3 ms·

Even though I largely disagree with Chomsky, he is extremely intelligent. That Hinton and Sejnowski behave like sore losers is a bit of a disappointment. The ar

by tgv 4y ago

Even though I largely disagree with Chomsky, he is extremely intelligent. That Hinton and Sejnowski behave like sore losers is a bit of a disappointment. The argument with the learning disabled student is a disgrace. It's almost as if they cannot imagine that the human brain has innate properties beyond "weight adaptation".

epups 4y ago
I did not see a problem with Hinton's argument here - perhaps you would care to expand? If we are evaluating LLM's purely on outcome alone, it seems only fair to compare them to humans that don't have fully functioning cognitive capabilities. We wouldn't discredit a schizophrenic person that hallucinates, we know that this person can also see reality for what it is, most of the time. I just don't think that finding absurd examples, usually after careful prompting, is a slam dunk against LLM's at all.
naasking 4y ago
> The argument with the learning disabled student is a disgrace. It's almost as if they cannot imagine that the human brain has innate properties beyond "weight adaptation". They can, it's just that Chomsky's dismissal of "weight adaptations" as insufficient is not actually justified by any evidence. The Transformer model augmented with memory is Turing complete, so it can in principle learn to compute anything using only weight adaptations.
- tgv 4y ago
  No, it cannot. Even CFGs aren't learnable.
  - naasking 4y ago
    * On the Turing Completeness of Modern Neural Network Architectures, https://arxiv.org/abs/1901.03429 https://arxiv.org/abs/1901.03429 * On the Computational Power of Transformers and its Implications in Sequence Modeling, https://arxiv.org/abs/2006.09286 https://arxiv.org/abs/2006.09286
    - tgv 4y ago
      A TM cannot learn an arbitrary CFG. It needs constraints. Those constraints must also exist in our brain. At birth. It's like an NN architecture. Our brains are much more complex than an artificial NN, though. So the mere thought that the two somehow share functionality after a wildly different learning process is somewhat far-fetched.
      - naasking 4y ago
        > A TM cannot learn an arbitrary CFG. It needs constraints. Well sure, because CFGs can have unbounded depth, and all real brains are finite. But they can learn bounded CFGs: https://arxiv.org/abs/2112.09174 https://arxiv.org/abs/2112.09174 > Our brains are much more complex than an artificial NN, though. Yes, brains have a few orders of magnitude more parameters, and lots of evolutionary baggage. It's not clear how many parameters would be needed without that evolutionary baggage. Time will tell. > So the mere thought that the two somehow share functionality after a wildly different learning process is somewhat far-fetched. I don't know what precisely you mean by "share functionality", but it doesn't seem like a stretch to suggest they are in some sense isomorphic given Turing equivalence. Rule 110 doesn't look like it should share much with Turing machines or the lambda calculus or SKI combinators, but these are all equivalent in a formal sense.
        tgv 4y ago
        That's not what learnable means, but it would only show there needs to be an a priori limit on the structure of learning. > in some sense isomorphic given Turing equivalence Then you're brain is just like any CPU, which is very far from what such a statement tries to express in the context of NNs.
        naasking 4y ago
        > Then you're brain is just like any CPU, which is very far from what such a statement tries to express in the context of NNs. Yes, the brain is like a CPU + a specific program running on that CPU. The properties of that program are where the isomorphism comes in. So to come full circle back to your original point, we don't yet know that "weight adaptations" are not a sufficient operational model for that program.
MrPatan 4y ago
Our vocal cords evolved for language, but the brain remained as rasa as a tabula can be.
- tgv 4y ago
  We are not born tabula rasa. Did you ever see a calf get born? It knows how to get onto its feet in minutes, find its mother and start feeding. That's hard-wired. Hormone secretion? Hard-wired. Reflexes? Hard-wired. Creating memories? Hard wired. Identification of sound direction? Hard-wired. Face recognition? Probably hard-wired too. Our CNS comes with a ton of instincts and basic processes. Why wouldn't that extend to other areas? In contrast, the world is too complicated to learn through exposure. Something has to structure our learning process. Language is (in principle) not learnable, only when you make enough assumptions. Those assumptions must be coded somewhere in your brain and probably genes. Tabula rasa sounds nice, but it is untenable and utterly lacks evidence.

