WEBVTT 1 00:00:00.100 --> 00:00:09.480 [scheerapparaat] Oh. Oh. Zo klinken 165.000 scheerbewegingen per minuut die je helpen om fris voor de dag te komen. 2 00:00:10.420 --> 00:00:18.620 De flexibele scheerkop van de Philips S9000 Prestige volgt de contouren van je gezicht en scheert tot op nul millimeter van je huid. 3 00:00:18.660 --> 00:00:29.000 Ervaar een ultiem gladde en comfortabele scheerbeurt, ook wanneer je nog niet helemaal wakker bent. Want dankzij de slimme sensoren past het scheerapparaat zich helemaal aan jou aan. 4 00:00:29.740 --> 00:00:39.320 Zo kun jij nog extra genieten van je kopje koffie voor je de deur uit gaat. De Philips S9000 Prestige is het ideale scheerapparaat om jouw dag goed mee te beginnen. 5 00:00:40.300 --> 00:00:51.000 Koop de Philips S9000 Prestige bij jouw favoriete winkel. Koffietijd voor mannen. De podcast waarin drie onbezonnen gasten je wekelijks een kijkje geven in hun zinderende studentenleven. 6 00:00:51.060 --> 00:01:00.540 We gaan geen onderwerp uit de weg en helpen je op het gebied van liefde. Zou je huilen tijdens de seks een, een, een afknapper vinden? Vriendschap en alle andere problemen die je tegenkomt in je studentenleven. 7 00:01:00.700 --> 00:01:14.320 Ze miste een beetje vo. [gelach] Elke maandag om 3.00 uur sharp op je favoriete podcastplatform. Ik heb met zoveel meiden getongd en met vier verschillende. Ik zo bram, dat was vier keer dezelfde. 8 00:01:14.340 --> 00:01:27.640 [piep] We luisteren deze podcast op eigen risico. Welkom bij Polky, de Nederlandse podcast over kunstmatige intelligentie waar we uitzoeken welke invloed AI gaat hebben op ons werk, ons leven en de samenleving. 9 00:01:28.240 --> 00:01:37.860 Bij mij is Wytse Hagen. Ik ben Alexander Klöpping en in deze aflevering hoor je een exclusief interview met een van 's werelds meest vooraanstaande experts op het gebied van AI, Wharton-professor Ethan Mollick. 10 00:01:38.440 --> 00:01:48.960 Omdat ik zijn nieuwste boek uitgeef, had ik de eer om te interviewen tussen zijn gesprekken door met wereldleiders en CEO's. Mensen die net als ik van hem willen weten: hoe ziet de wereld er straks uit met AI? 11 00:01:49.520 --> 00:01:52.220 En Tesla had vorige week haar ReRobot-event. 12 00:01:52.800 --> 00:02:05.220 Aangekondigd werden de Cybercab, een volledig autonome robottaxi zonder stuur of pedalen, een robovan, eigenlijk een soort van, uh, uit de plaat geslagen bus, maar dan in stijl van, nou ja, prachtig ouderwetse stijl, daar gaan we het over hebben. 13 00:02:05.720 --> 00:02:13.640 En een update van de humanoid robot die je in kan zetten om de afwas te doen thuis. Wytse is erin gedoken. Veel plezier met Polky. 14 00:02:23.740 --> 00:02:33.620 [muziek] Is het, uh, geworden wat je had verwacht ervan, Wytse, van het, van het event van, uh, Elon Musk? Ja, ik denk dat het aardig in de buurt kwam van wat ik had gedacht. 15 00:02:34.180 --> 00:02:39.100 Dat- Ik moest zeggen, ik zat naar het ding te kijken. Ik dacht: dat heeft Wytse toch knap zitten bespiegelen in zijn glazen bol. 16 00:02:39.140 --> 00:02:45.700 Nou, thanks, maar ik had echt wel, ik had me echt wel ingelezen en lopen, uh, graven in de krochten van leaks en dingen. 17 00:02:46.140 --> 00:02:53.560 En, uhm, ja, het ligt redelijk in de lijn der verwachting gezien de staat van andere robotica bedrijven en de andere zelfrijdende bedrijven. 18 00:02:53.860 --> 00:03:02.420 Het is wel trouwens interessant dat, uh, Waymo, waar we het over gehad hebben toen jij vroeg: joh, wat, wie zitten nu aan de frontlines? Daar is weer vijf miljard ingegaan vanuit Google. 19 00:03:02.740 --> 00:03:11.800 Of dat toevallig nu aangekondigd is, of dat zij ze, wou, even wouden zeggen: hey, wij zijn er ook nog, hè, en we gaan hartstikke goed, vind ik wel interessant. Zullen we er maar eens induiken? 20 00:03:11.840 --> 00:03:24.680 Wat is, misschien te beginnen bij de, bij de, bij de Cybercab. Een soort van Cybertruck, maar dan zonder pedalen en zonder stuur, waarbij het idee is dat je hem, uh, kan summonen en dan kun je erin gaan rijden. 21 00:03:24.840 --> 00:03:35.020 Wat, wat, uh, wat heb je gezien? Hij is klein. Het was al een beetje verwacht. Er zit eigenlijk alleen maar een, uh, een bankje voorin. Uh, het is een two person cab. 22 00:03:35.420 --> 00:03:44.020 Uh, ze hebben onderzoek gedaan dat de meeste ritjes blijkbaar met twee of één persoon is. Er is natuurlijk geen chauffeur, hè, dus, ja, als er twee stoelen zijn, kunnen er twee mensen in. 23 00:03:44.700 --> 00:03:55.920 Uh, en omdat hij wat kleiner is, het idee is echt, ja, dat het hele goedkope, in verhouding goedkope autootjes zijn die je in massa kan produceren en in die steden uit kan rollen. Ja, $ 30.000 moet ze gaan kosten. 24 00:03:56.460 --> 00:04:04.160 Ja, relatief goedkoop zou ik wel zeggen, hè. Uiteindelijk wor je naar een soort $ 10.000 toe. Voor een zelfrijdende auto. Ik schrik er niet echt van. Nee. 25 00:04:04.220 --> 00:04:08.600 En waarschijnlijk als je het deelt door het aantal ritjes dat hij moet rijden om zichzelf terug te betalen is het prima. 26 00:04:08.680 --> 00:04:20.390 Het is, uh, dan mag hij ook wel wat, een miljoen kilometer kunnen rijden, uh, zonder grote ingrepen. Maar oké, klein. Wat viel je verder nog op? Nou, de, het is wel interessant, want ik zat te kijken. 27 00:04:20.459 --> 00:04:24.140 Ik heb die video ook een paar keer teruggespoeld, dat ik dacht: wat zit er nou op die display? 28 00:04:24.180 --> 00:04:38.159 Ik kon dat, sowieso in, een beetje bijzonder dat er, wanneer Apple bijvoorbeeld een event doet, en Tesla eigenlijk ook in het verleden, is er daarna best wel een geavanceerde website beschikbaar waarop je gewoon even lekker kan scrollen om even, ja, even door die mooie 3D-renders te gaan. 29 00:04:38.460 --> 00:04:47.219 Dat is er nu eigenlijk niet. Die WeRobot-website ging een soort van uren later pas live na de livestream en uiteindelijk hebben we drie foto's, niet eens een video. 30 00:04:47.280 --> 00:04:56.580 Dus ik ben nog een beetje verward, dus ik ben vooral die livestream aan het terugspoelen. Niet de hele dag, maar ik heb het een paar keer gedaan, om te kijken, ja, hoe ziet die, hoe ziet dat dashboard er dan uit? 31 00:04:56.620 --> 00:04:57.520 Wat zit er op dat scherm? 32 00:04:57.580 --> 00:05:10.240 Maar het lijkt toch, en het geeft op zich ook niet, wel een, een, een soort Model 3 waar ze zoveel mogelijk hebben weggehaald om daar een soort heel klein, uh, uh, zelfrijdend ding van te maken met een groot tv'tje in het midden en daar moet het allemaal op gaan gebeuren, zeg maar. 33 00:05:10.350 --> 00:05:18.720 Best wel minimalistisch. En hoe gaat dit nou werken? Dan heb je een, een Tesla-app waarin je, waarin je er een kan bestellen en dan zoals een Uber-ritje er een kan aanvragen. 34 00:05:18.760 --> 00:05:21.140 Of hoe, hoe, hoe gaan we dit als consument gebruiken? 35 00:05:22.280 --> 00:05:32.930 Volgens mij hebben zij bedacht dat het vanuit de Tesla-app gaat, want ik begreep niet, uh, dealtjes met de andere grote, uh, taxibedrijven of, ja, de neotaxibedrijven dan, hè, zoals vroeger. 36 00:05:33.660 --> 00:05:40.760 Uh, dit moet een Tesla-ding worden, dus in de Tesla-app. Maar goed, ook dit, het zijn goede vragen. Het is allemaal nog erg vaag. Het is niet heel erg concreet. 37 00:05:40.780 --> 00:05:48.280 We hebben wel, er zitten in de, uh, ze doen natuurlijk, uh, quarterly reports, uh, omdat ze een publiek, uh, bedrijf zijn. 38 00:05:48.600 --> 00:05:55.180 Daar zitten dan pdf's bij met wat gave de-- een, een, een beetje een fancy deck met wat, uh, mooie 3D-renders erin. 39 00:05:55.520 --> 00:06:07.260 Daar hebben ooit wel eens screenshots in gezeten van hoe dat, uh, hoe die taxi-interface er dan uit zou zien om een soort van investeerders blij te maken. Ja. Dat is een soort Uber, een, een, een, uh, donkere Uber-app. 40 00:06:07.340 --> 00:06:20.276 Ja.Ja, ja, oké. Nou, dat busje hoeven we niet lang bij stil te staan. Het ziet eruit als, uh, ja, iets wat je, wat bij Tuschinski mooi zou kunnen voorrijden in Amsterdam. Zo'n, uh, prachtige art deco bioscoop. 41 00:06:20.286 --> 00:06:29.856 Het is een beetje een art deco versie van een bus, uh, waar twintig man in kunnen en, uh, waar je, uh. Maar is die niet juist, is dat niet juist zo raar? Want ik bedoel, er wordt, 42 00:06:30.776 --> 00:06:44.116 er, een van de commentaren op al die zelfrijdende auto's is van: is het, kunnen we niet gewoon beter openbaar vervoer hebben, uh, hè, in plaats van allemaal autootjes die dan op de snelweg aan elkaar gekoppeld worden als een soort asfalttrein. 43 00:06:44.436 --> 00:06:49.956 Ja, treintje, maar met die bus heb ik echt zoiets van, maar, maar nu, nu ben je gewoon een bus aan het maken. 44 00:06:49.976 --> 00:06:58.676 [lacht] Uh, ja, misschien moet je die dan gewoon aanbieden aan, uh, de connections van, uh, van Nederland of zo. Of, of Qbus. Die vond ik nog een beetje maf. 45 00:06:58.736 --> 00:07:05.316 Dat, dat voelt voor mij heel erg als een ding dat mensen binnen San Francisco van hun huis naar hun kantoor brengt. En dat, dat is het dan. 46 00:07:05.696 --> 00:07:13.596 Maar je, je hebt het de hele tijd over andere taxibedrijven en andere busbedrijven. Maar ik ga er toch van uit dat Tesla de hele keten gaat willen beheersen, toch? 47 00:07:13.616 --> 00:07:24.875 Dit is niet, het idee is, is neem ik aan niet dat je, dat Connection een Robo Van kan kopen, wel? Nee. Nou ja, nee. Maar ik bedoel tegelijkertijd de Semi, hè, dat is hun vrachtwagen. 