Rendered at 12:26:46 GMT+0000 (Coordinated Universal Time) with Cloudflare Workers.
nok22kon 4 hours ago [-]
Yann LeCun was saying 3 years ago that because token generation is auto-regressive, its mathematically impossible to generate a long stream of coherent tokens, because errors amplify exponentially.
and then models learned that they can back track and error correct
so much for "mathematically impossible..."
charcircuit 4 hours ago [-]
I think it was largely the introduction of tool calling that allowed models to mitigate the issue of errors amplifying exponentially since it allows the model to understand if what it generated is correct or has issues that it needs to address. This addresses the potential lack of or low quality of world model by being able to reference the current state of the world.
ravenstine 3 hours ago [-]
I've definitely realized this phenomenon after a few occasions of erroneously trying to rely purely on instructions to get an LLM to do a thing or take on a role, especially without persistent cloud-based sessions that have internal checklists and other opaque guidance. They're essentially poor at self-managing, but can do really well when they are limited in scope/context and are worked into a sort of state machine that guarantees they perform certain tasks predictably. They won't always do those tasks the exact way you expect them to, but at least they actually do them, and because of that they are more likely to have the correct prior context to perform the next task better. Because they are so prone to selectively ignoring directions, that can quickly send them down an incorrect path that compounds on itself.
shevy-java 3 hours ago [-]
You insinuate here AI "learned".
I doubt that this was AI self-improvement.
aswegs8 54 minutes ago [-]
Does that take anything away from the argument?
card_zero 11 minutes ago [-]
What argument, "a theory was wrong"? No, the inane central observation, the observation that a researcher was unable to predict a discovery before it was discovered, remains true despite the gratuitous insertion of a little bit of bullshit about AI learning.
I suppose it's additionally trying to imply something else, like "due to a pattern of researchers being unable to discover discoveries before they discover them, AGI is just around the corner".
rcxdude 3 hours ago [-]
Was there a particular change to the network or the way that it was trained that introduced the 'backtrack and error correct' mechanism?
infinite_spin 3 hours ago [-]
do you have a problem with this field of research being called "machine learning"?
TMWNN 4 hours ago [-]
> and then models learned that they can back track and error correct
You mean "Human developers learned ...", yes? Or was there really an all AI-driven, self-improving aspect to this?
rcxdude 3 hours ago [-]
Well, LLM networks don't have a 'back track and error correct' component in the design, AFAIK.
fatata123 2 hours ago [-]
[dead]
threethirtytwo 3 hours ago [-]
Stop attacking Yann. I would say like 90% of the HN crowd was parroting Yann too.
jiggawatts 4 hours ago [-]
Also, almost any argument against LLM intelligence also applies to humans.
I very commonly see someone make some small mistake and end up going in the wrong direction, “accumulating stupid” as they go, sometimes for years.
shevy-java 3 hours ago [-]
Humans can learn.
AI can not.
For those disagreeing: please explain how a static hardware can learn.
echoangle 3 hours ago [-]
By self-modifying the software. Currently the model harnesses only allow the model to modify its own prompt (which could be considered a really weak kind of learning), but theoretically, a model could design and train its own replacement and run that, continuously improving itself. I’m not sure if LLMs will be able to do that but the static hardware has nothing to do with it (since the bits on the harddrive aren’t static).
someonebaggy 3 hours ago [-]
idk, how does voice recognition learn my voice? How can I install programs when the hardware is static?
threethirtytwo 3 hours ago [-]
this is profoundly false. AI not only can learn, it is built entirely from learning. The field is called machine learning after all.
Not only that... AI is NOT only learning during the training phase... LLMs learn in real time the minute you talk to it. It learns something and saves those learnings in a context window or somewhere else if you want it to exist beyond the context window.
