LeCun has been giving the same talk with literally the exact same slides for the past 3 years. JEPA still hasn't delivered despite FAIR's substantial backing.
LeCun seems like an extremely smart person that suffers from an overgrown ego. I got that strong impression from seeing his Twitter feed - basically "smarter than thou".
Generative world models seem to be doing ok. Dreamer V4 looks promising. I’m not 100% sold on the necessity of EBMs.
Also I’m skeptical that self-supervised learning is sufficient for human level learning. Some of our ability is innate. I don’t believe it’s possible for statistical methods to learn language from raw audiovisual data the way children can.
Human DNA has under 1GB of information content in it. Most of which isn't even used in the brain. And the brain doesn't have a mechanism to read data out from the DNA efficiently.
This puts a severe limit on how much "innate knowledge" a human can possibly have.
Sure, human brain has a strong inductive bias. It also has a developmental plan, and it follows that plan. It guides its own learning, and ends up being better at self-supervised learning than even the very best of our AIs. But that guidance, that sequencing and that bias must all be created by the rules encoded in the DNA, and there's only this much data in the DNA.
It's quite possible that the human brain has a bunch of simple and clever learning tricks that, if we pried out and applied to our AIs, would give us x100 the learning rate and x1000 the sample efficiency. Or it could be that a single neuron in the human brain is worth 10000 neurons in an artificial neural network, and thus, the biggest part of the "secret" of human brain is just that it's hilariously overparameterized.
I don't know why people really dislike the idea of innate knowledge so much, it's obvious other animals have tons of it, why would we be any different.
The problem with assuming tons of innate knowledge is that it needs to be stored somewhere. DNA certainly contains enough information to determine the development of various different neuron types and which kinds of other neurons they connect to, but it certainly cannot specify weights for every individual synapse, except for animals with very low neuron counts.
So the existence of a sensorimotor feedback loop for a basic behavior is innate (e.g. moving forward to seek food), but the fine-tuning for reliabily executing this behavior while adapting to changing conditions (e.g. moving over difficult terrain with an injured limb after spotting a tasty plant) needs to be learned through interacting with the environment. (Stumbling around eating random stuff to find out what is edible.)
>certainly cannot specify weights for every individual synapse
That's not the only way to one could encode innate knowledge. Besides, we have demonstrated that animals have innate knowledge experimentally many times, the only reason we can't do this to humans is that it would be horrifically unethical.
>Stumbling around eating random stuff to find out what is edible
Plenty of animals have innate knowledge about what is and isn't edible: it's why, for example, tasty things generally speaking smell good and why things that are bad (rotting meat) smell horrific.
I'm not saying that there's no innate knowledge. This entire list of reflexes https://en.wikipedia.org/wiki/List_of_reflexes is essentially a list of innate knowledge in humans, many of which have been demonstrated in newborns, apparently without considering such experiments unethical.
I'm saying that there are limits to how much knowledge can be inherited. I.e. the question isn't "Where could innate knowledge be encoded other than in synapses?" but "Considering the extremely large number of synapses involved in complex behavior far exceeds genetic storage capacity, how are their weights determined?" And since we know that in addition to having innate behaviors, animals are also capable of learning (e.g. responding to artificial stimuli not found in nature), it stands to reason that most synapse weights must be set by a dynamic learning process.
> That's not the only way to one could encode innate knowledge.
Maybe sections could be read from DNA and broadcast as action potentials?
There's already ribosomes that go over RNA. You'd need a variant which instead of making amino acids, would read out the base pairs and make something that causes action potentials to happen based on the contents.
Some people just believe there is no innate knowledge or we dont need it if we just scale/learn better (in the direction of Bitter Lesson)
(ML) Academia is also heavily biased against it due to mainly two reasons:
- Its harder to publish, since if you learn Task X with innate Knowledge, its not as general, so reviewer can claims its just (feature) engineering - Which hurts acceptance, so people always try to frame their work as general as possible
- Historical reasons due to the conflict the symbolic community (which rely heavily on innate knowledge)
But generative models are always going to seem like they are doing ok. That's how they work. They are good at imitating and producing misleading demos.