48 00:07:25.416 --> 00:07:31.436 Die is dan weer helemaal wel gebrand door andere transportbedrijven. Dus het is niet alsof ze dat deal, dat soort deals nooit maken. 49 00:07:31.936 --> 00:07:44.896 En je hebt ook UFO, UFO, een taxi of een autohuurbedrijf die helemaal Tesla branded is, maar dan, hè, met een ander logo op Tesla's dus helemaal niet toestaan van een rebrand van hun eigen product. 50 00:07:44.936 --> 00:07:55.236 Nee, maar ik denk precies wat jij zegt. Het idee is dat Tesla dan ook de rol van Uber gaat spelen en niet de leverancier van Uber wordt. Dat is jouw punt, denk ik. En dat, dat is volgens mij waar. 51 00:07:55.436 --> 00:08:03.756 Maar goed, laten we naar de kern gaan en dat is hun, uh, humanoid, uh, robot. Ik zag filmpjes tegenover elkaar van de versie die ze drie jaar geleden lieten zien. 52 00:08:03.796 --> 00:08:07.796 Dat was iemand in een pak, uh, die dansjes aan het doen was. 53 00:08:08.176 --> 00:08:18.036 Wat toch een vrij verontrustende, verwarrende ervaring was van drie jaar geleden deze robot aangekondigd zien worden en dan daadwerkelijk een mens, uh, in die, in die dingen te laten zitten. 54 00:08:18.515 --> 00:08:29.976 Deze keer hadden ze een variant gemaakt waarin er daadwerkelijke robots waren die drankjes aan het maken waren, die selfie, waar je selfies mee kon nemen en waar je ook mee in gesprek kon gaan. 55 00:08:31.076 --> 00:08:35.836 Uh, en het werd gepresenteerd als in straks hebben we allemaal zo'n robot en die gaat de vaatwas voor je doen. 56 00:08:35.916 --> 00:08:42.476 En, uh, die kan de boodschappen uit de achterkant van de auto halen en die kan je in fabrieken inzetten om te werken eigenlijk. 57 00:08:42.516 --> 00:08:52.626 Het verhaal wat jij al een tijdje lang in deze podcast aan het, uh, aan het, uh, vertellen bent, dat werd hier in beeld gebracht. Uh, met een beetje fantasie zeg ik erbij. Uh, 58 00:08:53.896 --> 00:09:03.636 hoe, uh, hoe, hoe keek je naar dat shot waar, uh, waar, waar er twintig robots in een soort van Noord-Koreaanse formatie, uh, het, dat Hollywoodterrein opgelopen kwamen? 59 00:09:04.976 --> 00:09:11.646 Nou ja, die deed me heel erg denken aan, uh, de term visioneering. Dat is niet per se, uh, pessimistisch. 60 00:09:11.716 --> 00:09:23.216 Visioneering, dus engineering en vision, dat is een idee dat je zegt: je bent aan het engineeren, maar je gaat ook eigenlijk een visie neerzetten, dus daar mag dan wat smoke and mirrors bij zijn. Ja. 61 00:09:23.456 --> 00:09:35.036 Dus een deel van wat je laat zien is om het publiek te laten wennen, enthousiast te krijgen, hè, cynisch investeerdersgeld, maar meer optimistisch. Joh, die wereld moet erop voorbereid worden. 62 00:09:35.536 --> 00:09:47.116 We kunnen dan best wel in een ideale situatie, oftewel een Hollywoodstudio, hè, in een stadje gemaakt van karton, robots laten rondlopen die eigenlijk op afstand bestuurd worden door mensen met een VR-headset op. 63 00:09:47.166 --> 00:09:48.896 Want dat blijkt nu de realiteit te zijn. 64 00:09:49.386 --> 00:10:02.616 Maar dan nog, hè, want voor de luisteraars wat Alexander en ik hebben gezien in die live event is denk ik zestien of twintig robots, echt veel, die zonder, uh, tether, dus zonder kabel, uh, zonder mensen eromheen om te zorgen dat ze niet omvielen. 65 00:10:03.196 --> 00:10:13.516 Eigenlijk best wel, ja, serieus. Ik vond het wel impressive. Ik bedoel, ik had ze allemaal al alleen gezien in filmpjes en zo, maar als het dan daar in een livestream rondloopt op straat, het is echt wel anders. 66 00:10:13.856 --> 00:10:19.896 Dus ik denk dat dat- Tussen de mensen. Het was ook wel niet alleen maar de show, maar het was ook achteraf bij de borrel dat ze allerlei taakjes deden. 67 00:10:19.936 --> 00:10:27.496 Dus het was, dus wij, er is heel veel, veel beeldmateriaal uitgekomen van mensen die gewoon met die dingen interacteren en dat is toch best indrukwekkend. 68 00:10:27.556 --> 00:10:34.246 Ja, maar daar kregen we wel weer een beetje dat, uh, Rabbit R1 en, uh, Humane Pin effect. En wat bedoel ik daarmee? 69 00:10:34.756 --> 00:10:46.156 Is dat dan in het testen, hè, dus we zagen video, veel videootjes waar: kijk, ik heb een gesprek met zo'n robot. Dat ik denk, los of dat het een AI is. Ja, dat, dat vind ik wat minder interessant. 70 00:10:46.236 --> 00:10:53.116 Ja, ik kan ook bij mijn telefoon praten. Stop die speaker in een robot. Dus [gelach] dat is, ik bedoel dat, dat, daar kijk ik niet voor. 71 00:10:53.176 --> 00:10:58.116 Ik kijk voor: blijft dat ding staan tussen de mensen, uh, en kan die inderdaad een biertje tappen. 72 00:10:58.516 --> 00:11:07.975 Ik vind dat, dat vind ik dan het interessante eraan, uh, maar, dus een video dat iemand zegt: moet je kijken hoe menselijk die klinkt, dat ik denk: ja, goed, dat wist ik al, want Alexander en ik hebben dat live in de uitzending gedaan. 73 00:11:08.076 --> 00:11:13.456 Stop dat op twee poten en je hebt het ook. Dat bleek nou net het punt te zijn wat dus- Waar ze het gefaket hebben. 74 00:11:13.556 --> 00:11:20.856 Ja, dat waren toch teleoperated robots waar ik van denk, en daar ben ik niet de enige in, zeg dat gewoon, want het is nog steeds super bizar. Ja. 75 00:11:20.916 --> 00:11:30.896 Het, het feit dat er ergens nu in, in, in ruimtes mensen met VR-brillen op stonden en koptelefoons- Ja. Om die apparaten te besturen, dat is al, dat is op zichzelf toch al bizar. Ja. 76 00:11:31.036 --> 00:11:41.596 Uh, ja, fake dan niet een soort van ChatGPT daarin waar je mee kan praten. Het is zo'n harde dreigement, want dit is eigenlijk, dit is eigenlijk een beetje de, de, de vibe die rondom Musk komt te hangen. 77 00:11:41.636 --> 00:11:53.816 Als ik even de, de toch wel vette, zijn vette raketten, zijn rakettenhandeltje even buiten beschouwing laat, want dat is, ja, op, o-o-, dat is veel indrukwekkender zou ik zeggen, uh, zou ik zeggen. 78 00:11:53.856 --> 00:11:56.696 Daar is, daar is, daar is ook smoke, maar weinig mirrors. 79 00:11:57.466 --> 00:12:09.816 Uh, en in het geval van die, van die zelfrijdende auto, taxi, moeten we nog maar zien of die, die, die productiedatum 2026 gehaald gaat worden of die prijs gehaald gaat worden of überhaupt het gaat lukken om full self-driving te doen. 80 00:12:09.856 --> 00:12:11.056 Het zijn allemaal vraagtekens. 81 00:12:11.076 --> 00:12:27.880 En met die robots ookEn dan zitten we, moeten wij als soort van, nou toch mensen die, die het, het beste voor hebben, zou ik zeggen, met Tesla, die echt openstaan voor, uh, voor de, voor de innovatie en voor de, voor de technische, ja, knappe dingen die ze doen. 82 00:12:27.940 --> 00:12:36.660 Dat wij echt bijna, we moeten bijna een soort van, je krijgt allemaal vragen van mensen om je heen. Soort van: ja, ze hebben het gefaket, want iedereen hoort dat dan. Dat is dan wat blijft hangen bij mensen. 83 00:12:36.680 --> 00:12:40.999 En dan worden wij in een situatie gedwongen waarin wij moeten zeggen: ja, maar het is toch knap hoor. 84 00:12:41.040 --> 00:12:52.540 Want ja, ondanks dat ome Elon gefaket heeft dat die dingen praten, uh, ja, is toch wel knap dat die, dat die robot op zichzelf kan staan en, uh, weet je, dat er geen draadje aan hoeft te zitten, et cetera, et cetera. 85 00:12:52.680 --> 00:13:04.900 Moet je kijken hoe, hoe knap motorisch die vingers, uh, bewegen. Ja, het is, het is moeilijk om, om, uh, [lacht] het is moeilijk om, o-om, om het op te nemen voor, uh, voor de heer Musk. 86 00:13:05.780 --> 00:13:16.080 Ja, en ik denk dat in dat opzicht, uh, ik ben misschien net als jij ook een beetje in de war nu omdat er net weer gisteren een raket is geland die met twee stokjes uit de lucht gepakt is, terwijl dat ding zo groot is als een flatgebouw. 87 00:13:16.480 --> 00:13:26.100 Ik zat weer helemaal hyped. Dit wordt op maandag opgenomen, dus als je dit terugluistert, dat was afgelopen weekend. Ja. Ja, ik was helemaal hyped, uh, zaterdag dat ik dacht: uh, wat is er nou weer gebeurd? 88 00:13:26.580 --> 00:13:33.220 En dit was het. Ik ging het weer even allemaal geloven ook. Dat ik denk: ja, dit is toch wel heel gaaf, hè, weet je wel, zo. Maar goed, dan heb je dat event. 89 00:13:33.400 --> 00:13:37.320 Ik wil nog wel even terugkomen op wat je net zei, hè, met die robots en die smoke and mirrors. 90 00:13:37.940 --> 00:13:48.740 Wat ik wel had verwacht, wat er eigenlijk niet was, is die k-, om ook die laatste, uh, dat laatste puzzelstukje te leggen en een, desnoods was het een render geweest, vind ik prima, hè. 91 00:13:48.820 --> 00:13:57.720 Dus een volledige 3D-gegenereerde af, uh, video. Had even laten zien hoe die robots een deel van die taxi's in elkaar zetten. Had laten zien hoe ze de stoelen plaatsen. 