All of the above runs on static hardware. Don't understand how someone can say a profoundly wrong statement and get voted up.
guenthert 3 hours ago [-]
Correct me if I'm wrong, but if a profound insight is gathered in session 1 with user A and stored in context A1, this might be available to user A in session 2, if that still has access to context A1, but won't be available to user B in any of his/her sessions until that NN is retrained with input which includes at least some of the information from context A1.
fragmede 3 hours ago [-]
Also with the stochastic parrot thing. If you say just the right thing to the right human and the right time, they'll very predictibly say their favorite movie/book quote or song lyric, like some sort of parrot.
waldarbeiter 4 hours ago [-]
[dead]
armchairhacker 4 hours ago [-]
Besides "smart", the headline also conflates AI with LLMs. The real, non-clickbait title is "Yann LeCun, founder of AMI Labs, is developing a new AI system"
randsorex 4 hours ago [-]
It is just so bizarre compared to my everyday experience also.
I never ask Opus or Fable a question and think "what a stupid response".
Quite the opposite. It has actually raised the bar of what I consider an intelligent response to my inquiry. So much so that most responses from humans on most subjects to most forms of inquiry seem stupid and not really thought out.
acpdev 3 hours ago [-]
I’m sure you’re very intelligent and capable so I suspect we work in different problem spaces if you have not seen this, but I definitely think the responses are at times very very junior and I find myself having to explain first principles. Fable less so, but Opus routinely will make very naive assumptions about retry logic, protocols it supposedly has training material on, and it will very often miss the forest for the trees.
This isn’t exactly saying how stupid anyone is but I’d definitely have been concerned about a human’s reasoning ability and understanding of logic if they’d given me similar answers.
dlcarrier 3 hours ago [-]
Everyone nowadays seems to only think of AI as LLMs or maybe also stable diffusion. People want to ban games with AI in them, when by definition every NPC is following some kind of AI algorithm.
altmanaltman 3 hours ago [-]
They literally interview another person in it and mention a lot of other labs doing this kind of research including Google. Yes, he's the main starting interview but this is not really clickbait or a marketing piece.
dagss 5 hours ago [-]
The article seems to define "smart" as being good at spatial awareness and navigating a body through 3D space and such. Thus, a mice is smarter than an LLM.
That's the first time in my life I hear this definition. Until now, the word "smart" has meant doing exactly the things LLMs do, and mice don't.
I guess it is a sign we are re-evaluating what makes humans special.
JsonDemWitOster 4 hours ago [-]
> I guess it is a sign we are re-evaluating what makes humans special.
While we should be careful of a bias, it is also a good practice in the scientific method to review definitions that may have been not precise enough.
For example, initially, a "planet" was just a big body in space. Then when people started to see more and more nuances, the definition just refined, and some objects stopped being called "planet".
I would not be surprised if there is a bias that pushes some people to redefine "intelligence" away from machine, but I would not be surprised if there is a bias that pushes some people to ignore newly discovered nuance and put into the same "intelligence" bag things that are in fact very different. I personally can see how LLM are not really "intelligent", and I don't think it is a good idea to say: well, yesterday we said the minimum criteria is X, now that we noticed that X can be reached without really doing the real thing, let's just ignore that and pretend it is the same thing.
(
: the biggest clue for me is to use an early model, and see that it sometimes looks very intelligent, and then sometimes you can see that it gets it wrong in a way that shows that it never "understood" it at all. Newer models are better, but because it is an iteration on the same bases, the increase of performances cannot really due to replacing the things that "looked smart by aren't" by "real smart", but more replacing the things that "don't look smart" by "look smart by aren't")
JsonDemWitOster 3 hours ago [-]
Yeah I think if we are looking at it through that lens, the problem is in the term _intelligence_ in itself. Psychology and biology could not even pinpoint what exactly makes for _intelligence_. There isn't really a precise definition yet so it's just natural that definitions tend to shift.
I don't think we even need to go into tech and AI for an example. The intelligence or lack thereof of pets surprise us. Sometimes a cat is surprisingly smart when it is able to open a door to get food it wasn't supposed to. But then same cat gets bamboozled by walls and simple optical illusions. We generally expect that if something/a human is smart enough to do the former, then it shouldn't be dumb enough to fall for the latter.
Coming back to AI, this dissonance is how AI-generated images are detected for example. If a human can render something so well, you wouldn't expect them to make small but nonetheless elementary line art mistakes.
dagss 2 hours ago [-]
It's the same with human intelligence though. A human can be brilliant on some things and then we're puzzled why they are so idiotic in other areas.
Every time this comes up, people pick on any kind of flaws or inconsistencies of AI models, while at the same time giving a huge pass to the extreme variation in intelligence and stupidness displayed in human behaviour.
