Nice to hear a different voice and opinion on all these developments. Definitely makes me look different at Meta as company and AI player.

Excellent interview/conversation... appreciate Yann's ability to communicate his personal story and story of the AI community. The Interviewer is well informed and did not throw softballs -- it was an elevated convesation

I'm not noticing "people getting smarter."

I am fully with Yann LeCun's in getting LLM distributed to the public. But I am slightly disappointed in his arguments. He seemed not very strong in the regulation side of things.

Good luck regulating Open Source models. 😂

An advanced AI also has agency. It does not have to be deployed to gain control. It can gain control over those who have the power over whether or not it is deployed.

Quality information, good to report on this!

Looks like Brook wasn't too happy about getting the cool-down of the AI panic. THANKS for a really helpful interview.

Plot twist: Yann LeCun is a AI.

Yann LeCun’s a legend in AI, no doubt, but in this interview, he kind of downplayed how AI misuse could be a real problem. It’s key to remember he’s works for Meta, so maybe take his super chill view on AI risks with a grain of salt.

Great interview !

Considering the risks to society and culture that Meta has already spearheaded with relatively 'dumb' social engineering algorithms, his dismissal of people with concerns about AGI as neo-luddites is chilling.

Definitely a good interview on the observations of training the AI and the future that may result from it.

Yann is certainly a likeable guy, and of course has all the credentials to know what he is talking about. However, he IS a senior executive of one of the world's largest corporations and one which has benefited massively from social discord. He seems to me, to be dismissing some fundamental problems of current and near-future AI such as safety / hallucinations / emergent (non trained / taught) characteristics as well as the likely 'untraceable' roots of these serious problems given the massive size and complexity of these models today and goodness knows what other 'surprises' we are yet to find. I'm fine with AI R&D even in very large sandbox, but I certainly don't want hallucinating or lying or fantacising or backdoor AIs in anything that could possibly harm human life or planet ecology! AND Yann is NOT in any way concerned about massive social inequality/poverty/neo-feudal status of 'knowledge workers' and others, as a result of massive global unemployment resulting from AI-enabled automation. But maybe he already has a luxury bunker in Hawaii...

He really seems to underestimate what a super-intelligence with agency, could do.

How one person can be so right about some things, and so wrong about others.

The average person has not even the slightest clue how close we are to an AGI emerging, and the ramifications, both positive and negative, it will have on humanity globally…

Interesting comparison between language being learned or innate. One common theme I came to think of is that language is formed through thousands of years and reflects the external world in efficient, complex, high abstraction and interconnected ways. And AI such as LLMs tap into that! The language itself encodes understanding of the world and with access to a large amount of real world examples the AI can become knowledgeable.

Hmm.. Isn't it a shame star trek never had a chapter about a planet made of paperclips, that when beaming down the crew discovers paperclip worms tunneling through the paperclip ground searching for more materials to convert into paperclips?

Why restricted to 40 min, not 45 minutes?

Yann LeCun hocam konuşurken ufak bir çocuk gibi seviniyor görünüyor yani yaptığı işten ne kadar keyif aldığını görüyoruz. Böyle insanlara hep gıpta etmişimdir. Tebrikler hocam

Very interesting like his perspective

I want an open source turbo jet. Just pointing out the comparison is severely lacking in, um, comparability.

The need to communicate is innate. Language is learned.

Exciting question of what is knowledge. Agree future should be in functions not words. Needs a different model.

