Gary Marcus

Rank 13 of 47
|
Score 83
5/17/2025, 11:23:54 PM
X
In reply to:
Daniel Litt
@littmath
·
238d
@GaryMarcus @colin_fraser @MelMitchell1 @rao2z They’re clearly generalizing out of distribution. E.g. they can solve some novel math problems.
Gary Marcus
@GaryMarcus
·
238d
@littmath New situations is clearly the weakness in these approaches, and has been forever.

@colin_fraser has documented this endlessly, informally; I gave the intuitions behind that in 1998. The recent Apple reasoning paper, the studies of @MelMitchell1 and @rao2z etc reinforce it.

The
Daniel Litt
@littmath
·
238d
@GaryMarcus These seem fine.
Gary Marcus
@GaryMarcus
·
238d
@littmath Under some definitions of intelligence, maybe. What’s yours?
Daniel Litt
@littmath
·
238d
I don’t know how you can use frontier models and hold onto beliefs like this. Not sure about “conscious,” but for any reasonable meaning of “intelligent,” LLMs are already intelligent.

The statement critiques the capabilities of large language models (LLMs) in solving complex problems, particularly in novel situations. It engages in a technical discussion about the limitations of LLMs, focusing on their inability to reliably solve large multiplication problems without direct data augmentation. The tone is analytical and aims to provide a counterpoint to claims of LLMs' generalization abilities.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not cause harm and focuses on a technical critique. [+1]
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The statement promotes understanding by providing specific examples of LLM limitations. [+1]
  3. 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 engages in constructive criticism without personal attacks. [+1]