Hello, Starters!

We weren't kidding when we said September always brings surprises. Some might think it's because there are only three months left in the year, but those of us who follow AI innovation know it’s just part of the ongoing race.

Here’s what you’ll find today:

• OpenAI introduces o1

• Fei-Fei Li’s startup aims for a 3D AI breakthrough

• Google launches DataGemma

• Alibaba’s new open-source models

• Copilot gets to Microsoft 365

• And more.

HockeyStick Show

Curious about the tech industry? Join us on HockeyStick Show for insights from leading experts, valuable advice, and fresh perspectives. Subscribe now!

🤖 OpenAI introduces o1 (3 min)

After a lot of speculation, what we once knew as "Project Strawberry" is finally here for us to try. OpenAI has introduced o1-preview and o1-mini, their new model series focused on "thinking" before answering. In other words, these models are trained to mimic how humans solve problems, taking more time to think before arriving at a solution.

These models are designed for complex tasks, especially those difficult to tackle in science, coding, and maths. OpenAI expects them to have significant applications in fields like healthcare research and physics. However, the early versions of the models are limited to 50 queries per week for o1-preview and 50 queries per day for o1-mini. For other tasks, ChatGPT continues to leverage GPT-4.

The Daily Newsletter for Intellectually Curious Readers

If you're frustrated by one-sided reporting, our 5-minute newsletter is the missing piece. We sift through 100+ sources to bring you comprehensive, unbiased news—free from political agendas. Stay informed with factual coverage on the topics that matter.

Join for free today!

👀 Fei-Fei Li’s startup aims for a 3D AI breakthrough (2 min)

Fei-Fei Li is well known in the industry as the "godmother" of AI, and a few months ago, she announced she was working on a startup that plans to create AI models with the ability to understand the 3D world. The startup, which she named World Labs, raised $230 million in its first funding round, with names like Andreessen Horowitz, Intel, AMD, and Nvidia on the investor list.

The main focus of World Labs is "spatial intelligence," and they will train "large world models" to give AI a better understanding of 3D, allowing it to reason and act accordingly.

💥 Google launches DataGemma (3 min)

Google's solution to hallucinations has come, and it's called DataGemma. These open models have been trained with their Data Commons database, which contains over 240 billion rich data points from reliable sources like the United Nations, the World Health Organization, and more.

The way in which DataGemma works is by enhancing Google's Gemma models through the use of RIG (Retrieval-Interleaved Generation) and RAG (Retrieval-Augmented Generation) approaches in order to improve the accuracy.

🦾 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)🎁

🤖Alibaba is definitely catching up in the AI race. The Chinese company has recently released new open-source models and a text-to-video system based on its already popular Qwen 2.5 family. The models, whose sizes range from 0.5 to 72B parameters, showcase enhanced performance in maths and coding and support over 29 languages.

💼Microsoft keeps expanding Copilot's availability in a mix that aims to transform how users work through their Microsoft 365 apps. The announcement includes Copilot Pages, the integration of Copilot in Excel,PowerPoint, and Outlook, and lastly Copilot agents, which is designed to streamline business processes.

⚡️Quick links

Runway inks deal with Lionsgate in first team-up for AI provider and major movie studio

YouTube Shorts to integrate Veo, Google’s AI video model 

Slack now lets users add AI agents from Asana, Cohere, Adobe, Workday and more

OmniGen: Unified Image Generation

Mistral launches a free tier for developers to test its AI models

Thank you for reading!

Hello, Starters!

There's a lot of movement in the AI industry; we know that's the norm. But September is a month that promises a lot of advancements from both the open-source community and well-known giants. We'd better fasten our seatbelts!

Here’s what you’ll find today:

• Mistral introduces Pixtral 12B

• “Strawberry” is closer than we thought

• Learn about Apple’s Visual Intelligence

• Introducing: DeepSeek-V2.5

• Google turns notes into “podcasts”

• And more.

HockeyStick Show

Ready to explore the tech world with insights from top experts? HockeyStick Show is your perfect podcast for valuable advice and fresh perspectives. Don't miss out—subscribe now!

🖼️ Mistral introduces Pixtral 12B (1 min)

Mistral has made its entrance into the multimodal world with Pixtral 12B. This model, built on Nemo 12B and clearly harnessing 12 billion parameters, can answer queries about images shared through links or encoded ones and perform a variety of tasks related to pictures, such as captioning or identifying how many objects are in a photo.

Pixtral 12B is currently available on GitHub and Hugging Face, and it can easily be fine-tuned by users, as it's been developed under an Apache 2.0 licence.

🍓 “Strawberry” is closer than we thought (2 min)

The fruit-related theme continues as we receive more news about OpenAI's upcoming "Strawberry" model. Recent claims from two insiders who have tried it suggest a launch date closer than initially expected, with the model possibly making headlines in two weeks.

Strawberry will be integrated into ChatGPT, and what makes it stand out is a feature that allows it to "think" for 10 to 20 seconds before answering queries. This brief pause is designed to enhance its logical reasoning capabilities in complex areas like maths, programming, design, and new tasks.

🍎 Learn about Apple’s Visual Intelligence (1 min)

With the official announcement of the iPhone 16, we can finally say we're getting closer to Apple's full-on AI era. However, the excitement must wait until October, when iOS 18 becomes available with all the 'Apple Intelligence' features. One standout is Visual Intelligence, which resembles a concept we've already seen from OpenAI and Google.

Visual Intelligence, enabled through Camera Control, a new button for the iPhone 16 and 16 Pro, allows users to ask questions about their surroundings and gather information from pictures they've taken, such as dates and times from a flyer or identifying dog breeds.

🌎Every day, new open-source models compete for the top spot, so it's common to hear claims of different ones being the "leader." However, these claims are sometimes made too soon, and some models fail to live up to the hype. DeepSeek is taking its chance with the launch of DeepSeek-V2.5, now available on Hugging Face. Its benchmark results and performance earn it the "world's top."

🎙️Google is transforming NotebookLM through "Audio Overview," intending to translate complex information into easy-to-digest audio. Basically, your documents can be turned into a "podcast" where two AI hosts highlight the most relevant aspects of your annotations, dive deeper, and summarise the material to enhance comprehension.

⚡️Quick links

PingPong: A Benchmark for Role-Playing Language Models with User Emulation and Multi-Model Evaluation

Adobe says video generation is coming to Firefly this year

AI prompt engineering: A deep dive

AlphaProteo generates novel proteins for biology and health research

Introducing Chai-1: Decoding the molecular interactions of life

Thank you for reading!

Hello, Starters!

Have you noticed that we're meeting once a week now? Make sure to save a spot for Thursdays from now on as your starting day (see what we did there?). Sit back and enjoy all the latest from AI.

Here’s what you’ll find today:

• OpenAI Japan announces what’s “Next”

• Alibaba’s Qwen2-VL is a powerful visual model

• Cohere has improved its Command R series

• Anthropic introduces an enterprise plan

• Ilya Sutskever’s SSI gets backing

• And more.

HockeyStick Show

Want to learn more and hear from top experts? Check out HockeyStick Show, a podcast perfectly crafted for tech enthusiasts, where you'll gain valuable advice and fresh perspectives. Don't forget to subscribe!

🔜 OpenAI Japan announces what’s “Next” (1 min)

We're always on the lookout for what's coming from OpenAI, and insights into their moves sometimes come from where we least expect them. Japan is the place this time, as the company's CEO in the country, Tadao Nagasaki, announced at the KDDI Summit that a new version of their artificial intelligence models, labelled "GPT Next," is likely to be launched before this year ends.