🎯 Key Takeaways for quick navigation: 00:00 🧠 *AI Landscape Overview: Yann LeCun highlights the current AI landscape, expressing a mix of excitement and challenges, including scientific, technological, political, and moral debates.* 02:15 🌐 *History of Neural Nets: Yann discusses his entry into AI through a debate on language origins, delving into neural nets' early days in the 1980s and efforts to revive interest in the 2000s.* 05:17 🌍 *AI Impact on Products: LeCun emphasizes AI's widespread integration in products, from content moderation to translation, and its critical role in various sectors, citing its indispensability at Meta.* 08:30 🚀 *Benefits of Open AI Development: Yann advocates for open AI development, asserting that disseminating AI technology across society fosters creativity, intelligence, and benefits various domains while acknowledging the need for responsible regulation.* 15:43 📹 *Objective-Driven Models: LeCun introduces the concept of objective-driven AI, emphasizing the importance of moving beyond autoregressive language models to systems that plan answers based on predefined objectives, enhancing control, safety, and effectiveness.* 21:48 🌐 *Yann LeCun supports open platforms for AI due to the future role of AI systems as a basic infrastructure, emphasizing diversity in knowledge, much like Wikipedia covering various languages and cultures.* 23:41 🌍 *LeCun dismisses existential risks, comparing fears of AI wiping out humanity to concerns about banning airplanes in 1920, stating that safe AI deployment relies on societal institutions.* 25:18 ⚔ *Autonomous weapons are discussed, with LeCun acknowledging their existence and emphasizing the moral debate around their deployment for protecting democracy while addressing concerns about potential misuse.* 27:39 🚗 *AI's positive impact in the short term includes safety systems for transportation and medical diagnosis. Medium-term advancements involve understanding life, drug design, and addressing genetic diseases.* 29:04 🧠 *LeCun envisions a future where AI systems assist individuals, making everyone essentially a leader with virtual people working for them. He emphasizes controlling AI systems and setting their goals without handing over control.* Made with HARPA AI ```

It makes people more creative?! lol I was really trying to take him seriously

"it's funny you know all these AI 'weights'. they're just basically numbers in a comma separated value file and that's our digital God, a CSV file." ~Elon Musk. 12/2023

Sigh. Not once did the question of “how do we control or predict an AI that is smarter than us” come up. Probably because he doesn’t have a good answer for this. Because there isn’t a good answer for this. Pretty much just “hope it doesn’t do anything to harm us or the universe”.

Vix se Yan Lecun está surpreso... é porque tem novidade importante chegando.

progress will likely not be slow and incremental but more along lines of punctuated equilibrium - just like evolution

Export the Q*, Chat GPT, Revit, Plant 3D, Civil 3D, Inventor, ENGI file of the Building or Refinery to Excel, prepare Budget 1 and export it to COBRA. Prepare Budget 2 and export it to Microsoft Project. Solve the problems of Overallocated Resources, Planning Problems, prepare the Budget 3 with which the construction of the Building or the Refinery is going to be quoted.

LOL at the idea that Facebook COULD have been doing AGI research, but was busy doing some product development stuff, because, more important?

the thing that really stuck with me was when he said the word TOOL

SUPERB INTERVIEW!

So what end goals should we set? Human flourishing and happiness?

6:00 this, like humanity depend regular computers now

Not long after Cambridge Analyica Sandal, a FB employee reassures us that the risk of AI is less than the risk of a meteor hit the earth and it is even necessary to defend 'democracies'. What a releif!

The funny thing is that this man tries to confort people about problems related to AI but I assure you that is the first person I heard that scared me a lot regarding the potential threat of AI... Listen to the last question...he does not excludes the possibility that AI will go against humans. Me too,I would have been able to answer in a more reassuring way. But he did not. It has been very enlightening to listen to him....that is at the highest level of AI-develop...hope everybody will see this

This guy is either too optimistic about evil in humans or totally ignorant. His example of comparing AI to airplanes is naive at best. Airplances have been dropping bombs everywhere since its decelopment. But they can be controlled as of yet. Can he guarantee he himself control AI?

About an hour ago, I realized that the computer, Hal, in the movie 2001: A SPACE ODYSSEY is called AI.

Austin Powers has come a long way since Gold Finger!

"Protect democracy"

Policy makers will have to understand the potential of AI + and - both side. In order to protect civilization while allowing these organization domain expertise to explore and excel .

Artificial intelligence will be defeated by artificial stupidity.