92 00:13:57.980 --> 00:14:06.810 Had laten zien hoe die robots hem uiteindelijk schoonmaken met een paar doekjes voordat hij de fabriek uitrijdt. Had dat ook laten zien om, om, om, om, uh, die koppeling te maken. 93 00:14:06.860 --> 00:14:18.160 Maar wacht even, die humanoïde robot kan ook en wordt ook onderdeel van de productielijn, hè, want- Maar als onderdeel van de visioneering, want d-dit is niet hoe die taxi's nu gemaakt worden natuurlijk. 94 00:14:18.340 --> 00:14:25.330 Dat wordt gewoon door mensen in elkaar geschroefd. Als je dan een Bluetooth-speaker op twee benen zet, kan je net zo goed ook even dat filmpje laten zien. Ja, precies. 95 00:14:25.330 --> 00:14:30.660 Ik bedoel meer van, uhm, om, om die, om die visie te laten zien van: maar wacht even. 96 00:14:30.720 --> 00:14:43.080 En ook misschien zelfs Optimus in een andere, uhm, form factor, hè, waar we het over hadden voordat het event was van, had, had een andere, uh, een robot op twee wielen met hele grote armen in een fulfillmentcentrum laten zien. 97 00:14:43.160 --> 00:14:46.760 En dan had je gezegd: het is de, de Optimus Fulfillment 2 of zo. Mhmm. 98 00:14:46.800 --> 00:15:04.620 Want die, uh, dat is, vind ik het ingewikkeld, dat ik denk: nou, a-als Tesla daar straks een goed model in handen heeft, hè, dus dan heb ik het over het AI-model dat al die apparaten aanstuurt die in verschillende form factors, van auto's tot robots tot, uh, grasmaaiers, iets kan doen, dan is dat best wel substantieel. 99 00:15:04.740 --> 00:15:12.620 Ja. Als zij daar een beetje frontrunner in zijn. Dus ik had dat denk ik iets meer, uh, uh, daarop, uh, uh, gezet dan wat er nu, uh, uh, getoond is. 100 00:15:13.550 --> 00:15:31.120 Het was een visueel spektakel, uh, maar als dat zich moet vertalen tot, uh, tot publiekelijk enthousiasme of, uh, ja, investeerdersenthousiasme, dan zou ik zeggen dat dat medium gelukt is voor de hoeveelheid werk eigenlijk die er in de technologie is gestoken is en de aandacht in dit ev-, in dit event. 101 00:15:31.180 --> 00:15:40.229 Of zie je dat anders? Nee, nee, ik wou nog, ik, wa-, misschien heb ik het gemist hoor, maar ik, ik, ik kwam het niet echt tegen op NOS. Is toch een beetje de plek waar ik dan even doorheen ga scrollen. Ja. 102 00:15:40.240 --> 00:15:48.380 Ik heb, ik heb nu even gezocht. Ze hebben er wel wat over geschreven, maar het is geen frontpage geweest. Nee, zeker. Ik denk dat mensen, dat het in, in, uh, 103 00:15:49.580 --> 00:15:58.600 het is enorm gepolitiseerd sinds dat hij Twitter heeft overgenomen en ik denk echt dat dat invloed heeft op de manier waarop er over Tesla gepraat wordt en zelfs hoe er over SpaceX gepraat wordt. Ja. 104 00:15:58.640 --> 00:16:07.680 Dat, dat is toch echt een soort smet voor een heel groot deel van, van de mensen. Dat zou niet uit moeten maken. Ik, ik, ik heb net een boek gelezen die dat stelt. 105 00:16:07.720 --> 00:16:15.500 Je moet dat kunnen dicoppelen, heet dat, het, het, het uit elkaar trekken van het, de persoon en dan, uh, wat de persoon doet. 106 00:16:15.600 --> 00:16:25.180 Je moet die twee los van elkaar kunnen bekijken om, uh, een soort van interessante meningen te kunnen vormen. Uh, maar ik heb toch een beetje het idee dat, dat dat wel door elkaar loopt bij de meeste journalisten. 107 00:16:25.640 --> 00:16:31.280 Nou ja, dat zij zo. Zijn er nog dingen die je eraan wilt toevoegen? Of hebben we dit, hebben we dit wel, uh, gedegen behandeld? 108 00:16:32.479 --> 00:16:39.320 Uhm, nou, wat ik eraan toe wil voegen is, uh, d'r, d'r is dus niet echt een, een fijne marketingwebsite om te kijken wat er allemaal is. Ja. 109 00:16:39.800 --> 00:16:51.100 Uh, ik ben heel benieuwd wat er de komende maanden, uh, dan concreet naar buiten gaat komen en wat rond gaat rijden en of we video's gaan zien van, uh, robotaxi's in the wild. Ja. 110 00:16:51.140 --> 00:16:58.380 Dat gaat dan een beetje voor ons het antwoord geven op: hoeveel smoke and mirrors was dit nou eigenlijk? Ja. Ja, ja, ja, ja, ja, heel goed. Heel goed. 111 00:16:58.390 --> 00:17:01.620 [muziek] Straks het hoofdonderwerp, maar eerst een boodschap van onze sponsor. 112 00:17:01.860 --> 00:17:10.100 Datamakelaars verzamelen en verkopen jouw persoonlijke gegevens die vervolgens in handen komen van de hoogste bieder, van oplichters tot verzekeringsmaatschappijen. 113 00:17:10.120 --> 00:17:14.670 Maar gelukkig is daar Incogni, jouw digitale bodyguard, en met hen blijft je data beschermd. 114 00:17:15.040 --> 00:17:21.820 Incogni beperkt publieke toegang tot je privégegevens en zorgt ervoor dat jouw persoonlijke informatie niet zomaar kan worden doorverkocht. 115 00:17:22.000 --> 00:17:29.540 En daarmee verklein je het risico op identiteitsdiefstal en houd je de controle over je eigen data. Wil jij nou ook de baas blijven over je persoonlijke informatie? 116 00:17:29.580 --> 00:17:43.700 Bezoek dan nu incogni.com/pokepod en ontvang 60% korting op je Incogni-account. Dat is incogni.com/pokepod, omdat jouw privacy het waard is om te beschermen. Terug naar de show, het hoofdonderwerp. 117 00:17:43.960 --> 00:17:50.100 Uhm, ik zei het in de intro al, deze week een exclusief interview met een van de meest vooraanstaande experts op het gebied van AI. 118 00:17:50.480 --> 00:17:58.020 Uh, en ja, dit klinkt een beetje wc-eend omdat ik zijn boek uitgeef, maar deze man wordt, ja, we geven het boek niet voor niks uit, zullen we maar zeggen. 119 00:17:58.060 --> 00:18:10.880 Hij wordt wereldwijd gezien als, uh, een belangrijke stem in het debat over AI. En, uhm, ik heb een exclusief interview met hem gehad. Het had ook heel wat voeten in de aarde, uh, om met hem te mogen praten tussendoor. 120 00:18:11.020 --> 00:18:18.600 Uh, en dat is omdat de man onwaarschijnlijk, uh, druk is sinds zijn boek is uitgekomen en het boek in heel veel talen vertaald wordt. 121 00:18:19.140 --> 00:18:36.396 Uhm, ik heb een gesprek met hem gehad, uh, waar ik geprobeerd hebZo veel mogelijk thema's die wij in deze podcast, uh, bespreken van, uh, wat AI voor onze banen betekent, de economie, het onderwijs, de manier waarop we media consumeren, uhm, nou ja, langs te laten komen en zijn reflectie erop te krijgen. 122 00:18:36.956 --> 00:18:42.896 En Ethan vertelt ook wat hij hoort van medewerkers van grote AI-bedrijven als hij ze privé spreekt. 123 00:18:43.136 --> 00:18:50.556 Over het algemeen krijg je dan andere dingen te horen dan dat je met de marketingafdeling, uh, spreekt van, van die grote AI-bedrijven. Dus dat gaan we van hem horen. 124 00:18:51.036 --> 00:19:03.776 Uh, welke cutting edge AI-tools hij zelf gebruikt en welke nieuwe golf aan AI-systemen eraan zit te komen. En ik waarschuw je alvast, hij praat ontzettend snel. Dus in een half uur krijg je veel informatie te verstouwen. 125 00:19:03.866 --> 00:19:16.236 Ik, uh, zou het niet erg vinden als je hem op 0,75 luistert, maar dat moet je lekker zelf weten. Ik zou zeggen: veel plezier met dit interview met Ethan Mollick. Thanks for doing this, Ethan. Thank you for having me. 126 00:19:16.776 --> 00:19:21.156 Uhm, you are in frequent contact with people in the I-AI industry. 127 00:19:22.096 --> 00:19:32.266 I guess my first question is: what are the topics that the people, the discussion that these people are having that haven't yet reached mainstream media or public awareness? 128 00:19:33.116 --> 00:19:36.096 Uhm, I'm interested in what they are talking about to you. 129 00:19:36.256 --> 00:19:47.846 So what, what insights can you share about what engineers or professionals at AI companies are currently thinking about that haven't yet reached, um, the public awareness? That's a terrific question. 130 00:19:48.376 --> 00:19:57.336 Uh, and I would say generally the answer is not-- They're, the things they're saying publicly, but I think people don't believe them when they say it publicly. So I'll emphasize privately- Mm. 131 00:19:57.386 --> 00:20:06.236 ...that they believe, whether or not this is true or not, that they have found the key to scaling up to somewhere near human level or beyond intelligence, just doing what they're doing now with more computers. 132 00:20:06.686 --> 00:20:11.706 They keep saying that to everybody, and everyone's like, "Eh, they might be making it up." They might be making it up, but they believe that. 133 00:20:12.026 --> 00:20:21.036 And the second thing is, you know, that they also think that agents, the idea of AIs that are autonomous and can do stuff, is right around the corner. So these are things that they firmly believe. 134 00:20:21.056 --> 00:20:26.646 Now, they, they are telling everybody at every meeting about this, but I think everyone's like, "Oh, they're just trying to raise funds." They are- Mm-hmm....true believers. 135 00:20:26.696 --> 00:20:34.226 Not everyone, but enough people that I talk to are true believers that I take that as a signal. I don't necessarily think they're right, but I also don't necessarily think they're wrong. 136 00:20:34.756 --> 00:20:46.776 What should we, we, we take from this as a society? 'Cause on, on one end, you can be cynical about it and say they just say that for fundraising. But let's think for a second about what it would mean if they are right. 