Gary Marcus

Rank 21 of 47
|
Score 52
6/28/2025, 9:08:40 PM
X
In reply to:
Lawrence Chan
@justanotherlaw
·
53d
@GaryMarcus @ohabryka I don’t think you intended readers to take away this claim! I think your point is that we can’t easily get >0.999999 reliability via scaling due to how the transformer architecture (and more broadly, modern deep learning) works. Thus, each failure is evidence that we’re not at
Lawrence Chan
@justanotherlaw
·
53d
@GaryMarcus @ohabryka Yes, I agree that under a careful narrow reading, you were correct. GPT-2 probably doesn’t have a world model by any reasonable use of the word, and GPT-3 fails basic physics questions without careful prompting. But under a more cursory reading, the claims that these failings
Gary Marcus
@GaryMarcus
·
53d
@justanotherlaw @ohabryka rather, instead, i highlighted specific problems that have not in fact yet been solved.
Lawrence Chan
@justanotherlaw
·
53d
@GaryMarcus @ohabryka While I think your specific predictions have been above average, I do think people would’ve been quite wrong about how AI in 2025 was like if they internalized your general mood of “AI won’t work” in 2019-20. Oli is probably referring to that.
Gary Marcus
@GaryMarcus
·
53d
@justanotherlaw @ohabryka it was completely unfair, and his nitpicking exercise didn’t change that.
Lawrence Chan
@justanotherlaw
·
53d
@GaryMarcus @ohabryka I do think it’s quite unfair to say that Gary is “exceptionally bad”! He’s doing better than the horde of people who confidently asserted that AIs wouldn’t be able do particular things that they did.

The statement is part of a broader discussion on AI reliability and the interpretation of claims about AI capabilities. It engages with public discourse on the topic of AI development and its limitations, particularly in the context of deep learning and transformer architectures.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not appear to cause harm with words or actions. It is a question seeking clarification, which is neutral in tone.
  2. Principle 4:
    I will engage in constructive criticism and dialogue with those in disagreement and will not engage in personal attacks or ad hominem arguments.
    The statement uses the term 'strawman,' which could imply a misrepresentation of an argument. However, it does not engage in personal attacks or ad hominem arguments, maintaining a focus on the discussion topic. [-1]