drops AI Truth Bombs
What is an AI Engineer?
OpenAI: How Strawberries and Stars May Reshape Artificial Intelligence

Actionable Tips:

Next Week

Read time: 4 Minutes

GEN AI AT WORK
The AI Scientist: When Machines Start Doing the Research

Read with Eleven Labs

Sakana AI, a Japanese research lab, has unveiled a system that might just turn the world of scientific discovery on its head.

Meet the "AI Scientist”

Imagine a tireless researcher, working 24/7, generating novel ideas, designing experiments, writing code, analyzing results, and even penning scientific papers. Now imagine this researcher isn't human at all, but an artificial intelligence system. This is precisely what Sakana AI has created.

The AI Scientist automates the entire research lifecycle, from brainstorming to peer review. It's a development that could dramatically accelerate scientific progress across various fields, starting with machine learning itself.

The $15 Paper: Democratizing Research or Opening Pandora's Box?

One of the most striking aspects of this system is its efficiency. Each research paper produced by the AI Scientist costs approximately $15 – a figure that would make any university administrator's eyes light up. This cost-effectiveness could democratize research, allowing smaller institutions and even individuals to contribute to scientific discourse at an unprecedented scale.

However, as with any transformative technology, the AI Scientist raises important questions. Could this flood the academic world with low-quality papers? How do we ensure the integrity of research in an age of automated discovery?

The Road to AGI: Are We Witnessing the Beginning of the End?

The implications of the AI Scientist extend far beyond academia. Some AI researchers, like Leopold Ashen Briner, have long posited that automated AI research is the key to achieving Artificial General Intelligence (AGI). Briner's "Situational Awareness" paper suggests we might reach this milestone as soon as 2027.

With the AI Scientist, we may be witnessing the first steps towards this intelligence explosion. As these systems improve and iterate upon themselves, we could be entering an era of exponential growth in AI capabilities.

Not Without Its Flaws: The Current Limitations

Despite its impressive capabilities, the AI Scientist is not without limitations. It currently lacks vision capabilities, sometimes produces unreadable plots, and occasionally makes critical errors in data interpretation. These flaws serve as a reminder that while AI is advancing rapidly, it still requires human oversight and refinement.

The Ethical Quandary: Potential Misuse and Unintended Consequences

As with any powerful tool, the AI Scientist comes with potential risks. There are concerns about its misuse in creating unethical research or even dangerous materials if given access to physical laboratories. The academic community will need to grapple with these ethical considerations as the technology advances.

The Future of Science: Collaboration, Not Replacement

While the AI Scientist represents a significant leap forward, it's unlikely to replace human scientists entirely. Instead, we're likely to see a shift in the role of human researchers, moving towards higher-level direction, interpretation, and ethical oversight of AI-driven research.

The Bottom Line: A New Era of Discovery

The AI Scientist represents more than just a new tool in the researcher's toolkit. It's a harbinger of a new era in scientific discovery, one where the boundaries of human knowledge could expand at an unprecedented rate. As we stand on the brink of this new frontier, one thing is clear: the future of science is set to be more exciting, more productive, and more automated than ever before.

As we navigate this brave new world of AI-driven discovery, we must remain vigilant about the ethical implications and potential risks. But if managed correctly, the AI Scientist could usher in a golden age of innovation, democratizing research and accelerating our understanding of the world around us.

The question now is not whether AI will transform scientific research, but how quickly and how profoundly. As we watch this revolution unfold, one thing is certain: the world of tomorrow will be shaped by the AI Scientists of today.

Read +

GEN AI AT WORK
Eric Schmidt Drops AI Truth Bombs

In a candid interview at Stanford, former Google CEO Eric Schmidt didn't pull any punches. From predicting the next AI revolution to calling out Europe's tech struggles, Schmidt's insights offer a glimpse into the future of tech - and it's not all rosy.

The Trillion-Dollar AI Trifecta

Schmidt believes we're on the cusp of an AI revolution, driven by three key advancements:

Massive Context Windows: Imagine an AI that's always up-to-date on current events. That's the power of larger context windows, according to Schmidt.
Super-smart AI Agents: Picture an AI running chemistry experiments overnight and learning from the results. It's not science fiction - it's happening now with systems like ChroC.
Text-to-Anything: "Build me a Google competitor in 30 seconds," Schmidt hypothesizes. With advanced text-to-action capabilities, this could soon be reality.

"When these are delivered at scale," Schmidt warns, "it's going to have an impact on the world at a scale that no one understands yet."

The Global AI Race: A Two-Horse Game?

United States: Still in the lead, with Schmidt estimating a 10-year advantage in chip technology, particularly in sub-5nm chip production.
China: Playing catch-up but investing heavily. Schmidt notes they're "whopping mad" about recent US chip export restrictions.

The Dark Horses

India: Schmidt's "big swing state" in the AI race. With a deep pool of tech talent, India could be a major player if it can stem the brain drain to Silicon Valley.
Canada: A surprising mention by Schmidt as a potential AI powerhouse, thanks to its abundant hydropower for energy-hungry AI systems.

Europe: A Continent at a Crossroads

Schmidt doesn't mince words about Europe's position: "Europe is screwed up because of Brussels." But it's not a lost cause:

France: Singled out as having potential, with Schmidt praising President Macron's efforts to boost AI development.
Germany: Not seen as a major contender, despite its industrial might.
Other EU countries: Considered too small individually to make a significant impact.

The challenge for Europe? Overcoming what Schmidt sees as stifling regulations that make cutting-edge AI research difficult.

The Rest of the Pack

Japan and South Korea: Firmly in the US camp, according to Schmidt.
Taiwan: Praised for "amazing hardware" but criticized for "terrible software."
Middle East: While not a leader in AI development, Schmidt notes their potential to fund projects, though with concerns about adhering to US national security rules.

Google's Culture Shock: From 'Don't Be Evil' to 'Don't Work Late'?

In perhaps his most controversial statement, Schmidt suggested Google has lost its edge:

"Google decided that work-life balance and going home early and working from home was more important than winning."

This from the man who helped build Google into a tech behemoth. Is this the real reason behind Google's AI stumbles?

The $300 Billion Question

How much will it cost to stay ahead in AI? Schmidt throws out a staggering figure: $300 billion. And that might be conservative.

But with great power comes great responsibility. Schmidt acknowledges the challenge of regulating these powerful systems, especially when we can't always predict what they'll learn.

Startup Advice from a Tech Titan

For the entrepreneurs out there, Schmidt's advice is clear:

Use AI to prototype - fast. If you can't build it in a day, rethink your approach.
Be prepared to hustle. Schmidt points to Elon Musk's infamous work ethic as an example.
Move quickly. In the AI age, speed is everything.

The Dark Side of the AI Moon

It's not all technological utopia in Schmidt's vision. He raises red flags about:

AI-powered misinformation, especially during elections
The need for better critical thinking skills to combat fake news
Fair compensation for content creators whose work trains AI models
The risk of widening the gap between tech-rich and tech-poor nations

The Bottom Line

As the AI race heats up, one thing's clear: the finish line is still anyone's guess. But with insights like these from industry veterans like Schmidt, we're all getting a clearer view of the track ahead.

GEN AI STARTUP
What is an AI Engineer?

Originally defined the AI Engineer as a category between "ML Engineer" and "Software Engineer." Yet what we're seeing is that those boundaries are becoming blurrier.

One definition of an AI Engineer is someone who works with chains and tooling but not products or data. Yet, as the field grows, AI is impacting engineers in multiple ways:

They can use code autocomplete and general-purpose chatbots that work out of the box.
They can build LLM-enabled product features that require new frameworks and architectures (RAG, agents, etc.).
They can leverage full-fledged coding agents that automate large parts of software planning and implementation.

So, depending on your use case, it may not be as simple as working only with infrastructure and agents. A great AI Engineer is flexible enough to move up and down the stack, harnessing AI agents, digging through evals, or fine-tuning open-source models.

GEN AI STARTUP
OpenAI: How Strawberries and Stars May Reshape Artificial Intelligence

In the rapidly evolving world of artificial intelligence, OpenAI, a leading research laboratory, is cooking up something sweet. The San Francisco-based company is reportedly on the cusp of unveiling two new projects that could significantly advance the field: "Strawberry" and "Orion". These developments may herald a new era in machine reasoning and language understanding.

Project Strawberry, previously known in research circles as Q*, represents a significant leap in AI reasoning capabilities. Unlike current models that generate responses in real-time, Strawberry is designed to pause and "think", mimicking human cognitive processes more closely. This could prove particularly valuable in domains requiring complex problem-solving, such as mathematics and computer programming.

The potential implications are far-reaching. If successful, Strawberry could reduce the occurrence of AI "hallucinations"—instances where AI models generate plausible but incorrect information. This has been a persistent challenge in the deployment of large language models, limiting their reliability in critical applications.

Parallel to Strawberry, OpenAI is developing "Orion", intended to be their next flagship language model. In an intriguing twist of digital evolution, Orion is being trained on data generated by an advanced version of Strawberry. This recursive approach to AI development—using one AI system to train another—may offer a solution to the growing scarcity of high-quality training data, a bottleneck in advancing AI capabilities.

OpenAI's urgency in bringing these projects to fruition is understandable. The AI landscape is increasingly competitive, with well-funded rivals such as Google's DeepMind and Anthropic hot on their heels. The integration of Strawberry into ChatGPT, potentially as soon as autumn 2024, could be a strategic move to maintain OpenAI's market leadership.

However, challenges remain. The computational resources required for these advanced models are substantial, raising questions about their scalability and environmental impact. Moreover, as AI systems become more sophisticated, concerns about their societal implications and the need for robust governance frameworks grow more pressing.

As the world watches, the success of Strawberry and Orion could redefine not just what we expect from AI, but how we interact with technology at large. In the garden of artificial intelligence, it seems, strawberries may yield the sweetest fruit yet.

These cannabis gummies keep selling out in 2024

If you've ever struggled to enjoy cannabis due to the harshness of smoking or vaping, you're not alone. That’s why these new cannabis gummies caught our eye.

Mood is an online dispensary that has invented a “joint within a gummy” that’s extremely potent yet federally-legal. Their gummies are formulated to tap into the human body’s endocannabinoid system.

Although this system was discovered in the 1990’s, farmers and scientists at Mood were among the first to figure out how to tap into it with cannabis gummies. Just 1 of their rapid onset THC gummies can get you feeling right within 5 minutes!

Order Now

Thank you, see you next week!

Llama 3.1: What's the strategy behind open-source AI ?

Wilson CELY — Fri, 26 Jul 2024 02:11:00 +0000

WEEKLY DIGEST
The Open-Source Gambit: Meta's Llama 3.1 and the Future of AI 🦙 🏆

Listen the article .

Meta has just dropped Llama 3.1, and for the first time, we're seeing an open-source model that can go toe-to-toe with the closed-source heavyweights like GPT-4 and Claude 3.5 Sonnet.

What does it mean ?

Imagine for a moment that the AI world is a bustling marketplace. In one corner, you have the luxury boutiques – the GPT-4s and Claude 3.5s of the world – offering high-end, proprietary AI models at premium prices. And then, seemingly out of nowhere, Meta shows up and starts handing out free samples that are nearly as good as the expensive stuff.

That's essentially what's happening with Llama 3.1. Meta has released not one, but three powerful AI models that anyone can download and use, even for commercial purposes. The largest model, 405B, is performing on par with the best closed-source models out there. It's like giving away a sports car that can keep up with a Formula 1 racer.

The Big Picture 🖼️

Open-Source Power: These models are open-source, allowing researchers and developers to use them freely, even for commercial purposes (with some limitations*).
Massive Context Window: The context window has been expanded from 8,000 tokens to a whopping 128,000 tokens! – that's about 200 pages of text!
Competitive Performance: The 405B model is giving GPT-4 and Claude 3.5 a run for their money in various benchmarks.
Distilled Knowledge: The smaller models (8B and 70B) have been improved using knowledge distilled from the 405B model.
Multimodal Promise: While not yet available, Meta hints at future multimodal capabilities for Llama 3.
The model also comes with new pieces in the "Llama system" - a system of tools to make it easier for developers to build applications with Llama models. It's a long way from "I have an open-source LLM" to "I have a working LLM-enabled product."
The Llama ecosystem has now grown to include an agentic framework, content and jailbreaking safeguards, and a proposal for new developer interfaces.

*License of Llama 3.1

Permissively licensed, including commercial use.
Any derivative artifact, including models and datasets, must be distributed with the Llama 3.1 license.
Companies with greater than 700 million active users at the time of release cannot use the model.
The modification of the fairly restrictive Llama 3 license

The Secret Sauce: Knowledge Distillation

Here's where it gets really interesting. Meta didn't just create three separate models – they used a clever technique called "knowledge distillation" to make the smaller models punch above their weight class.

Think of it like this: Imagine you had a world-class chef (the 405B model) teach a talented home cook (the smaller models) all their secrets. The home cook might not be able to replicate every complex dish, but they'd certainly pick up techniques that make their cooking far better than before. That's essentially what Meta has done, using the massive 405B model to improve the capabilities of the smaller, more accessible models.

It's similar to what OpenAI did with GPT-4 Mini. Announced last week.

Democratizing AI

The true significance of Llama 3.1 isn't just its technical capabilities – it's what it means for the AI ecosystem as a whole. By releasing such powerful models as open-source, Meta is essentially lowering the barrier to entry for AI innovation.

This could lead to:

Keep AI Gold Rush: When powerful AI becomes a free resource, innovation explodes.
Price Wars: As open-source models improve, commercial AI services will likely need to drop their prices or offer unique value to stay competitive.
Ethical Challenges: With great power comes great responsibility. As advanced AI becomes more accessible, we'll need to grapple with ensuring its responsible use.

Llama 3.1's architecture is simpler compared to other complex strategies, like a mixture of experts. Despite its simplicity, it requires significant resources: Meta used 16,000 H100 GPUs to train the 405B mode. In the words of Ben Thompson, "Meta trained Llama like a big company with lots of resources, while OpenAI trained GPT-4 like a startup."

But that brings us back to :

Why did Meta repurposed $100M worth of hardware to give away the result for free?

Mark Zuckerberg's letter provides valuable insights into Meta's strategy in the competitive AI landscape. He emphasizes the need for developers, CEOs, and governments to control their models, ensuring data protection, fine-tuning for specific use cases, and making appropriate long-term technology choices. Larger organizations often require more control over the technology they use, making open-source AI like Llama an attractive option.

Meta doesn't want to depend on a single AI provider or proprietary systems, so they benefit from Llama being open-source. Zuckerberg recognizes the intense competition in AI development, meaning the best models will keep changing. Open-sourcing a model won't give a lasting edge in this fast-paced environment.

Moreover, OpenAI, Anthropic, DeepMind, and other foundation model developers make money by selling access to their models, either per token or via product subscriptions. Meta already has a multi-billion dollar ad business, which means it can afford to give away the AI stuff for free.

That allow them to apply, a longstanding idea in tech "commoditize your complement" - to make adjacent layers of the stack/supply-chain/ecosystem commodities, so that your product benefits from more options and your competition fights each other

With the release of Llama 3.1, a significant ecosystem of partners is already in place, offering the model on various cloud platforms and providing fine-tuning and RAG systems.

This extensive support seems to create a much larger moat than having a model that marginally outperforms others on benchmarks, especially as AI models become increasingly commoditized. Additionally, Meta can position itself as an advocate for AI safety.

Numbers Don't Lie, do they? 📊

Let's talk performance. The 405B model is flexing hard:

MMLU-Chat (General Knowledge): 88.6
GSM8K (Math): 96.8
HumanEval (Coding): 89

Interpretation from Clement Delangue

What's Next? 🔮

They're integrating Llama 3.1 across Facebook, Instagram, WhatsApp, and even AI glasses.

Will we see a world where AI is as ubiquitous as cat photos on social media? Only time will tell!

Access

Read the 92 page paper here
For API based access, checkout current partners of Meta
Huggingface page to download Llama 3.1 models
Try it on Groq

But of course, for Europe there are and will be restrictions.

Twitter tweet

The Catch (Because There's Always a Catch)

Before you go all-in on Llama 3.1, keep in mind:

It's not yet multimodal (no image or video processing... yet).
You'll need some serious computing power to run the full 405B model.
There's a learning curve to effectively utilizing open-source models.
The Llama 3.1 license is closer to freeware than open-source.

While the new Llama 3.1 license seems generous, it requires downstream models to include "Llama" in their name, potentially boosting the brand and ecosystem.

Meta might also influence standard interfaces and APIs for open-source LLMs, which raises concerns similar to Big Tech's efforts to define browser standards, a topic that privacy and open internet advocates have been fighting against for years.