Creativity is the same. Human artists are "inspired" by earlier arts, perhaps following and slightly changing "trends" they participate in -- which is somehow seen as totally different from what AIs are doing.
cauch 19 minutes ago [-]
> It's the same with human intelligence though.
No, this is not the same observation. In "basic LLM", the answer is not "confused" or "fail to understand", the answer is "inconsistent with the understanding mechanism". It is not that they "fail to understand while trying to understand", it is that there is not understanding mechanism at all.
Humans can have different level of intelligence, depending on the individuals, the subjects, even circumstantial situations (someone being tired, someone being distracted, or just bad luck). But they never make the same kind of mistakes I've observed with "basic LLM", where they do "non sequitur" that does not make sense at all but has all the characteristic of imitating something said by someone who understood.
I still even see it sometimes with Claude. It says logical stuff, and then suddenly something that does not make sense and it snaps me back to reality: none of this, including the correct things, are the result of understanding the underlying concepts, it is just that the correct things are more probable to generate, and that suddenly, a nonsensical happen to also be probable for a given configuration.
JsonDemWitOster 1 hours ago [-]
My problem with AI is the sheer variance of its stupid-smart spectrum. While it's true that human intelligence is not deterministic or predictable, the inconsistency exists in a much narrower band of variance which makes failure modes foreseeable. Thus I would much prefer a system with humans in the loop with processes in place for idiot-proofing.
This is true for "lateral" (I lack a better term) fields of intelligence as well. You don't ask a philosophy professor advice for the rashes on your skin; you see a doctor for that. And yet both the professor and the doctor could be expected to accurately identify from a picture that you do have rashes on your skin. An AI (and I mean in the general sense, not only transformer LLMs) could give you a pretty accurate rundown of Plato and still think the same picture is a beautiful sunrise.
(I don't even kid. Just this morning, an AI labeled a GIF from _Friends_ as a 1950s magazine ad for white bread. Just what in the failure mode is that?)
You can't idiot-proof AI without knowing what's in the training data set and even then you run into question of scale.
linzhangrun 4 hours ago [-]
It depends on how you define "smart".
For me, "smart" means doing things less based on instinct. Things humans can do but mice cannot, things mathematicians can do but normal people cannot, etc.
Considering the unit distance conjecture was disproved by OAI's model last month, I think maybe LLMs should count as "smart".
heohk 9 hours ago [-]
It's inference. It's really good at generating stuff when the example base is extensive. Like for non-esoteric coding.
ramon156 6 hours ago [-]
Is a brain also inference? I know that an LLM works very different from the brain, but I wonder what makes a brain more capable of thinking. Is it the long term context? Is it a different type of neuron activation?
agenticup 4 hours ago [-]
i guess inference engineering, like dpsark or dflash specific speculative decoding technqiues
arisAlexis 5 hours ago [-]
Ha before reading the article I thought "this must be an interview of Lecun". A bitter scientist that hates he was left behind the revolution.
dgellow 5 hours ago [-]
In what way was he left behind? If he wanted to actually work on LLMs all the AI labs would fight to get him
MrScruff 3 hours ago [-]
Considering all of the great research that has come from his labs (eg. DINO, Segment Anything) I don’t think that’s fair (no pun intended).
imtringued 5 hours ago [-]
Left behind how? It's been transformers since 2016 and not much actual progress in basic architectures has happened 10 years later. I'm honestly struggling to see how you can be left behind in this field.
menaerus 3 hours ago [-]
Obviously, transformers architecture is just one of the ingredients. Otherwise we wouldn't be seeing competing labs in this race. I also read all his interviews as a marketing material.
nok22kon 4 hours ago [-]
and CPUs have the same basic architecture since 2000. no progress happened, right?
shevy-java 3 hours ago [-]
What's next is more AI spam-slop. I already noticed
this on youtube. Tons of short videos are AI sloppified,
making youtube worse in the process.
JsonDemWitOster 1 hours ago [-]
> What's next is more AI spam-slop. I already noticed this on youtube.
I hope not but your observation on YouTube is spot-on. It's really frustrating. I've managed to keep good hygiene for my shorts feed. I practice zero-tolerance for braindead content; one strike is all it takes for the "never recommend" button.