137 00:20:47.196 --> 00:20:56.436 What are we underappreciating here, uh, when we just read accounts in the newspapers? Well, I mean, one of the major messages in my book is that you have to just use these systems. 138 00:20:56.496 --> 00:21:04.196 Like, it's a really complicated, uh, way of basically trying to show you it's not that hard and it's kind of fun, and the implications will become clearer to you as you use them, right? 139 00:21:04.556 --> 00:21:11.836 And this is a general purpose technology, so it's gonna have very uneven effects. It's not gonna change everything everywhere all at once. It's gonna make changes in different places. 140 00:21:12.476 --> 00:21:23.985 It depends on how it interacts with humans. And so I, I think that too much of our anxiety is either around the idea of, um, long-term science fictional harms, which enough people are worried about. 141 00:21:24.006 --> 00:21:29.736 I mean, the Nobel Prize winner is worried about AI becoming super intelligent and murdering us all. So we should obviously spend some worry about that. 142 00:21:29.796 --> 00:21:37.276 But I think that that tends to take away the focus from the fact that we're also living through, regardless of what happens in the long-term future, a short-term exponential change. 143 00:21:37.656 --> 00:21:43.736 And adjusting that requires getting involved in this. This is not a hard technology to use, weirdly. It is very easy to use. 144 00:21:43.796 --> 00:21:52.196 If you are a good teacher, a good parent, a good manager, a good interviewer, you're gonna be able to figure out ways to use this technology because it's like kind of working with a person. 145 00:21:53.356 --> 00:22:01.116 And y-you talk about, about this in the book where you say the moment that th-this thought really hit you was the moment that you had three sleepless nights. 146 00:22:01.596 --> 00:22:08.536 Combining that idea and, and taking the doomerism aside for a moment, let's, let's not, n-not, not do that. 147 00:22:08.656 --> 00:22:18.276 But if you, if you combine the sort of sense that these people at AI engineer-- at AI companies share with you privately, uh, combined with this, this idea, 148 00:22:19.736 --> 00:22:27.186 w-what do you, what do you envision for the next couple of years? What scenarios do you have in your mind where you really, you know, that, that you really 149 00:22:28.196 --> 00:22:36.536 t-think, think a lot about and, and, and don't think is, is being, uh, talked about enough? Uh, right. Uh, and I end the book with some scenarios. 150 00:22:36.676 --> 00:22:43.616 I, I would say just, uh, on the three sleepless nights issue, we're starting to get some empirical evidence that that's true, that people when they first use AI are kind of unhappy. 151 00:22:44.196 --> 00:22:50.036 Uh, and then after they've used it enough, they have started to feel good. So there is this kind of like one-two punch, um, of that. Um, 152 00:22:50.976 --> 00:22:55.476 I would say, you know, there are four scenarios at the end of the book about what could happen in the future, and one of them is a static future. 153 00:22:55.656 --> 00:23:02.896 The other is, uh, linear improvement, exponential improvement, which is what we've been seeing, or else, um, you know, AGI, a machine smarter than a human at every task. 154 00:23:03.446 --> 00:23:15.516 And what I would say is, since I wrote the book, scenario one of sta-a static world has actually become less likely. Mm. Part of the reason for that is two things. One is more evidence of what's called the scaling law. 155 00:23:15.536 --> 00:23:23.936 It's not really a law, it's an observation, but the idea that the more computing power and the more data you put into a sys- uh, into an AI, the smarter it gets. That seems to be holding. 156 00:23:24.416 --> 00:23:31.636 So that means the, for the next couple generations of AI systems, the next, you know, two to four years, we're gonna keep seeing exponential improvement. 157 00:23:31.776 --> 00:23:42.536 And then there was this breakthrough, um, OpenAI announced a new model called o1. And what o1 did that's really interesting is they showed that if you just let the AI think for longer, it comes up with better answers. 158 00:23:42.596 --> 00:23:48.616 Mm-hmm. So what that means is that you don't have to just scale this by throwing more data at it. You could also just give it more time to think. Mm-hmm. 159 00:23:48.716 --> 00:23:53.596 And that means that it's harder to see an endpoint to this technology in the near future given that revelation. 160 00:23:54.856 --> 00:24:01.116 I know that in your book you emphasize that we shouldn't look at AI progress with a doom set-- doomsday mindset, so I tried to take that into account. 161 00:24:01.136 --> 00:24:08.036 But it's, it's-- a-and you say that it's more helpful to, to think about how it can assist us and how it can benefit us. 162 00:24:08.676 --> 00:24:17.216 But let's explore that perspective a little bit later and f- and for now, I, I think a lot of people, on a lot of people's minds is consider-- concerns about the job market. 163 00:24:17.266 --> 00:24:29.956 And economists always note that past technological innovations leading to job loss wereAlways compensated by the creation of new jobs. Do you believe that that pattern will continue this time? 164 00:24:29.996 --> 00:24:39.716 And do you feel that the economy is equipped sort of to handle the growth of more advanced AI systems in the short term? So I don't think we know. 165 00:24:39.776 --> 00:24:49.556 I mean, I think there is a little bit of Pollyanna-ish, like looking to the positive side, we're economists, that jobs will always be increased through technology. And that's been true for physical labor replacement. 166 00:24:49.716 --> 00:24:55.366 We don't know if it's true for intellectual labor replacement in the same way. It has been for small scale in a lot of cases, but not always. 167 00:24:56.076 --> 00:25:04.586 Um, and, you know, the other thing I point out is like living through the Industrial Revolution wasn't exactly a, you know, whatever the appropriate expression is for, you know, a bag of chips for something great. 168 00:25:04.856 --> 00:25:13.356 Mm-hmm. Uh, because like, you know, there was massive job displacement. People moved towns, cities rose and fell. Like, even if things are good, there can be disruption. 169 00:25:13.496 --> 00:25:19.145 I think part of the issue is the disruption is going to be very uneven here. It's going to happen across society in different ways. 170 00:25:19.176 --> 00:25:29.186 And, I, I mean, let's be clear, the AI companies themselves have been very explicit that they think they'll be able to replace most human labor, so we'll all be able to do other stuff, and the AI will make infinite bounty for all of us. 171 00:25:29.236 --> 00:25:36.896 Like, we don't really have a sense of what that science fiction future looks like. I think over the next three to five years, what we'll see is rolling change in industries. 172 00:25:36.936 --> 00:25:40.576 And a lot of that comes down to organizations choosing what to do. 173 00:25:40.756 --> 00:25:53.436 I, I often warn companies that if they think about this as technology for the last five or ten, you know, uh, for a few decades has been about replacing labor. So if I get a, you know, higher efficiency, I fire people. 174 00:25:53.476 --> 00:25:53.676 Mm-hmm. 175 00:25:53.816 --> 00:26:05.686 I think that doesn't work when you're also on the edge of a technological boom, which is, you know, a, a, the image I like to give people is let's say that you were a Dutch beer manufacturer in the early 1800s, and, um, along comes steam power. 176 00:26:05.886 --> 00:26:06.596 You have two choices. 177 00:26:06.616 --> 00:26:16.626 You can either fire most of your people, make more money per barrel of beer as in your local community, or you can expand and go worldwide and hire 100,000 people and be the equivalent of Guinness or something. 178 00:26:17.076 --> 00:26:27.356 I think that companies that are more forward-looking about what the future possibilities of this technology can do will gain a lot of benefit. But I don't know if that balances out job loss. Yeah. So nobody really knows. 179 00:26:27.576 --> 00:26:38.076 It sounds, sounds like a coping mechanism in some way that we tell ourselves that. If, if we believe what the engineers are telling you, then, then it might be a little bit more dire. 180 00:26:38.276 --> 00:26:44.396 A-and you must take this into account when, when you're, when you're thinking about this at night. Yeah, I mean, I do. 181 00:26:44.556 --> 00:26:54.736 I, I-- you know, part of this is I think that we're underestimating technology, but I think there's a lot... And one of the key findings of the book that I discuss is the idea that AI is jagged. 