But this is not the first time we've heard about "Next," as OpenAI's CTO Kevin Scott claimed earlier this year that there would be a more powerful model by the end of 2024. Initially, we were led to believe this was "Strawberry," which we now know is in the works, but if rumours turn out to be true, "GPT Next" could be linked to "Orion," the successor to GPT-4, which will be powered by data generated from Strawberry, with increased reliability and reasoning.

Daily News for Curious Minds

Be the smartest person in the room by reading 1440! Dive into 1440, where 4 million Americans find their daily, fact-based news fix. We navigate through 100+ sources to deliver a comprehensive roundup from every corner of the internet – politics, global events, business, and culture, all in a quick, 5-minute newsletter. It's completely free and devoid of bias or political influence, ensuring you get the facts straight. Subscribe to 1440 today.

Sign up now!

👀 Alibaba’s Qwen2-VL is a powerful visual model (1 min)

Alibaba Cloud has recently introduced Qwen2-VL, and this vision-language model showcases enhanced visual understanding, video comprehension, and multilingual text-image processing. Yet, what sets it apart from current offerings is that it can analyse long-duration videos, up to 20 minutes, for example, and provide answers about their content.

Qwen2-VL supports several languages, with English, Chinese, and Vietnamese among them, and its benchmark results have nothing to envy leading models like Llama 3.1 or GPT-4. For Alibaba, this could mark significant progress in the field of visual AI.

🧑‍💻 Cohere has improved its Command R series (2 min)

Cohere keeps improving its Command R series. The RAG-optimised large language models now have updated versions that display enhanced capabilities for multilingual applications, and structured data analysis, as well as major improvements in coding, maths, logic, and latency.

Command R and Command R+ are directed at developers and companies, and they're available through Cohere's hosted API and Amazon SageMaker.

🦾 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)🎁

💼Although Anthropic's approach has always been directed at business users, they've decided to enhance Claude's capabilities with a new subscription plan aimed at enterprises and make it even more useful. The enterprise version allows the upload of proprietary data to the chatbot, from which it can analyse, answer queries, create graphics and pages, and overall be a customised AI assistant.

🤖OpenAI's former chief scientist Ilya Sutskever announced his exit from the company in May, and he shared details about its upcoming project, an AI startup called Safe Superintelligence (SSI). Now reports claim that the startup has raised $1 billion in funding, which will further nurture Sutskever's goal of developing "safe" AI systems.

⚡️Quick links

Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoders

Apple, Nvidia in talks to join OpenAI funding round, media reports say

Generative Verifiers: Reward Modeling as Next-Token Prediction

Ask Claude: Amazon turns to Anthropic's AI for Alexa revamp

Midjourney says it’s ‘getting into hardware’

Bill Gates has a good feeling about AI

Thank you for reading!

Hello, Starters!

Have you missed us? No worries! We're back again with all the latest happenings in the AI realm. We've just made a few tweaks to make sure you're getting the best of the best!

Here’s what you’ll find today:

• OpenAI is working on two new AI models

• Meta unveils “Sapiens”

• There are new versions of Gemini 1.5

• Developing a tool to predict dementia with AI

• Google presents Gemini’s latest features

• And more.

All your news. None of the bias.

Be the smartest person in the room by reading 1440! Dive into 1440, where 3.5 million readers find their daily, fact-based news fix. We navigate through 100+ sources to deliver a comprehensive roundup from every corner of the internet – politics, global events, business, and culture, all in a quick, 5-minute newsletter. It's completely free and devoid of bias or political influence, ensuring you get the facts straight.

Subscribe to 1440 today.

🍓 OpenAI is working on two new AI models (2 min)

The buzz surrounding OpenAI and its secret projects hasn't stopped for months, and every hint we've gotten only leads to more hype. But, according to insiders, fall may be the time when we're finally able to witness how these models will disrupt AI. We're talking about the upcoming and long-awaited "Strawberry" and "Orion," a flagship language model.

Reports claim that Strawberry has been designed to excel at solving maths and programming problems, while Orion is set to outperform GPT-4 and will be trained with high-quality synthetic data, most of which is generated by Strawberry.

However, it's likely that the full version of Strawberry won't arrive this year.

💥 Meta unveils “Sapiens” (1 min)

Meta's latest push into AI involves "Sapiens," a family of AI models developed for human image analysis. It comes in four sizes, with the largest being Sapiens-2B, which was trained on 1024 x 1024 pixel images. Overall, the models are able to perform tasks like body segmentation and 2D pose estimation.

The aim of Sapiens is similar to that of Meta's Segment Anything Model. Both tools are progressing human-centric vision models while also promoting innovation in the AI community, as Sapiens is currently available on GitHub.

🤖 There are new versions of Gemini 1.5 (3 min)

Gemini 1.5 arrived back in May, and although it's difficult to keep track of Google's moves as they're moving faster than anyone in the field, their latest announcement involves three new variants of this model. They've presented a small Gemini 1.5 Flash-8B, an improved version of Gemini 1.5 Flash, and an even more powerful Gemini 1.5 Pro.

Reportedly, the 1.5 Flash and 1.5 Pro showcase enhanced results in maths, coding, and complex prompts, with one of Google's AI Studio product leads claiming that the model (1.5 Flash) is the best in the world for developers at the moment. Maybe it's worth a try!

👀Google doesn't stop with the Gemini enhancements, as they've also announced the rollout of "Gems." Through this feature, users can create customised AI experts on any topic they want. On the other hand, they're broadening the availability of Imagen 3. Both features are coming to the Gemini Advanced, Business, and Enterprise tiers.

🧑‍⚕️With the help of AI, researchers from the University of Edinburgh and the University of Dundee are analysing over a million brain scans in order to develop a tool to help doctors determine the risk of dementia in patients and hopefully lead to better diagnoses and treatments earlier.

⚡️Quick links

Diffusion Models Are Real-Time Game Engines

Dolphin: Long Context as a New Modality for Energy-Efficient On-Device Language Models

The AI Code Editor

Artifacts are now generally available

Meet Einstein SDR and Einstein Sales Coach: Two New Autonomous AI Sales Agents to Scale Your Sales Team

Why AI can’t spell ‘strawberry’

Thank you for reading!

Hello, Starters!

We're not running out of new tools to try, as the AI scene keeps improving and more offerings keep surfacing. One thing is for sure: some leading models have to be on the lookout if they don't want to lose their spot at the top!

Here’s what you’ll find today:

• Ideogram’s latest text-to-image model is here

• Improving LLMs' reasoning abilities through prompts

• AI21 has unveiled Jamba 1.5

• Google introduces its "Prompt Gallery"

• These two CEOs are unsure of the EU’s AI rules

• And more.

🖼️ Ideogram’s latest text-to-image model is here (1 min)

Image generation continues to be one of the fastest-growing areas of AI, with plenty of tools to choose from. Ideogram has made the move with a new iteration of its text-to-image model. Ideogram 2.0 is available for free and showcases enhanced image-text alignment, accuracy, and five new image styles.

Besides this, the company has also launched an iOS app and a beta API. Users can opt for premium features through their multiple subscription plans. It is noteworthy that the model, allegedly, outshines similar competitors, like Midjourney and DALL-E 3.

🧠 Improving LLMs' reasoning abilities through prompts (1 min)

Research from China's Guilin University of Electronic Technology has introduced a technique based on the GSM8K dataset that allows large language models to filter irrelevant information and subsequently improve their reasoning process.

The GSMIR dataset has been tested with GPT-3.5-Turbo and GPT-3.5-Turbo-16k. Researchers intentionally fed irrelevant sentences into a series of elementary school maths problems, and both models were able to identify them 74.9% of the time through the "Analysis to Filtration Prompting" (ATF) method.