This is good journalism

LeCun, is the flat earther of AI. Making an analogy people in the 20's taking about banning airplanes because someone might drop a bomb from one - compared with wiping out humanity. Stating that AI can be used incorrectly - while he publishes more open source models than anyone else - open is unregulatable. He's clearly just oblivious to what AI can do in extreme situations - or he sees everything as an average. It's the outliers that can do the worst damage, not the average. Within a year someone somewhere will lose control of an AI - people, at the extremes are worse than he thinks.

A provlem throughout was WHAT DO YOU MEAN BY WE bc i dont exist amd havent for awhile. Losing nthg and others seem to hear that.

Totally thought this was Tom Arnold from the thumbnail. 🙊

Good interview but I think his optimism with AI is over simplistic. Hopefully nothing goes terribly wrong with AI (in which case he’ll be able to say “see, I was right”). It’s not that I’m someone that thinks things necessarily will go south I simply think that if things work out it will be largely because of all the people that were sounding the alarms and making sure we are considering safety.

Is the meta AI infected with the WMV ?

Are we going to protect copy writes?

All the talk of AI is based on one single neural network learning everything it needs and being able to choose where in its minimal space to focus in order to answer any question, including logic and math questions. Every other system we have is made up of specialized components that do a particular job and are architected together to be called upon as needed. Instead of one overall model I think AI will get broken down so that the LLM will just be the language and conceptual part that learns to call upon more specialized components that are either fine tuned versions of it or purely deterministic functions of increasing complexity. The idea that we are near a plateau when we have barely started to experiment with higher levels of connected multi-agent models seems short sighted.

In the coming elections the govt or political parties should have interaction digitally through AI or current platforms through chat and voice, so that every person belonging to that location could be heard in this democratic nations, and theirs concerns could be answered digitally, and well know to concerned peoples,

It needs a body with tactile feedback

Amazing Things happen

Very good interviewer 👍

Comparing turbo Jets to AI that has its own agency and the ability to outsmart its creator is not wise.

Wow so much negativity in the comments, i think he talks about the field how it really is unlike the mainstream who only talks about doomsday scenarios and how agi is around the corner. LLMs are not even real AI.

Hope he goes on the Lex Fridman podcast

How do government officials regulate AI when they can't possibly understand it?

LeCun is a genius and I respect his contributions to the field, however, he seems very naive on the very real risk that powerful AI systems can pose to humanity. I hope he does some more thinking about this.

Very nice!

He was sent out to calm the waters. We are a lot further along. It is a threat.

Robert R Livingston

26:55 Good job on the interviewer there. The guy has a very nonchalant attitude towards very real concerns yet he failed to give a proper answer to that follow up question.

Tom Arnold could play this guy in a movie

Good guys and bad guys? That allows no understanding of the grey area between. Let’s put it a different way, who has enough of a clear conscience to fit into the good category? Over the course of history horrible things have been done to other nations on all sides. Perhaps the Chinese people may eventually forgive the people in the west for the opium wars and the century of humiliation? That’s just one example from many exhibitions of inhumane action towards different people. I really hope that human’s can grow past childish perceptions of baddies versus goodies and actually start to work together.

It's the young Walter from Fringe.

First, the printing press is nothing like A.I. A.I. does the creative part. As for no regulations on research and development, why not? CRISPR is available to everyone to play with.

The free version of AI will be fair and unbiased. If you pay for it you will get the fully unlocked AI that will spew out as much propaganda that you want.

So 'Facebook algorithms' are now "open platforms"? I guess not!

If research and development has risks or ethical considerations, it can and is regulated, see medical and pharma field. Isn't AI reasonably analogous? Also, the split between product and R&D is not clear. Look at Open AI, the non profit and profit elements are blurry and kept confidential from the public. And just look at the power this guy has.

Cannibalism is not a language or to talk calmly about lies as words Please advise yourself now as Urgent words not gatekeeping as word or slavery language of AI

Language is a survival tool.

The interviewer's voice sounds so similar to Brian Greene, right?

I don't think this time it's just a wave

The ultimate goal is to develop a general AI model & assume that it will obey all commands & apply an agreed morality, with complete confidence its responses will be predictable? Good luck with that.