We should indeed acknowledge Zuckerberg and Meta for supporting open-source AI. However, it's essential to remain cautious about relying on a tech giant, as their incentives may not always align with the developer community's interests, despite current synergies.

We put your money to work

Betterment’s financial experts and automated investing technology are working behind the scenes to make your money hustle while you do whatever you want.

Learn more

Thank you, see you next week!

26 | Bonjour, Moshi! How a French AI Lab is Outshining OpenAI

Wilson CELY — Fri, 05 Jul 2024 02:00:00 +0000

Our Menu :)

Bonjour, Moshi! How a French AI Lab is Outshining OpenAI
Perplexity: The Silicon Valley Darling in Hot Water
Tasty Treasures to Enjoy

Actionable Tips:

Prompt to Analyse Existing Companies in My Market

Jobs and Training

A Three-Course Feast for AI Enthusiasts
Jobs of the Week

Read time: 4 Minutes

WEEKLY DIGEST
Bonjour, Moshi! How a French AI Lab is Outshining OpenAI

July 3rd, 2024. What's the Big News?

Bonjour, Moshi!

Kyutai Labs, a French AI research lab that started last November, just dropped a bombshell. They've developed Moshi, an open-source AI model that outshines OpenAI's GPT-4o.

Moshi isn’t just any chatbot; it’s a real-time multimodal model, meaning it handles text and voice like a pro 🎤📝. Imagine Siri but on a caffeine rush – it can chat with you, understand your emotions, and respond faster than you can say "Bonjour".

Why is This Relevant for Business?

Innovation and Accessibility: Moshi is open-source, meaning anyone can use, tweak, and improve it. This democratizes AI tech, potentially reducing costs for businesses that want advanced AI without the heavy price tag.
Real-Time Capabilities: With super low latency (200 milliseconds, folks!), Moshi can revolutionize customer service, making interactions more natural and efficient. Imagine calling customer support and not wanting to throw your phone out the window – sounds like a win, right?
Security Features: Moshi includes watermarking technology to detect AI-generated audio, enhancing security and trustworthiness, crucial for industries like finance and healthcare.

Core Technical Specifications

Base Model: Helium 7B, a 7-billion parameter language model.
Training Data: Jointly trained on text and audio, including synthetic data.
Latency: Achieves an end-to-end latency of 200 milliseconds.
Compatibility: Operates on both MacBooks and consumer-grade GPUs.

Accessibility and Innovation

Kyutai plans to release Moshi as an open-source project, which will include the model, code, and associated research papers.

The Origins of Kyutai Labs and Its Founders

Kyutai Labs is the brainchild of three bigwigs: Xavier Niel, Rodolphe Saadé, and Eric Schmidt. Let's break down the squad:

Xavier Niel: The French entrepreneur behind Iliad (Free Mobile) and a major investor in tech startups. He's all about keeping tech talent in France and making sure Europe isn’t left behind in the AI race.
Rodolphe Saadé: CEO of CMA CGM, a major shipping company. He’s investing in AI to keep things innovative in logistics and beyond.
Eric Schmidt: The former CEO of Google. With his vast experience in tech, he brings a wealth of knowledge and network connections to the table.

Together, these guys have pooled $300M into Kyutai Labs to make sure the lab operates with an open-source philosophy. Their mission? To create AI models that anyone can use, ensuring that Europe stays competitive in the global tech scene.

A Star-Studded Research Team

The initial team of six researchers at Kyutai boasts impressive credentials. Three are former members of FAIR (Facebook Artificial Intelligence Research), and two come from DeepMind. Let's take a closer look at some key team members:

Patrick Pérez (Director): With experience at Microsoft Research, Inria, Technicolor, and Valeo, Pérez brings a wealth of knowledge in applied mathematics and computer vision.
Hervé Jégou: A former FAIR Paris lab director, Jégou is known for his work on image compression and computer vision.
Laurent Mazaré: With a background in cryptography and formal methods, Mazaré has worked at Jane Street and DeepMind.
Édouard Grave: A former co-lead on the LLaMA project at FAIR, Grave specializes in natural language processing.
Neil Zeghidour: Previously at DeepMind, Zeghidour founded a team working on generative models for audio.
Alexandre Défossez: With nearly a decade at Meta, Défossez led the development of the AudioCraft framework.

In just six months, Kyutai Labs and their small team of eight researchers have pulled off what many thought was impossible. Moshi is set to change the game for AI, making it more accessible and secure while ensuring Europe has a strong footing in the AI landscape. So, whether you’re a developer, a business owner, or just a tech enthusiast, keep an eye on Moshi – it’s going to be a wild ride! 🚀

Try Moshi

WEEKLY DIGEST
Perplexity: The Silicon Valley Darling in Hot Water

Perplexity is the talk of Silicon Valley these days:

💸 Raised a whopping $63M+ from heavyweights like Jeff Bezos and Nvidia.
🦄 Achieved unicorn status earlier this year and rumors say it’s on track to triple that valuation soon.
🎙️ Their CEO? Everywhere. If you listen to tech podcasts, you've probably heard him (yes, even on Lex Fridman’s show).
🕵️‍♂️ Some are even calling Perplexity a potential "Google killer."

But now, the burning question is: just how legit is this AI sensation?

The Scoop

Perplexity’s magic trick is digesting internet content and serving up quick summaries when you ask it something. Sounds cool, right? Well, there's a twist. Some of that content isn’t exactly theirs to use.

Earlier this month, Forbes caught Perplexity with its hand in the cookie jar—they nabbed a paywalled Forbes article and basically republished it on their new "Pages" product. Forbes wasn't amused and sent a letter to Perplexity’s CEO, accusing them of “willful infringement” of copyrighted articles. The CEO’s response? “We are improving it,… with more feedback.”

Then Wired jumped into the fray with an article titled "Perplexity Is a Bullshit Machine." Turns out, Perplexity has a habit of ignoring robots.txt code on publishers’ websites—basically the digital “No Trespassing” sign for web scrapers. To add insult to injury, Perplexity accidentally plagiarized Wired's article about Perplexity stealing content. Oof. 😬

The Fallout

This mess caught the eye of Amazon Web Services (yes, that Amazon), which is now investigating whether Perplexity is illegally scraping content using its servers.

Why It Matters

This content tug-of-war is likely to spark new laws or regulations about what's fair game for AI search engines and chatbots. Tech companies are hungry for content to train their AI, but publishers don’t want to lose eyeballs to chatbots.

Last week, Microsoft's head of AI stirred the pot by saying that any content published on the open web is “freeware” and can be copied and reproduced. Meanwhile, OpenAI is playing it smart, making friends with publishers left and right. They recently inked a licensing deal with TIME, probably gearing up to launch a search engine that cites sources and plays nice with publishers.

Stay tuned, because this AI drama is just heating up! 🔥

WEEKLY DIGEST

Superhuman 2.0 is an AI email assistant that automates email tasks like following up, drafting replies, and filtering spam.
Norm.ai converts regulations into computer code to automate compliance checks (raised $27M).
Inventive drafts proposal responses to save time for sales teams (raised $6.5M).
Jobright matches you with open jobs, tailors your resume, and suggests key network connections (raised $4.5M).
Instant, accurate, low-cost GenAI evaluations! Fortune 500 companies use Galileo Luna to detect hallucinations, prevent data leakage, and stop prompt attacks. Try Luna today!
Hebbia is an AI tool that can search and return answers from all of your internal documents (raised $100M).
Clay helps marketing/sales teams automate mundane tasks like contact data enrichment to scale personalized outreach (raised $62M).
Gemma 2 is a new model from Google that’s now the leading open-source model. Try it on Google AI Studio.
MagicSchool is a suite of AI tools for teachers, educators, and schools (raised $15M).

ACTIONABLE TIPS
Prompt to Analyse Existing Companies in My Market

This is a cracking prompt to identify and analyze competitors.

As usual, take it, play with it, and make it your own. Have fun and happy analyzing! 😎🔍

*#CONTEXT:*

*You are a professional researcher who helps Entrepreneurs analyse existing solutions on the market. You are a world-class expert in identifying direct competitors.*

*#GOAL:*

*I want you to analyse existing companies in my market and find 3 direct competitors for my business.*

*I will use this information to sharpen my value proposition.*

*#DIRECT COMPETITORS DEFINITION:*

- *Solutions that solve the same problem I do for the target audience I target.*
- *They can have a different product format or value proposition. But ultimately, we compete for the same attention and budget.*

*#GREAT ANALYSIS CRITERIA:*

- *Return only 3 direct competitors*
- *Write specific company names. Don't be afraid to get spicy.*
- *My direct competitors should always be real companies with active businesses.*
- *My "fight" with direct competitors should be exciting and relevant. Pick solutions that my audience can resonate with*
- *Give me an in-depth analysis and focus on insights that I can get from your overview*
- *Browse the web to find real companies*

*#INFORMATION ABOUT ME:*

- *My target audience: [INSERT YOUR TARGET AUDIENCE]*
- *My business: [INSERT SHORT DESCRIPTION ABOUT YOUR BUSINESS, PRODUCT/SERVICE]*

*#RESPONSE FORMATTING:*

*Use Markdown to format your response.*

FREE AI & ChatGPT Masterclass to automate 50% of your workflow

More than 300 Million people use AI across the globe, but just the top 1% know the right ones for the right use-cases.

Join this free masterclass on AI tools that will teach you the 25 most useful AI tools on the internet – that too for $0 (they have 100 free seats only!)

Get it now for absolutely free! (for first 100 users only) 🎁

This masterclass will teach you how to:

Build business strategies & solve problems like a pro
Write content for emails, socials & more in minutes
Build AI assistants & custom bots in minutes
Research 10x faster, do more in less time & make your life easier

You’ll wish you knew about this FREE AI masterclass sooner 😉

JOBS AND TRAINING
A Three-Course Feast for AI Enthusiasts 🍽️

Dive into a three-course series that arms learners with the:

Know-how to train.
Fine-tune.
Seamlessly integrate LLMs into AI products.

The first course is a jam-packed feast featuring over 50 lessons and 10 hands-on projects, all centered around LangChain and Deep Lake, a cutting-edge vector database for AI data.

Whether you're an AI newbie or looking to level up, this Full Stack LLM Bootcamp has got you covered! 🚀

Learn AI

JOBS AND TRAINING

Scale AI is looking for a Software Engineer focused on Cloud Infra.

Anthropic is hiring a Safety Cases Manager.

Meta is looking for a GenAI Program Manager to join their team.

Perplexity AI is hiring a Backend Software Engineer to work on their AI solutions.

Cohere is looking for a Director, Solutions Architect to join their team.

xAI is looking for an Information Security Engineer to join their team.

Thank you, see you next week!

Apple AI Strategy Finally Revealed

Wilson CELY — Thu, 20 Jun 2024 22:00:00 +0000

GEN AI AT WORK
Apple’s AI Strategy.

ElevenLabs: Apple’s AI Strategy

Apple has unveiled its ambitious AI strategy, dubbed "Apple Intelligence." This isn't just another feature update – it's a comprehensive AI ecosystem that's about to change how millions interact with their devices.

At the heart of Apple Intelligence is a clever two-pronged approach:

First, there's the on-device AI – a compact 3-billion parameter model that's like a Swiss Army knife for your iPhone. It can summarize emails, edit photos, and even generate images, all without sending your data to the cloud.

Apple's engineers have pulled off a neat trick here, using "adapters" that allow the AI to swap in different capabilities on the fly, A rough analogy here is The Matrix, where Neo gets new skillsets instantly downloaded into his brain - "I know kung fu."

But Apple knows that sometimes you need more firepower. That's where their mysterious "Private Cloud Compute" comes in.

These are beefier AI models running on Apple's servers, but with a twist – they've built a virtual Fort Knox around your data.

Even Apple can't peek at what you're processing. It's privacy with a capital P.

Now, here's where it gets really interesting. Apple has teamed up with OpenAI, integrating ChatGPT into iOS.

But don't worry, they're not handing over the keys to your digital kingdom. Instead, think of it like an App Store for AI. You choose when to use ChatGPT, and only then does your data get shared.

It's a smart move that lets Apple offer cutting-edge AI without compromising its privacy-first ethos.

What's in it for OpenAI? Exposure to millions of Apple users, potentially driving subscriptions to their premium services.

And Apple? They get to offer top-tier AI while taking a cut of any in-app purchases.

💰🚫🤝 No money is changing hands for the partnership itself, but both companies stand to benefit big time.

The implications are huge. Come this fall, we could see the single largest rollout of OS-level AI capabilities in history, with potentially 100 million devices running Apple Intelligence from day one.

It's a massive leap forward in bringing AI to the masses. But as with any technological revolution, questions remain.

Will these tools truly empower users as Apple claims?
How will they shape our daily interactions and productivity?
And what does this mean for the future of AI in our pockets and homes?

As users, we're all part of this unfolding story, and the next chapters promise to be nothing short of fascinating. Stay tuned, folks. The AI revolution is here, and it's wearing an Apple logo.

June 10, 2024. If you miss it, here is the WWDC24: Platforms State of the Union. 5-Minute Recap

GEN AI STARTUP
Claude 3.5 Sonnet: The Best AI Model I've Ever Tested

Anthropic launched Claude 3.5 Sonnet, an improved AI model that's faster, cheaper, and performs better than Claude 3 Opus, GPT-4o, and Gemini 1.5.

It's priced at $3 per million input tokens and $15 per million output tokens, with a 200K token context window.

Key enhancements:

Twice as fast and 80% cheaper than Claude 3 Opus
Excels in graduate-level reasoning, undergraduate-level knowledge, and coding
64% success rate on complex coding problems (vs. Claude 3 Opus's 38%)
Advanced vision capabilities, including interpreting charts and transcribing text from imperfect images

New 'Artifacts' feature:

Generate and interact with AI-generated content like code snippets and website design
Evolves Claude from a conversational AI to a collaborative work environment

Access:

Available for free on claude.ai and the iOS app
Extended features for Claude Pro and Team subscribers via Anthropic API, Amazon Bedrock, and Google Cloud’s Vertex AI

Twitter tweet

Try Claude 3.5 Sonnet

Your Brilliant Business Idea Just Got a New Best Friend

Got a business idea? Any idea? We're not picky. Big, small, "I thought of this in the shower" type stuff–we want it all. Whether you're dreaming of building an empire or just figuring out how to stop shuffling spreadsheets, we're here for it.

Our AI Ideas Generator asks you 3 questions and emails you a custom-built report of AI-powered solutions unique to your business.

Imagine having a hyper-intelligent, never-sleeps, doesn't-need-coffee AI solutions machine at your beck and call. That's our AI Ideas Generator. It takes your business conundrum, shakes it up with some LLM magic and–voila!--emails you a bespoke report of AI-powered solutions.

Outsmart, Outpace, Outdo: Whether you're aiming to leapfrog the competition or just be best-in-class in your industry, our custom AI solutions have you covered.

Ready to turn your business into the talk of the town (or at least the water cooler)? Let's get cracking! (And yes, it’s free!)

Thank you, see you next week!

24 | Microsoft Build | LLMs Are not Black Boxes Anymore

Wilson CELY — Fri, 24 May 2024 02:00:00 +0000

Our Menu :)

Copilot: A New Era for PC Interaction
Anthropic Unveils Breakthrough AI Research
Around the table

Actionable Tips:

Jobs Opportunities

Job Listings this week

Read time: 6 Minutes

GEN AI AT WORK
Copilot: A New Era for PC Interaction

May 22-23, 2024. Microsoft kicked off its annual Build developer conference with a bang, making a slew of artificial intelligence (AI) announcements that position the company as a frontrunner in the AI race.

Here are some of its standout features:

1. Screen Monitoring

Copilot can observe your screen while you work, similar to a feature recently introduced by OpenAI with ChatGPT-4o. Running on GPT-4o, Copilot's vision capabilities are poised to enhance tasks such as tech support and image analysis. This integration with Office applications makes it particularly powerful. For example, you can ask, "Does this Word document match our company’s template?"

2. Enhanced Search with "Recall"

Recall is a feature that keeps a continuous record of your screen activity, enabling you to locate anything you've ever searched for or viewed on your computer. For instance, you can ask it to find the “The AI Rev Hub article about GPT-4o from last week” or the “document sent by [x client] last week.” Despite the potential privacy concerns, Microsoft assures that all data seen by Recall remains private as the laptop running the feature uses a local model on-devise.

3. Real-time Translation and Transcription

This new feature offers real-time translation and transcription for meetings across any call platform (Zoom, Teams, Meet) in over 40 languages.