Agree with LeCun that current ai doesn’t exhibit anything close to actual intelligence.
I think the solution lies into cracking the core algorithms used by nature to build the brain. Too bad it’s such an inscrutable hairball of analog spaghetti code.
I think that LeCun has correctly identified that LLM is only one type of intelligence and that AGI/AMI needs to combine multiple other types … hierarchical goal setting, attention/focus management, and so on.
Seems that he is able to garner support for his ideas and to make progress at the leading edge - yes a little bit hard to take the “I know better” style, but then many innovations are driven by narcissism.
There is a lot of "transformer LLMs are flawed" going around, and a lot of alternative architectures being proposed, or even trained and demonstrated. But so far? There's nothing that would actually outperform transformer LLMs at their strengths. Most alternatives are sidegrades at best.
For how "naive" transformer LLMs seem, they sure set a high bar.
Saying "I know better" is quite easy. Backing that up is really hard.
> Some people were technical, but they didn't do technical work for many months, or longer, and now are no longer technical, they fell behind, but still think they are.
This seems like the same exact talk LeCun has been giving for years, basically pushing JEPA, world models, and attacking contemporary LLMs. Maybe he’s right but it also seems like he’s wrong in terms of timing or impact. LLMs have been going strong for longer than he expected, and providing more value than expected.
This is also my read; JEPA is a genuinely interesting concept, but he's been hawking it for several years, and nothing has come of it in the domains in which LLMs are successful. Hoping that changes at some point!
Yeah, he was quite vocal in his opinion that they would plateau earlier than they did and that little value would be derived from them because they're just stochastic parrots. Agree with him that they're probably not sufficient for AGI, but, at least in my experience, they're adding a lot of value and they're continuously performing better in a range of tasks that he wasn't expecting them to.
LeCun has been giving the same talk with literally the exact same slides for the past 3 years. JEPA still hasn't delivered despite FAIR's substantial backing.
LeCun seems like an extremely smart person that suffers from an overgrown ego. I got that strong impression from seeing his Twitter feed - basically "smarter than thou".
Generative world models seem to be doing ok. Dreamer V4 looks promising. I’m not 100% sold on the necessity of EBMs.
Also I’m skeptical that self-supervised learning is sufficient for human level learning. Some of our ability is innate. I don’t believe it’s possible for statistical methods to learn language from raw audiovisual data the way children can.
Human DNA has under 1GB of information content in it. Most of which isn't even used in the brain. And the brain doesn't have a mechanism to read data out from the DNA efficiently.
This puts a severe limit on how much "innate knowledge" a human can possibly have.
Sure, human brain has a strong inductive bias. It also has a developmental plan, and it follows that plan. It guides its own learning, and ends up being better at self-supervised learning than even the very best of our AIs. But that guidance, that sequencing and that bias must all be created by the rules encoded in the DNA, and there's only this much data in the DNA.
It's quite possible that the human brain has a bunch of simple and clever learning tricks that, if we pried out and applied to our AIs, would give us x100 the learning rate and x1000 the sample efficiency. Or it could be that a single neuron in the human brain is worth 10000 neurons in an artificial neural network, and thus, the biggest part of the "secret" of human brain is just that it's hilariously overparameterized.
I don't know why people really dislike the idea of innate knowledge so much, it's obvious other animals have tons of it, why would we be any different.
The problem with assuming tons of innate knowledge is that it needs to be stored somewhere. DNA certainly contains enough information to determine the development of various different neuron types and which kinds of other neurons they connect to, but it certainly cannot specify weights for every individual synapse, except for animals with very low neuron counts.
So the existence of a sensorimotor feedback loop for a basic behavior is innate (e.g. moving forward to seek food), but the fine-tuning for reliabily executing this behavior while adapting to changing conditions (e.g. moving over difficult terrain with an injured limb after spotting a tasty plant) needs to be learned through interacting with the environment. (Stumbling around eating random stuff to find out what is edible.)