🤖 AI21 has unveiled Jamba 1.5 (2 min)

A combination of transformers and the Structured State Space (SSM) model approach has brought Jamba 1.5 to life. This model, developed by AI21, is an upgraded version of the one they showcased in March, and it includes Jamba 1.5 Mini and Jamba 1.5 large build.

Jamba 1.5 stands out as both of the models have a large context window (256K), which makes them a perfect fit for the development of agentic AI systems. Also, they're available under an open licence, suiting enterprises and developers alike.

🖼️Putting more tools in the hands of developers, Google has added a new feature to their AI Studio platform. Prompt Gallery, as they decided to call it, is integrated with pre-built prompts that get the most out of their Gemini models, covering a broad range of applications, like coding, productivity tools, and even a maths tutor.

🏎️Mark Zuckerberg and Daniel Ek, from Meta and Spotify respectively, have shared their thoughts on Europe's AI regulations. Earlier this year, rules were drafted to be applied to AI, ensuring its safe and responsible development. However, both CEOs believe these regulations can have an impact on the growing landscape of open-source.

⚡️Quick links

Google appoints former Character.AI founder as co-lead of its AI models

Adobe’s Magic Fixup is an AI Cut-and-Paste Photo Editor Trained on Videos

How LLMs Work, Explained Without Math

The Midjourney web experience is now open to everyone

Neuralink says it may have fixed its brain implant problem

BeyondMath’s ‘digital wind tunnel’ puts a physics-based AI simulation to work on F1 cars

Thank you for reading!

Hello, Starters!

Innovation in the AI era always comes from the hands of developers, who play a key role in the industry, so it's unsurprising that every day, even more companies keep putting the right tools at their disposal.

Here’s what you’ll find today:

• You can now fine-tune GPT4-o

• Google Research presents HeAR

• Nvidia’s on-device model for gaming

• OpenAI announces collaboration with Condé Nast

• Microsoft’s latest Phi models

• And more.

⚙️ You can now fine-tune GPT4-o (2 min)

Sky's the limit now for developers as OpenAI has finally launched a fine-tuning feature for GPT-4o. This will allow the customization of the model for an infinite list of use cases, ensuring users full ownership of their data and promoting the surge of innovative applications by offering 1M training tokens per day for free next month.

Fine-tuning is available in all paid tiers and includes a fine-tuning option for GPT-4o mini, which will also have free training tokens. As an example of the breakthroughs developers can create with GPT-4o at hand, OpenAI shares the Genie AI software engineering assistant from Cosine, which recently achieved great results in the SWE-bench benchmark.

👂Google Research presents HeAR (2 min)

Google Research's latest model showcases how AI can transform the health field. HeAR (Health Acoustic Representations) is a bioacoustic foundation model, trained on 300 million pieces of audio data, which can provide revealing insights into patients' well-being through audio analysis. Through this approach, over time, a sound can lead to a diagnosis.

HeAR also includes a cough model, which has been trained on 100 million cough sounds. As the model is available for custom research, companies like India's Salcit Technologies have used it to develop an application for early tuberculosis detection.

🎮 Nvidia’s on-device model for gaming (2 min)

Developed with a focus on gaming, Nvidia has presented Nemotron-4 4B Instruct, an on-device small language model that's part of the company's ACE suite. Nemotron harnesses retrieval-augmented generation and function-calling to improve the performance of game characters.

Mecha BREAK, a game developed by Amazing Seasun Games, is currently using the model, showcasing how Nemotron enhances role-playing, creating dynamic gameplay that allows seamless interactions between characters and players.

🤝OpenAI has revealed that Condé Nast is joining their list of partnerships, allowing content from leading publications to appear in ChatGPT, and now in their SearchGPT prototype. Through this alliance, which includes brands like Vogue, Wired, The New Yorker, and more, OpenAI continues to forge a path for AI in journalism.

🤖Microsoft introduced new versions of their Phi models to their family. Phi 3.5 is multimodal, with an open licence, and comes in mini, MoE, and Vision versions with 3.82B, 41.9B, and 4.15B parameters each. They've improved in tasks like reasoning, and image and video analysis, surpassing leading models like Gemini 1.5 or Meta's Llama 3.1 in certain benchmarks.

⚡️Quick links

New McAfee tool can detect AI-generated audio

Nvidia researchers create AI that predicts thunderstorms with kilometer-scale precision

Dream Machine 1.5 is here

To Code, or Not To Code? Exploring Impact of Code in Pre-training

Gemini in Gmail can now help polish up your drafts

The AI wave continues to grow on software development teams

Thank you for reading!

Hello, Starters!

Catching up with an ever-growing technology is not a simple task, especially at the moment of establishing preventative measures and regulations in case it gets out of hand. This is why most companies emphasise their focus on responsible AI development.

Here’s what you’ll find today:

• Build your own robot with Hugging Face

• Researchers propose a way to distinguish humans from AI

• OpenResearcher promotes AI in scientific research

• AMD tries to catch up with Nvidia’s AI reign

• Google’s “Image 3” is now available

• And more.

🤖 Build your own robot with Hugging Face (3 min)

Hugging Face's mission involves the democratisation of AI development through openness, and they've taken this effort to a field that's set to bring major advancements to the AI community: robotics. This started with the launch of their LeRobot platform a few months ago, and now it continues to grow as they've released a tutorial that guides developers through building and training AI-powered robots.

Now, developers can build their own robots and teach them to move with just a laptop—an activity that, long ago, was only available to corporations and institutions is now in the hands of those who are curious enough. We can expect big innovations to come from this approach.

🆔 Researchers propose a way to distinguish humans from AI (1 min)

Many predict that a time will come when it will be difficult to differentiate humans from AI bots on the internet. Although this remains a possibility, it hasn't happened yet, and at the moment, researchers are looking for ways to prevent it from happening. A group that includes members from OpenAI, Microsoft, MIT, and others is developing a method called "Personhood Credentials" to tackle the problem.

What they hope to accomplish with this is a way to diminish the surge of AI bots, as well as avoid identity theft as they become more capable. The digital credentials will confirm that the holder is a human without revealing more information.

📄 OpenResearcher promotes AI in scientific research (2 min)

OpenResearcher is the result of a joint effort by the Generative AI Research Lab (GAIR). As its name reveals, this is an open-source project with a focus on scientific research, a domain that is being transformed by the use of AI. The application works as an assistant and easily answers researchers' questions.

By combining retrieval augmentation from the internet with base knowledge from ArXiv, OpenResearcher can ask reorienting questions to guide the analysis. It also includes customised tools that accelerate the research process.

💸AMD is demonstrating that Nvidia isn't the only player in the AI arena. They've recently announced the acquisition of ZT Systems, a company specialising in building custom computing infrastructure for AI "hyperscalers." The deal has a valuation of $4.9 billion, and it seeks to improve the speed at which AMD develops its technologies.

🖼️After announcing it back in May and with a limited rollout following, Google has made its Imagen 3 generator available to US users through their AI Test Kitchen. The image generator has received great praise from Google, as it creates images with higher quality, better detail, and richer lighting. They've also shared a research paper that goes in-depth about the system.

⚡️Quick links

Transformers in music recommendation

RAGChecker: a fine-grained evaluation framework for diagnosing retrieval and generation modules in RAG

MeshFormer: High-Quality Mesh Generation with 3D-Guided Reconstruction Model

Meta’s Self-Taught Evaluator enables LLMs to create their own training data

Microsoft will start training AI models with user data from Copilot, Bing, and MSN in October

Tesla is hiring people to do the robot

Thank you for reading!

Hello, Starters!