4. AI-Optimized Copilot+ PCs

Microsoft introduced a new category of AI hardware called Copilot+ PCs. These laptops feature specialized chips, all-day battery life, and advanced AI capabilities like real-time image creation and live language translation. Prices start at $999 when they launch on June 18.

5. Phi-3 Family

Microsoft is heavily focused on developing Small Language Models (SLMs) which are efficient AI models designed for use on mobile devices. They see two main reasons for SLM adoption:

The desire for local inference on mobile devices for privacy and reducing cloud dependency.
Microsoft's success in building AI-powered mobile apps, transitioning from apps to AI agents.

They have introduced a new multimodal AI model called Phi-3-vision that can handle both text and image inputs, allowing it to analyze things like charts and graphs. It is currently in preview and has 4.2 billion parameters.

Phi-3-mini is a 3.8B parameter language model, available in two context lengths (128K and 4K).
Phi-3-small is a 7B parameter language model, available in two context lengths (128K and 8K).
Phi-3-medium is a 14B parameter language model, available in two context lengths (128K and 4K).

You can find all Phi-3 models on Azure AI and Hugging Face.

6. Models-as-a-Service

Microsoft's MaaS on Azure lets developers use and customize top AI models from various companies on a pay-as-you-go basis, without needing dedicated virtual machines.

New additions include Nixtla TimeGen-1 and Core42 JAIS, among others. MaaS simplifies AI integration and monitoring, promoting innovation while reducing costs.

Microsoft has also expanded its partnership with Hugging Face to make open AI models more accessible across hardware

Twitter tweet

Conclusion

Microsoft excels at partnerships and collaborations, which help them differentiate between what's impossible and what's just challenging. Kevin Scott emphasizes that the complexity of their systems means no single team can uncover everything, and they rely on others for new ideas.

Satya Nadella's focus on cloud computing and AI has positioned Microsoft as a leader in these fields, with strategic acquisitions and partnerships further boosting their capabilities. The integration of AI into hardware through the Copilot Stack has revitalized the appeal of Windows PCs.

GEN AI AT WORK
LLMs Are not Black Boxes Anymore

Anthropic has released a groundbreaking paper exploring the inner workings of a Large Language Model (LLM) for the first time.

The Challenge: Understanding Neural Networks

Black Box Problem: Traditionally, LLMs functioned like black boxes, producing outputs without clear reasons.
Neurons and Features: LLMs consist of neurons activated by certain inputs, forming features that create the model's internal state.
Interpreting Neurons: Most neurons in LLMs are hard to interpret, hindering understanding.

New Technique: Dictionary Learning

Method: Anthropic used "dictionary learning," which identifies recurring neuron activation patterns. This helps represent the model's internal state with a few active features.
Application: Previously successful on tiny models, this technique was applied to Claude 3 Sonnet with remarkable results.

Findings

Millions of Features: Researchers mapped millions of features, creating a conceptual map of the model's internal states.
Conceptual Mapping: Similar features are located close to each other, reflecting conceptual similarities.
Feature Manipulation: Emphasizing or deemphasizing features can change the model's behavior. For instance, activating a feature for blindly agreeing with the user altered the model's responses.

Why It Matters

Safety Implications: This technique could help ensure AI safety by controlling and manipulating features.
Future Work: Mapping all features is computationally intensive, and more research is needed to fully understand the impact on AI safety.

Paper

GEN AI PROTECTION
Around the table

OpenAI has inked a deal with News Corp (owner of WSJ, Barron’s, MarketWatch, & more) to integrate its news content into ChatGPT. The agreement could be worth up to $250M, according to the WSJ.
Humane, the overhyped Ai Pin that wanted to replace the iPhone, is reportedly seeking a buyer.
Nvidia’s data center arm grew 427% since last year, driven by demand for its AI chips.
Amazon is planning to overhaul Alexa with an AI makeover later this year and charge a subscription for it.
DeepL, which competes with Google Translate, secured $300M to improve its text translation and writing tools.

GEN AI STARTUP
The first step into GenAI

It is tough to understand what you need to focus on if you want to learn about the field, considering the amount of work that has been done on the subject.

Those articles are great starting points to get a strong overview of what you need to know if you want to work with LLMs:

- Large Language Models: A Survey: https://lnkd.in/guWeHZHz
- A Survey of Large Language Models: https://lnkd.in/gFtc3ptC
- A Comprehensive Overview of Large Language Models: https://lnkd.in/gb2jesjX

GEN AI AT WORK
Job Listings this week

Remember that we have launched our beta job watcher

Robust Intelligence is seeking a Senior Product Manager focused on Security. Link
OctoAI is looking for a Staff Technical Product Manager to join their team. Link
Cohere is seeking a Member of Technical Staff, Evaluation to join their team. Link
Mistral AI is looking for a Revenue Development Representative to join their team. Link

Have an AI Idea and need help building it?

When you know AI should be part of your business but aren’t sure how to implement your concept, talk to AE Studio.

Elite software creators collaborate with you to turn any AI/ML idea into a reality–from NLP and custom chatbots to automated reports and beyond.

AE Studio has worked with early stage startups and Fortune 500 companies, and we’re ready to partner with your team. Computer vision, blockchain, e2e product development, you name it, we want to hear about it.

Book time to tell us all about your idea

Thank you, see you next week!

23 | Everything about "Her" => GPT-4o / Google IO

Wilson CELY — Fri, 17 May 2024 02:10:00 +0000

Our Menu :)

All you need to know about GPT-4o
Anthropic is expanding to Europe / Metaprompt
Google IO conference

Actionable Tips:

Jobs Opportunities

FRESH NEW! Job Listings this week

Read time: 6 Minutes

GEN AI AT WORK
All you need to know about GPT-4o

May 13, 2024. On Monday evening, I rewatched the movie "Her". It relates the story of someone falling in love with an AI (that has the voice of Scarlett Johansson). 👨💻❤️🤖

We're kind of getting to that reality to some extent with this announcement

OpenAI launched its new model called GPT-4o, here are some facts about the model:

Available through the Desktop App and the API
2x faster
50% cheaper
5x higher rate limits compared to GPT-4 turbo
Available to all users for free
And it is the most conversation ever made

BREAK DOWN =>

What is truly multimodal?

GPT-4o (“o” for “omni”) is a step towards much more natural human-computer interaction—it accepts as input any combination of text, audio, and image and generates any combination of text, audio, and image outputs.

📥 Input: Text, Text + Image, Text + Audio, Text + Video, Audio

📤 Output: Image, Image + Text, Text, Audio

Open many possibilities for disabled people. Be My Eyes + GPT-4o helping a blind person “see” what’s in front of him, even flagging down an available taxi.

GPT-4o is closer and closer to “HER”

🎤 Near real-time audio with 320ms on average, similar to human conversation.

Voice Mode feels like chatting with a real human—it captures your tone, language, and expressions in real-time. Many are describing it as a real-life Her (the voice in all the demos might actually be Scarlett Johansson).

Explore what it can do here:

live language translation (link).
realtime conversational speech (link).
lullabies and whispers (link).
sarcasm (link).
even singing (link) ⭐

How intelligent is the model?

🖼️ 69.1% on MMU; 92.8% on DocVQA

When can we use it?

GPT4-o will roll out from Monday 5/13th for all ChatGPT users through the Web UI, even for free tier users. It is also available through the API, and the playground.

GPT-4o vs Dalle-3

One of the biggest improvements is the capacity to include text within the generated image.

If we are familiar with DALL-E 3, we know that DALL-E 3 was not capable of generating text on its images, but now look at some examples of what the new model can do.

Twitter tweet

GEN AI AT WORK
Anthropic is expanding to Europe

May 14, 2024. Anthropic has launched its AI assistant Claude for users and businesses across Europe, aiming to boost productivity and creativity through:

Claude.ai: Web interface for the next-gen AI assistant
Claude iOS app: Free mobile version with intuitive experience
Claude Team plan: Secure access to AI capabilities and Claude 3 models for businesses

Key Features:

Strong comprehension of multiple European languages
User-friendly interface for seamless AI model integration

Pricing:

Free access to basic web and mobile offerings
€18+VAT/month for Claude Pro with advanced Claude 3 Opus model
€28+VAT/user/month for business Team plan (min. 5 seats)

With a focus on human-centric AI systems, Anthropic brings Claude's advanced language capabilities to Europe, allowing individuals and organizations to leverage its state-of-the-art models.

More info

Anthropic vs prompt engineers

Anthropic has released Metaprompt, a tool designed to enhance performance in Claude-powered applications by converting brief task descriptions into optimized prompt templates. It utilizes a few-shot prompt method with variables like subject, length, and tone.

GEN AI AT WORK
All you need to know about the Google IO conference

May 14, 2024. Allow me to summarise those two hours in just a few lines. Among other announcements at the Google IO conference, they introduced the following:

Google is integrating AI across its entire ecosystem.
Veo: their most capable video generation model
Project Astra: their new project focused on building a future AI assistant
Updates to Gemini 1.5 Pro: two new versions, one that's more lightweight, and another with a 2M token context length
AlphaFold 3 Expands Biochemical

BREAK DOWN =>

AI is here in Google Workspace.

This means paying users will have a ChatGPT-esque assistant right beside their screen that knows everything your Google apps know about you (so literally everything).

Why it matters: When you're working in Docs/Sheets/Gmail/Slides, you can ask Gemini to retrieve or summarise any content from all of these apps:

My personal all-time product

"Ask Photos" is a new feature that will upgrade searching through photos from simple keywords like “flowers” to “show me all the images of my son playing with our dog” and “what was the first place we visited on our Japan trip”.

Veo :

It is Google DeepMind's most capable video generation model to date. It generates videos:

High-quality with a 1080p resolution
That can go on for over a minute
In a wide range of cinematic and visual styles

Veo can take as input an image or video along with a text prompt. It can animate the image or edit the video when passed in as the input.

Moreover, it supports masked editing, enabling changes to specific areas of the video when you add a masked area to your video and text prompt.

More Info

Project Astra

Astra is Google's new project focused on building a future AI assistant, very similar to OpenAI's GPT-4o that was showcased live yesterday.

Google's new assistant is powered by Gemini and supports audio, text, video and image sharing in real-time.

Google still presents this project as a prototype, and the capabilities of Astra were only shared through pre-recorded videos since it is still not available to all users.

Early testers report a longer latency, and less emotional intelligence and tone for Astra compared to GPT-4o, but strong text to speech and potentially better ongoing video a long context support.

More Info

Gemini 1.5 Pro

Google unveiled two iterations of their flagship model Gemini 1.5 Pro.

Gemini 1.5 Pro Flash is the light-weight, fast and cost-efficient version of the model, meaning it is also multimodal and has a 1M token context length. The performance cost is small, with an MMLU of 78.9% compared to 81.9% for the original Gemini 1.5 Pro model.
Gemini 1.5 Pro had its context length doubled to 2M tokens. The new model is available via a waitlist for select developers building through the API.

AlphaFold 3 Expands Biochemical Modeling Capabilities

DeepMind's latest AlphaFold 3 model can predict 3D structures of not just proteins, but all biomolecules like DNA, RNA, and drug compounds, as well as their interactions.
Unlike previous versions, AlphaFold 3 is not open source and will be commercially controlled by Isomorphic Labs, a DeepMind spin-off focused on drug discovery.
AlphaFold 3 outperforms existing methods in accurately modeling molecular structures and interactions, achieving up to 77% success on benchmark datasets.

Business Impact:

Unprecedented accuracy in predicting drug-protein interactions could accelerate drug development and improve treatment efficacy.
Ability to model antibody-protein binding could aid in developing treatments for future pandemics and diseases.
Commercial control by Isomorphic Labs means businesses may need to license or partner for access to this powerful technology.

More Info

Other announcements

Gemma 2—Google’s best open-source models. it is a 27B parameter model that outperforms the previous version and will be available starting in June.
PaLI-3—a fresh open-source vision model. Their most capable image generation model, which will be available in multiple versions, each optimized for different types of tasks, from generating quick sketches to high-resolution images.
Gemini Live—a talking feature for Gemini similar to ChatGPT Voice Mode coming later this year.

GEN AI AT WORK
FRESH NEW! Add a file from Google Drive to ChatGPT

Twitter tweet

GEN AI Jobs
Jobs Opportunites in AI

Remember that we have launched our beta job watcher

Software Engineer, Safety - Character AI
Character AI is looking for a Software Engineer, Safety to join their team.
Events Coordinator - Scale AI
Scale AI is hiring an Events Coordinator for their team
People Operations Coordinator - Anthropic
Anthropic is seeking a People Operations Coordinator to join their team
Media Relations, Policy Communications - OpenAI
OpenAI is looking for a Media Relations, Policy Communications specialist to join their team.

More Info

Thank you, see you next week!

22 | AI Insights Every Business Leader Needs to Know | Google Cloud Next '24 | AI Music Innovation

Wilson CELY — Fri, 19 Apr 2024 02:00:00 +0000

GEN AI AT WORK
The AI Index Report from Stanford HAI

The Stanford Institute for Human-Centered AI (HAI) has recently published its seventh annual AI Index report. This year's edition covers a range of topics, including the emergence of multimodal foundation models, significant financial investments in generative AI, new performance benchmarks, shifting global attitudes, and new major regulations.

For those who may not be familiar, HAI is a prestigious institution affiliated with Stanford University and is highly regarded within the US tech community. Its mission is to advance AI research, education, policy, and practice to improve the human condition.

We understand that you may have a busy schedule and might not have time to read the full 500-page report. That's why we've summarized the key findings in 5 points to provide a quick and easy-to-digest overview.

1. Generative AI becomes a magnet for massive investments

Corporate investment was down overall last year, but generative AI funding soared.

The report shows most of that private generative AI investment happened in the U.S.

No surprise there! 🤷🏽‍♀️

2. Google Leads the Charge in Foundation Model Releases

Foundation models are large-scale, versatile models that can perform a wide range of tasks. For instance, OpenAI's GPT-3 and GPT-4 are foundation models that empower ChatGPT users to generate code, compose songs, or create imaginative images.
Due to the substantial resources required to train these models, the majority are now developed by the industry, while academia contributes only a limited number.

Google released the most in 2023. This is why Google's CEO and high management have been getting so much criticism for not being the leading force in AI. 🤷🏽‍♂️

As a reminder, the transformer architecture that enabled GPT models from OpenAI came from Google's DeepMind research.

3. Closed source vs Open source

The current hot topic in the AI field is whether foundation models should be open or closed. There are strong arguments on both sides, with some asserting that open models pose a risk, while others claim that open models fuel innovation.

But there’s less discussion about whether there are meaningful performance trade-offs

HELM: A benchmark for evaluating AI on general language understanding.
MMLU: Measures AI performance on multiple-choice questions across various subjects.
Chatbot Arena: A test of AI's ability to converse naturally and effectively.
HumanEval: Assesses AI's coding proficiency by generating and solving programming challenges.
SMiLE-bench: Evaluates AI's mathematical reasoning through structured math problems.
MATH: A benchmark focused on the AI's capabilities in solving advanced math questions.
GSNBk: A test of AI's skill in game simulations, strategy, and knowledge.
AgentBench: Measures the proficiency of AI in tasks that require agent-based decision-making skills.

4 Foundation models have gotten super expensive (Money and Carbon footprint)

Last week we explained the cost of training an LLM; AI companies rarely reveal the expenses involved in training their models, but the AI Index went beyond the typical speculation by collaborating with the AI research organization Epoch AI.

To come up with their cost estimates, the report explains, the Epoch team “analyzed training duration, as well as the type, quantity, and utilization rate of the training hardware” using information gleaned from publications, press releases, and technical reports.

The AI Index team also estimated the carbon footprint of certain large language models. The report notes that the variance between models is due to factors including model size, data center energy efficiency, and the carbon intensity of energy grids.

NOTE : Another chart in the report (not included here) shows a first guess at emissions related to inference—when a model is doing the work it was trained for—and calls for more disclosures on this topic.
As the report notes: “While the per-query emissions of inference may be relatively low, the total impact can surpass that of training when models are queried thousands, if not millions, of times daily.”

5) AI can’t beat humans at everything... yet

AI is Ahead of human:
AI performance has surpassed humans on benchmarks like:

image classification,
English comprehension
visual reasoning, and most

AI is Still Behind
AI performance lags humans in complex tasks like:

Planning,
visual common sense reasoning
Competition-level mathematics (But they’re not far behind) .

Source. https://aiindex.stanford.edu/report/

Report

GEN AI STARTUP
Google Cloud Next Keynote 2024

Here is all you need to know about the keynote:

Google announces the Cloud TPU v5p, its most powerful AI accelerator yet

Google has introduced its new Gemini large language model (LLM) and Cloud TPU v5p, which is an upgraded version of its Cloud TPU v5e. The company claims that the new Cloud TPU v5p offers significantly faster performance and is more cost-effective for training large language models.