182 00:26:54.796 --> 00:27:03.396 It's really good at some stuff and really bad at other things. Mm-hmm. So I don't know if you've played with, um, and this being a podcast, have you played with NotebookLM yet? Sure. Right? I did it. 183 00:27:03.496 --> 00:27:13.196 I fed my entire book into Notebook, um, and it did a really nice podcast based on it. But, you know, and as great as that is though, there's still a lot of like detail missing that would have been useful or important. 184 00:27:13.276 --> 00:27:16.816 Will it fill in the detail just as it gets smarter? We don't know a hundred percent at this point. 185 00:27:17.416 --> 00:27:25.916 So I think there's a jaggedness there that means not a total replacement for humans, but some fields it will beat us, just, just like it does in chess right now. Mm-hmm. 186 00:27:26.156 --> 00:27:35.236 I, I think there is some coping and some hiding our head in the sand, but I also think that total doom feels unlikely as well. Sure. So we're gonna live somewhere in the middle. We'll muddle through. 187 00:27:35.616 --> 00:27:43.356 I'd like people to muddle through in a positive way where they're showing positive impacts of this technology. It's gonna have a lot of good and bad. Like, we can't be close-eyed to the bad stuff. 188 00:27:43.396 --> 00:27:51.836 But there is agency on the good stuff as well. What, what skills and sort of competencies do you advise companies to, to gain in this? 189 00:27:51.876 --> 00:27:59.276 What are, what are essential skills in this AI-dominated, uh, economy that we're looking at for the next couple of years? So 190 00:28:00.276 --> 00:28:08.156 I think the thing that gets overemphasized, I discuss in the book a well-- as well, is the idea of prompt crafting or prompt engineering, that you have to get really good at prompting the AI. 191 00:28:08.896 --> 00:28:15.116 I think that that has limited value and will be decreasing in value as these systems get smarter. So they'll just be able to do things for you. Mm-hmm. 192 00:28:15.816 --> 00:28:22.276 I think that where we're gonna see real value in increase, um, is actually people who are experts at their job. 193 00:28:22.316 --> 00:28:31.996 That's the skill right now because you-- if you're an expert, you can work with the AI for a few seconds and really learn quickly. Is it any good at this? Is it bad at this? Where does it tend to lie to me? 194 00:28:32.096 --> 00:28:42.845 Where is it-- where are, you know, where is it getting better? Where is it not getting better? And is it a multiplier of your expertise? Hmm. I think that people are too eager to try and find some secret to AI. 195 00:28:43.056 --> 00:28:50.676 The real secret, I, you know... I've become very convinced since the book that one piece of advice I gave turned out to be exactly the right advice, which is just use the damn thing. 196 00:28:50.776 --> 00:28:58.016 Ten hours is my minimum amount of time, and if you do that, if you can push through to ten hours, you're as good an AI expert in your field as anyone's going to be right now. 197 00:28:58.556 --> 00:29:11.316 You look at it like it's sort of like a multiplier effect on your, on your, on your expertise. Yeah. Well, and a replacement. I mean, jobs, a-and we discuss in the book, jobs are, um, are bundles of tasks. 198 00:29:11.436 --> 00:29:19.576 You do more than one thing in your job, right? So, you know, you do, you, you know, run a podcast, you run a publishing company, you do all the, like, you do a lot of things all at once. 199 00:29:20.236 --> 00:29:28.676 Some of those job bundles will be replaced by AI. Some will be supplemented by AI. Whatever-- for right now, whatever you're best at, you're almost certainly better than AI at that. Mm-hmm. 200 00:29:28.726 --> 00:29:30.756 So does the AI do the tedious parts of your job, right? 201 00:29:31.116 --> 00:29:37.996 Would, would it have arranged for the two of us as we tried to find a time to talk when our two agents have just talked to each other and worked that all out without worrying about it, right? 202 00:29:38.436 --> 00:29:46.776 And, you know, would that have-- that would have changed your job bundle and mine, but would that have been a good use of, of our, you know, of our time that we gave up? So I think there is change that's gonna happen. 203 00:29:47.356 --> 00:29:51.826 Um, and I think some of it's gonna be more profound than others. Like, I am worried about some giant sectors of the economy. 204 00:29:51.916 --> 00:30:04.956 Customer service call centers are something to be concerned about because AI's pretty good at customer service right now, and there's a lot of money in it, and it's a low, you know, it's a low prestige field for a lot of people, and there's a lot of, you know, effort to replace them. 205 00:30:05.076 --> 00:30:10.235 I think stock photography is another area that's gonna take hit. So there's some areas that are gonna take more damage than others. 206 00:30:10.336 --> 00:30:19.996 I think on the other hand, people are saying Hollywood's gonna be destroyed, don't really know how much effort goes into making a movie that tells a coherent story, and that when you're the top person in a field, it's not easy to do. 207 00:30:20.096 --> 00:30:35.055 So I mean, I think that it is a co-intelligence, really, and a, a multiplier for people by and large.Sometimes I walk in, in this district in Amsterdam, it's called the Zuidas, and all the financial companies and lawyers and, uh, the banks are situated there. 208 00:30:35.676 --> 00:30:45.636 And sometimes I walk through these cro-crowds, and they are not customer service people. They're the, they're the high-paid lawyers. And sometimes I think, like, are they still gonna be here in five years? 209 00:30:45.656 --> 00:30:56.056 What do you think about, um, those high, high-paying jobs, consultants, that, that kind of stuff? Is- It is a really interesting question, right? 210 00:30:56.196 --> 00:31:01.996 Like, there are things that-- So, you know, in this, one of the studies I talk about in the book, we actually do this with a high-end consulting company. 211 00:31:02.056 --> 00:31:06.436 We find the AI is at the eightieth percentile of doing consultant tasks and lots of tasks. 212 00:31:06.456 --> 00:31:14.136 However, there's also lots of glue and intuition and other pieces that may be more flexible than the AI requires, so the job of consultants changes. 213 00:31:15.156 --> 00:31:26.236 Law, I don't know how law is done in, in, um, in, in, in, uh, your country, but in the US they charge by the hour. Mm-hmm. Um, for, for-- that's how legal firms work. Yeah, they have the tendency to do that here as well. 214 00:31:26.436 --> 00:31:32.676 Yeah. That's going to have to change because it turns out, like, right now, all the client-- When you talk to a big law firm, all their clients are saying, "Don't use AI." 215 00:31:32.996 --> 00:31:41.996 That's gonna switch very quickly to, "If you're not using AI to do all the basic law stuff, why are you charging us for this?" But then the remaining work the AI doesn't do becomes very high-value work. 216 00:31:42.536 --> 00:31:49.936 So, you know, I'm more worried about the pipeline of new people into these jobs than mass unemployment from people currently in these jobs. Mm-hmm. Right? 217 00:31:50.056 --> 00:31:58.936 Because when I talk to companies, what's happening right now is, you know, how do you learn to be a lawyer or, you know, or a banker or a consultant? 218 00:31:59.056 --> 00:32:07.366 You go to a s- you go to school and we teach you to be a generalist, and then you go to the, to that company, and you end up becoming an apprentice, right? 219 00:32:07.536 --> 00:32:13.355 You do repeated work over and over again while people yell at you if you do things wrong. You gain the experience. And, um, that's how you learn the job. 220 00:32:13.656 --> 00:32:23.996 Now, what's happening right now is everyone's delegating that intern work to AI 'cause AI's already better than your interns, and your interns are also all using AI to give answers back to you, and it's completely severed the talent pipeline. 221 00:32:24.716 --> 00:32:33.296 That, to me, is the bigger concern that I have about the next generation of talent. Is this, is this something that you, you think about a lot when you're teaching your students? 222 00:32:33.376 --> 00:32:42.296 How, how they, in, in times where they don't-- [chuckles] where I guess they need the experience but feel that they don't need the experience because ChatGPT can do it for them. 223 00:32:42.856 --> 00:32:52.466 What, what, what, h-how do you get them to, to do the rough handiwork? Well, so school will solve the problem. It's a cha- it's a chaotic mess right now, right? 224 00:32:52.896 --> 00:33:01.246 So I, we, I don't have studies from the Netherlands, but we know in, like, Denmark and the US basically, everybody's using AI. All the students are cheating, right? Cheating is not detectable. Same here. 225 00:33:01.256 --> 00:33:06.276 It's all falling apart. That'll-- We'll, we'll get through that, right? My classes are a hundred percent AI classes. 