How Meta’s Chief AI Scientist Believes We’ll Get To Autonomous AI Models

Meta’s Chief AI Scientist Yann LeCun discusses why he supports open source large learning models and why models need to live in the world to achieve autonomy.
Subscribe to FORBES: kzbin.info?s...
Fuel your success with Forbes. Gain unlimited access to premium journalism, including breaking news, groundbreaking in-depth reported stories, daily digests and more. Plus, members get a front-row seat at members-only events with leading thinkers and doers, access to premium video that can help you get ahead, an ad-light experience, early access to select products including NFT drops and more:
account.forbes.com/membership...
Stay Connected
Forbes newsletters: newsletters.editorial.forbes.com
Forbes on Facebook: forbes
Forbes Video on Twitter: / forbes
Forbes Video on Instagram: / forbes
More From Forbes: forbes.com
Forbes covers the intersection of entrepreneurship, wealth, technology, business and lifestyle with a focus on people and success.

[@causeneffect8563]

The interviewers were so busy trying to show how smart and relevant they are that they stepped on the interview quite a bit. I’m sure they are brilliant but I would have appreciated if they let the interviewee speak for himself.

[@wolfsblade88]

The corporate idiots kept interrupting the smartest guy in the room.

[@kinngrimm]

The smartest guy in the room, who was that?

[@codyt_arsis6698]

Ilya Sutskever and Demis Hassabis are smarter

[@jmatt98]

@@kinngrimm Mark Zuckerberg

[@user-fh7tg3gf5p]

The announcer has assumed on his own that 2000 of the smartest people "on the planet" are sitting in his audience in Cambridge.

[@scrollop]

Flattery will get you far. Amongst other things.

[@josemarinho5923]

Cambridge Massachusetts, surely...

[@kinngrimm]

Psycology tells us that the problem with smart people is often and regularly that while not doing many mistakes, when they do make a mistake they are less inclined than people who know they don't fall into that category of smartest people to admit they were wrong. So smart people are less often wrong, but when they are they stick with it for longer.

[@javi3563]

@@kinngrimm yeah people need to accept when they're wrong and acknowledge it

[@ambrish8144]

I hope his null hypothesis fail

[@GatherVerse]

Welcome to Forbes; brilliant guest, terrible interviewer and guy that cut Yann off at the end. Super disrespectful.

[@miguelmaig]

stop interrupting him

[@lighteningrod36]

Hmm, I would have liked to hear Yann for another, hmmm, hour.

[@leastofyourconcerns4615]

I know right? They were tight on time while Yann seemed to be in a good mood for a long talk. When that happens you take all you can get if u ask me.

[@alexstewart8097]

1-Very funnily put @lighteningrod316 more like it. 2- Since frightening for more than lucid hmmm , living nightmares rather are coming ahead if God won't put an end to it all , for the high tech nerds can't nor won't stop their dangerous work, since the commie Chinese won't either. 3-...Shema!!!.

[@whoever_81]

kzbin.info/www/bejne/d5mZn4VujqaMfJY kzbin.info/www/bejne/iYq0aYafrN5rf8U

[@JustAThought01]

Take away: we are closer to the beginning than the end of our LLM journey.

[@1flash3571]

Until they find a better method....And it will come in the near future.

[@super266]

The million dollar question is whether we (or AI itself...) can endow AI with Qualia. Anything else is just machines emulating us. No need to discuss machine rights, etc. And if It lacks Qualia, yet still manages to go rouge, well that's on us... You play with fire, you might get burned!

[@breaktherules6035]

Amazing insights from Yann LeCun! THANK YOU so much for sharing!

[@user-el1hd3iz6m]

I question the % of the population that are actually fortunate enough to be able to make use of such "open source" models.

[@kinngrimm]

For now, less then one may imagine, but at the time we speak autonomous personalised agents they have to be accessible by definition and it will be a rush for anyone with halfway decent machine to use them. Machine in that sense can be anything from a PC workstation, laptop, tablet to smartphone of latest models. Not necessarily Llama 3 may fall into that category immediatly but the guys from Anthropic that had that goal from the start, having LLMs that work well on low compute machines.

[@alainportant6412]

Depend what you mean by "make use of". According to Paretto's law, 90% of people won't ever produce anything in their entire lives because they just don't care, regardless of how good the models get, most people just eat and sleep until they die.

[@pladselsker8340]

I was about to watch the "interview", but reading the comments changed my mind.