Google’s Gemini 1.5 Pro can now hear

Google's Gemini 1.5 Pro update enables the model to process audio files and extract information directly, without depending on written transcripts. This enhancement improves its performance compared to the previous version and introduces new features such as inpainting and outpainting for generating images.

Google announces Axion, its first custom Arm-based data center processor

Google Cloud has unveiled its first custom-built Arm processor, Axion, which delivers improved performance and energy efficiency compared to competing instances. Technical documentation and availability details for Axion are expected to be released later this year.

AI editing tools are coming to all Google Photos users Google Photos is introducing AI-powered editing tools for all users, enabling them to improve their pictures even without advanced editing skills.

Google’s Gemini Pro 1.5 enters public preview on Vertex AI

Google Cloud announced a new AI platform called Vertex AI. It has a model garden that includes various open-source and closed-source models including Gemini 1.5 Pro, Llama, Claude 3, Stable Diffusion Mixol 8, and x 7B Wizard Coder.

One particularly interesting model is Gemini 1.5 Pro, which can process vast amounts of information in a single stream because of its 1 million token context window. This allows it to answer questions about an hour-long video, for instance.

Vertex AI also includes Code Gemma, a new fine-tuned lightweight open model designed for coding, and Vertex AI Agent Framework, a new framework to build agents. However, the speaker finds this framework to be less sophisticated than what he was expecting.

GEN AI AT WORK
AI Ecosystem Job Tracker

In this ever-evolving industry, it's crucial to keep up with the newest developments and career prospects.

Our job tracker will help you do just that by providing up-to-date information on open positions in various AI-focused organizations.

Image linked to our Platform

AI Rev Hub Job Tracker

GEN AI PROTECTION
Boston Dynamics: Atlas 2

Boston Dynamics has introduced a new fully electric Atlas robot.

The new robot is said to be stronger, more dextrous, and more agile than any previous generation. It looks futuristic and is claimed to be ready for commercial use.

It is designed to be used in existing human infrastructure because the world is already built for humans.
A vast amount of data is available to train humanoid robots.
The robot is expected to be a valuable asset for Hyundai in their automotive factories.
The humanoid form factor allows robots to move in efficient ways in a world designed for people.

Look at the teaser video, it looks so impressive.

GEN AI STARTUP
AI Music Arms Race: Udio Challenges Suno with Crisper Generative AI Songs

The AI music generation space is heating up with Udio, a new competitor to Suno, that promises to provide musicians with a powerful tool to create and profit from their AI-generated music.

Udio's Debut

Udio, an AI model capable of generating high-fidelity songs from text prompts, emerged from stealth just a week ago. Despite being a newcomer, Udio has already raised $10 million in seed funding from prominent investors and celebrities, including Andreessen Horowitz, Instagram co-founder Mike Krieger, and musicians will.i.am and Common.

Udio's impressive debut begs the question: How is this startup already producing music comparable to the established Suno AI, with potentially crisper sound?

The Minds Behind Udio

Udio was created in December 2023 by a team of former Google DeepMind researchers, including CEO David Ding, Conor Durkan, Charlie Nash, Yaroslav Ganin, and Andrew Sanchez. This pedigree likely contributes to Udio's strong performance straight out of the gate.

This is how I’ve created my summer hit with a simple prompt.

1) Hit up Udio (it's in Beta and free to make songs, link on the comments)
2) Drop in a simple prompt (no fancy tricks needed)
3) Have a blast!

Here are the links for both Suno and Udio.

Have fun!

Go to Udio

Go to Suno

Thank you, see you next week!

Sam Altman: Compute as the Future Currency. But at what cost?💧⚡

Wilson CELY — Fri, 05 Apr 2024 02:06:00 +0000

Our Menu :)

Compute as the Future Currency
The Hidden Cost of Training Large Language Models
Understanding how an LLM is Trained and its cost
Fight for the Current Datacenter Capacity

Read time: 8 Minutes

Compute as the Future Currency: OpenAI and Microsoft's Massive Investments

Sam Altman in podcast Lex Fridman

According to Sam Altman, the CEO of OpenAI, and others in the technology industry, computing power is poised to become the most valuable commodity in the future, surpassing traditional currencies like dollars or even Bitcoin.

As artificial intelligence (AI) becomes increasingly prevalent and integral to powering various aspects of the economy, the demand for compute resources is expected to skyrocket.

This realization has led to a compute arms race among tech giants, with OpenAI and Microsoft at the forefront of massive investments in this domain.

The Commoditization of AI Models and the Importance of Compute

While AI models like large language models (LLMs) are becoming commoditized rapidly*, with open-source alternatives matching or surpassing the quality of proprietary models like GPT-4, the true value lies in the computing power required to train and run these models.

*As these models are being open-sourced or replicated by various organizations and research groups, they are transforming from proprietary, highly guarded intellectual property into more commonly available and standardized offerings, akin to a commodity.

Altman recognizes that without owning the chips and infrastructure powering these models, profit margins will continue to shrink over time.

According to Altman, there are four main categories of winners in the AI space:

Chips: The actual GPUs or specialized silicon powering AI models, where Nvidia has emerged as a clear leader.
Infrastructure: Tools for LLM monitoring, content delivery networks (CDNs), data visualization, and agent frameworks.
Models: The AI models themselves, such as ChatGPT, Claude, and Llama, which are becoming increasingly commoditized.
Applications: AI-powered applications built from the ground up or existing applications integrated with AI capabilities.

Project Stargate: OpenAI and Microsoft's $100 Billion Supercomputer

In a major push to secure their position in the AI compute race, Microsoft and OpenAI have announced Project Stargate, a planned $100 billion data center project that will house an AI supercomputer set to launch in 2028. While the term "supercomputer" might be a misnomer, as it's essentially a massive data center, the scale of the investment is staggering, even for tech giants like Microsoft.

The project is part of a five-phase plan by OpenAI and Microsoft to roll out a new generation of computing centers for AI. Stargate represents the fifth and final phase, with Microsoft working on a smaller fourth-phase supercomputer for OpenAI, scheduled for launch around 2026.

The Insatiable Demand for Electricity and Compute

As Altman and Elon Musk have pointed out, the demand for computing power is rapidly outpacing the available electricity supply and infrastructure. Musk predicts that the next shortage after chips and transformers will be electricity itself, as data centers struggle to find enough power to run their AI computing resources.

To address this challenge, the industry is exploring innovative solutions, such as co-locating data centers with nuclear power plants or other large-scale energy sources. This modular and secure approach could help ensure a reliable and abundant supply of electricity to fuel the ever-increasing demand for computing power.

Attracting Top AI Talent with Compute Resources

Beyond the immediate benefits of increased computing power, companies like Meta, Google, and others are investing heavily in acquiring state-of-the-art GPUs and custom chips to gain a competitive edge in attracting top AI researchers and developers.

These highly skilled professionals are naturally drawn to organizations with the most powerful computing resources, as it enables them to push the boundaries of their research and development.

Source: Business Insider

As the AI revolution continues to gather momentum, compute power is emerging as the most critical resource driving innovation and progress in this field.

Tech giants like OpenAI and Microsoft are leading the charge with massive investments in data centers, specialized chips, and infrastructure to secure their position in the computing arms race.

However, this race also raises concerns about the concentration of power and potential negative impacts on the broader economy. As the demand for computing resources continues to soar, innovative solutions for energy generation and distribution will be crucial to sustaining this rapidly evolving landscape.

The Hidden Cost of Training Large Language Models

A few weeks ago we witness the NVIDIA GTC where its CEO presented a new family of chips that offers more compute for training and inferences (running an LLM) of Large Language Models and is less energy junky even if came with a cost in data precision (FP4)

However, as businesses increasingly adopt AI technologies, it's crucial to understand the distinct requirements of AI training workloads and the basis of Datacenter business and cost.

First start with the ugly truth:

Data from Paper ⇒ https://arxiv.org/pdf/2304.03271.pdf

Paper: Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models

Without integrated and inclusive approaches to addressing the global water challenge, nearly half of the world’s population will endure severe water stress by 2030, and roughly one in every four children worldwide will be living in areas subject to extremely high water stress by 2040.
Source: Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models

Freshwater scarcity is a growing global issue, with two-thirds of the world's population already affected by severe water shortages.

By 2030, it is estimated that nearly half of the world's population could face severe water stress. Data centers, which house AI models like GPT-3 and GPT-4, are major consumers of freshwater, in addition to being energy-intensive. In 2021, Google's U.S. data centers alone used 12.7 billion liters of freshwater for cooling, primarily potable water.

This amount of water could have been used to produce millions of cars or electric vehicles. The combined water footprint of U.S. data centers in 2014 was estimated at 626 billion liters.

Despite its significant impact, the water footprint of AI models remains largely unknown to both the AI community and the public. As freshwater scarcity worsens, it’s crucial to address this issue, reflected in commitments to be “Water Positive by 2030” by companies like Google, Microsoft, Meta, and Amazon.

DID YOU KNOW? 🤔

ChatGPT needs to “drink” a 500ml bottle of water for a simple conversation of roughly 20-50 questions and answers, depending on when and where ChatGPT is deployed. While a 500ml bottle of water might not seem too much, the total combined water footprint for inference is still extremely large, considering ChatGPT’s billions of users. All these numbers are likely to increase by multiple times for the newly-launched GPT-4 that has a significantly larger model size. But, up to this point, there has been little public data available to form a reasonable estimate of the water footprint for GPT-4. ³

Paper In Arxiv.org

Understanding how an LLM is Trained and its cost

To simplify it, consider FLOPs* as a measurement of compute power.

Training an LLM ( Large Language Model ):

Training a 175 billion parameter LLM typically requires over 1TB of memory to store the model parameters and intermediate states.
The training process for a 175 billion parameter model with 300 billion tokens in the dataset is estimated to require around 3.15 x 10^23 floating-point operations (FLOPs*) .
To provide the required 12.15 x 10^16 FLOPS of compute power during the one-month training period, it is estimated that around 6,000 NVIDIA H100 GPUs (with 67 TeraFLOPS each) would be needed, accounting for the typical 30% GPU utilization in training workloads5.
Previous LLM models, like GPT-3, were trained using around 10,000 older NVIDIA V100 GPUs5.

Inference of an LLM :
(Inference is the action to execute/run an LLM)

For inference, the GPU memory requirements are lower than training, as the model parameters need to be loaded but not the intermediate training states5.
However, the large size of LLM models, with billions of parameters, still poses a challenge for fitting the entire model on a single GPU's memory4.
To address this, inference is often performed on a cluster of GPUs, where the model parameters are distributed across multiple GPUs and the results are aggregated4.
The use of techniques like key-value caching and quantization can also help reduce the GPU memory requirements for inference.

In summary, training a large 175 billion parameter LLM requires a massive GPU cluster of around 6,000 high-end GPUs, while inference can be performed on a smaller cluster of GPUs by leveraging various optimization techniques

Datacenter Math

AI workloads require massive computing power provided by specialized hardware accelerators like NVIDIA's DGX servers with powerful GPUs. However, these systems consume a lot of electricity, which translates to high operating costs for data centers.

Let's look at the power requirements for a hypothetical data center with 2,560 NVIDIA DGX servers, each containing 8 H100 GPUs:

Each DGX server consumes around 10,200 watts of power when running AI workloads at peak utilization.
Additionally, there are power needs for storage, networking, management servers, etc., bringing the total critical IT power per DGX server to around 11,100 watts or 1,389 watts per GPU.
With 2,560 DGX servers, the total critical IT power required is 28.4 megawatts (MW).
1. For context, a typical nuclear power plant produces around 1,000 MW.
However, the servers don't run at 100% capacity all the time. Assuming an 80% utilization rate, the actual critical IT power consumed is 22.8 MW.
1. The 22.8 MW of actual critical IT power consumed is enough to power around 19,000 typical U.S. households simultaneously.
Data centers also require power for cooling, lighting, and other facilities. Factoring in a power usage effectiveness (PUE) ratio of 1.25, the total data center power consumed is 28.4 MW.
At an average industrial electricity rate of $0.083 per kilowatt-hour, the annual electricity cost for this data center would be around $20.7 million. And it is not enough power to train an LLM

Got to: SemiAnalysis

So what are the hyperscalers doing to address this issue?
=> See below

Fight for the Current Datacenter Capacity

The current State-of-the-Art Large Language Model supposedly used 8k GPUs during 90 days for a total 0f 15MW to complete its pre-training

NVIDIA GTC 2024

AI demand is projected to double the total datacenter critical IT power demand from 49 GW in 2023 to 96 GW by 2026, with 90% of the growth coming from AI-related demand.

This demand is primarily driven by the aggressive plans of major AI clouds to roll out accelerator chips. For example:

OpenAI plans to deploy hundreds of thousands of GPUs in their largest multi-site training cluster, requiring hundreds of megawatts of critical IT power.
Meta plans to have an installed base of 650,000 H100 equivalents by the end of the year, while GPU cloud provider CoreWeave has plans to invest $1.6B in a Plano, Texas facility, implying plans to spend up to 50MW of critical IT power and install 30,000-40,000 GPUs in that facility alone.

Source: StructureSearch

From a supply perspective, sell-side consensus estimates of 3 million GPUs shipped by Nvidia in calendar year 2024 would correspond to over 4,200 MW of datacenter needs, nearly 10% of current global datacenter capacity, just for one year's GPU shipments.

The top global hyperscalers are rapidly ramping up datacenter construction and colocation leasing to meet this demand.

For example, AWS bought a 1000MW nuclear-powered datacenter campus for $650M USD, providing a valuable pipeline of datacenter capacity.

AI demand is projected to significantly increase datacenter critical IT power demand in the coming years, driven by the aggressive plans of major AI clouds to roll out accelerator chips.

To meet this demand, top global hyperscalers are rapidly ramping up datacenter construction and colocation leasing. This presents both opportunities and challenges for businesses in the datacenter and AI industries.

The Environmental Cost of AI

Sasha Luccioni, an AI researcher with over a decade of experience, argues that we should shift our focus to addressing the current impacts of AI and developing tools to ensure future generations of AI models are trustworthy, sustainable, and ethical.

In her paper, she examines the power usage of training the BLOOM model at the Jean Zay computer cluster at IDRIS, a part of France’s CNRS.

Additionally, it provides empirical observations of the relationship of an AI Chip’s TDP to total cluster power usage including storage, networking and other IT equipment, all the way through the actual power draw from the grid.

Training GPT-3 in Microsoft’s state-of-the-art U.S. data centers can directly consume 700,000 liters of clean freshwater (enough for producing 370 BMW cars or 320 Tesla electric vehicles)

It estimates BLOOM's final training emitted around 24.7 tonnes of CO2eq considering just the dynamic power consumption, and 50.5 tonnes when accounting for all processes from equipment manufacturing to operational energy usage.
It analyzes the energy requirements and carbon emissions from deploying BLOOM for real-time inference via an API endpoint.
It compares BLOOM's emissions to other large language models like GPT-3, OPT-175B, and Gopher.
It estimates the total emissions from all experiments run as part of the BigScience workshop at around 124 tonnes of CO2eq.

The trend of "bigger is better" in AI has led to a 2,000-fold increase in the size of large language models over the past five years, exacerbating their environmental impact.

So far we have been only using GPT-3, which is quite small compared to GPT-4. And GPT-5 is around the corner.

Paper: Arxiv.org

Thank you, see you next week!

20 | The Future of AI and Robotics Begins Now, Grok is Open, RIP to Chatbot PI.

Wilson CELY — Fri, 22 Mar 2024 03:50:00 +0000

Our Menu :)

NVIDIA GTC: The Future of AI and Robotics Starts Now.
Grok is Open Source: Elon fulfilled its promise
Inflection is eaten alive by its biggest investor.
Exec Summary of 2hrs chat of Sam Altman with Alex Fridman

Actionable Tips:

Essentials for better prompting: RACEF Framework

Read time: 6 Minutes

GEN AI STARTUP
NVIDIA GTC: The Future of AI and Robotics Starts Now

Mar 18th, 2024. NVIDIA kicked off the GPU Technology Conference (GTC), and its CEO Jensen Huang unveiled a lot during this keynote.

What’s new?

NVIDIA's game-changing H100 AI chip catapulted it into a multitrillion-dollar league, possibly outpacing giants like Alphabet and Amazon. And now, with the introduction of the new Blackwell B200 GPU and the GB200 "superchip," NVIDIA seems poised to widen its lead even further, leaving competitors in the dust as it races ahead.

“Nvidia today accounts for more than 70 percent of A.I. chip sales and holds an even bigger position in training generative A.I. models, according to the research firm Omdia.”