The open-source community has become one of the must-watches in AI development, as the most promising offerings are coming from it, and its growth ensures we will have a never-ending list of innovations in the upcoming years.

Here’s what you’ll find today:

• Nous introduces HERMES 3

• LLMs have their own understanding of reality

• Geekbench AI’s benchmark for ML

• Midjourney’s latest features

• AI restores the voice of an ALS patient

• And more.

🤖 Nous introduces HERMES 3 (4 min)

Hermes 3 is the latest breakthrough in open-source. This model, developed by Nous Research in collaboration with Lambda Labs, is a fine-tuned version of Meta's Llama 3.1 framework and is available in three parameter sizes: 8B, 70B, and 405B showcasing performance that actually surpasses Llama. It has been trained with synthesised data and a mix of reinforcement learning from human feedback and Neural Magic's FP8 method.

This model has been designed with a user-first mindset, which goes beyond a creative approach, displaying great adaptability to following instructions, long-term context retention, agentic functions, and even internal monologue abilities.

💭 LLMs have their own understanding of reality (5 min)

Research from MIT's CSAIL is allowing us to have a glimpse into the thought process of LLMs. By training a model on over 1 million random puzzles and giving it solutions without showing how these solutions work, the model was consequently led to come up with its own.

Although the research was conducted with a small model and a simple programming language, the results are promising, as they showcase ways in which LLMs can develop a better understanding of language, which will further improve their learning capabilities.

📏 Geekbench AI’s benchmark for ML (2 min)

Most companies are looking for ways to improve their current evaluation methods to ensure the development of AI technologies goes right. Geekbench is one of them, and they've introduced its benchmarking suite specially designed for machine learning, deep learning, and most AI workloads.

Geekbench AI is aimed at developers, as they can harness it to ensure their apps are working optimally across platforms and for engineers or general users who want to analyse how a specific device uses AI.

🖼️Making image generation easier and getting one step ahead of new offerings, Midjourney has recently updated its website with tools that allow users to customise its AI creations even further. The website now includes a revamped editor with a canvas extension tool, which creates a larger version of the image with new visuals, and inpainting, which transforms selected areas through prompts.

🧠A 45-year-old ALS patient has regained his voice thanks to AI. Research by UC Davis has implemented a brain-computer interface and neural sensors to capture the brain commands involved in speech and translate them into words, which are processed by AI text-to-speech software based on old recordings of the patient's voice to make it sound exactly like him.

⚡️Quick links

These Living Computers Are Made from Human Neurons

Google brings AI answers in Search to new countries

Gen-3 Alpha Turbo Image to Video is now available

How to Prune and Distill Llama-3.1 8B to an NVIDIA Llama-3.1-Minitron 4B Model

OpenAI is fresh out of SearchGPT

xGen-MM (BLIP-3): A Family of Open Large Multimodal Models

Thank you for reading!

Hello, Starters!

When you realise that companies are currently competing with one another to come up with breakthroughs faster, you understand that calling it the "AI race" isn't just a catchy headline.

Here’s what you’ll find today:

• Gemini Live comes to transform smartphones

• OpenAI introduces “SWE-bench Verified”

• Anthropic presents Prompt Caching

• There’s a new GPT-4o model for ChatGPT

• MIT’s AI risk repository

• And more.

💥 Gemini Live comes to transform smartphones (5 min)

As the wait for OpenAI's advanced Voice Mode and Apple's revamp of Siri continues, Google has decided to take the lead with Gemini Live. This mobile AI assistant takes over your phone and allows you to have a full-on conversational experience where you can ask questions, command it to set reminders, find emails, play music, and more, even when your phone is locked, as it keeps running in the background.

This is not the first time that Google has taken an advantage over other competitors; they've already made it clear in their conferences that infusing AI into their products is one of their main goals. And, unlike OpenAI, they've already started rolling out Gemini Live to Gemini Advanced subscribers on Android, with iOS coming up in the next few weeks.

🏅OpenAI introduces “SWE-bench Verified” (12 min)

SWE-bench is a popular benchmark in the AI field, as it tests models' capabilities in solving software issues found on GitHub, a key part of their development as autonomous systems. As part of their Preparedness Framework project, OpenAI has worked with the authors of SWE-bench to update the benchmark and make it more reliable and accurate.

This is how SWE-bench Verified was born. This upgraded benchmark is a subset of SWE-bench's original test set, and it includes 500 verified samples that have been catalogued as non-problematic by a group of software developers.

📄 Anthropic presents Prompt Caching (2 min)

Anthropic has launched "Prompt Caching" for developers in its API. This feature allows users to cache a context they're frequently using in case they need to look back at that information constantly in other requests. With this approach, they're reducing costs by up to 90%, and it is already available in beta for Claude 3.5 Sonnet and Claude 3 Haiku.

As Anthropic points out, prompt caching has proven to be useful in situations where large amounts of context are involved, such as conversational agents, coding assistants, detailed instruction sets, and more.

🤖Last week, OpenAI launched a version of GPT-4o optimised for API usage, but now they've also disclosed that there's an updated version of the model powering ChatGPT, and that's been tested on LMSYS as "anonymous-chatbot," showcasing great performance at tasks like coding, instruction-following, and hard prompts.

📚A group of researchers at MIT is currently working on an AI "risk repository," a library or database aimed at companies, individuals, or even governments as a guideline to design regulations for AI technologies, as every aspect in which AI could be used is filled with very different kinds of risks, which should be cautiously analysed.

⚡️Quick links

Unreasonably Effective AI with Demis Hassabis

What are AI agents and why do they matter?

Apple Aiming to Launch Tabletop Robotic Home Device as Soon as 2026

Mutual Reasoning Makes Smaller LLMs Stronger Problem-Solvers

AI could be the breakthrough that allows humanoid robots to jump from science fiction to reality

Thank you for reading!

Hello, Starters!

There's a recurring theme in our newsletter that we think you've already noticed, and it not only happens here, it is happening in the whole AI scene: we keep getting more advanced AI systems. Are you curious about which ones we'll be covering?

Here’s what you’ll find today:

• Sakana AI boosts research with The AI Scientist

• xAI releases Grok-2 in beta

• Cosine unveils Genie 

• Google’s upcoming Pixel series gets an AI touch

• Meta and Oxford present VFusion3D

• And more.

🧑‍🔬 Sakana AI boosts research with The AI Scientist (9 min)

Sakana AI has an approach to AI technologies that distinguishes them from other companies in the industry, and as they move forward with their nature-inspired offerings, the novelty doesn't wear off. They've presented "The AI Scientist," an AI system designed to streamline the research process.

This agent is capable of generating research ideas, writing code, performing experiments, summarising results, evaluating, and then presenting them in a scientific paper with near-human accuracy. However, they emphasise that this doesn't diminish the role of scientists, as it is rather a proposition of how the integration of AI technologies into the mix can open the door to more innovation.

🤖 xAI releases Grok-2 in beta (3 min)

Moving forward in the AI arena is Grok, xAI's outstanding AI model which in just a few months has turned into a notable rival for most leading models. Grok-2 has recently been released as a beta on X, alongside Grok-2 mini. Both models have demonstrated increased performance in reasoning, reading, maths, science, and coding.

Grok-2 went into testing in disguise on LMSYS's leaderboard, and results showcase the model outperforming Claude 3.5 Sonnet and GPT-4-Turbo. The announcement also included a collaboration with Black Forest Labs, with its FLUX.1 model enhancing Grok's capabilities in image generation.

🧑‍💻 Cosine unveils Genie (7 min)

Cosine has introduced "Genie," an AI model that they claim is the world's first AI Software Engineering colleague. Genie has been trained on a special dataset focused on human reasoning, which allows it to mimic the way human engineers tackle problems. It has performed great results on SWE-bench, which tests LLMs' capabilities on GitHub software issues.