226 00:33:06.536 --> 00:33:13.716 Other classes, like if you need to learn a foreign language, you're going to end up doing a lot more in-class work and writing essays in class. Like, we'll solve the problem in class. 227 00:33:14.396 --> 00:33:20.236 The question is what do we do about the on-the-job training piece that used to be apprenticeship? And I think it has to be two things. 228 00:33:20.336 --> 00:33:27.776 Companies either have to start to figure out how are they going to build an actual education system rather than pretending they have one, right? Or hoping it just happens. 229 00:33:28.136 --> 00:33:37.196 Or we'll have to start taking that into universities more. And I don't know which two answer will, which of the two answers will follow. Don't you think there's a role for AI in sort of simulating these experiences? 230 00:33:37.376 --> 00:33:46.656 Oh, absolutely. And I've got a-- If you, uh, google, uh, Moloch and simulation, there's a few you can play with right there. They're all free. Um, yeah, I mean, I think it can supercharge education. 231 00:33:46.736 --> 00:33:54.976 Once we're done destroying it, we'll be able to rebuild it much better. I mean, it is a good tutor. It's a good tool. The problem is that students use it in a way that's lazy and hurts them. Mm-hmm. 232 00:33:55.006 --> 00:33:56.796 I mean, I speculate about this in the book. It turns out it's true. 233 00:33:56.836 --> 00:34:09.776 There's a nice study by my colleagues in Turkey that finds that students who are given AI to use and just use it naively to try and solve problems, um, you know, to help them with a, they do much better on homework but much worse on the test 'cause the AI just does the work for them and they think they're learning. 234 00:34:10.556 --> 00:34:20.476 If you do this in a more directed way with a tutor, with an AI tutor, you end up doing better on the homework and don't take a hit on the test. So there is a future here, but it can't just be kind of naive use. 235 00:34:22.316 --> 00:34:28.156 You, you, you focus on the, the, on universities and, and sort of education to prepare people for the job market. 236 00:34:28.676 --> 00:34:40.436 But one step before, maybe in high school or in elementary school, what, what, what-- You're very certain that, that first under sort of education gets crushed and then gets built from the ashes again. 237 00:34:40.616 --> 00:34:49.596 How, how, how do you feel like a high school curriculum will, will look differently when, uh, when we build it up again? Curri- I mean, so curriculum is harder than pedagogy. 238 00:34:49.636 --> 00:34:53.815 And by the way, cri- I'm being a little overdramatic, right? [laughs] Teaching changes very little over the centuries. 239 00:34:53.856 --> 00:35:01.656 Like, most of us are just, most teachers are just gonna ignore AI and hope it goes away, and you'll do badly the test if you use AI, and they'll sort, it'll sort out the way it always sorts out, right? 240 00:35:02.196 --> 00:35:09.586 Um, I do think that, you know, we've known long before AI, for example, that lectures are a bad way to teach, right? Right. Or at least a bad main way to teach. 241 00:35:09.616 --> 00:35:17.756 You need to do active learning, people who participate in the learning. Like, so this will hopefully push us to doing more of that flipped classrooms where- Mm... you learn outside of class and apply inside of class. 242 00:35:17.796 --> 00:35:26.216 So I'm a little facetious, a little joking when I say destroy all of education, but it is undermining a lot of assignments and other kinds of things. Those are rebuildable, right? 243 00:35:26.276 --> 00:35:34.656 The curriculum question is a much bigger one. My argument is still that we need to keep learning what we've learned because you need a base of expertise to build knowledge on top of. 244 00:35:34.816 --> 00:35:39.565 So you still need to know basic math and, you know, and, and basic literature and history. 245 00:35:39.596 --> 00:35:49.096 Those things might become even more important, especially humanities, because it turns out the AI is trained on all of our cultural heritage, and if you know the cultural heritage, you can make the AI do much more interesting things. 246 00:35:49.376 --> 00:36:02.796 Mm-hmm. And how do you think you can motivate people to, to do that in these new times? I mean, I think that part of this is trying to get people to actually face what this thing is. 247 00:36:02.936 --> 00:36:09.276 I mean, not to loop back to the idea that I told you before of that you need your ten hours with this, you need your crisis, but I think that's the key. 248 00:36:09.656 --> 00:36:20.135 Like, people are naturally averse to kind of-- Once you've seen the shape of the hyper object, as they joke about online, once you've seen what this thing looks like, you start to realize, "Oof, things are gonna change," and then you, they're motivated. 249 00:36:20.716 --> 00:36:24.136 But otherwise, I think we like to pretend that nothing is happening. Mm-hmm. 250 00:36:24.836 --> 00:36:42.264 Sometimes I, I feel like, you know, my, my job changed a lot because I can now dictate something and then let, and Claude rewrite the text like it's me and-I can sort of nudge it in some way so that something comes out that is quite presentable and to me looks good enough. 251 00:36:42.284 --> 00:36:51.244 But I have this in, built-in sense of experience, sort of being able to judge what the thing comes up with to say if it's okay, yes or no. 252 00:36:51.694 --> 00:37:09.324 And I'm thinking about kids that, you know, are twelve right now and then grow up not having that sort of, um, the, the reference, the reference of knowing what is good or what isn't good, what is good context, what is, you know, not just the facts. 253 00:37:09.404 --> 00:37:20.184 It's not just the sort of basic curriculum that I'm, I'm worried about. I'm worried about sort of what their, how their judgment is going to form, uh, for the, for the stuff that they do in school. 254 00:37:20.424 --> 00:37:29.504 How do you think about that? I mean, that's where I was talking about building that base of expertise being so important because we do need those basics. You cannot learn to be good at your job. 255 00:37:29.544 --> 00:37:39.424 You would not be a good interviewer if you hadn't had a huge amount of background in reading material and understanding and talking to people, understanding what's interesting, what isn't, how to pose a question, how to learn. 256 00:37:39.944 --> 00:37:45.324 So that's-- the great thing is we have students in school. We can get them to do this stuff. That's what school is supposed to do. 257 00:37:45.744 --> 00:37:56.144 It might mean taking that more seriously, but we do have to kind of grind people through those basics, right? The question is then what happens afterwards? How do you get that last mile of skill of practical use? 258 00:37:56.164 --> 00:38:05.394 And that may need to be something we bring into schools as well. Okay. I'm trying to make the AI sort of this, this sort of thing that's happening to humanity now. 259 00:38:05.524 --> 00:38:09.544 I try to make it super concrete because it makes it possible to sort of visualize it. 260 00:38:09.584 --> 00:38:23.124 And education is one of those things where I think you help to visualize it by saying we can now flip the classroom, and we can do these mentoring things that we always knew would be good for our kids, and we couldn't because of the, the lack of people. 261 00:38:23.144 --> 00:38:30.924 And I try to zoom in on a different-- c-couple of different parts of what society is. Now, media is also one of those parts that really shape us as a, as a society. 262 00:38:31.324 --> 00:38:35.124 How do you, how do you envision media consumption, consumption changing? 263 00:38:35.164 --> 00:38:45.584 And do you sort of anticipate that, that synthetic media, like the stuff being generated by NotebookLM, will play a significant role in the, in the consumption that we have? 264 00:38:45.623 --> 00:38:59.654 And in what way do you expect that to, to change? So I think that is a fairly profound question. Um, I think that right now, right, and it's not just NotebookLM, it's also all the fake information on there. 265 00:38:59.764 --> 00:39:08.934 Like, I can create, you know, fake images of anybody. We are-- we're already living in a problematic, troubled information environment, right? Everyone's in their own bubbles. No one's interacting with each other. 266 00:39:09.954 --> 00:39:14.084 I am worried that this underlines that rather than anything else. You get like a source of truth. 267 00:39:14.144 --> 00:39:24.824 Now, on the other hand, the AI, um, there's a study that shows that the only way to-- the only thing we actually found that robustly lowers belief in conspiracy theories is actually talking to the AI. Yeah. 268 00:39:24.924 --> 00:39:32.024 Um, it lowers conspiracy theories three months later. And that also suggests it's really good at persuasion and for people with deeply held beliefs, which is also troubling. 