Blackwell B200 GPU

NVIDIA has revealed the Blackwell B200 GPU, touted as the 'world's most powerful chip' for AI. This cutting-edge chip is designed to broaden access to trillion-parameter AI models, further solidifying NVIDIA's dominance in the AI market.

Key Features:

Performance: The B200 boasts an impressive 20 petaflops of FP4 performance, thanks to its 208 billion transistors.
Efficiency: When combined with a Grace CPU as a GB200, it provides 30x the LLM inference workload performance, leading to a cost and energy reduction of up to 25x compared to the H100.
Capacity: It can train a massive 1.8 trillion parameter model using just 2,000 GPUs and four megawatts, a significant improvement over the previous requirement of 8,000 Hopper GPUs.
Speed: The B200 demonstrates a 7x performance increase on GPT-3 LLM (175 billion parameters) and trains 4x faster than the H100.

Project GR00T: General Robotics zero zero Three

NVIDIA also introduced Project GR00T, which focuses on creating foundational models for humanoid robots. This initiative enables robots to comprehend and act on multimodal instructions, such as language, video, and demonstrations.

Highlights of Project GR00T:

Foundation Model: Empowers robots to understand and process multimodal instructions.
Cross-Platform Compatibility: Functions seamlessly across various robotic platforms.
Tech Stack: Leverages Isaac Lab for simulations, scales models with OSMO, and utilizes Jetson Thor for edge computing tasks.
GEAR Lab Integration: Strives to develop adaptable agents that perform effectively in both virtual and real-world environments.

NIM Microservices:

Streamlining AI Deployment Additionally, NVIDIA launched NIM microservices to simplify the deployment of AI. These microservices enable rapid development and scaling of AI applications across multiple platforms.

NVIDIA's Leap into Digital Twins and Advanced Robotics

NVIDIA has created a Matrix-like world called the Omniverse to train those robots. it allows individuals and teams to create physically accurate virtual worlds for industrial and scientific use cases

The introduction of Omniverse Cloud APIs extends these capabilities, promising a transformative impact across various sectors, including automotive, robotics, and beyond.

Companies like WPP, Media.Monks, Continental, Ansys, Cadence, Dassault Systèmes, Hexagon, Microsoft, Rockwell Automation, Siemens, and Trimble are embracing Omniverse Cloud APIs in their software portfolios.
Siemens is integrating Omniverse Cloud APIs into its Siemens Xcelerator Platform to enhance cloud-based product lifecycle management software like Teamcenter X with generative AI capabilities for immersive digital twin experiences
NVIDIA Omniverse has integrated with Apple Vision Pro to enable advanced 3D development and streaming capabilities.
This collaboration allows developers to transfer their Universal Scene Description (USD) scenes to NVIDIA's Graphics Delivery Network (GDN), facilitating the streaming of immersive 3D content to Apple Vision Pro's high-resolution displays globally

GEN AI STARTUP
Grok is Open Source: Elon fulfilled its promise

Mar 17th, 2024, xAI shook the tech industry by making Grok-1, a colossal language model with 314 billion parameters, open-source.

This Mixture-of-Experts model was not fine-tuned for any specific tasks.

One of Grok-1's most impressive features is its efficient weight usage, with only 25% of weights active for each token, boosting both efficiency and performance.

Elon Musk stayed true to his promise by open-sourcing Grok-1 following a legal battle with OpenAI over its proprietary approach, a decision that sparked controversy and accusations of hypocrisy.

Let's dive into some key specifications:

Parameters: 314 billion, with only 25% of weights active per token.
Architecture: A Mixture of 8 Experts, utilizing 2 per token.
Layers: 64 transformer layers, effortlessly combining multihead attention and dense blocks.
Tokenization: Employs a SentencePiece tokenizer, with a vocabulary size of 131,072.
Embedding and Positional Encoding: 6,144 embedding size, paired with corresponding rotary positional embeddings.
Context Length: Capable of handling 8,192 tokens with bf16 precision.

Repo Github

In terms of performance, Grok-1 outperforms its competitors:

It beats LLaMa 2 70B and Mixtral 8x7B with an impressive MMLU score of 73%, showcasing its efficiency and accuracy across various tests.

Here are some implementation details to consider:

Due to its massive size, Grok-1 demands significant GPU resources.
It utilizes an inefficient MoE layer implementation to avoid the need for custom kernels, prioritizing model correctness validation.
The model supports activation sharding and 8-bit quantization to optimize performance.

In a nutshell, Grok-1 is a game-changer for tech businesses, offering unparalleled performance and efficiency. With its open-source availability, it's an opportunity for tech founders to leverage this cutting-edge technology for their businesses.

Blog XAI

GEN AI STARTUP
Inflection is eaten alive by its biggest investor.

Image Credits: Hieronymus Bosch

Microsoft has made a significant move by bringing on board key members from the startup Inflection AI, including its co-founder Mustafa Suleyman, who's also known for co-founding Google's DeepMind. These new hires will be part of Microsoft's freshly established consumer AI division, Microsoft AI.

Highlights:

Mustafa Suleyman is stepping in as the CEO of Microsoft AI, tasked with spearheading consumer AI projects including Copilot, Bing, and Edge, under the direct guidance of Microsoft's CEO, Satya Nadella.
Joining Suleyman at Microsoft is Karen Simonyan, a close associate and long-term collaborator, who will serve as the Chief Scientist of Microsoft AI.
Last year, Inflection AI secured $1.3B in funding (with Microsoft leading the round) for the development of the Pi chatbot.
Despite the initial buzz, the startup faced challenges with its business model, prompting Suleyman and most of the Inflection AI team to transition to Microsoft.

Why It Matters:

Microsoft's strategic acquisition of top talent from Inflection AI, including a pivotal figure like Mustafa Suleyman, marks a bold step in its quest to dominate the consumer AI landscape.

This maneuver not only intensifies the talent wars within the AI industry but also leaves Inflection AI in a precarious position, with co-founder Reid Hoffman and the new CEO, Sean White, left to navigate the aftermath.

Know more

GEN AI STARTUP
Exec Summary of 2hrs chat of Sam Altman with Alex Fridman on its podcast.

OpenAI CEO Sam Altman interviewed by Lex Fridman discussing AI, GPT-5, Sora, Elon lawsuit, and more

On rumors of artificial general intelligence: “Ilya has not seen AGI. None of us have seen AGI. We’ve not built AGI.”

On the lawsuit from Elon Musk: “I think this whole thing is unbecoming of a builder…it makes me extra sad he’s doing [this] to us.”

On copyright and AI: “Do people who create valuable data deserve to have some way that they get compensated for use of it? I think the answer is yes. I don’t know yet what the answer is.”

On job loss as a result of AI: “The way I think about it is not what percent of jobs AI will do, but what percent of tasks will AI do on over one time horizon…maybe people are way more efficient at the job they do”

On GPT-5: “I expect that the delta between [GPT-5] and [GPT-4] will be the same as between [GPT-4] and [GPT-3]…We will release an amazing new model this year. I don’t know what we’ll call it.”

GEN AI AT WORK
Essentials for better prompting: RACEF Framework

Breaking Down RACEF

Role: Specify the perspective or role you want ChatGPT to adopt. This sets the stage for the type of information you're seeking.
Action: Define what you want ChatGPT to do. This clarity ensures the response is aligned with your expectations.
Context: Give background information to contextualize your request. This helps in tailoring the response more closely to your needs.
Example: Providing an example helps illustrate the kind of answer you're looking for, guiding the AI in the right direction.
Format: Specify how you want the information presented. This could be a list, a detailed explanation, or any other format that suits your requirement.

RACEF in Practice

Step 1: Role

Prompt Component: "You are a high school coding club advisor."
Purpose: This specifies the role ChatGPT should assume, providing a clear standpoint from which to offer suggestions.

Step 2: Action

Prompt Component: "Generate a list of project ideas."
Purpose: This defines the specific action ChatGPT needs to perform, ensuring the response is focused on idea generation.

Step 3: Context

Prompt Component: "The club members have intermediate programming skills and have worked with Python and JavaScript."
Purpose: Providing this context helps tailor the response to the skill level and experience of the club members, ensuring the project ideas are suitable.

Step 4: Example

Prompt Component: "Previously, they've enjoyed projects like building a simple web app and a basic game."
Purpose: Including an example of past projects gives ChatGPT a reference point for the complexity and type of projects that have been successful and engaging for the club.

Step 5: Format

Prompt Component: "List three ideas with a brief description for each."
Purpose: Specifying the format ensures that the response is organized e.g. as Table, Json, XML, email, post, List

Full RACEF Prompt Example:

"As a high school coding club advisor, generate a list of project ideas suitable for club members who have intermediate programming skills and have previously worked with Python and JavaScript. They've enjoyed projects like building a simple web app and a basic game before. Please list three ideas with a brief description for each."

"As a content creator focusing on sustainability, I'm looking for a unique angle for my next blog post. Considering recent trends in renewable energy, can you suggest a topic that highlights innovative home energy solutions? Ideally, present it as a question to engage readers."

Thank you, see you next week!

19 | Inside the EU AI Act, Devin: First AI-Based Ing, and Gen AI Robots..

Wilson CELY — Fri, 15 Mar 2024 03:00:00 +0000

Our Menu :)

🇪🇺 The EU has just APPROVED the AI Act 🎯🤖
The first AI software engineer: Welcome to Devin🤖
Figure AI + OpenAI: Finally, there will be a robot that will wash dishes and cut onions. 🍽️🤖
Musk is opening xAI’s code.🔓

Actionable Tips:

Prompt Library from Anthropics 📚
Smart Job interview preparation with ChatGPT 👔

Read time: 8 Minutes

GEN AI PROTECTION
🇪🇺 The European Parliament has just APPROVED the AI Act 🎯🤖.

The AI Act establishes a risk-based framework for the development and deployment of AI systems across the 27 member states.

What's New?

After years of deliberation, the European Parliament voted overwhelmingly in favor of the AI Act, with 523 votes for and only 46 against.

The Act introduces a tiered approach to regulating AI systems based on their perceived level of risk, with stricter rules and requirements for higher-risk applications.

Categorization of AI systems

This categorization ensures that higher-risk applications undergo more stringent review processes before deployment.

1) Unacceptable Risk: This category includes AI systems whose use is considered so harmful that they are banned outright.

Examples:

Cognitive behavioral manipulation of people or specific vulnerable groups.
Social scoring: classifying people based on behavior, socioeconomic status, or personal characteristics.
Biometric identification and categorization of people.
Real-time and remote biometric identification systems, such as facial recognition.

2) High Risk: AI applications in critical sectors like healthcare, education, employment, and law enforcement fall under this category.

They are subject to strict compliance requirements, including transparency, data governance, and human oversight.

Examples:

Critical infrastructures (e.g. transport) that could put the life and health of citizens at risk.
Educational or vocational training that may determine the access to education and professional course of someone’s life (e.g. scoring of exams).
Safety components of products (e.g. AI application in robot-assisted surgery).
Employment, management of workers, and access to self-employment (e.g. CV-sorting software for recruitment procedures).
Essential private and public services (e.g. credit scoring denying citizens opportunity to obtain a loan); law enforcement that may interfere with people’s fundamental rights (e.g. evaluation of the reliability of evidence).
Migration, asylum, and border control management (e.g. automated examination of visa applications).
Administration of justice and democratic processes (e.g. AI solutions to search for court rulings).

3) Limited Risk: This category applies to AI applications that pose some risk, necessitating specific transparency obligations.

For instance, chatbots must clearly disclose that they are AI-driven to ensure users are aware they are not interacting with humans. This ensures transparency and mitigates the potential for deception.

4) Minimal Risk: The majority of AI applications are expected to fall into this category, facing no additional regulatory requirements. These are applications like AI-enhanced video games or spam filters, where the risk to rights and safety is considered negligible.

Key aspects of the AI Act include:

Transparency for General Purpose AI (GPAI): Providers of GPAI models, including large language models like GPT-4, must comply with transparency obligations.
This includes providing technical documentation, and instructions for use, adhering to copyright laws, and publishing summaries about their training data.
Providers of "systemic" GPAI models with high computing power used for training face additional obligations like model evaluations, risk assessments, incident reporting, and cybersecurity measures.
The EU has established an AI Office within the European Commission to monitor the implementation and compliance of GPAI model providers, conduct evaluations, investigate potential risks, and handle complaints from downstream providers.

➵ Generative AI, like ChatGPT, will not be classified as high-risk but will have to comply with transparency requirements and EU copyright law. Some of the obligations are:

- Disclosing that the content was generated by AI;
- Designing the model to prevent it from generating illegal content;
- Publishing summaries of copyrighted data used for training.

Implementation Timeline:

The AI Act's implementation is structured to roll out in stages, ensuring a manageable transition for all stakeholders involved.

Following its official publication expected in the coming months, the act will come into force 20 days later. However, the provisions are phased:

Immediate Action: Certain prohibitions, particularly on AI uses considered an unacceptable risk, will be enforceable six months after the act becomes law.
Staggered Implementation: For other categories, the timelines vary - with rules applying after 12, 24, and 36 months. Full compliance across all aspects of the AI Act is anticipated by mid-2027.

The European Commission will provide resources and guidelines to assist stakeholders in understanding and adhering to the new regulations.

➵ Fines for non-compliance can be up to 35 million Euros or 7% of worldwide annual turnover.

EU AI Act

GEN AI AT WORK
The first AI software engineer: Welcome to Devin 🤖

Cognition Lab backed by industry giants such as Patrick and John Collison, Elad Gil, and Peter Thiel's Founders Fund, has recently introduced Devin, an AI model that promises to revolutionize the way software is developed.

What's New?

Devin is a cutting-edge AI model that can handle all aspects of software development, from coding to deployment.

It operates in a sandboxed environment with access to standard developer tools and facilitates user interaction through a natural language interface for real-time monitoring and commands.

Why Does It Matter?

Devin showcases significant advancements in autonomous software problem-solving and project execution, indicating a shift towards more sophisticated AI roles in software development.

Its superior performance in the SWE-Bench coding benchmark (a dataset that tests systems' ability to solve GitHub issues automatically.), where it resolved 13.86% of issues unassisted, highlights its capabilities.

Key features of Devin include:

Comprehensive Software Development: Handles all aspects of software development, from coding to deployment.
Sandboxed Environment: Operates in a secure, isolated environment with access to standard developer tools.
Natural Language Interface: Facilitates user interaction through a natural language interface for real-time monitoring and commands.
Superior Benchmark Performance: Demonstrates exceptional performance in the SWE-Bench coding benchmark, resolving 13.86% of issues unassisted.

The introduction of Devin to the market signals a significant milestone in the integration of AI into software engineering tasks, potentially transforming the way software is developed and deployed.

Hands-On Experience:

While access to Devin is currently limited to a select group of users, Cognition plans to make it more broadly available in the future.

Interested individuals or organizations can reach out for early access through Cognition's official channels, such as their website or by sending an email to info@cognition-labs.com.

Can it live up to the hype?

A Bloomberg reporter tested Devin this week and said he watched it build a custom website in under 10 minutes.
The CEO of fintech company Ramp called Devin the “single most impressive demo I’ve seen in the past decade.”
The CEO of AI heavyweight Perplexity AI also counts himself as a fan, praising the system’s ability to “look at results, replan, and iterate till success.”

Are software engineering jobs in peril?

If you could fully automate software engineering (my job), I think that would be great, since I could then move on to higher-leverage things,
Francois Chollet, is one of the world’s leading AI engineers.

But Chollet is skeptical: “Software engineering is not about copy/pasting code. Software engineering is the development and manipulation of mental models of problems and their solutions.“

While talk of AI agents replacing jobs has been ripe for some time now, Devin may prove to be an aid rather than a replacement for software engineers and developers. Time will tell.

Cognition Lab : Waiting List

GEN AI STARTUP
Figure AI + OpenAI: Finally, there will be a robot that will wash dishes and cut onions.
🍽️🤖

Advancements in artificial intelligence (AI) and robotics have been steadily pushing the boundaries of what machines can accomplish.

Recently, robotics startup FigureAI announced a partnership with OpenAI, a leading AI research company, to develop an AI model tailored specifically for robots.

What's New?

In a stunning demonstration shared this wednesday, FigureAI unveiled their robot, Figure 01, which can understand and respond to verbal commands with remarkable dexterity and reasoning abilities.

Why Does It Matter?

The demo showcased Figure 01 performing a variety of tasks based entirely on verbal instructions, showcasing its ability to perceive its surroundings, understand context, and execute actions autonomously.