This approach makes Genie stand out against other coding assistants, as it can autonomously perform coding tasks and can write in up to 15 coding languages, like JavaScript, Python, and Java.

Power your competitive advantage with intelligent automation from ELEKS

ELEKS' intelligent automation service transforms your business operations through data-driven solutions. We automate complex tasks, streamlining processes to increase productivity and reduce operational costs. Our tailored solutions adapt to your changing needs and help you unlock new growth opportunities by freeing your team to focus on high-value tasks.

The result? Enhanced customer satisfaction, improved client retention, and a stronger market position.

Talk to our automation specialists

📱This year's Made by Google is another example of the company's efforts in AI, which many claim are more noticeable than those of fellow competitors like Apple. Google’s upcoming Pixel phones have Gemini integration as their main characteristic, while Android users can also expect a new update strongly focused on AI.

☎️A research effort between Meta and the University of Oxford has brought VFusion3D into the spotlight. This AI model can transform images or text prompts into high-quality 3D objects, which could possibly aid not only designers but also developers and artists in fields like gaming, virtual reality, and other creative industries.

⚡️Quick links

The Pixel 9 phones are big cameras and screens soldered onto Gemini AI ambitions

LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs

How to use AI coding tools to learn a new programming language

This AI understands video games, and it could make cheating easier than ever

The next iPhone SE could have Apple Intelligence

Thank you for reading!

Hello, Starters!

The weekend is over, and here at Starter AI, we're more than ready to start our daily journey into the AI realm. We're always discovering something new; if you don't believe us, just keep reading.

Here’s what you’ll find today:

• Google DeepMind’s robotic table tennis player

• Alibaba aims to improve maths reasoning

• Character.AI presents Prompt Poet

• Free ChatGPT users can access DALL-E 3

• AI robocallers are becoming regulated

• And more.

🏓 Google DeepMind’s robotic table tennis player (4 min)

The possibility of robots achieving human-like capabilities doesn't seem as far if we direct our attention to the work of Google DeepMind. They've developed a robot that's achieved a 45% winning rate at table tennis by training it in a simulation using reinforcement learning.

To start the training process, they gathered a small dataset of human-human play that served as a base, alongside techniques like "zero-shot." As the robot played more matches, its capabilities improved, allowing it to adapt in real time to new opponents.

🧮 Alibaba aims to improve maths reasoning (3 min)

Maths is one of the most demanding tasks for large language models, and as companies keep advancing their AI technologies, they're seeking ways to improve their reasoning in this area. Alibaba is up for this challenge with its Qwen2 series, as they've announced the integration of specialised maths models, aiming to promote AI development and help in solving complex mathematical problems.

Joining the Qwen2 family are Qwen2-Math and Qwen2-Math-Instruct, with the latter coming in three sizes: 1.5B, 7B, and 72B. These models are showcasing great results across the most used maths benchmarks, surpassing both open-source and closed-source models.

📝 Character.AI presents Prompt Poet (5 min)

Character.AI is one of Google's latest acquisitions and the minds behind "Prompt Poet," a tool that focuses on prompt engineering and how its precise design can lead to better results when using large language models. Although models are already optimised to understand natural language and inputs in a conversational tone, advanced techniques give users more control.

Prompt Poet uses a low-code approach based mostly on Python, making it accessible to developers and non-technical users. It streamlines the process of designing prompts that contain complex data and an abundance of context, resulting in engaging prompts that lead to the desired outcome.

🖼️OpenAI is granting free users access to its DALL-E 3 image generator, a service only available through a ChatGPT subscription. They can now ask the chatbot to show them what "something looks like," and DALL-E will create a visual. The only limitation is that they can only make two images per day.

☎️The rise of AI technologies has prompted most countries to set regulations. Recently, the U.S. Federal Communications Commission has proposed a rule directed at robocallers, which demands that they disclose if they're using AI for phone calls and text messages, as users can be prone to fraud or other scams.

⚡️Quick links

GPT-4o System Card

RUBICON: Evaluating conversations between humans and AI systems

Google cuts Gemini 1.5 Flash prices and adds new PDF features

Anthropic and Caylent partner to slash AI deployment times in half

OpenAI is a bigger threat to Google than US regulators

Thank you for reading!

Hello, Starters!

Did you notice we've already got to Thursday? Neither do we! The pace at which AI keeps evolving sometimes makes us lose our sense of time. But since there are no signs of it slowing down, we might as well continue to catch up!

Here’s what you’ll find today:

• YouTube tests Gemini integration

• Meet the startup that wants to predict space weather with AI

• OpenAI has more announcements for us

• Sam Altman teases “Strawberry”

• Apple Intelligence might come with subscriptions

• And more.

💥 YouTube tests Gemini integration (1 min)

Google keeps going with its goal of infusing AI into all of its offerings, and now it's YouTube's turn. It's been announced that the platform is testing a Gemini integration to help creators brainstorm ideas for their channels, mostly for videos, titles, and thumbnails.

The feature is currently available for selected users. This strategy aims to position YouTube as a solid competitor against other video platforms while promoting Google's AI tools at the same time.

🚀 Meet the startup that wants to predict space weather with AI (6 min)

How AI can aid innovation is endless, and we could properly say that the sky isn't the limit. Perceptive Space, a Canadian startup, is working on a platform powered by AI to predict space weather, as it's been proven that the technology brings several benefits to traditional forecasting techniques.

Perceptive Space has recently emerged from stealth and plans to use funding to continue its development. They leverage AI algorithms to analyse decades' worth of space weather and operational data to predict how the conditions can affect satellite operations or other infrastructure.

🤖 OpenAI has more announcements for us (1 min)

There's more to find out about OpenAI's moves. First, they're making GPT-4o more affordable for developers. The latest version, "gpt-4o-2024-08-06," can reportedly save 50% on input tokens and 33% on output tokens. Also, more output tokens are supported with this upgrade. Second, the company has made it clear that this year's DevDay conference will not include the announcement of new products.

This has been a busy week for OpenAI, yet they continue to move forward, and it seems that their recent focus is on enhancing the developer experience.

🍓Sam Altman's followers on X are already used to him hinting at OpenAI's moves through cryptic posts. So, it's no surprise that one of his most recent, in which he shared a harmless photo of a strawberry, has brought a sudden buzz. Many believe that it is related to the company's upcoming model.

🍎According to insiders, Apple's dive into AI might come with a price tag. Although the full rollout of the features isn't happening yet, and regions like China and Europe have to wait a while for access, there's a high probability that the company will charge between $10 and $20 for advanced features.

⚡️Quick links

Language Model Can Listen While Speaking

Apple Intelligence may come to EU after all…but only for Mac

People are returning Humane AI Pins faster than the company can sell them

OpenAI invests in a webcam company turned AI startup

RLHF is just barely RL

Thank you for reading!

Hello, Starters!

There's no doubt that one of the biggest players in the AI race is OpenAI; we don't need to remind you of that. However, being at the top comes with its challenges, and this year they've faced plenty.

Here’s what you’ll find today:

• Oxford researchers present MedSAM-2

• John Schulman departs OpenAI

• Figure introduces its Figure 02 robot

• OpenAI releases “Structured outputs in the API”

• ByteDance launches an AI video generator

• And more.

🩺 Oxford researchers present MedSAM-2 (2 min)

Meta's open approach with SAM 2 promised to spark innovation, and we're witnessing that with the introduction of MedSAM-2, which comes from the hands of University of Oxford researchers. This AI model can process entire series of medical images with enhanced segmentation in 2D and 3D.