But with the World Cup, the situation is just pandemonium. Football has always been breeding ground for low-effort content on the platform: unofficial highlights, cropped to death to avoid copyright, and always, for some reason, with a blaring background music. But now it's reached peak slop chaos: AI voiceovers, dubious anecdotes ("...and that kid, was Ronaldo"), and STILL the horribly blaring background music. The algorithm makes no distinction between quality and slop content of a topic. It's all just about the topic. So all it takes is for me to view a short related to football (even one of thoughtful commentary) for the slop to come in hordes.
At least I can now share in this forum's disdain for shorts. Kill that shit with fire man.
feverzsj 6 hours ago [-]
AI winter.
SwellJoe 4 hours ago [-]
We're past the point where there's a feasible argument that there is an AI winter coming.
The models work remarkably well for several classes of problem that seemed impossible a few years ago. They're not going away. There will still absolutely be a lot of ups and down and crazy stuff that happens in AI, but it won't be that AI almost completely stops being developed/funded for a decade or more. The biggest risk, I think, is regulatory capture; it's what Anthropic and OpenAI seem to be aiming for with their scaremongering about how capable and dangerous their models are. That'll put a damper on the industry for everyone except the companies that bribe the right people.
card_zero 1 minutes ago [-]
They're not going away, in the same sense that Henry Winkler is still alive and working.
karahime 5 hours ago [-]
Not likely. Take with whatever grain of salt you'd like, but that was largely a property of development being academicized and subject to things like grant cycles, research topic fashionability trends, and institutional structure. It would be wrong to assume it's some baked in thing that's guaranteed to happen independent of how development looks.
JsonDemWitOster 4 hours ago [-]
But _AI today_ is heavily subsidized by investor capital in the same way investors subsidized social/mobile/big data/VR/blockchain in the past. It's not unlikely "AI" would get a soft taboo in the same way as if you just presented a mobile-first, big-data driven, VR social media app today.
Which, judging by the terrible PR optics AI has nowadays, could unfortunately seep into academia too. Fund grants wouldn't want their names associated with anything with "AI" in its name even if it's a return to expert systems.
karahime 4 hours ago [-]
You're mixing different things. Mobile first is integrated into new services to the point that they either are mobile first, or they have a design system which includes mobile as a surface. VR has a wide user base (MQ2 sold as well as the original Xbox) and is involved in both manufacturing design and simulation, and is hardly an academic taboo, even if the "main" topic of discussion is elsewhere right now. Blockchains are being integrated into financial infrastructure even as some people make snarky commentary about it. Sometimes optics is just an optical illusion.
JsonDemWitOster 3 hours ago [-]
Fair enough. Mobile and social became ubiquitous and are now table stakes. But my problem with VR and blockchain---even allowing for the fact/assumption that they are still relevant---is that they never lived up to their hype. They never became ubiquitous as mobile and social. They don't inspire investor confidence like they did in the past, if at all. AI, if it survives the public and regulatory backlash, could be headed to the same understudy role.
I'm using "AI" broadly here even if the current investor darling is just LLMs because, well, the term AI has been front and center of all promotions and investors and the general consumer public isn't really a discerning bunch. So I stand by my prediction that a "soft taboo" is likely where investors and consumers shy away from anything even remotely AI. The consumer backlash has arguably already started.
karahime 3 hours ago [-]
The vast consumer adoption and ongoing involvement seems to point the other way, though. I think a lot of the appearance of backlash is on (specifically anglophone, mostly) social media, which is going through a somewhat reactionary phase regardless.
JsonDemWitOster 3 hours ago [-]
Out of curiosity which markets exactly do you see with a positive AI outlook?
Also, I think you are downplaying the "anglophone" social media backlash too much for a couple reasons. _Anglophone_ social media is huge, even global. Everyone participates in anglophone social media even non-English speakers (who post in broken English, or comment in their native language in English-language content). So there is anglophone social media in all markets; it's not difficult to be aware of and espouse American public sentiment.