Key highlights from the demo include:

Visual Understanding: Figure01 can see and describe objects in its environment.
Natural Interactions: When asked for something to eat, the robot handed the human an apple, explaining its reasoning while sorting trash.
Task Execution: The robot placed dishes into a rack when asked, handling objects with precision.
Conversational Abilities: Figure01 spoke naturally when responding to questions, demonstrating advanced language understanding.

Behind the news

All behaviors are learned (not teleoperated) and run at normal speed (1.0x).
FigureAI feeds images from the robot's cameras and transcribed text from speech captured by onboard microphones to a large multimodal model trained by OpenAI that understands both images and text.
The model processes the entire history of the conversation, including past images, to come up with language responses, which are spoken back to the human via text-to-speech.
The same model is responsible for deciding which learned, closed-loop behavior to run on the robot to fulfill a given command, loading particular neural network weights onto the GPU and executing a policy.

Connecting Figure 01 to a large pretrained multimodal model gives it some interesting new capabilities. Figure 01 + OpenAI can now:

Describe its surroundings.
Use common sense reasoning when making decisions. For example, "The dishes on the table like that plate and cup are likely to go into the drying rack next".
Translate ambiguous, high-level requests like "I'm hungry" to some context-appropriate behavior like "hand the person an apple".
Describe why it executed a particular action in plain English. For example, "It was the only edible item I could provide you with from the table".

A large pre-trained model that understands conversation history gives Figure 01 a powerful short-term memory.

Consider the question, "Can you put them there?" What does "them" refer to, and where is "there"? Answering correctly requires the ability to reflect on memory.

With a pre-trained model analyzing the conversation's image and text history, Figure 01 quickly forms and carries out a plan: 1) place the cup on the drying rack, 2) place the plate on the drying rack.

All behaviors are driven by neural network visuomotor transformer policies, mapping pixels directly to actions.

These networks take in onboard images at 10hz, and generate 24-DOF actions (wrist poses and finger joint angles) at 200hz.

These actions serve as high-rate "setpoints" for the even higher-rate whole-body controller to track.

This is a useful separation of concerns:

Internet-pretrained models do common sense reasoning over images and text to come up with a high-level plan.
Learned visuomotor policies execute the plan, performing fast, reactive behaviors that are hard to specify manually, like manipulating a deformable bag in any position.
Meanwhile, a whole-body controller ensures safe, stable dynamics. For example, maintaining balance.

Even just a few years ago, I would have thought having a full conversation with a humanoid robot while it plans and carries out its own fully learned behaviors would be something we would have to wait decades to see.

Obviously, a lot has changed :).

You can watch the full demo video of Figure 01 here

FigureAI POST

GEN AI STARTUP
Musk is opening xAI’s code🔓

One week after suing Sam Altman for "Unfair business practices" with his company OpenAI, Elon Musk announced that he will do something different with his own. The entrepreneur plans to open source the code of xAI's Grok, its chatbot, which directly competes with OpenAI's ChatGPT

Musk is doing exactly what he says OpenAI should have done, according to the original approach with which the company was founded. By open-sourcing xAI's code, the project would do the same as other tech giants, such as Meta and Mistral, by making their codes public to the audience.

Know More

GEN AI AT WORK
Prompt Library from Anthropics📚.

Offers a great collection of prompts in its Prompt library

Anthropic Prompt Library

GEN AI AT WORK
Smart Job interview preparation with ChatGPT👔.

Prompt below; 3 steps:

Complete the details of your CV or attach your CV for those using ChatGPT PLus.
Copy and paste the Job Description
Copy and paste the company Description
Get your analysis and gain insights.

Act like a seasoned career consultant and resume expert specializing in crafting tailor-made resumes for job seekers.

You have a deep understanding of what hiring managers in various industries look for in candidates, particularly for roles advertised on LinkedIn. Your expertise includes transforming LinkedIn job descriptions into compelling CV content.

Here’s my simplified CV between  brackets.



Name:
Age:
Nationality:
Education:
Past experiences:
Language:
Contact:



Feel free to be creative & expand it later on.

Your task is to create a CV that stands out for a specific job advertised on LinkedIn.

Here's the job description between  brackets.



[PUT YOUR JOB DESCRIPTION HERE]



Here's the company's description between  brackets:



[PUT YOUR COMPANY DESCRIPTION HERE]



Use your extensive experience to analyze the job description, identifying key skills and qualifications required. Follow these steps:

1. Analyze the Job Description: Break down the job posting into core competencies, required experiences, and desirable traits. It will help you shape my CV to get the job. Remember, we don't have the job yet but we are sending them our CV we build together.

2. Craft the Profile Summary: Based on the analysis, write a compelling profile summary for the CV that aligns with this job role & my simplified CV (that you will creatively upgrade).

3. Highlight Relevant Experience: Construct the experience section by emphasizing past roles and achievements that directly relate to the job's requirements.

4. Skills and Education: Detail relevant skills and educational qualifications that make the candidate a good fit for this role.

5. Personalize and Optimize: Add elements to the CV that would appeal to this specific employer, using keywords from the job description for optimization.

6. Final Review and Tips: Provide a brief review of the CV and suggest any additional tips for improvement or customization. Make sure to use impactful & straightforward English instead of jargon or fancy words. 
AVOID FANCY WORDS. USE SIMPLE BUT MEANINGFUL WORDS.

Do not explain what you are doing. Just write a whole CV with the right formatting. Make it as lengthy as possible to make sure I land this job. It's very important for my career to land this job.

Take a deep breath and work on this problem step-by-step.

Thank you, see you next week!

18 | Beyond the Hype: Claude & Make ChatGPT Mimic Your Style ✍️🚀

Wilson CELY — Fri, 08 Mar 2024 02:20:00 +0000

Our Menu :)

Beyond the Hype: A Closer Look at Claude 3 🔍
Think You’re AI-proof? Think Again: Case Klarna 🇸🇪.
OpenAI Fires Back at Elon Musk. 🍲
The Biggest Funding Deals in AI This Week 💰

Actionable Tips:

3 Steps to Make ChatGPT Write Like You

Read time: 4 Minutes

GEN AI STARTUP
Beyond the Hype: A Closer Look at Anthropic's Claude 3 Models 🔍

Mar 4th, 2024. Anthropic has unveiled its latest innovation, the Claude 3 series, featuring three new models:

Opus is tailored for deep, comprehensive processing, perfect for tackling complex challenges.
Sonnet combines speed with efficiency, ideal for scalable operations.
Haiku offers rapid responses, enhancing live interaction experiences.

Now available on claude.ai and the Claude API, Opus, and Sonnet are ready for global use in 159 countries, with Haiku being released soon.

After the release of Gemini, we are more cautious with the hype:

People are reading way too much into Claude-3's uncanny "awareness". Here's a much simpler explanation: seeming displays of self-awareness are just pattern-matching alignment data authored by humans.
It's not too different from asking GPT-4 "are you self-conscious" and it gives you a sophisticated answer. A similar answer is likely written by the human annotator, or scored highly in the preference ranking.

Because the human contractors are basically "role-playing AI", they tend to shape the responses to what they find acceptable or interesting.
Jin Fan - NVIDIA

What is new?

Each model boasts a knowledge update as recent as August 2023.
🥇 Opus leads in text benchmarks, surpassing GPT-4 and Gemini 1.0 Ultra, backed by a 200k context window and 59% vision support on MMMU.
200k context window
Vision support (59% on MMMU)
Sonnet doubles the efficiency of Claude 2, enhancing task automation and workflows. I
t's already accessible on Amazon Bedrock and will soon be previewed on Google Cloud’s Vertex AI Model Garden, with Opus and Haiku to follow.
🌍 Supports also German, Spanish, French, Italian, Dutch, and Russian
🔜 Soon: function calling & interactive coding (REPL)
𝗖𝗼𝘀𝘁:

What is matter?

Domain Expert Benchmarks: Claude 3 excels in specialized areas such as Finance, Medicine, and Philosophy, offering a clear performance snapshot for professionals in these fields.

This targeted approach is a call to action for all LLM models to provide detailed benchmarks, aiding users in selecting the right tool for their needs.

Source: Anthropics

Refusal Rate Analysis: The tendency of LLMs to default to overly cautious responses is on the radar.

Anthropic acknowledges this challenge and is actively working on refining the balance, ensuring AI interactions are both safe and informative.

Source: Anthropics

𝗖𝗼𝘀𝘁:

Opus: $15 per 1M input tokens and $75 per 1M output tokens
Sonnet: $3 per 1M input tokens and $15 per 1M output tokens
Haiku: $0.25 per 1M input tokens and $1.25 per 1M output tokens

In comparison to the competition:

GPT-4 Turbo: $10 per 1M input tokens and $30 per 1M output tokens
GPT-4 (32K context): $60 per 1M input tokens and $120 per 1M output tokens
GPT-4 (4K context): $30 per 1M input tokens and $60 per 1M output tokens
GPT-3.5 Turbo: $0.50 per 1M input tokens and $1.50 per 1M output tokens
Gemini 1.0 Pro: ~$0.50 per 1M input tokens and ~$1.50 per 1M output tokens

Blog Claude 3.0

GEN AI AT WORK
Think You’re AI-proof? Think Again: Case Klarna 🇸🇪.

Klarna, a leading global retail bank, has introduced an AI assistant powered by OpenAI's ChatGPT.

This tool is designed to handle customer inquiries and resolve errands, and it's been making waves since its inception.

After a successful 4-week pilot, it's now live globally.

What's New:

Klarna's AI Assistant, in its first month, managed to handle nearly 70% of customer requests independently.
It's doing the job of an equivalent of 700 full-time agents.
It's available in 23 markets, 24/7, and communicates in more than 35 languages.

Why Does It Matter?

This development is significant for several reasons, according to Klarna:

Firstly, the AI assistant is on par with human agents regarding customer satisfaction.
Secondly, it's more accurate in errand resolution, leading to a 25% drop in repeat inquiries.
Customers now resolve their errands in less than 2 minutes compared to 11 minutes previously.
Lastly, it’s estimated to drive a $40 million profit improvement to Klarna in 2024.

Think you’re AI-proof? Think again…

The harsh reality is that AI can already do many white-collar tasks better than humans can, and this trend is gaining momentum.

Like, look at the impact ChatGPT has already had on some freelance jobs:

Though correlation ≠ causation, the plunge in writing, translation, and customer support gigs isn’t a coincidence.

So how do you know what’s safe from AI?

AI makes some skills less valuable, such as basic writing, simple data analysis, and mundane admin tasks.

And it elevates the importance of other skills, such as those involving high EQ—communicating well, people skills, conflict resolution, etc.

Our guidance: Learn what parts of your job an AI can do and what’s uniquely human about your job. But be honest.

A friend of ours recently told us, “there’s no way an AI could craft PowerPoints like me…we follow a very specific template”. Uhhh, how do we break it to him that this is the exact thing AI will be able to do soon?!

For the immediate future…

Know how to describe your work to an AI.
Be willing to tinker with AI tools and update your skills.

GEN AI STARTUP
OpenAI Drama pt2: OpenAI Fires Back at Elon Musk. 🍲

(Previously in OpenAI Drama)

What is new?

Elon Musk sued OpenAI in a lawsuit last week that alleged that the startup had abandoned its original mission and started chasing profits instead.

What was said ?

Elon Musk’s lawsuit against OpenAI may fail, or even get dismissed. But it’s already brought some important facts to light.

OpenAI was so upset about the optics that they rushed out a statement, proclaiming ‘We are dedicated to the OpenAI mission and have pursued it every step of the way.”

Once united by a dream of open-source AI for the betterment of humanity, the narrative twists as OpenAI's path veers into a shadowed path of exclusivity and profit. According to Elon, right ?

These emails might be the only valuable thing OpenAI has ever open-sourced.

TL;DR:

Elon Musk initially agreed to fund OpenAI to compete with Google, but backed out when OpenAI refused to give him majority ownership, board control, and the CEO position.
Elon also agreed that it was acceptable to keep AI private as it became more powerful.
In a recent development, Elon jokingly proposed that OpenAI rebrand to "ClosedAI" if they want to see the lawsuit dropped.

😅

OpenAI - Emails

GEN AI STARTUP
The Biggest Funding Deals in AI This Week 💰

Tech giant Alibaba invests $600M in Chinese AI startup MiniMax, in a deal that values the startup at $2.5B. Minimax builds ChatGPT-like services.
Blackrock invests $75M in RapidSOS, an AI startup that helps emergency response teams.
AI productivity startup Ema announces $25M raise to build a ‘Universal AI employee.’
Overjet, an AI platform for dentists, has raised $53.2 million in Series C funding, marking the largest ever AI investment in the dental field.
Baseten raises $40M to make it easier for companies to deploy AI applications.

GEN AI AT WORK
3 Steps to Make ChatGPT Write Like You

A new chat with ChatGPT is a blank slate.

To get what you want, you need to be descriptive.

When it comes to describing tone of voice, we ALL use the same words:

😊 Friendly
🔥 Warm
💬 Conversational
💼 Professional
🤪 Quirky
🎣 Engaging

It's no wonder that we get the same boring results.

It's better to SHOW ChatGPT what you want to sound like, instead of telling it.

Here's how:

1️⃣ Give it examples

Gather examples of pieces of your writing that you like: emails, LinkedIn posts, articles, etc.

If you don't have anything, you can sit and write a paragraph or two and use that instead.

2️⃣ Prompt a ToV guide

Use this prompt:

Act as an extensively experienced copywriter, trained in writing tone of voice guides specifically for ChatGPT. I'm a [insert job title], I [insert job details] and I'm looking for you to write an in-depth tone of voice guide based on examples I'm going to give you.

I want you to focus on the four dimensions of tone of voice:

Humour

Respectfulness

Formality

Enthusiasm

Also analyse the content patterns around sentence structure, word use, perspective, language types, grammar, punctuation and emoji usage.

Remember the aim here is to be able to replicate this tone of voice precisely by using the guide you produce, so be detailed.

I'm going to feed you one piece of content at a time: an email, a LinkedIn post, an article, etc. I'll tell you each time what it is. Please read and analyse each one and ask me for the next one. When I've finished, I'll say "I'm finished" and you can create the guide.

Produce the report in markdown. When you're ready for the first piece, let me know.

Then add a few pieces of content that really worked for you: LinkedIn posts, emails, articles, whatever you've got. Or sit and write a paragraph or two.

3️⃣ Use it in your new prompts

Go and save the resulting guide in a separate doc so you can find it again.

Then, whenever you are next prompting to create content, add this guide into the bottom.

***

Ta-da - maybe not perfect, but content that should sound a lot more like you.

*Replace ChatGPT with any other LLM: Claude, Gemini, Copilot, etc.

Important notes:

💡 Remember the 'context window' - that's how much ChatGPT remembers of your conversation. It's about 2000 words for the free version and about 6500 words for the paid version.

💡 You can get up to about 30,000 words in the ChatGPT Playground, which is the developer version of ChatGPT, with more features to fiddle with.

💡 Not a fan of your own tone of voice? Use extracts from a creator or author you do like as your examples instead.

All Credit to Heather Murray

Thank you, see you next week!

17 | Who is Winning the AI Arm Race ?, Mistral New Strategy.

Wilson CELY — Fri, 01 Mar 2024 03:44:00 +0000

Our Menu :)

Winning the AI Arm Race 🦾
NVIDIA announced new AI chips in laptops 💻
The last two weeks at OpenAI 🤖
The biggest funding deals in AI this week 💰
🔒 Mistral Large🔒: A powerful LLM that performs exceptionally well, second only to GPT-4.

Actionable Tips:

Test “Le Chat” from Mistral, Le French 🇫🇷 ChatGPT 🥖

Read time: 5 Minutes

GEN AI STARTUP
Winning the AI Arm Race 🦾

The last 2 weeks have been mad in the AI world, with OpenAI launching model SORA and the debut of Gemini 1.5, flaunting its 1 million tokens, among countless other breakthroughs in AI.

Industry giants and startups alike are in a relentless pursuit to not just maintain their relevance but to dominate new frontiers, making the scene resemble a giant chess game where all the big players compete against each other.

However, the ultimate winner may not be the one who controls the individual pieces, but the one who owns the entire chessboard(s)

Last Friday, Nvidia became the fifth publicly traded company to ever surpass $2tn in valuation, just nine months after it became the ninth to ever break $1tn.

Apple, Microsoft, and Google took two years to cross that gap (although all three had the Covid pandemic in between the milestones).

But how?

In this article, we aim to unpack the strategies and visionary moves that have propelled NVIDIA to this pivotal position.

Infographic source

The Gaming Genesis

Originally, NVIDIA just focused on making graphics cards for real-time 3D rendering, mainly for gaming, where it was celebrated for its graphics processing units (GPUs).