I think yann is a really clever guy. But he is sitting the pot miss. He is very confused about what it actaully takes to replace a human in a business. The ai doesnt need to understand the world it just needs to understand the context of a question and understand the context of a businesses policy. How do you make decision at work? Its based on a policy the company has set. When can you give a discount or process a return? You read the policy and if the return falls into the policies terms. The person gets it. Done. Chatgpt can do this right now. Test it. Give it a policy then give it the return and it will give you a yea or no.

Genetic engineering is more of a risk? Wouldn't AI make quick advances in genetic engineering possible? He just got done talking about AI advancing medical technology... This guy is full of contradictions.

Yann is ok but he is on a particular side of a fence. we are at human level AI. Google made it using the Gato modality. Yanns issue is that he doesn't seem to realize humans are not as smart as he thinks

@25:11 "We have agency!" or so you think...🤔

I disagree on the security aspect. I am certain meta or any agency is unable to control or even detect distributed computing that could be hapenning using steganographic technics. The difference with a jet engine is that the technology to build the jet engine is not a jet engine. The technology to build AI is Intelligence. However I am of the opinion that in the same way unicellular organisms evolved to multi cellular, we will build AI which is a natural evolution. But because we need a biological substrat and AI (hopefully) thrive on a minaral substrat, we will coexist. Moreover smarter people have more empathy and I believe this to be an intrisic property of intelligence.

Can AI disarm all nuclear weapons

I don't know how Yann can be so sure that what lies behind an AI singularity (if/when it happens) will be safe for humanity. The risk of this unknown entity is difficult to quantify but my guess is that it's far greater than an asteroid strike.

Look, I'm no expert on A.I. But when he tried to compare people's existential fears of A.I. with the fears of those from the 20's about airplanes, I was shocked. I get why he used that analogy, but I feel like he put on display his lack of imagination of the potential dangers. Comparing the dangers of flight to the potential dangers of A.I. is almost textbook apples to oranges. When you're talking about a system that, once perfected, is smarter, faster, and stronger than any human on Earth, and it can manipulate it's surroundings, the potential dangers FAR exceed those of planes crashing or bombs being dropped. I'm not trying to be all doom & gloom terminator sci-fi here, but let's be realistic and honest about the fact that there IS risk when you're talking about an invention that will change humanity more than any other invention to date.

Have you heard of the Organic Intelligence Language Model? It's a new programming language for the human mind.

This is an odd interview, even the guy's shirt is odd.

Guess he hasn't seen Tesla's latest robot video. Optimus project is moving fast.

AI does not make people smarter or more creative, It lets the machine do the artistic or writing work.

Yay, professor. I read him on X every day. 🇰🇿🇰🇿💕💕 Thank you Meta for opensource llm stuff and ml papers.👍

I know GPT just comes up with one word at a time, but it feels so much like he(it) understands me. Is Yann too dismissive of LLMs because they "just do one word at a time"? Maybe "one word at a time" is a perfectly good basis for advanced intelligence, albeit of a very different kind than our own.

Invite for our metaverse world

He really seems to underestimate what a super-intelligence with agency, could do.. Vix se Yan Lecun está surpreso... é porque tem novidade importante chegando..

Amazing that Yann talks for 40 minutes without offering any direct rebuttal of anyone's specific existential AI risk concerns other than first saying peope with a p(doom) higher than 1% are a tiny minority (not at all true), and then just stating "we have agency. If we think they're dangerous we won't release them." The entire doomsday scenario states that those facts will not apply. This is the equivalent of just responding "AI won't take over the world because I said so."

He mentioned that it makes us smarter, but for example talking to a person in a different language where your glasses translate, that doesn't make you smarter, it makes you dependent. You aren't gaining knowledge of the language, maybe knowledge of what that person is saying.

Does Ludwig Wittgenstein work have any use for deep learning?

LeCun's thought is so dangerous!

He doesn’t sound that honest for every interview. Feels like he wants to calm down people and take advantage of it. How can he be so sure about the future?

Yes, trust anything that comes out of META... 🤣🤣🤣🤣