>certainly cannot specify weights for every individual synapse
That's not the only way to one could encode innate knowledge. Besides, we have demonstrated that animals have innate knowledge experimentally many times, the only reason we can't do this to humans is that it would be horrifically unethical.
>Stumbling around eating random stuff to find out what is edible
Plenty of animals have innate knowledge about what is and isn't edible: it's why, for example, tasty things generally speaking smell good and why things that are bad (rotting meat) smell horrific.
I'm not saying that there's no innate knowledge. This entire list of reflexes https://en.wikipedia.org/wiki/List_of_reflexes is essentially a list of innate knowledge in humans, many of which have been demonstrated in newborns, apparently without considering such experiments unethical.
I'm saying that there are limits to how much knowledge can be inherited. I.e. the question isn't "Where could innate knowledge be encoded other than in synapses?" but "Considering the extremely large number of synapses involved in complex behavior far exceeds genetic storage capacity, how are their weights determined?" And since we know that in addition to having innate behaviors, animals are also capable of learning (e.g. responding to artificial stimuli not found in nature), it stands to reason that most synapse weights must be set by a dynamic learning process.
> That's not the only way to one could encode innate knowledge.
Maybe sections could be read from DNA and broadcast as action potentials?
There's already ribosomes that go over RNA. You'd need a variant which instead of making amino acids, would read out the base pairs and make something that causes action potentials to happen based on the contents.
Various reasons
Some people just believe there is no innate knowledge or we dont need it if we just scale/learn better (in the direction of Bitter Lesson)
(ML) Academia is also heavily biased against it due to mainly two reasons: - Its harder to publish, since if you learn Task X with innate Knowledge, its not as general, so reviewer can claims its just (feature) engineering - Which hurts acceptance, so people always try to frame their work as general as possible - Historical reasons due to the conflict the symbolic community (which rely heavily on innate knowledge)
But generative models are always going to seem like they are doing ok. That's how they work. They are good at imitating and producing misleading demos.
Agree with LeCun that current ai doesn’t exhibit anything close to actual intelligence.
I think the solution lies into cracking the core algorithms used by nature to build the brain. Too bad it’s such an inscrutable hairball of analog spaghetti code.
I think that LeCun has correctly identified that LLM is only one type of intelligence and that AGI/AMI needs to combine multiple other types … hierarchical goal setting, attention/focus management, and so on.
Seems that he is able to garner support for his ideas and to make progress at the leading edge - yes a little bit hard to take the “I know better” style, but then many innovations are driven by narcissism.
There is a lot of "transformer LLMs are flawed" going around, and a lot of alternative architectures being proposed, or even trained and demonstrated. But so far? There's nothing that would actually outperform transformer LLMs at their strengths. Most alternatives are sidegrades at best.
For how "naive" transformer LLMs seem, they sure set a high bar.
Saying "I know better" is quite easy. Backing that up is really hard.
To quote Zuck:
> Some people were technical, but they didn't do technical work for many months, or longer, and now are no longer technical, they fell behind, but still think they are.
Where is this from?
https://youtu.be/WuTJkFvw70o?t=2340
This seems like the same exact talk LeCun has been giving for years, basically pushing JEPA, world models, and attacking contemporary LLMs. Maybe he’s right but it also seems like he’s wrong in terms of timing or impact. LLMs have been going strong for longer than he expected, and providing more value than expected.
This is also my read; JEPA is a genuinely interesting concept, but he's been hawking it for several years, and nothing has come of it in the domains in which LLMs are successful. Hoping that changes at some point!
>LLMs have been going strong for longer than he expected
Have they? They still seem to be a dead end toward AGI.
Yeah, he was quite vocal in his opinion that they would plateau earlier than they did and that little value would be derived from them because they're just stochastic parrots. Agree with him that they're probably not sufficient for AGI, but, at least in my experience, they're adding a lot of value and they're continuously performing better in a range of tasks that he wasn't expecting them to.
2 more years bro.
Give it up Yann…LLMs won.
They won for now...