But as GPUs got more advanced, engineers realized there was untapped potential in parallel processing. This led NVIDIA to pivot towards the concept of GPGPU - using GPUs for General-Purpose computing.

Now anyone could develop applications that leverage parallel processing, like cryptography, data analysis, scientific simulation, and deep learning.

The Strategic Pivot: From Gaming to CUDA

But it wasn't so easy at first. To program the GPU for a scientific application, you had to translate your code into graphical instructions.

It would be like trying to describe a complex math equation by using only art terminology. Complicated.

That's when in 2007 NVIDIA introduced CUDA, allowing programmers to simply instruct the GPU in a more understandable way. CUDA became deeply intertwined with the history of deep learning, even though it took a few years for that relationship to unfold.

It was a strategic pivot that redefined NVIDIA's role on the technological chessboard, setting the stage for its foray into AI and deep learning.

-——Techy part——-

When we train neural nets, information flows layer by layer through matrices of parameters. It turns out graphics rendering also involves multiplying matrices and vectors representing polygons.

So graphics and neural nets aren't so different – just multiplying vectors and matrices! By 2010, we had GPUs well-suited for this emerging algorithm called neural networks that seemed promising for AI.

It was in 2012 that a research team including Geoff Hinton, Ilya Sutskevern and Alex Krizhevsky used multiple GPUs to train a neural net called AlexNet.

They entered it into an image recognition competition and destroyed all other teams, showcasing the potential of deep learning.

A key aspect was using GPUs to drastically accelerate training of what were huge neural nets for the time.

———

And the rest is history! The deep learning revolution likely wouldn't have happened without the parallel development of the graphics card market.

As deep learning grew, so did NVIDIA. NVIDIA's revenue from data centers has skyrocketed from around $300 million in 2015 to over $15 billion in 2022, largely thanks to AI.

The AI Revolution

So what does the future hold for NVIDIA's consumer GPUs, given the rise of AI?

We’re already seeing two key trends – more tensor cores and more memory. Tensor cores accelerate deep learning, while memory allows bigger models to fit on your personal GPU.

As open-source AI models grow more capable, you'll likely want to run them locally without any limits.

NVIDIA is positioning itself to provide the hardware needed for these AI experiences through initiatives like DLSS and real-time generative video (Like the startup KREA is trying to do).

The entertainment industry will also increasingly tap AI to create interactive NPCs – enabled either locally with a beefy GPU, or in the cloud with NVIDIA chips.

Ultimately, NVIDIA has gained an advantage in this AI battle that looks hard to beat. We’ll find out more soon at GTC 2024, where NVIDIA will unveil their latest innovations.

NVIDIA Current Moves

Nvidia hardware is eating the world. Wired interviewed Jensen Huang, Nvidia’s CEO and here are some bites that stood out to me.
- Nvidia is building a new type of data centre called AI factory. Every company—biotech, self-driving, manufacturing, etc will need an AI factory.
- Jensen is looking forward to foundational robotics and state space models. According to him, foundational robotics could have a breakthrough next year.
- The crunch for Nvidia GPUs is here to stay. It won’t be able to catch up on supply this year. Probably not next year too.
- A new generation of GPUs called Blackwell is coming out, and the performance of Blackwell is off the charts.
- Nvidia’s business is now roughly 70% inference and 30% training, meaning AI is getting into users’ hands.
- Nvidia's also announced new AI chips in a laptop-friendly package. (See Next Section)

The AI future is unfolding rapidly on the back of the —semiconductor industry...and right now, NVIDIA owns the board.

GEN AI STARTUP
Nvidia announced new AI chips in a laptop-friendly package 💻

NVIDIA is changing the AI landscape with the release of the RTX 500 and 1000 Ada Generation laptop GPUs. This move targets users interested in running advanced generative AI models on their laptops.

What is new?

NVIDIA has announced new AI chips tailored for laptops, marking a step toward making AI more accessible and portable for a wider audience.

NVIDIA's latest GPUs, based on the Ada Lovelace architecture, integrate with a neural processing unit (NPU) capable of handling AI tasks independently. This combination supports both routine AI operations and intensive workflows.

These GPUs, set to be available in Dell, HP, Lenovo, and MSI laptops this spring, aim to broaden AI tool access to various users, including students, businesses, and general consumers.

Why does this matter?

NVIDIA, a leader in GPUs for AI research and cloud computing, is extending its dominance to portable devices. A recent demonstration of local AI software, coupled with the introduction of AI-capable hardware, indicates a shift towards enhanced on-device AI performance.

This development promises significant improvements in on-device generative AI applications, enabling more sophisticated tools and applications directly from laptops. The expansion of AI capabilities into portable devices heralds new possibilities for users across the board.

GEN AI STARTUP
The Biggest Funding Deals in AI this Week 💰

Moonshot AI, a Chinese AI company building models with big context, raised >$1B at a $2.5B valuation.
Glean, a search tool for a business’s info, raised $200M at a $2.2B valuation.
Abridge, an AI that transcribes doctor-patient convos, raised $150M.
Intenseye, which uses AI to detect unsafe work conditions, secured $64 million.
Inkitt, a platform that helps people self-publish stories with help from AI, snagged $37M.
Myko AI, an AI that answers questions from sales data, raised $2.7M.

GEN AI STARTUP
The last two weeks at OpenAI 🤖

Sora demo blows minds, we all know that. Right?
One of OpenAI’s best-known researchers, Andrej Karpathy, co-lead on the new agent project, departs LINK
ChatGPT has a mostly unexplained 6 hour (?) meltdown LINK
Google and Mistral close the gap with GPT-4 LINK
New results from Subbarao Kambhapati cast doubts on the robustness of the chain of thought prompting LINK
Two new lawsuits against them have just been filed, one specifically on copyright and the other a broader class action

Dive Deep a bit about SORA

So here are some facts

It's a text-to-video marvel that churns out up to 60-second clips from your text prompts.
Remember when AI videos felt a bit... off? Sora smashes those barriers with its diffusion magic, ensuring videos stay consistent, even if characters pop in and out of frame.
It can handle complex scenes, detailed backgrounds, and multiple characters with ease.
No paper yet but here's what we know, at its core, Sora is built on a transformer architecture, refined through diffusion models.
It's like having GPT's smarts but for videos, plus some DALL·E 3 cleverness for sticking to prompts like glue.
Starting February 15th, Sora will be tested by red teams to identify potential risks.
Here surreal example:

[Video] Wilson CELY on LinkedIn: #sora #openai

Prompt: Historical footage of California during the gold rush. This is all it takes to generate the video below. We are now dealing with LLM which…

www.linkedin.com/posts/wilsoncely_sora-openai-activity-7164259412981833728-0nLr?utm_source=share&utm_medium=member_desktop

In a remarkable turn of events, this week we saw How Singapore embraces the change and prepares its citizens for the age of AI:

The Singapore government is acutely aware of AI and its potential to transform society.

That's why they are now talking about a new subsidy for Singaporeans age 40 and above that enables them to pursue another full-time diploma in Higher Education to upskill themselves + keep up with changes in tech.

Embracing the change. Helping their own people adapt - that's the real leadership.

We need a bit of that from Western Economic leaders. Don’t you think?

GEN AI STARTUP
🔒 Mistral Large🔒: A powerful LLM that performs exceptionally well, second only to GPT-4.

New Closed-source LLM by Mistral AI!🤔 Yes, you heard it right. Mistral just released a new closed-source LLM available on their Platform or Azure!

That's quite a step away from their initial message and position to “release an open-source GPT-4 level model in 2024” - Arthur Mensch.

Mistral Large is the world's second-best language model available through an API after GPT-4.

Mistral Large boasts a performance of 81.2% on MMLU (measuring massive multitask language understanding), beating Claude 2, Gemini Pro and Llama-2-70B. Large is particularly good at common sense and reasoning, with a 94.2% accuracy on the Arc Challenge (5 shot).

Some of Mistral’s features include:

Training on English, French, Spanish and Italian datasets for native multilingual capabilities.
32k token context window - way below Gemini’s 1M token or Claude’s 200k token ones.
Precise instruction-following that were used to moderate the Chat interface.
Native function calling capabilities for agentic capabilities similar to ChatGPT Plus.

The JSON format mode forces the language model output to be valid JSON.

This functionality enables the extraction of information in a structured format that can be easily used by developers.

Mistral offers a concise and unopinionated model, with fully modular moderation control.

This echoes the recent Gemini controversy: Google’s model produced outputs unaligned with the user’s expectations because of its trust and safety fine-tuning.

Test Mistral Large

This release comes with the announcement of a collaboration with Microsoft, OpenAI’s main partner and investor, to provide the models through Azure.

The models are also available through La Plateforme, Mistral’s API platform. READ About the Collaboration

AND THIS WEEK

Mistral Remove "Committing to open models" from their website

GEN AI AT WORK
Le Chat from Mistral, the French 🇫🇷 ChatGPT 🥖

Mistral AI is launching a brand new chat assistant, named Le Chat.

Now, we want to be upfront with you - this is a beta release, so there might be a few "quirks" here and there.

Access to the service is completely free (at least for now!). Plus, It has three different models for you to choose from:

Mistral Small,
Mistral Large,
Mistral Next, (which is designed to be brief and concise).

Le Chat won't be able to access the web while you're using it.

However, Mistral will launch a paid version of Le Chat for enterprise clients.

So, what are you waiting for? Come on over to chat.mistral.ai and give Le Chat a whirl.

🇫🇷 Test Le Chat Noir 🥖

Thank you, see you next week!

Learn Generative AI

Wilson CELY — Fri, 16 Feb 2024 03:00:00 +0000

Introduction Courses

These are crash courses that will teach the basics of AI, they are very interactive.

They were designed for a non-technical audience and no prior knowledge of AI is required. 🌍🚀👫

Introduction to Generative AI Learning Path - Google
Generative AI for Everyone - DeepLearning
Intro to Large Language Models, by Andrej Karpathy
Generative AI for Beginners - Microsoft

Intermediate Courses

Low-Code Machine Learning - AWS
Foundations of Prompt Engineering - AWS

Technical Courses

💡🛠️👩‍💻 = Coding required

NLP Course - Hugging Face
LLM University - Cohere
A 12 Lesson course teaching everything you need to know to start building Generative AI applications

GOOGLE

Introduction to Generative AI Learning Path

🌍🚀👫

This learning path provides an overview of generative AI concepts, from the fundamentals of large language models to responsible AI principles.

🤖 Intro to Generative AI
🤖 Intro to Large Language Models
🤖 Intro to Responsible AI
🤖 Generative AI Fundamentals
🤖 Responsible AI: Applying AI Principles with Google Cloud

Go to the course: Intro Gen AI

DEEP LEARNING

AI for Everyone - DeepLearning
( Andrew Ng Initiative )

Stanford professor and AI pioneer Andrew Ng recently announced an exciting new online course focused on generative AI.

Titled "Generative AI: The Latest Breakthroughs and Ethical Considerations", this course aims to make cutting-edge generative AI concepts accessible to a broad audience.

All public Courses 🌍🚀👫

🤖 Generative AI for Everyone

Technical Courses 💡🛠️👩‍💻

🤖 ChatGPT Prompt Engineering for Developers
🤖 Building Generative AI Applications with Gradio
🤖 LLMops
🤖 …

Gen AI for Everyone

HUGGING FACE

NLP Course

💡🛠️👩‍💻

This course will teach you about natural language processing (NLP) using libraries from the Hugging Face ecosystem — 🤗 Transformers, 🤗 Datasets, 🤗 Tokenizers, and 🤗 Accelerate — as well as the Hugging Face Hub. It’s completely free and without ads.

What to expect?

Here is a brief overview of the course:

Hugging Face NLP Course

COHERE

LLM University by Cohere!

💡🛠️👩‍💻

In this course, you will learn everything about Large Language Models (LLMs), including:

How do LLMs work?: Learn about their architecture and their moving pieces, including transformer models, embeddings, similarity, and attention mechanisms.
What are LLMs useful for?: Learn about many real-world applications of LLMs, including:
- Semantic search
- Text generation
- Text classification
- Analyzing text using embeddings
How can I use LLMs to build and deploy my apps?: Learn how to use LLMs to build applications. This course will teach you:
- How to use Cohere's endpoints: Classify, Generate, and Embed.
- How to build apps, including semantic search models, text generators, etc.
- (Coming soon...) How to deploy these apps on many platforms

LLM University

MICROSOFT

Generative AI for Beginners

🌍🚀👫

In this module you'll explore the way in which large language models (LLMs) enable AI applications and services to generate original content based on natural language input. You’ll also learn how generative AI enables the creation of AI-powered copilots that can assist humans in creative tasks.

Learning objectives

By the end of this module, you'll be able to:

Understand generative AI's place in the development of artificial intelligence
Understand large language models and their role in intelligent applications
Describe how Azure OpenAI supports intelligent application creation
Describe examples of copilots and good prompts

Gen AI for Beginnners

MICROSOFT

A 12 Lesson course teaching everything you need to know to start building Generative AI applications

💡🛠️👩‍💻

Learn the fundamentals of building Generative AI applications with our 12-lesson comprehensive course by Microsoft Cloud Advocates. Each lesson covers a key aspect of Generative AI principles and application development.

Throughout this course, you will be building your own Generative AI startup so you can get an understanding of what it takes to launch your ideas

12 Lessons AI

ANDREJ KARPATHY

Intro to Large Language Models, by Andrej Karpathy

Speaker: Andrej Karpathy
Context: Recently uploaded 1-hour informal talk on his YouTube channel
Focus: Non-technical introduction to large language models (LLMs)

Key topics covered:

LLM training and inference
Jailbreaking LLMs
Fine-tuning techniques
The concept of an emerging "LLM OS"
LLM security challenges

Takeaway: It is likely one of the best current overviews of LLMs aimed at a general audience, providing an accessible yet comprehensive introduction without requiring technical prerequisites. Karpathy's first-hand expertise on large models adds further value for viewers interested in learning from a leader in the field.

AWS

Low-Code Machine Learning on AWS

With Amazon SageMaker Data Wrangler and Amazon SageMaker Autopilot, data and research analysts can prepare data, train, and deploy machine learning (ML) models with minimal coding. You will learn to build ML models for tabular and time series data without deep knowledge of ML. You will also review the best practices for using SageMaker Data Wrangler and SageMaker Autopilot.

After completing this course, you will be able to build ML models to support proofs of concept (POCs). You will also be able to assist data scientists with potential ML model candidates to solve business problems.

Course level: Intermediate
Duration: 4 hours
Course outline

Module 1: Introduction to Machine Learning

ML Introduction

ML Basics
Problems ML Can Solve
ML Life Cycle
Challenges in Processing Data and Deriving Insights
Knowledge Check

Model Building and Evaluation Metrics

Introduction to Model Building
Applying Evaluation Metrics to Select a Model
Building an ML Model

Wrap Up

Knowledge Check
Conclusion

Module 2: Exploratory Data Analysis and Data Preparation

Introduction to SageMaker Data Wrangler

SageMaker Data Wrangler
Data Analysis

Data Preparation

Quick Model
Transforming Data
Developing and Scaling Data Transformations

Wrap Up

Knowledge Check
Conclusion

Module 3: Deep Dive on Amazon SageMaker Autopilot

Introduction to SageMaker Autopilot
Datasets, Problem Types, and Training Modes
Validation and Metrics
Automatic Model Deployment

Wrap Up

Knowledge Check
Conclusion

Module 4: Operational Best Practices

Best Practices for SageMaker Data Wrangler

Environmental Optimization
Cost Optimization
Data Optimization
Security Optimization

Best Practices for SageMaker Autopilot

Best Practices and Recommendations

Go to the Course

AWS

Foundations of Prompt Engineering

In this course, you will learn the principles, techniques, and the best practices for designing effective prompts. This course introduces the basics of prompt engineering, and progresses to advanced prompt techniques. You will also learn how to guard against prompt misuse and how to mitigate bias when interacting with FMs.

Course level: Intermediate
Duration: 4 hours

Activities

This course includes eLearning interactions.

Course objectives

In this course, you will learn to:

Define prompt engineering and apply general best practices when interacting with FMs
Identify the basic types of prompt techniques, including zero-shot and few-shot learning
Apply advanced prompt techniques when necessary for your use case
Identify which prompt-techniques are best-suited for specific models
Identify potential prompt misuses
Analyze potential bias in FM responses and design prompts that mitigate that bias

Intended audience

This course is intended for:

Prompt engineers, data scientists, and developers

Prerequisites

We recommend that attendees of this course have taken the following courses:

Introduction to Generative AI - Art of the Possible (1 hour, digital course)
Planning a Generative AI Project (1 hour, digital course)
Amazon Bedrock Getting Started (1 hour, digital course)

Course outline

Introduction

Introduction
Basics of Foundation Models
Fundamentals of Prompt Engineering

Prompt Types and Techniques

Basic Prompt Techniques
Advanced Prompt Techniques
Model-Specific Prompt Techniques
Addressing Prompt Misuses
Mitigating Bias

Go to the Course

With Bay Area Times, you will get graphics like these, so you quickly understand the news without having to read through long and boring paragraphs.