by tgv 4y ago

epups 4y ago

I did not see a problem with Hinton's argument here - perhaps you would care to expand? If we are evaluating LLM's purely on outcome alone, it seems only fair to compare them to humans that don't have fully functioning cognitive capabilities. We wouldn't discredit a schizophrenic person that hallucinates, we know that this person can also see reality for what it is, most of the time. I just don't think that finding absurd examples, usually after careful prompting, is a slam dunk against LLM's at all.

naasking 4y ago

> The argument with the learning disabled student is a disgrace. It's almost as if they cannot imagine that the human brain has innate properties beyond "weight adaptation". They can, it's just that Chomsky's dismissal of "weight adaptations" as insufficient is not actually justified by any evidence. The Transformer model augmented with memory is Turing complete, so it can in principle learn to compute anything using only weight adaptations.

tgv 4y ago
No, it cannot. Even CFGs aren't learnable.
- naasking 4y ago
  * On the Turing Completeness of Modern Neural Network Architectures, https://arxiv.org/abs/1901.03429 https://arxiv.org/abs/1901.03429 * On the Computational Power of Transformers and its Implications in Sequence Modeling, https://arxiv.org/abs/2006.09286 https://arxiv.org/abs/2006.09286
  - tgv 4y ago
    A TM cannot learn an arbitrary CFG. It needs constraints. Those constraints must also exist in our brain. At birth. It's like an NN architecture. Our brains are much more complex than an artificial NN, though. So the mere thought that the two somehow share functionality after a wildly different learning process is somewhat far-fetched.
    - naasking 4y ago
      > A TM cannot learn an arbitrary CFG. It needs constraints. Well sure, because CFGs can have unbounded depth, and all real brains are finite. But they can learn bounded CFGs: https://arxiv.org/abs/2112.09174 https://arxiv.org/abs/2112.09174 > Our brains are much more complex than an artificial NN, though. Yes, brains have a few orders of magnitude more parameters, and lots of evolutionary baggage. It's not clear how many parameters would be needed without that evolutionary baggage. Time will tell. > So the mere thought that the two somehow share functionality after a wildly different learning process is somewhat far-fetched. I don't know what precisely you mean by "share functionality", but it doesn't seem like a stretch to suggest they are in some sense isomorphic given Turing equivalence. Rule 110 doesn't look like it should share much with Turing machines or the lambda calculus or SKI combinators, but these are all equivalent in a formal sense.
      - tgv 4y ago
        That's not what learnable means, but it would only show there needs to be an a priori limit on the structure of learning. > in some sense isomorphic given Turing equivalence Then you're brain is just like any CPU, which is very far from what such a statement tries to express in the context of NNs.
        naasking 4y ago
        > Then you're brain is just like any CPU, which is very far from what such a statement tries to express in the context of NNs. Yes, the brain is like a CPU + a specific program running on that CPU. The properties of that program are where the isomorphism comes in. So to come full circle back to your original point, we don't yet know that "weight adaptations" are not a sufficient operational model for that program.

MrPatan 4y ago

Our vocal cords evolved for language, but the brain remained as rasa as a tabula can be.

tgv 4y ago
We are not born tabula rasa. Did you ever see a calf get born? It knows how to get onto its feet in minutes, find its mother and start feeding. That's hard-wired. Hormone secretion? Hard-wired. Reflexes? Hard-wired. Creating memories? Hard wired. Identification of sound direction? Hard-wired. Face recognition? Probably hard-wired too. Our CNS comes with a ton of instincts and basic processes. Why wouldn't that extend to other areas? In contrast, the world is too complicated to learn through exposure. Something has to structure our learning process. Language is (in principle) not learnable, only when you make enough assumptions. Those assumptions must be coded somewhere in your brain and probably genes. Tabula rasa sounds nice, but it is untenable and utterly lacks evidence.