MedSAM-2 achieves this by treating images like video sequences. Results across medical datasets reveal that it surpasses similar segmentation models. Also, a relevant aspect is that the model remembers reliable predictions through a "Confident Memory Back," which streamlines the process.

👋 John Schulman departs OpenAI (1 min)

At this point, OpenAI is familiar with making headlines. This year, although it has brought major announcements by the company, it has also got its sudden twists, and some of those are related to the departure of its members. John Schulman, one of the co-founders, recently announced that he's leaving the company to join Anthropic.

According to Schulman, the main reason behind his decision is that he wants to keep focusing on AI alignment, and he believes that Anthropic is a place that can offer new perspectives on these topics.

Moreover, Greg Brockman, president and co-founder of the company, is also taking time off from OpenAI. However, he has claimed that it is more of a "sabbatical" to "recharge."

🤖 Figure introduces its Figure 02 robot (1 min)

Figure, the robotics company, has presented its latest advancement in humanoids, Figure 02, a robot that harnesses the power of AI to potentially make a change in places like factories or even homes. This new version distinguishes itself from "Figure 01," launched last year, as it includes a speech recognition and output system based on custom AI models to enhance human interaction.

Both hardware and software have been updated in Figure 02. It includes six RGB cameras controlled by a Vision Language Model, a set of hands whose force is comparable to humans, and improved battery capacity.

🧑‍💻As a way of making developers' lives easier, OpenAI has released "Structured outputs in the API," a feature that reduces the reliance on other tools and increases interoperability between systems, as it assures that model-generated outputs align with JSON Schemas, even complex ones, an aspect most LLMs struggle with.

🎥The announcement of Sora prompted a surge of AI video generation tools, and ByteDance, the Chinese company behind TikTok, is also taking its step into the field with "Jimeng AI," a text-to-video model with several subscription plans that allow users to generate 168 videos per month.

⚡️Quick links

Reddit to test AI-powered search result pages

Amazon Titan Image Generator v2 is now available in Amazon Bedrock

NVIDIA’s AI team reportedly scraped YouTube, Netflix videos without permission

LG, Samsung eye South Korea's AI textbooks as edtech springboard

Elon Musk sues OpenAI again

Thank you for reading!

Hello, Starters!

We're already getting too used to our classic models, chatbots, and image generators. It's always good to look around and see what new offers are coming up in the AI scene, and we assure you, they're worth your time!

Here’s what you’ll find today:

• GitHub Models are here for devs

• MIT’s novel AI calibration method

• Black Forest Labs unveils FLUX

• OpenAI is working on a ChatGPT detection tool

• Nvidia’s latest chips are getting delayed

• And more.

🧑‍💻 GitHub Models are here for devs (4 min)

Developers are now getting help from GitHub to become the AI engineers of the future. The platform has recently launched a built-in playground where users can access leading models in the industry and harness them to develop their AI applications. The offerings go from GPT-4o to Mistral Large 2, providing an innovative perspective for coding.

This approach is just another example of the democratisation of AI. GitHub users will be able to employ models from companies like Meta, Mistral, and Microsoft, to name a few, to uncover novel ways in which AI can aid in the development process, leading to groundbreaking advancements as we inch closer to AGI.

🌡️ MIT’s novel AI calibration method (3 min)

Although most models have become improved enough for us to rely on them, there are still certain tasks or queries where we might want to double-check the information they're providing. That's why a research group at MIT has come across a calibration method, dubbed Thermometer, that can prevent models from giving incorrect outputs.

"Thermometer" consists of building a smaller model that assists the main one and provides the right "temperature" to improve its accuracy when facing a new task. This technique is based on the 'temperature scaling' calibration method and shows great results in reducing computational costs.

🖼️ Black Forest Labs unveils FLUX (3 min)

Black Forest Labs is a German startup founded by some of the minds behind the technology of Stable Diffusion, and they've brought a new contender to the image generation arena: FLUX.1. This text-to-image model is built in a hybrid architecture that mixes both transformer and diffusion techniques to reach incredible results.

According to the company, FLUX.1 surpasses offerings like Midjourney or DALL-E, as it showcases improved image quality and correlation to text prompts. Also, results note that the model is slightly better at producing hands, an issue that constantly haunts most AI image generators.

💬According to reports from the Wall Street Journal, OpenAI has an AI-text detector in the works, distinct from the one they retired last year. This version provides better results at discovering changes like paraphrasing; however, the company isn't completely convinced about its launch, as it may hurt its business model and unfairly affect certain users.

⏳Being the main provider in the AI era comes at a cost. Many companies rely on Nvidia for their innovations, and they'll now have to wait until early 2025 to get ahold of the company's Blackwell chips, as sources claim they've encountered design issues late in the production process, pushing back the initial shipping date.

⚡️Quick links

Neuralink implanted second trial patient with brain chip, Musk says

Stability AI speeds up 3D image generation with Stable Fast 3D

Microsoft says OpenAI is now a competitor in AI and search

The EU’s AI Act is now in force

Meta is reportedly offering millions to use Hollywood voices in AI projects

Thank you for reading!

Hello, Starters!

AI is everywhere, there's no denying it. But also, it is coming to places we have never thought it would get before. Can you imagine getting an AI-powered accessory?

Here’s what you’ll find today:

• Google expands its Gemma 2 family

• Meet your new AI “Friend”

• Canva acquires Leonardo.ai

• Midjourney shares its latest updates

• First look at Apple’s AI approach

• And more.

🤖 Google expands its Gemma 2 family (4 min)

Google is also making its bets on openness with the launch of three new models to expand its Gemma 2 family. The company's main focus is on responsible development, and Gemma 2 2B, ShieldGemma, and Gemma Scope are a great example of how they're prioritising safety and democratising access for developers to nurture innovation.

Besides their enhanced performance, the models also have pretty interesting features. Gemma 2 2B is a lightweight model with built-in safety advancements, ShieldGemma works as a safety content classifier, filtering malicious content from prompts and generated outputs, and Gemma Scope aids developers as an interpretability tool, giving insights into the inner workings of the models.

❤️ Meet your new AI “Friend” (2 min)

Wearables are still a big part of the upcoming innovations we might see with AI. Despite some of the offerings we've seen not surviving past the hype phase, it's good to keep track of how the technology can be translated into an everyday use object. And, focusing on one of the main problems of this era, loneliness, we've been introduced to "Friend."

Created by Avi Schiffmann, this necklace serves as its name notes: a friend. It listens to you and sends its responses via text through an app. Friend doesn't store data, and its abilities are not necessarily focused on creating code, writing emails, or coming up with clever answers to queries; it is based on the sole idea of companionship.

🎨 Canva acquires Leonardo.ai (2 min)

Canva is broadening its commitment to AI with the recent acquisition of Leonardo.ai, an Australian startup focused on training AI models for image generation. This collaboration plans to give more control to creatives and promote the adoption of Gen-AI.

Despite the acquisition, Leonardo.ai will continue to work as an independent platform. However, users can expect most of its tools to start rolling out on Canva.

🍎Developers are the first ones to get a glimpse of Apple's approach to AI. Although iOS 18.1 isn't arriving yet, beta testing has started, demonstrating how the company has harnessed AI mostly for productivity enhancements, like writing aids and Siri's complete revamp.

💥The latest version of Midjourney (V6.1) is live. Some upgrades include improved image quality, accuracy in prompts, more coherent results for hands, arms, and animals, and 25% faster performance for standard image jobs. According to them, this version will make everything "generally more beautiful."