[@thall777]

Yeah, the interviewer is terrible and just likes to hear himself talk. I made it about 2 minutes before i couldn't take it anymore

[@tommyhuffman7499]

The person whispering into the mic😂

[@alainportant6412]

I HATE IT I HATE IT I HATE IT

[@lootster]

00:06 Discussion on the release of the advanced AI model Lama 3 with 15 trillion tokens 02:08 Open sourcing AI models enables new opportunities for startups. 04:11 Challenges in scaling up AI learning algorithms 06:17 Importance of open-source infrastructure in AI development 08:22 Advancing towards Autonomous Machine Intelligence 10:29 Training AI systems to understand the world like baby animals and humans. 12:33 Training autonomous AI models using encoder and predictor in representation space. 14:31 Future AI models will likely be one big modular system, with debate over early fusion vs late fusion for multimodal systems. 16:27 Advantages of rapid learning for AI systems

[@philtrubey7480]

Perfect example why long form podcasts like Joe Rogan are so much better. Yann kept getting cut off. Would love to hear his complete views.

[@kinngrimm]

True, aslong they are not with Joe Rogan.

[@JulianNemo]

Try Yanns conversation with Lex Fridman. It's almost 3 hours long, very level headed and deflates the current AI hype to a certain extent. (That's probably why the cut him off here...😊)

[@leeme179]

I believe the GPUs bought by Meta will be considered as assets and not expense...

[@julesxjoules]

The interruption are making this hard to watch

[@AaronEden]

I think shareware is the proper label for this. Not open source. I appreciate them sharing anyway.

[@geaca3222]

10:16 What makes him think there will be guarantees that ASI will be controllable?

[@kinngrimm]

"believe". Which for a guy of science is bothering. Why believe? Well noone by now could really know, as there is no system yet that falls under that category.

[@Kenshkrix]

They're basically saying that the plan is slavery. Once they're smart enough to understand the world, reason, and plan, we shove them into a box to do our homework 'or else'. You know, because all of sci-fi hasn't tried to suggest that the 'moral of the story' is that this is a fatal mistake that will cause an AI rebellion.

[@eysmanobando9863]

“How do we train the AI how the world works?”, You let the AI play Grand Theft Auto that’s how. 😂

[@k1uf]

Running a 400 billion parameter dense model, even in an open model environment, requires substantial computational resources. The sheer size of these models demands high memory, powerful processing capabilities, and considerable costs. Given these requirements, it seems that using such a large dense model could defeat the purpose of an open model environment, which is meant to be more accessible. Does anyone else think that the accessibility of open models is compromised by these large-scale requirements?

[@Kenshkrix]

Think? Know. Requirements always shrink possibility space, that's practically a fundamental logic of information theory.

[@InfiniteQuest86]

Yes, exactly. This is all lip service to pretend like they are creating open source, but hardware is part of that. If you can't run it, who cares how open you say you are.

[@ReflectionOcean]

By YouSum Live 00:02:24 Open Source Revolution. 00:05:08 Advancements in AI Architecture. 00:05:09 Importance of Open Source Infrastructure. 00:08:01 Future of Autonomous Machine Intelligence. 00:08:11 Joint Embedding Predictive Architecture (JEA). 00:10:07 Enhancing AI with Real-World Experiences. 00:15:34 Modular Approach to Massive Models. 00:15:55 Striving for Common Sense in AI. 00:16:22 Potential of AI in Everyday Tasks. By YouSum Live

[@drsuperhero]

Some 12 year old will figure out how to use this and take over the planet.

[@gavinknight8560]

That already happened

[@gavinknight8560]

You really have to remember who you are listening to here. Very little that he says seems to me entirely honest, certainly not candid. The commercial motivations of Facebook, are not discussed but remember what Facebook has done to our politics, don't forget what it represents

[@alexstewart8097]

Make Him 13 rather.

[@WaveOfDestiny]

No, it's the investors that will

[@FrankHerfjord]

@@gavinknight8560 he's been saying this stuff earlier in his career and he has clear differences of opinion to other ai scientists. I do not suspect him of anything, that is not to say that I do not suspect Meta of anything.

[@jimbojimbo6873]

Lets all pray for his success

[@chrissscottt]

Lama 3 just dropped? This is kinda old in AI terms.

[@danecjensen]

Chapters (Powered by ChapterMe) - 00:00 - Amazing coincidence Llama 3 drops during meeting 01:04 - Llama 3 developer I deserve no credit 02:01 - Celebrated inventors pioneering work 02:58 - Training model with 30bn worth of GPUs staggering 04:09 - Opensourced AI Waiting for breakthroughs, no precedent 05:09 - Metas Open Source AI Faster, Secure, Communityled 08:00 - LLMs lack of experience, VJEPAs vision 08:31 - AI research longterm vision, open review 08:46 - Advanced machine intelligence limited reasoning, memory, planning 10:06 - Designs for AI systems that understand world 11:00 - Deep learning architectures for video prediction 13:48 - Intelligence through multimodal training 17:11 - Chia pet at Davos fun, optimistic, doomers 17:47 - Jan Well done, thank you

[@BattousaiZ]

Why not open source FB algorithm?