269 00:39:32.064 --> 00:39:39.004 But, like, this is a good learning tool, right? It is a good tool for interrogating things. It's very good at pulling out information. 270 00:39:39.324 --> 00:39:48.304 I, I-- you know, this feels very cheesy to say, but I think that it really is about how we as humans use these systems. And that's what I'm pessimistic about in the information environment. Yeah. 271 00:39:48.494 --> 00:39:54.244 Like, it would do a really good job. Explain the pros and cons of all of the, of this politician's arguments. It will do a nice job with that. 272 00:39:54.584 --> 00:40:03.204 Help me think through why someone might support them or why they might not support them. Like, give me empathy. It's great at those kind of things. Are people gonna ask? That's a separate question. 273 00:40:03.284 --> 00:40:12.024 Yeah, but that's an important question, though. [laughing] What do you think? I agree. And that's a, that's a person question, right? Not an AI question. I don't know. 274 00:40:12.084 --> 00:40:36.844 There's-- people are part of, of a, a system with economic incentives and, you know, with companies, um, you know, optimizing for profit and, and, um, when you just take that into account, it-- and, and, and combine it with the technical sort of advances that we can now make, having an TikTok feed that's completely synthetically s- generated is not a 275 00:40:37.784 --> 00:40:42.174 weird thing to think about anymore. No, that, that's going to happen. That's a, that is a bad thing that is going to happen, right? Yeah. 276 00:40:42.244 --> 00:40:48.284 Um, you know, we-- uh, but what I'm saying is we've already built the systems that are gonna supercharge the bad effects. Exactly. Right? 277 00:40:48.584 --> 00:41:02.124 So the issue we have is that we have an engagement-based economy and a personalized-based economy, and we're about to set off a, you know, uh, supercharge that with AI. And that, that is ominous. So [laughing] 278 00:41:02.244 --> 00:41:11.134 I guess my question is- So I can't-- I, I, I'm not claiming I can solve-- like, I think some of these are long-standing sets of issues, right? And some of these are, are, like I said, social problems. 279 00:41:11.344 --> 00:41:25.924 So in this event, what, you know, the logical answer, and we're filled with a world of deepfakes and no common understanding, let's return to traditional media and, you know, listen to the people who were the authorities twenty years ago because they at least were-- had a professional standard. 280 00:41:26.324 --> 00:41:33.124 I mean, that's one solution. Do I think that's gonna happen? I, I don't know. The polit-- you know, it doesn't feel like we're gonna move back from that partisan. 281 00:41:33.204 --> 00:41:43.544 Maybe people just descend into complete bubbles and-- but that's so annoying that they end up dealing with each other in real life more and reject this AI world. Maybe, you know, the-- like, I don't know, right? 282 00:41:43.644 --> 00:41:51.684 I don't-- like, what I'm saying is that there's a policy piece to this too. There's a deeper set of questions about what this means and how we handle it. Mm-hmm. 283 00:41:53.544 --> 00:42:03.504 Like, the cr-- I can point out the crisis, and I can give pathways to solutions, but, like, this is where agency matters. I really worry about people viewing AI as something happening to them. Hmm. 284 00:42:03.644 --> 00:42:11.984 And to some extent, no one asked for this technology. I totally get it. But we also get to shape what this is. The AI labs do not have a view of how this is going to affect media. 285 00:42:12.004 --> 00:42:19.664 They just haven't thought, they never thought about how it's gonna affect education. This is all news to them. So who gets to shape this? Well, if we can start showing them how to do this stuff right. 286 00:42:19.744 --> 00:42:22.004 I mean, you know, it's funny, you talk about NotebookLM, right? 287 00:42:22.104 --> 00:42:29.784 I've spoken to a couple, um, radio hosts and interviewers like yourself, and most of them are like, "This is pretty good, but it messes up a few things that I would definitely do differently." 288 00:42:29.804 --> 00:42:35.304 More, you know, and there's-- So, like, we should be seizing control of that and saying, "Okay, how do we build this kind of tool?" It's not, not even... 289 00:42:35.344 --> 00:42:43.544 These are not-- We have the ability to do these kind of things now. These technologies are not that hard to use. And what I deeply worry about is that we're abrogating our responsibility to shape that world. 290 00:42:43.584 --> 00:42:52.684 So you're asking me, who has nothing to do with media, right, what would I do? You know, what would I do? The real question is, okay, so you are a national government. You are a national broadcaster. 291 00:42:53.204 --> 00:43:02.848 You are, you know, you have the head of TikTok in front of your congressional committee.You have to make some, take some agency. People in TikTok need to decide, is this good or bad? 292 00:43:02.908 --> 00:43:14.508 Or decide if they're gonna quit en masse. Like, we have to make some decisions, and I, I agree with you, it won't all work out for the best if we don't make choices. Okay, let's see. 293 00:43:14.708 --> 00:43:23.538 We have five minutes left, so I'm gonna, uh- These are great questions, by the way. Thank you, sir. So let's switch gears a bit. How do you-- 294 00:43:24.438 --> 00:43:36.367 To, to end with a very practical, uh, practical way of looking at this, and maybe phili- Oh, sorry. With a philosoph- [laughs] No worries. I mean, what is there? Okay. Say it again. Bless you, sir. Thank you. 295 00:43:36.428 --> 00:43:41.818 Let's switch gears a bit. How do you personally integrate these tools in your daily, uh, daily work? 296 00:43:41.908 --> 00:43:50.988 What is the, what is the stuff that feels cutting edge to you that you, you, that you toyed with and it makes you really excited in the last couple of weeks? 297 00:43:51.088 --> 00:43:54.848 I mean, so, you know, we've talked about it, and, you know, you are an early adopter. 298 00:43:55.008 --> 00:44:01.588 So let's just go back and say the things we've already mentioned, the things you've mentioned that you're using already are going to blow people's minds. 299 00:44:01.648 --> 00:44:11.078 So if you have not used NotebookLM from Google, you should do that. It is a great tool where you just upload documents. Please not an illegal copy of my book, but you can take anything else you want. 300 00:44:11.078 --> 00:44:20.708 [laughs] Uh, and load that in, and then it will create a radio show for you. It will let you interact with those documents. Absolutely mind-blowing. One of those things that moves, moves the needle for people. 301 00:44:20.788 --> 00:44:29.648 I think advanced voice mode for ChatGPT just came out, and I should make a good point. I talked about AI labs, I'm not paid by any of them, so this is not endorsements, right? This is just my current feeling. 302 00:44:29.748 --> 00:44:41.358 Um, if you haven't used advanced voice mode yet, which is, um, if you get the premium access to ChatGPT, you can use it. That will, I think, probably ma- have a big effect on you. And I think- How do you use that? 303 00:44:41.358 --> 00:44:45.128 How do you use the, the voice thing from, from ChatGPT? Right now, I mean, I, I talk to it when I'm hands-off. 304 00:44:45.328 --> 00:44:51.198 I, I'm not the kind of person who builds a relationship with AI, so I'm not that interested in having deep, meaningful conversations with it. Mm-hmm. 305 00:44:51.198 --> 00:44:58.288 But I find it very-- it's a very fun way to get some quick interaction going on, right? And to ask questions and other stuff. It's like what you wanted Siri to be. 306 00:44:58.408 --> 00:45:06.977 Now, other people I know have very deep conversations with the AI. I've talked to a, a quantum physicist at Harvard who talks to the AI and says he gets all of his good ideas talking to it. 307 00:45:06.988 --> 00:45:12.558 I'm like, "Is it good at physics?" He's like, "No, but it's really good at asking questions." Really? So you'll have to figure out your own use cases, right? 308 00:45:12.928 --> 00:45:22.618 And then the third tool I would recommend to everybody is just try Claude Sonnet, because if you like words, you're gonna be very impressed by how just clever it is, right? Mm-hmm. 309 00:45:22.968 --> 00:45:32.688 And by the way, we're on the cusp of a whole new range of AI tools being released, and we've got pretty heavy rumors going on. So I-- The-- One of the things about AI is that it's generational. 310 00:45:32.748 --> 00:45:36.768 The larger the AI systems, the stronger they are. It takes a while for a new generation to keep up. 311 00:45:36.888 --> 00:45:50.148 We now have a whole bunch of what I call generation two AI systems, like GPT-4, like Claude Sonnet, and generation three will be released soon, which will be like GPT-5, um, Claude, uh, 3.5, Opus, a bunch of other of these tools. 