Even if you narrowly define anglophone socmed to correspond to the geo-cultural anglosphere, I think it's not surprising at all that the bulk of backlash is focused there because the leading AI companies are based there as well.
selestify 54 minutes ago [-]
The Chinese market for one seems pretty optimistic about AI and the presence of AI in apps.
nananana9 3 hours ago [-]
> MQ2 sold as well as the original Xbox
I'd be interested in the "retention rate" for these two products. I wouldn't be surprised if the average original Xbox was used 2 orders of magnitude more than the average Meta Quest, which is collecting dust on some shelf.
I'd wager the typical MQ2 owner is someone with 20 hours of Beat Saber on it and 5000 hours total on Steam or PS.
bmacho 5 hours ago [-]
Human winter.
dosisking 4 hours ago [-]
AI climate change.
throw2007 4 hours ago [-]
Its definitely not as dumb as MAGA crowd
nananana9 3 hours ago [-]
Was it worth it to log out and create a new account just to post this?
and then models learned that they can back track and error correct
so much for "mathematically impossible..."
I doubt that this was AI self-improvement.
I suppose it's additionally trying to imply something else, like "due to a pattern of researchers being unable to discover discoveries before they discover them, AGI is just around the corner".
You mean "Human developers learned ...", yes? Or was there really an all AI-driven, self-improving aspect to this?
I very commonly see someone make some small mistake and end up going in the wrong direction, “accumulating stupid” as they go, sometimes for years.
AI can not.
For those disagreeing: please explain how a static hardware can learn.
Not only that... AI is NOT only learning during the training phase... LLMs learn in real time the minute you talk to it. It learns something and saves those learnings in a context window or somewhere else if you want it to exist beyond the context window.
All of the above runs on static hardware. Don't understand how someone can say a profoundly wrong statement and get voted up.
I never ask Opus or Fable a question and think "what a stupid response".
Quite the opposite. It has actually raised the bar of what I consider an intelligent response to my inquiry. So much so that most responses from humans on most subjects to most forms of inquiry seem stupid and not really thought out.
This isn’t exactly saying how stupid anyone is but I’d definitely have been concerned about a human’s reasoning ability and understanding of logic if they’d given me similar answers.
That's the first time in my life I hear this definition. Until now, the word "smart" has meant doing exactly the things LLMs do, and mice don't.
I guess it is a sign we are re-evaluating what makes humans special.
Always has been: https://en.wikipedia.org/wiki/AI_effect
Tangentially: https://en.wikipedia.org/wiki/Moravec%27s_paradox
For example, initially, a "planet" was just a big body in space. Then when people started to see more and more nuances, the definition just refined, and some objects stopped being called "planet".
I would not be surprised if there is a bias that pushes some people to redefine "intelligence" away from machine, but I would not be surprised if there is a bias that pushes some people to ignore newly discovered nuance and put into the same "intelligence" bag things that are in fact very different. I personally can see how LLM are not really "intelligent", and I don't think it is a good idea to say: well, yesterday we said the minimum criteria is X, now that we noticed that X can be reached without really doing the real thing, let's just ignore that and pretend it is the same thing.
(
: the biggest clue for me is to use an early model, and see that it sometimes looks very intelligent, and then sometimes you can see that it gets it wrong in a way that shows that it never "understood" it at all. Newer models are better, but because it is an iteration on the same bases, the increase of performances cannot really due to replacing the things that "looked smart by aren't" by "real smart", but more replacing the things that "don't look smart" by "look smart by aren't")I don't think we even need to go into tech and AI for an example. The intelligence or lack thereof of pets surprise us. Sometimes a cat is surprisingly smart when it is able to open a door to get food it wasn't supposed to. But then same cat gets bamboozled by walls and simple optical illusions. We generally expect that if something/a human is smart enough to do the former, then it shouldn't be dumb enough to fall for the latter.
Coming back to AI, this dissonance is how AI-generated images are detected for example. If a human can render something so well, you wouldn't expect them to make small but nonetheless elementary line art mistakes.
Every time this comes up, people pick on any kind of flaws or inconsistencies of AI models, while at the same time giving a huge pass to the extreme variation in intelligence and stupidness displayed in human behaviour.
Creativity is the same. Human artists are "inspired" by earlier arts, perhaps following and slightly changing "trends" they participate in -- which is somehow seen as totally different from what AIs are doing.