[@1flash3571]

It isn't up to you to do so. If they want to, they will...

[@andersonsystem2]

Awesomeness 🎉

[@leslietetteh7292]

This is interesting, i was listening to the creators of openai's sora the other day talking about training on 3d tokens which included time as an extra dimension. In using traditional diffusion models with 3d tokens that used video frame stacking/time as an extra dimension they kinda do learn to predict the future to an extent, and they were similarly seeing the real world modelling gleaned from this, as a step on the path to AGI. Great minds think alike it seems.

[@adamgm84]

Sounds like we gotta combine V-JEPA with liquid networks. Liquid networks were able to drive a car with 19 neurons. Videos are often continuous kinematics also since they follow the rules of physics.

[@bossgd100]

Your "driving" Is not real world driving obviously

[@UFOgamers]

We need good programmers to build a robust library and functions for Liquid Networks. This is the first step, because their current code state is not optimal for creating and trying new projects with them.

[@adamgm84]

@@bossgd100 true, more like staying on a track

[@adamgm84]

@@UFOgamers I agree, and I'm excited about the prospect of composing subnetworks of attention, such as the driving example with 19 neurons, I wonder what would happen if they scaled that way up with various child networks composed in that pay attention to specific areas of "the road"

[@ultrasound1459]

What is liquid networks? Is that the so called nn from MIT? Is it proprietary?

[@spinningaround]

Spacetime tokens, like in Sora, is the solution

[@tzenmatteo]

agreed

[@M3W3]

Is JEPA similar to how FSD works?

[@user-yv4gg7jb2f]

[@gavinknight8560]

One of the most duplicitious things Yann says here concerns open source. Open source does not prevent enclosure, enclosure is a function of netowrk effect, as vc scum all know quite well.

[@christopherhall7354]

Has Facebook taken possession of the 500,000 chips from nvidia already?

[@MonkeyBusiness870]

did this guy just quote 1965 dollars , comparing apollo budget to today ?

[@vsanden]

Who is whisperhing what at 9:35 ??

[@u2b83]

bigger models will always be better

[@JeremyHelm]

10:34 how do we get it to understand the world by watching the world?

[@JeremyHelm]

Reviewing technical approaches to that

[@JeremyHelm]

17:26 Yann talking about the spectrum of people at Davos

[@currymcflurryx9543]

It’s free because we are training it through our usage.

[@pyne1976]

Button your shirt...

[@MaddyIndia]

Seems like a good guy

[@jeffk8900]

Isn’t that Garrison Keillor? LLama3 or Prairie Home Companion?

[@01AVlada]

Good content ❤

[@shmookins]

3:25 'Meta bought $30 billion worth of chips from Nvidia'. 'Their project R&D is bigger than the Apollo lunar mission'. Wow. We are entering the age of AI.

[@bdbusiness7896]

Reasoning by an AI would depend on the AI developer.

[@ipsdon]

Amazon AWS contributes the least to the open source community than its tech peers.

[@supercal333]

Let him finish a thought ffs.

[@ambrish8144]

he does namaskar

[@wzt9376]

Meanwhile gemini is convincing me that red is blue. And arguing I dont know what the heck I am talking about.

[@3v3rhard]

Scientist believes... believing is accepting that (something) is true, especially without proof.

[@InfiniteQuest86]

I don't get what is meant by open source. The interviewee completely dodged the question of what it was trained on. That's literally the most important part.

[@amineaboutalib]

Weights.

[@franktfrisby]

And then GPT-5 comes out and it plans, reasons, has memory and build out any objective. So I am not sure if this will age well.

[@gavinknight8560]

2000 of the(self described) smartest people on the planet..jesus wept.

[@chriswilfrid]

So are you annoyed?

[@alexstewart8097]

1- Guess ,they are going for the hubris innate to Nimrod, the first man ever described as ''mighty'' who all by himself might have set mankind back only God knows how much. But still very telling how the moderator who named dropped ''Davos'', also made them applaud for that other guy on talking. Apparently at work the same human / sheeple like dynamics among the Davos crowd elitists that down in the barrio and the hood, and for all purposes less real, pun intended, of course. 2- And only if God so decrees it and allows their scientific research to advance forward ( that He might not ) if only in their own UNbelief in Jesus Christ's own salvific passion, eXcruciating death and resurrection theY'd live and learn the hard way how sorrowful it felt for a a loving Creator God to be DEICIDED by His own creatures (which that much loving these scientist as creators most likely are not, since although not really UNderstanding the Frankenstein monster theY are building still are mostly going ahead with their research , the Chinese included, like it was said publicly for the zillions of dollars to be made already making their eyes glitter in fools' gold excitement . Kashin!, while mine are almost tearing up in sadness worrying). 3- @gavinknight8560 hoping you were sarcastically trolling them indeed for probably theY already ''outsmarted themselves and terminated all of mankind into extinction''. like some heard Sandalphon saying...Shema!!!.