⚡️Quick links

Runway announces even faster, cheaper AI video model Gen-3 Alpha Turbo

Apple says its AI models were trained on Google’s custom chips

Create Your Own Custom AI With AI Studio

Investors Are Suddenly Getting Very Concerned That AI Isn't Making Any Serious Money

Zuckerberg says Meta will need 10x more computing power to train Llama 4 than Llama 3

Thank you for reading!

Hello, Starters!

July has been a very busy month in tech. Companies have been continuously sharing their latest advancements, and it seems that they don't plan to stop for now. We're considering calling this the summer of AI.

Here’s what you’ll find today:

• Meta releases SAM 2

• OpenAI starts to rollout “Voice Mode”

• Baidu’s research on hallucinations

• Runway’s new Image to Video model

• Getty and Shutterstock’s AI work with Nvidia

• And more.

🤖 Meta releases SAM 2 (1 min)

Meta continues to thrill the AI community, this time by presenting a new generation of its Segment Anything Model. SAM 2 is a unified model that allows seamless real-time object segmentation in images and videos. As expected, they're also sharing the code and weights for the model under an Apache 2.0 licence, promoting once again openness in development.

Segmentation is based on identifying the pixels corresponding to a specific object of interest in an image, mostly used in computer vision. Meta points out that SAM 2 will be a great aid to bring innovations in this field, as well as allow users to create new video effects, among other applications.

🗣️ OpenAI starts to rollout “Voice Mode” (2 min)

A number of ups and downs have followed OpenAI's announcement of its advanced “Voice Mode” feature, which is one of the most expected this year. After delaying its release last month to improve its safety, they've just started giving access to selected users of ChatGPT Plus.

Advanced Voice Mode is scheduled for a gradual release that OpenAI anticipates doing this fall. However, the demo that left most users in awe will not be fully blown out yet, like the video or screen-sharing capabilities, which will be coming at a later date.

🧠 Baidu’s research on hallucinations (4 min)

Hallucinations are still one of the main issues most models face, and finding solutions to them has been a constant focus for companies that develop AI technologies. Baidu is one of them, and they've recently shared a paper that contains its research on how a "self-reasoning" framework can help AI systems stop giving incorrect information.

This method consists of allowing the models to evaluate their own knowledge by focusing on their reasoning trajectories and putting them through a series of processes that improve their decision-making and enable the system to critically analyse its outputs.

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 😉

Register & save your seat now! (valid for next 24 hours only!)

🎥After announcing Gen-3 Alpha in June, Runway is now adding a new Image to Video feature to the tool which empowers users with even more creative freedom. Images can be used on their own or with additional prompts to generate videos from 5 to 10 seconds.

💥Both Shutterstock and Getty Images are diving into generative AI with the help of Nvidia. From Shutterstock's side, its latest service allows users to create 3D assets through images or text prompts, and Getty has made enhancements to their image generator.

⚡️Quick links

GPT-4o Long Output

AI and The Next Computing Platforms With Jensen Huang and Mark Zuckerberg

AMD is becoming an AI chip company, just like Nvidia

Samsung bullish on AI demand as profit soars on higher chip prices

Researchers at Stanford Present RelBench: An Open Benchmark for Deep Learning on Relational Databases

Thank you for reading!

Hello, Starters!

How we look up answers online is expected to change radically in the AI era. There have been attempts by several companies, and now, unsurprisingly, OpenAI has joined the trend.

Here’s what you’ll find today:

• DeepMind’s models win a silver medal in maths

• OpenAI announces SearchGPT

• Google upgrades its Gemini chatbot

• Apple’s AI may get delayed

• Udio releases a new version of its AI model

• And more.

🥈 DeepMind’s models win a silver medal in maths (6 min)

AGI can be achieved by developing models that solve advanced maths problems. Google DeepMind is aware of this, and that's why most of its efforts have been directed towards the creation of models like AlphaProof and AlphaGeometry 2, which have recently showcased amazing results by solving four of six problems from the International Mathematical Olympiad (IMO), one of the main benchmarks for these capabilities.

As has been discussed several times, most AI models are still in the very early stages of development, struggling to catch up with general maths problems due to reduced reasoning skills and limitations on training data. Yet, the transformations we expect from AI don't seem as far off as we witness these advances.

🔍 OpenAI announces SearchGPT (2 min)

After months of speculation, OpenAI has finally released SearchGPT, their take on an AI-powered search experience. The feature is a prototype, and it's currently available to selected users, although there's a possibility to join the waitlist. The company expects to integrate these capabilities into ChatGPT once the testing process ends.

Everything points to the next aspect AI will change: how we search the web, as we've seen with offerings like Perplexity, Google, and Bing. Now, here's OpenAI's approach, which is based on quick and direct responses, with links to relevant sources as part of their ongoing partnerships with publishers. SearchGPT is not related to the training of OpenAI's foundation models.

🤖 Google upgrades its Gemini chatbot (2 min)

Gemini users can now enjoy an improved experience as Google has decided to make 1.5 Flash its new base model. With this integration, the multimodal model will now be available for Gemini's free tier and will broaden its availability to over 40 languages and 230 countries.

1.5 Flash will provide faster and more insightful responses, as Google points out that many of the main use cases for Gemini focus on productivity and streamlining tasks. Also, the company has expanded the context window for the chatbot, which now goes up to 32K tokens.

🍎It seems that Apple Intelligence won't be rolled out at the same time as iOS 18 as initially expected. According to reports, the company's widely anticipated foray into AI will be released for customers in October. However, beta testers could be getting a glimpse of it this week.

🎶Udio has introduced a new version of its novel AI music generator, with tools to enhance music production and catch the attention of creatives and music experts. The latest upgrades include improved audio quality, more human-like voices, and a revamped user experience.

⚡️Quick links

Microsoft unveils serverless fine-tuning for its Phi-3 small language model

Study reveals major weaknesses in AI's ability to understand diagrams and abstract visuals

Generative AI requires massive amounts of power and water, and the aging U.S. grid can’t handle the load

Google Deepmind Researchers Introduce Jumprelu Sparse Autoencoders

Paris Olympics broadcasters diverge on AI approach

Thank you for reading!

Hello, Starters!

It is incredible to think about the times we're living in. Since AI became more prominent in the industry, there hasn't been a day when we took a break from innovation, and we're just at the very beginning!

Here’s what you’ll find today:

• Mistral presents Large 2

• Bing steps up its AI game

• AI models are prone to collapsing

• OpenAI’s Rule-Based Rewards for model safety

• Stability AI unveils Stable Video 4D

• And more.

💥 Mistral presents Large 2 (3 min)

With open source having its moment in the industry, another big player has decided to make a move too. Mistral has unveiled Large 2, a 123B parameter model with a 128k context window and multilingual support that also includes more than 80 coding languages like Python, Java, and C. However, its "openness" is restricted to non-commercial use.

The benchmark results shared by Mistral showcase the model nearly matching Meta's Llama 3.1 405B. Large 2 has enhanced performance in tasks like reasoning, code generation, and maths.

🔍 Bing steps up its AI game (1 min)

Microsoft has given Bing, its popular search engine, a new AI revamp. They're introducing a twist to the search experience by mixing Gen-AI and LLMs. The feature is being launched on a small batch of user queries and includes an AI-generated response, a link to the source, and regular search results.

After Google's attempt at AI-powered search didn't go well, it's not surprising that fellow tech companies are trying the approach with a bit more caution.

🤖 AI models are prone to collapsing (17 min)

Model collapse has been a recurrent topic as AI technologies keep on developing. Scientists are constantly sharing insights into what would happen if the data on which models are trained became redundant to the point where it starts affecting them. In other words, as AI-generated text becomes more prominent online, and companies keep getting their data from websites, models are prone to affecting themselves with their own content.