No, this is not the same observation. In "basic LLM", the answer is not "confused" or "fail to understand", the answer is "inconsistent with the understanding mechanism". It is not that they "fail to understand while trying to understand", it is that there is not understanding mechanism at all.
Humans can have different level of intelligence, depending on the individuals, the subjects, even circumstantial situations (someone being tired, someone being distracted, or just bad luck). But they never make the same kind of mistakes I've observed with "basic LLM", where they do "non sequitur" that does not make sense at all but has all the characteristic of imitating something said by someone who understood.
I still even see it sometimes with Claude. It says logical stuff, and then suddenly something that does not make sense and it snaps me back to reality: none of this, including the correct things, are the result of understanding the underlying concepts, it is just that the correct things are more probable to generate, and that suddenly, a nonsensical happen to also be probable for a given configuration.
This is true for "lateral" (I lack a better term) fields of intelligence as well. You don't ask a philosophy professor advice for the rashes on your skin; you see a doctor for that. And yet both the professor and the doctor could be expected to accurately identify from a picture that you do have rashes on your skin. An AI (and I mean in the general sense, not only transformer LLMs) could give you a pretty accurate rundown of Plato and still think the same picture is a beautiful sunrise.
(I don't even kid. Just this morning, an AI labeled a GIF from _Friends_ as a 1950s magazine ad for white bread. Just what in the failure mode is that?)
You can't idiot-proof AI without knowing what's in the training data set and even then you run into question of scale.
For me, "smart" means doing things less based on instinct. Things humans can do but mice cannot, things mathematicians can do but normal people cannot, etc.
Considering the unit distance conjecture was disproved by OAI's model last month, I think maybe LLMs should count as "smart".
I hope not but your observation on YouTube is spot-on. It's really frustrating. I've managed to keep good hygiene for my shorts feed. I practice zero-tolerance for braindead content; one strike is all it takes for the "never recommend" button.
But with the World Cup, the situation is just pandemonium. Football has always been breeding ground for low-effort content on the platform: unofficial highlights, cropped to death to avoid copyright, and always, for some reason, with a blaring background music. But now it's reached peak slop chaos: AI voiceovers, dubious anecdotes ("...and that kid, was Ronaldo"), and STILL the horribly blaring background music. The algorithm makes no distinction between quality and slop content of a topic. It's all just about the topic. So all it takes is for me to view a short related to football (even one of thoughtful commentary) for the slop to come in hordes.
At least I can now share in this forum's disdain for shorts. Kill that shit with fire man.
The models work remarkably well for several classes of problem that seemed impossible a few years ago. They're not going away. There will still absolutely be a lot of ups and down and crazy stuff that happens in AI, but it won't be that AI almost completely stops being developed/funded for a decade or more. The biggest risk, I think, is regulatory capture; it's what Anthropic and OpenAI seem to be aiming for with their scaremongering about how capable and dangerous their models are. That'll put a damper on the industry for everyone except the companies that bribe the right people.
Which, judging by the terrible PR optics AI has nowadays, could unfortunately seep into academia too. Fund grants wouldn't want their names associated with anything with "AI" in its name even if it's a return to expert systems.
I'm using "AI" broadly here even if the current investor darling is just LLMs because, well, the term AI has been front and center of all promotions and investors and the general consumer public isn't really a discerning bunch. So I stand by my prediction that a "soft taboo" is likely where investors and consumers shy away from anything even remotely AI. The consumer backlash has arguably already started.
Also, I think you are downplaying the "anglophone" social media backlash too much for a couple reasons. _Anglophone_ social media is huge, even global. Everyone participates in anglophone social media even non-English speakers (who post in broken English, or comment in their native language in English-language content). So there is anglophone social media in all markets; it's not difficult to be aware of and espouse American public sentiment.
Even if you narrowly define anglophone socmed to correspond to the geo-cultural anglosphere, I think it's not surprising at all that the bulk of backlash is focused there because the leading AI companies are based there as well.
I'd be interested in the "retention rate" for these two products. I wouldn't be surprised if the average original Xbox was used 2 orders of magnitude more than the average Meta Quest, which is collecting dust on some shelf.
I'd wager the typical MQ2 owner is someone with 20 hours of Beat Saber on it and 5000 hours total on Steam or PS.