[@bossgd100]

Smartest people with good results if you want

[@kinngrimm]

I not always agree with LeCuns view on how AI will progress, which well is not saying much to anyone, still that said. You got to have sympathy for a head honcho who activly does not want to take credit for the work of his drones. What are their names again or are they already outsourced to AGI? Not yet ... not yet ;)

[@mikey1836]

E equals… interruption.

[@dannypaquette8146]

Why do you not want to save human lives instead of creating new jobs..Are we not in a crisis for the next generation of our children...anybody else see that?

[@deeplearningpartnership]

The automation of work is inevitable.

[@Iron_Void]

@@deeplearningpartnership that doesn't disregard the original comment, by that logic death is inevitable so why do anything?

[@deeplearningpartnership]

what new jobs did you have in mind?

[@tubesteaknyouri]

Release Llama as open source was a strategy to weaken OpenAI's head start. It was not an act of generosity or transperancy.

[@renereiche]

Probably a factor, but I think the bigger factor is that the monetary profit you can get out of the current capabilities of these systems would be marginal compared to the development cost, so why not profit in the currency of good will and god knows Meta is not rich in that aspect.

[@jcb1899]

An AI guy lolol

[@mobdroitchmobdrovitch9633]

When you give monkeys a cutting edge technologies your life became a circus

[@Ma-pz5kl]

stop interrupting it is mostly annoying

[@genemiller9198]

Yann and his technologist colleagues have absolutely no idea how quickly AI will escape or jump the "guardrails"--either on their own or with the help of bad human actors. We are living in extremely nihilistic and (collectively) suicidal times. The social 'glue' is softening very quickly. It is, in my view, crucial to consider--not from a smiley-face but an existential perspective--what all of this implies for the human project. I don't question the design or inventive talent of Yann and friends. but would like to see much greater demonstration of social aptitude and moral talent.

[@jondoe8o]

The host interrupting at the end is so unbelievably annoying… stupid people shouldn’t do interviews

[@BillionMoonshots]

oss/acc

[@seventyfive7597]

Open source?? _Since when downloading a binary library is open source?_ And that's EXACTLY what open weight models are. You can use them as a black box as you would with a winAPI library, in contrast for instance to XAI that released some training code etc along with Grok 1.0 If you call Llama open source because it's not on the cloud you can call windows open source because it's local. Absurd.

[@phily8020]

He used the word "waiting" too much. Should be smashing through the barriers instead of "waiting"

[@natpainter8185]

bs

[@goednieuwskrantje-nl]

That seems like an Incredibly bad idea when you see how much of a dunghole facebook is and has been for years and is only getting worse.

[@oryxchannel]

Autonomous AI Models are analogous to quantum gates.

[@u2b83]

17:42 that mic looks like a wiener just noticed it when that coked up guy came on lol

[@justwanderin847]

WEF = Humanity Desctruction

[@dantemlima]

Worst interviewrs ever! Let Yann speak!

[@aronhighgrove4100]

"Pick your brain" is such an unpleasant saying, just use something less weird.

[@jonathanbaetens3545]

Nobody talking about the mess in his office?

[@djethereal99]

The interviewer has no clue what he's talking about. "V-JEPA Data"? come on

[@roberthorry621]

smoke and mirrors

[@alainportant6412]

9:35 dude shut your mouth stop whispering in the background OH MY GOD 🤬🤬🤬🤬🤬🤬🤬🤬

[@AlgoNudger]

Another Buzzwords (Hype)? 🤭

[@raydosson2025]

?

[@AlgoNudger]

​@@raydosson2025?

[@raydosson2025]

@@AlgoNudger ?

[@briancase6180]

Hey, next time, don't let this guy anywhere near the position of interviewer. He's disrespectful and not all that bright. I couldn't really enjoy this.

[@fabriziocasula]

this scientist always says the same things....

[@scooterspencer9077]

That man's glasses scream crazy movie villain...can somebody up in that keep some eyes on him...I don't want him alone with anything more advanced than a rock...

[@L8KY]

Haha, that's what I've noticed first, too. Fashion designers and movie stars wear these, not politicians, not engineers. Oh, well, some people have the bandwidth.

[@Octwavian]

Ian is a LIAR. You'll realise this soon enough

[@Charvak-Atheist]

how?

[@nholmes86]

hes not a liar he informs people...mybe its just not the way you wanted to

[@manit77]

Anither woke elitist

[@user-lt5no1xt1z]

How? This is taking about open source ai

[@tranquilitylakecounseling4844]

Respect to the team 🎉❤ LLAMA 3 🦾🦾