This is where the importance of valuable data resides, and why we're also witnessing plenty of attempts by most tech giants to obtain it from different sources. The effects of "model collapse" are something that must be on the lookout for as technology keeps advancing.

📄As part of its commitment to transparency and responsible development, OpenAI has shared research on Rule-Based Rewards, a method that they've been employing in the training of GPT-4 to GPT-4o Mini to ensure that they're aligned with their safety demands. RBRs are based on step-by-step rules to monitor the model's outputs.

🎥Stability AI revolutionises video generation with the launch of Stable Video 4D, standing apart from other models with its ability to process video input and then create new videos from different angles based on the original one. This tool could be a great aid in fields like gaming or AR/VR.

⚡️Quick links

AI-powered medical diagnosis gets a transparency boost with new 'Chain of Diagnosis' method

Reddit is now blocking major search engines and AI bots

Customise GPT-4o mini for your application with fine-tuning

Nvidia AI Releases Minitron 4B and 8B

Tandem drifting Toyotas show how AI might help drivers on slippery roads

Thank you for reading!

Hello, Starters!

AI keeps growing, and so does its influence. One thing that many tech leaders have in mind is that there's something that will assure constant innovation... and that is open source.

Here’s what you’ll find today:

• Meta unveils Llama 3.1

• Elon Musk announces xAI’s Supercluster

• Visual HayStacks: A new benchmark

• Adobe’s latest AI tools

• Mark Zuckerberg’s thoughts on open source

• And more.

🦙 Meta unveils Llama 3.1 (15 min)

Meta is turning the world of open source upside down with the release of Llama 3.1, which not only includes upgrades to its 8B and 70B models but also introduces a 405B version that they've described as the first frontier-level open-source AI model, with capabilities almost on par with GPT-4o and Claude 3.5 Sonnet.

Open source is a key part of AI advancements. By granting users access to powerful systems like Llama 3.1 405B, Meta ensures a never-ending stream of innovation, as the decentralisation of AI technologies opens up new development possibilities.

🤖 Elon Musk announces xAI’s Supercluster (2 min)

Memphis, Tennessee is the selected place for what appears to be an upcoming AI breakthrough. xAI's CEO, Elon Musk, shared through X that they've started training the "most powerful AI training cluster" in the world, also known as the Memphis Supercluster. Musk's main goal for the cluster is to train "the world's most powerful AI" by December of this year.

The initiative also counts on the presence of Nvidia, as the cluster began its training with 100,000 liquid-cooled H100 GPUs on a single RDMA (Remote Direct Memory Access) fabric. It's a matter of time to see how xAI performs this ambitious feat.

💥 Visual HayStacks: A new benchmark (3 min)

A study from UC Berkeley's Artificial Intelligence Research has discovered that most models fall short when it comes to filtering relevant information from large image sets. Addressing this weakness has led them to develop "Visual Haystacks," a benchmark that allows them to evaluate an AI model's performance when processing extensive collections of images.

The benchmark draws inspiration from the "needle in a haystack" saying, and is based on two tasks, a "single needle" challenge and a "multi-needle" challenge. The "needle" is the relevant image or images on which the model should focus.

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 😉

Register & save your seat now! (valid for next 24 hours only!)

🎨Continuing with its infusion of AI into most of its applications, Adobe has recently announced new Firefly tools aimed at Photoshop and Illustrator. The features focus on image generation and allow users to create textures or images by just using prompts. Adobe claims to have taken the necessary steps to maintain a mindful AI approach.

🌐Following the launching of Meta's Llama 3.1 model, Mark Zuckerberg shared his thoughts on open source and how he believes that it will lead to major developments as it becomes the industry standard in the future, cataloguing it as a successful approach that Meta plans to stick with in the long run, and a positive platform for developers and the AI community.

⚡️Quick links

Model mixes AI and physics to do global forecasts

Want to spot a deepfake? Look for the stars in their eyes

Eleven Labs new Turbo 2.5 model 

Musk launches poll asking if Tesla should invest $5 bln in xAI, early votes in favour

Sakana AI drops image models to generate Japan’s traditional ukiyo-e artwork

Thank you for reading!

Hello, Starters!

As we've stated before, major companies in the AI league are shifting their focus to small models due to their cost-effective potential. This trend has been ongoing, and we expect it to keep growing... So, let's be on the lookout!

Here’s what you’ll find today:

• OpenAI introduces GPT-4o mini

• LLM safeguards are not as strong as we think

• Mistral and Nvidia present NeMo

• Apple released new DCLM models

• Big names in the industry focus on AI safety

• And more.

🤖 OpenAI introduces GPT-4o mini (4 min)

After hinting at its release in the LMSYS chatbot arena, OpenAI has finally made its new model official. We're now ready to experience the power of GPT-4o mini. There's much to be said about the model, but the focus remains on two key aspects: its size and its cost efficiency. Clearly, OpenAI acknowledges the industry trend of smaller models, as it makes AI technologies more approachable and affordable.

GPT-4o mini is now available for free to ChatGPT users as a replacement for GPT-3.5. With a context window of 128K tokens and updated knowledge up to October 2023, it stands as a strong competitor for similar models like Claude Haiku and Gemini Flash.

🔓 LLM safeguards are not as strong as we think (2 min)

There's a considerable flaw that affects most language models and could potentially lead to harmful outcomes. Researchers from the École Polytechnique Fédérale de Lausanne (EPFL) reveal in a study that it is easy to bypass the safeguards of AI systems by just rewriting malicious queries in the past tense.

The paper explains that by writing in the past tense, models like GPT-4o are likely to answer queries that in other circumstances would be blocked by safety measures. This leads to the belief that AI models consider past actions less harmful than future scenarios, a vulnerability that calls for attention as LLM technology keeps developing.

💥 Mistral and Nvidia present NeMo (1 min)

Mistral and Nvidia have joined forces to introduce Mistral NeMo, a compact model with 12B parameters and a context window of 128k, that is directed at researchers and enterprises. Its size allows it to have enhanced performance without the need for fancy systems, as it can easily run on Nvidia's RTX GPUs.

Mistral NeMo is multilingual, and recent comparisons show it surpassing models like Gemma 2 9B and Llama 3 8B.

Learn AI-led Business & startup strategies, tools, & hacks worth a Million Dollars (free AI Masterclass) 🚀

This incredible 3-hour Crash Course on AI & ChatGPT (worth $399) designed for founders & entrepreneurs will help you 10x your business, revenue, team management & more.

It has been taken by 1 Million+ founders & entrepreneurs across the globe, who have been able to:

• Automate 50% of their workflow & scale your business

• Make quick & smarter decisions for their company using AI-led data insights

• Write emails, content & more in seconds using AI

• Solve complex problems, research 10x faster & save 16 hours every week

Register & save your seat now (100 free seats only)

🍎Apple has already been in the small model trend for a while; however, they're stepping up to the game with new additions to its family of open DCLM models. They've recently released two versions with 7B and 1.4B parameters, which are securing great spots in most benchmarks.

🚨A group that includes familiar names like Amazon, Nvidia, Google, OpenAI, and more, is teaming up in the creation of CoSAI (Coalition for Secure AI). This project focuses on providing users with the necessary information, tools, and frameworks to increase the safety aspects of AI.

⚡️Quick links

Google’s Gemini AI will be all over the Paris Olympics broadcast

Nvidia preparing version of new flagship AI chip for Chinese market

Autonomous AI workers that talk to each other will arrive in 2025

Google Deepmind develops open-source AI to tackle biases in evaluating language models

OpenAI holds talks with Broadcom about developing new AI chip

Thank you for reading!