312 00:45:50.188 --> 00:46:00.088 What do you expect of these? What are the rumors? I mean, the rumors are- Let's talk rumors, Ethan. The rumors are exactly what you'd expect, which is that these are better systems. 313 00:46:00.148 --> 00:46:08.008 I mean, the interesting thing I'm running into, and we saw this with this o1 release that just came out- Mm. This high-end system. Like, a lot of people are like, "Well, I don't know what to do with this. 314 00:46:08.088 --> 00:46:11.708 I don't have that many problems that require a PhD student to solve my problems for me." 315 00:46:12.257 --> 00:46:24.148 [laughs] So like, you know, one of the world's best mathematicians, Terence Tao, who, who's ever lived, said that, you know, he views o1, the new model, as, um, a, a mediocre but not incompetent grad student in math. 316 00:46:24.438 --> 00:46:28.597 And speaking as one of the best mathematicians who's ever lived- It's, it's a compliment. This is not a meaningless statement, right? Yeah. 317 00:46:28.628 --> 00:46:36.988 So like part of the question I have is like, okay, these systems get smarter and smarter. You know, like does it change your life if you already find the model smart enough? That's a hard question. 318 00:46:37.288 --> 00:46:49.248 I guess the assumption is that then it can, it can not make those mistakes anymore, that it can be m- so creative that we, you know, are actually surprised when a thing comes up with, with something that- But, but that already is happening, right? 319 00:46:49.278 --> 00:46:56.878 Mm. I mean, first of all, mistakes are overrated as a concern. Like, hallucinations are a real issue, but if you're using it as a co-intelligence, it's less of an issue because you know when it's BS'ing or not. 320 00:46:56.928 --> 00:47:04.728 My standard's always been best available human. Is it smarter, or is it more or less accurate than the best human you have access to? And that's something, again, that you learn how to use. 321 00:47:04.768 --> 00:47:12.788 And I think the other kind of, you know, the other question about this stuff is like, okay, it's already beating humans at creativity tests. For most people, it's more creative than them already. 322 00:47:13.208 --> 00:47:15.507 How has that changed their lives? So that's what I think about a lot. 323 00:47:16.887 --> 00:47:30.448 Do you have also this tendency that, that you combine, you know, the progress that you see with, um, NotebookLM, with the stuff that you see from the video generation, Pika, the Chinese models that we now see generating these, these videos? 324 00:47:31.138 --> 00:47:45.338 The amount of, uh, you know, the, the limitation of the, the lag time in the way that we talk to AIs. It feels like we can see this path where there's different building blocks that's, that are now in isolation- Yeah... 325 00:47:45.348 --> 00:47:54.328 progressing. But there's gonna be this path where it combines, and you combine with, with just real world applications, with people doing jobs or leading their lives- Yes... 326 00:47:54.338 --> 00:48:05.888 or being entertained or communicating with friends. What do you feel is sort of underappreciated, um, when you combine these things, when, when people talk to you in interviews? 327 00:48:06.068 --> 00:48:19.088 What, what, what do, what don't people get? I don't think they get how much on the cusp we are of AI doing meaningful, obviously meaningful work as a stand-in for people. Like one of the things that makes these... 328 00:48:19.168 --> 00:48:24.468 Like people still think of this like software, where what it does is you need to write all these APIs and code around it. 329 00:48:24.528 --> 00:48:32.488 What the companies are building towards is something they can see and talk and interact with the world and click buttons on a mouse and follow directions. 330 00:48:32.508 --> 00:48:40.008 And the, as you said, I mean, and I wanna point out, you are far more advanced. Anyone who's listening to this is like, "What are you talking about?" Please try the kind of tools we talked about for a while. 331 00:48:40.088 --> 00:48:48.378 You will start to get the picture that, uh, that Alexander's bringing to the table, 'cause I think you're, you're on the cutting edge here, what you're telling me. And I think that's the image I'm seeing too, right? 332 00:48:48.408 --> 00:48:59.238 Which is these pieces are starting to come together in a way that feels like working with an actual person, as opposed to chatting with a chatbot and then feeding it a picture and then having a, you know, and then having an image generator work. 333 00:48:59.268 --> 00:49:10.956 It's all supposed to clearly become one thing soon.Is there something that you wish could have been in the book that, that hasn't, uh, been in the book? Uh, like, uh, is it already time for, for a second book, Ethan? 334 00:49:11.036 --> 00:49:16.436 Your book is selling incredibly well here in the Netherlands. What, what i-, what is missing from the book? That's wonderful to hear. 335 00:49:16.516 --> 00:49:23.756 I don't think, you know, so because we've been in generation two for a while, I don't think anything is fundamentally different. I think more evidence has come out supporting the points of the book. 336 00:49:23.956 --> 00:49:35.036 I haven't found anything that I have written that I would change because it's inaccurate. I think I, un-- I think I did not do enough to see how much of a role agents will play in the future. 337 00:49:35.076 --> 00:49:43.206 Autonomous AI systems that I mentioned those, but I think that I thought they were further away than they are right now, and I think that they're gonna be closer than we think. 338 00:49:43.776 --> 00:49:54.256 I think the other thing that, as I mentioned earlier, is I thought the scaling law might run its course sooner than it has, because when I talked to people a year and a half ago or whatever, I was writing the book they were more pessimistic about. 339 00:49:54.516 --> 00:50:01.936 It was more divided about whether the AI could, you know, keep getting AI smarter by just adding more computing power. That question seems to have gone away, and the answer seems to be yes. 340 00:50:02.016 --> 00:50:12.696 So I think that if anything, I think I underestimate the level of acceleration that we might be facing. You've been very generous with your time. Thank you so much for doing it, Ethan. Thank you very much. 341 00:50:13.516 --> 00:50:26.136 [muziek] Dan nog een tip voor de luisteraars. Uhm, een essay met de titel Machines of Loving Grace. O, ja. Uh, na-- ja, van, uh, Dario Amodei. De co-founder van Anthropic. 342 00:50:26.836 --> 00:50:32.236 Yes, en dat, dat is eigenlijk, de subtekst is: How AI could transform the world for, for the better. 343 00:50:32.836 --> 00:50:40.846 Een optimistisch verhaal over AI heb ik ook wel eens nodig en ik werd wel uitgedaagd door deze essay en de discussies eromheen. 344 00:50:40.846 --> 00:50:50.736 Het is best wel, het heeft een discussie gecreëerd van: oké, inderdaad, dit zijn ook allemaal mooie zaken als het gaat om medicijnen uitvinden en mensen uit stomme banen halen en dat soort dingen. 345 00:50:51.316 --> 00:51:02.536 Uhm, ja, ik, ik, ik tip hem even ter uitdaging voor de cynisten en ter, uhm, [lacht] bevestiging van de optimisten. Stuur dit door in je omgeving als er mensen bezig zijn rondom dit onderwerp. 346 00:51:02.616 --> 00:51:13.776 Het is een leuke, uh, uh, lijst van mogelijke dingen die gaan gebeuren die wel heel erg goed en leuk zijn. Ja, nou, ook helemaal in het verlengde van het gesprek wat ik met, uh, met Ethan Mollick had. 347 00:51:14.296 --> 00:51:28.076 Uh, dat artikel, waar kunnen mensen dat vinden? Uh, in de shownotes. Uh, het staat ook op de website van Dario Amodei zelf en dat is darioamodei.com. Oké, en het, tie-- artikel heet, voor als mensen dat willen googelen? 348 00:51:28.496 --> 00:51:38.676 Machines of Loving Grace. Ja. Uh, naar, en het, uh, gedicht All Watched Over by Machines of Loving Grace. Ook een tip. Kan je zelf even googelen. [muziek] Nou, heel goed. 349 00:51:38.796 --> 00:51:52.316 Uh, het boek Co-intelligentie kun je bestellen op co-intelligentie.nl. Dit was Pokkie. Met dank aan Danny Vermeulen voor de edit. Uh, vergeet je niet te abonneren op onze nieuwsbrief. Dat kan via ai-report.email. 350 00:51:52.336 --> 00:52:00.916 Het was hem weer, Wytse. Volgende week zijn we weer met een gewone aflevering. Het was een beetje een rare aflevering, maar kan ook wel een keer. Als we er maar gewoon iedere week zijn. Daar gaat het om, toch? 351 00:52:00.956 --> 00:52:09.216 Nou, uh, sorry hoor, ik vind dat we enorm veel waarde-- ik bedoel alleen maar te zeggen: we hebben geen, uh, [Wytse lacht] geen nieuwsaflevering gedaan. [lacht] Sorry. Ik ben geen excuses aan het maken hier. 352 00:52:09.316 --> 00:52:19.636 Oké, wacht, ik heb dat interview helemaal nog niet gehoord. Ik ben net zo fresh als de luisteraar, dus dat wordt vast een goed interview geweest. Ga jij echt de podcast terugluisteren of wil jij hem ook horen? Oké. Dag! 353 00:52:20.376 --> 00:52:20.616 Dag.