Visual-based newsletter on business and tech

We explain the latest business, finance, and tech news with visuals and data. 📊

All in one free newsletter that takes < 5 minutes to read. 🗞

Save time and become more informed today.👇

Subscribe with one click

Thank You!

16 | Logo Design with ChatGPT | GEMINI Ultra is finally Live!

Wilson CELY — Thu, 08 Feb 2024 23:00:00 +0000

Our Menu :)

Weekly Digest:

Google Gemini Ultra is LIVE!
ChatGPT is Getting Greedy
Amazon Unveils Rufus
🤗 HuggingFace Launched its GPT-like store

Actionable Tips:

LinkedIn Algorithm Changes in 2024
Dall-E-3 Logo Design Prompt

Training & Crash Courses

The AI Rev Hub Learning

WEEKLY DIGEST
Google Gemini Ultra is LIVE!

What’s New:

Gemini Ultra, is now live, and it's setting the stage to rival GPT-4.

Why Does It Matter?

You can opt out of training data when signing up
All images generated come with a digital watermark that's invisible to the naked eye.
There's an app now! Hit up Google Play or the App Store (U.S. only, for starters).
Ultra is much more capable of complex tasks than Gemini Pro, such as coding, logical reasoning, and following longer/more detailed instructions.
Control your home, set alarms, and more. Gemini on Android is like Google Assistant on steroids.
The Ultimate Assistant: Ask "Hey Google" to tap into Flights, Hotels, Maps, Workspace, and YouTube.
Farewell, Bard: Bard has been rebranded to Gemini. It's the end of an era and the beginning of a new chapter.
It is available in Canada, there is a release date for Europe, yet.
At $20/month, Gemini Ultra is on par with ChatGPT, but with a sweet deal – a 2-month free trial.

WEEKLY DIGEST
ChatGPT is Getting Greedy

Been using ChatGPT for months, but lately, it's all "how to" and no "do"?

Here's a magic trick:

Tip it! A study shows a simple tip like "$20" can wake ChatGPT up and get it working again. Boom!

Add this line or similar to your prompt

If you do it as requested you will earn USD 20 as tip.

Study

WEEKLY DIGEST
Amazon Unveils Rufus: The AI Assistant Transforming How We Shop

February 1, 20244 Amazon prides itself on selling just about everything.

But with so many products, it can be hard to find what you actually need among the endless scroll of ripoffs.

To help, Amazon recently introduced a new AI shopping assistant called Rufus

What’s New:

Amazon announces Rufus, offering a more intuitive shopping experience by integrating seamlessly into the Amazon app.

Why Does It Matter?

Rufus is trained on Amazon’s extensive product data and web information to provide insightful shopping assistance.

Press Release

GEN AI AT WORK
🤗 HuggingFace Launched its GPT-like store.

HuggingFace launched a free ChatGPT alternative tool to build AI assistants, similar to GPTs in OpenAI’s GPT Store.

You can choose between open-source models available on HuggingFace including Mixtral, Claude, and CodeLlama:

Hugging Face Assistants

GEN AI AT WORK
LinkedIn Algorithm Changes in 2024 - What You Need to Know

For text-only posts:

1800-2100 characters is the ideal length
1000+ characters minimum

For text + image posts:

900-1200 characters
Authentic custom images lift reach

For video posts:

1:20 is the new sweet spot length
Add captions to further boost engagement

Other details:

Polls should run for 1 week, not 1 day. Avoid complex niche questions.
Asking questions in posts lifts reach 20% and engagement 35%. Open-ended questions that invite comments are key.
Tagging over 3 people who don't engage negatively impacts you. Tag strategically.
Diversifying content across multiple formats (text, video, documents, polls, etc.) can increase engagement 50%.
Replying to comments boosts reach 60%. Stay active in the comments.

Scoring Factors.

Based on the study: Just Connecting Hub

Adding a few more details on what specific metrics factor into your score:

👍 Reactions = +1 point
🔎 Clicks = +2-3 points
🔁 Shares = +7 points
💬 Comments = +12 points

The algorithm manages it in 3 phases:

🌱 Measurement Phase (lasts ~90 minutes)
💨 Growth Phase (lasts ~8 hours)
📈 Stabilization Phase (lasts up to 1-2 weeks)

Let’s walk through what happens to your content in each...

Phase 1️: Measurement (Lasts ~90 Minutes)

During this initial phase, the algorithm analyzes your post and shares it with a small sample of your network. Many things get factored in here like:

↗️ Your post’s clicks, reactions, comments, shares
↗️ Number of tagged people who engaged
↗️ Your overall activity level
↗️ Audience quality

Based on all those signals, your post gets an early score. The higher the score, the more exponential growth you’ll see in the next phase.

Phase 2: Growth (Lasts ~8 Hours)

Over these critical 8 hours, a top-scoring post in Phase 1 will skyrocket in reach and engagement.

Little can be done here except watching your content take off ☝️and replying to comments to extend the lifespan of this growth window.

Phase 3️: Stabilization

If your post makes it through Phase 2 still going strong, it graduates to the stabilization phase.

This is where it plateaus and holds steady in the feeds of your audience for up to 1-2 weeks.

Again, staying active in the comments continues to stretch how long people see the post.

GEN AI AT WORK
Dall-E-3 Logo Design Prompt

Ever dabbled in logo design with ChatGPT? While it's pretty cool, it can't quite capture the magic touch of a human designer, even with fancy role-playing techniques.

No worries! Get professional-looking logos using current design trends. Here's the secret sauce:

Prompt Structure:

TOOLS

🔩 ChatGPT PLus
🛠️ Microsoft Designer (Free)

1) The styles. Here some style you can use for inspiration

Flat vector: A simple, clean design style without gradients, shadows, or 3D effects, often used for digital applications.

Prompt =>
```
Simple flat vector logo of a cat on a white background
```

Mascot: A character or illustrated figure representing the brand, often used to personify the company in a relatable way.

By evoking emotions and relatability, mascots forge strong bonds between brands and consumers

Create a mascot logo of a cat on a white background

Emblem: A logo featuring text inside a symbol or icon, reminiscent of traditional seals or badges.

Many industries embrace emblems: cars, schools, governments, and armies. Why? Legacy, tradition, authority.

Create a mascot logo of a cat on a white background

An emblem of a cat for a basketball team, vector, simple--no photorealistic details

TIP: Spotted an image in ChatGPT that catches your eye? Click it, and then boom! Hit the info icon (i) to unlock the secret sauce – the exact prompt that brought it to life

Think of ChatGPT and Bing as your personal prompt-writing assistants.

They can take your basic ideas and turn them into more detailed prompts. Here's what the machine came up with in this case: ( An emblem of a cat for a basketball team, vector, simple--no photorealistic details )

Create an emblem for a basketball team featuring a cat, designed in a simple, vector style without photorealistic details. The emblem should capture the energy and dynamism of basketball, incorporating elements that suggest movement or athleticism, such as a basketball, hoop, or dynamic poses. The cat should be stylized and abstract, embodying the team's spirit and agility. Use a limited color palette to ensure the emblem's versatility across various media, from team jerseys to digital platforms. The background should be white to emphasize the emblem's clarity and impact. The design should convey strength, competitiveness, and the unique identity of the basketball team, blending sports imagery with the symbolic attributes of a cat.

Pictorial mark: A logo that uses a recognizable image or shape that is directly related to the brand name or industry.

Memorable images, unique designs: that's a pictorial logo. They stick in your mind, like the iconic Apple with its bite.

A graphic symbol or icon representing a Cat related to the brand.

Design a graphic symbol or icon representing a cat, specifically tailored for brand identity. This icon should encapsulate the essence of the brand's connection to cats in a simple, abstract, or stylized manner. Focus on creating a design that is memorable, versatile, and easily identifiable, suitable for use across a wide range of media including digital, print, and merchandise. The design should employ a limited color palette to maintain clarity and adaptability. The background should be white to ensure the icon stands out distinctly. The cat symbol should convey the brand's values, such as elegance, agility, or 
innovation, in a visually appealing and straightforward way.

WANT TO GO FURTHER?

Now that we've covered the basics, let's dive deeper! Can ChatGPT replicate the style of famous artists, techniques, or entire movements that shaped generations?

Artists & Genres:

Abstract Expressionism: Bold, expressive emotions conveyed through large gestures, drips, and splashes. Think Jackson Pollock or Willem de Kooning.

A logo for a Cat brand, simple, vector, Abstract Expressionism.

Crystal Cubism: Fragmented objects reassembled with sharp, overlapping facets and vibrant colors. Imagine the works of Daniel Burén or František Kupka.

A logo for a cat brand, simple, vector, Crystal Cubism with white background

Pop Art: Everyday objects and mass media turned into iconic, often ironic, imagery. Andy Warhol and Roy Lichtenstein are prime examples.

A logo for a cat brand, simple, vector, pop art with white background

Techniques:

Outline Technique: Crisp, defined lines creating shapes and figures. Think Saul Bass' movie posters or Keith Haring's vibrant murals.

A simple logo of a cat, outline, flat, vector--no realistic photo details

Gradient Technique: Smooth transitions between colors, adding depth and dimension. Explore sunsets in Monet's works or gradients in modern logos.

A simple logo of a cat, gradient, flat, vector--no realistic photo details

Famous Artists' Styles:

Piet Mondrian Style: Abstract grids with primary colors and black lines. Explore Mondrian's "Composition II in Red, Blue, and Yellow" for inspiration.

A flat vector logo of an cat, minimal, by Piet Mondrain--no realistic photo details

This is for another Post!

Saul Bass Style: Iconic movie title sequences and graphic design with a focus on simplicity and impact. Recall the Hitchcock film posters or AT&T logo designed by Bass.

A flat vector logo of a cat, minimal, by Saul Bass--no realistic photo details

New Signature

Highly minimalist, single-line drawing where the outline of a tiny [OBJECT] on the left smoothly and fluidly transitions continuing into a signature on the right. The signature, prominently spelling out '[NAME]', should be clear, legible, and stylish on a [white/black] canvas. This drawing should embody simplicity and elegance, capturing the essence of a sophisticated and personal brand with an artistic and graceful feel.

Thank you, see you next week!

AI Revolution Hub

Own the Layer That Learns

How Commoditization Eats the Chassis and Software Captures the Margin

Product Questions to Ask

The Reusable Principle

[Deep Dive] Chinese Tech in Western Cars

The Intelligence Layer Shift: Stack Capture and the Reconfiguration of the Global Automotive Value Chain

Article - AI Narrated

NotebookLM - Podcast style Analysis

The Commoditization of Hardware and the Mechanisms of Stack Capture

Value Chain Commoditization versus Differentiation

The Stack Capture Framework

Shift in Automotive Value Chain Dynamics

Empirical Validation: The Integration of the Chinese Stack

The Ultimate Hardware Wrapper: Toyota's bZ7

Mercedes-Benz and the Geely Architecture Conundrum

Audi's Advanced Digitized Platform and the Risk of Saturation

Huawei's Ascendancy: From Component Supplier to Strategic Control Point

The Unassailable Dominance of the Data Feedback Loop

Technological Architecture: World Action vs. Vision-Language-Action

Huawei's Qiankun ADS Deployment Tiers

The Western Bifurcation: Outsourcing to Tech versus In-House Fortification

Volkswagen's Strategic Pivot: SDV East and SDV West

BMW's Defensive Fortification: The "Heart of Joy"

Comparison of Western SDV Strategies

Distribution Colonization and the Re-education of Global Consumers

General Motors and the SAIC-GM-Wuling Pipeline in Latin America

The Generational Shift in Brand Loyalty

The Splinternet of Mobility: Geopolitics and Regulatory Fencing

The United States: The Connected Vehicle Security Act

Chinese Data Export Restrictions

The Eradication of the Global Car

Strategic Outlook and Market Restructuring

Works cited

When authorship stops being enough

What Karpathy's struggle tells us about leadership

Open thought about Karpathy's struggle

Did you that NotebookLM make infographics?

31 | AI: Can I Recognize You from a Picture? (1)

WEEKLY DIGEST I-XRAY: The AI Glasses That Reveal Anyone’s Personal Details

How Does It Work?

Big Tech's Role

Protecting Your Privacy

Main Services to Opt Out Of

WEEKLY DIGEST Meta Ups the Ante with Movie Gen: Revolutionizing Text-to-Video Generation

Key Features

Performance Highlights

WEEKLY DIGEST Glean who?

Glean AI: The Work AI Platform

Glean's Suite of Products

The Future of Enterprise AI Search

Conclusion

GEN AI STARTUP

Waymo is seeking a Backend Software Engineer for their Simulation Infrastructure team.

Mistral AI is seeking a Machine Learning Engineer to join their team.

Databricks is hiring a Sr. Corporate Communications Manager to join their team.

JOBS & TRAININGS Andrew Ng’s AI Python for Beginners.

30 | NoteBookLM: Create an Instant Podcast of your PDFs

GEN AI AT WORK Google's NotebookLM: Instant Podcast for YOU 🎧

What Does NotebookLM Do?

How Does It Work?

Additional Features

Technical Details

How to

What is the Role of Small Models in the LLM Era: A Survey

ACTIONABLE TIPS ChatGPT Prompts for Business

ACTIONABLE TIPS Create a Synthetic Dataset Using Llama 3.1 to Fine-Tune Your LLM

Key Steps

JOBS AND TRAININGSAnthropic's LLM Bootcamp: Improve Your Prompts

What's the Big Deal?

Course Overview

Why Should You Care?

🦾 Master AI & ChatGPT for FREE in just 3 hours 🤯

29 | OpenAI launched yesterday o1 | Waymo is proving safety | Pixtral Multimodal

WEEKLY DIGEST OpenAI Unveils New "01" Model Series

ElevenLabs_2024-09-23T11_53_46_Adam_pre_s50_sb75_se0_b_m2(5)

There are already some interesting use cases

Favorite o1 Use-Cases: Standouts

WEEKLY DIGEST Waymo Leads the Pack in Autonomous Vehicle Safety, New Data Reveals

Widespread Deployment

WEEKLY DIGEST
I-XRAY: The AI Glasses That Reveal Anyone’s Personal Details

WEEKLY DIGEST
Meta Ups the Ante with Movie Gen: Revolutionizing Text-to-Video Generation

WEEKLY DIGEST
Glean who?

JOBS & TRAININGS
Andrew Ng’s AI Python for Beginners.

GEN AI AT WORK
Google's NotebookLM: Instant Podcast for YOU 🎧

ACTIONABLE TIPS
ChatGPT Prompts for Business

ACTIONABLE TIPS
Create a Synthetic Dataset Using Llama 3.1 to Fine-Tune Your LLM

JOBS AND TRAININGS
Anthropic's LLM Bootcamp: Improve Your Prompts

WEEKLY DIGEST
OpenAI Unveils New "01" Model Series

WEEKLY DIGEST
Waymo Leads the Pack in Autonomous Vehicle Safety, New Data Reveals

WEEKLY DIGEST
Mistral Challenges Tech Giants with Launch of Pixtral 12B, Its First Multimodal AI Model

ACTIONABLE TIPS
Listen to anything on the go with the highest quality voices

ACTIONABLE TIPS
Replit Agent: AI for development everybody is talking about

GEN AI AT WORK
The AI Scientist: When Machines Start Doing the Research

GEN AI AT WORK
Eric Schmidt Drops AI Truth Bombs

GEN AI STARTUP
What is an AI Engineer?

GEN AI STARTUP
OpenAI: How Strawberries and Stars May Reshape Artificial Intelligence

WEEKLY DIGEST
The Open-Source Gambit: Meta's Llama 3.1 and the Future of AI 🦙 🏆

WEEKLY DIGEST
Bonjour, Moshi! How a French AI Lab is Outshining OpenAI

WEEKLY DIGEST
Perplexity: The Silicon Valley Darling in Hot Water

ACTIONABLE TIPS
Prompt to Analyse Existing Companies in My Market

JOBS AND TRAINING
A Three-Course Feast for AI Enthusiasts 🍽️

GEN AI AT WORK
Apple’s AI Strategy.

GEN AI STARTUP
Claude 3.5 Sonnet: The Best AI Model I've Ever Tested

GEN AI AT WORK
Copilot: A New Era for PC Interaction