korrents

Gary Marcus

@gary-marcus · 10 positions · 0 changes of mind

Cognitive scientist, professor emeritus of psychology and neural science at NYU, and the most persistent public critic of large language models. Author of The Algebraic Mind and Rebooting AI; writes the Marcus on AI newsletter.

Gary Marcus did not write this page.

We collected these quotes from things they published elsewhere, and every quote links to where it was said. They have no account here and have not endorsed this site. Quotes are word for word; the short line under each one is our own restatement, not their wording. Their own site. Is this you? Claim it or ask us to remove it. Or tell us what is wrong here.

  1. Our reading

    Almost every major figure in AI has now arrived at the critique of large language models he began making in 2019.

    Their wordsBut one by one, almost every major thinker in AI has come around to the critique of LLMs that I began presenting in 2019.

    ↗Game over for pure LLMs. Even Turing Award Winner Rich Sutton has gotten off the bus.garymarcus.substack.com

    LLMs

    Interesting
  2. 7 weeks earlier

  3. Our reading

    GPT-5 was an ordinary incremental advance, and the gap between what was promised for it and what shipped is the story of the field.

    Their wordsPeople had grown to expect miracles, but GPT-5 is just the latest incremental advance.

    ↗GPT-5: Overdue, overhyped and underwhelming. And that’s not the worst of it.garymarcus.substack.com

    OpenAI

    Interesting
  4. 2 months earlier

  5. Our reading

    There is no route to AI alignment or safety that goes around reliability: a system that cannot reliably follow a known algorithm cannot be made safe.

    Their wordsAnd good luck getting to “alignment” or “safety” without reliabilty.

    ↗A knockout blow for LLMs?garymarcus.substack.com 1st of 6 in this piece

    AI alignment

    Interesting

  7. Our reading

    AGI worth the name would combine human adaptiveness with machine reliability, not reproduce human performance including human failures.

    Their wordsThe vision of AGI I have always had is one that combines the strengths of humans with the strength of machines, overcoming the weaknesses of humans.

    ↗A knockout blow for LLMs?garymarcus.substack.com 2nd of 6 in this piece

    AGI

    Interesting

  9. Our reading

    A language model is no substitute for a well-specified conventional algorithm, so it cannot simply be dropped into a complex problem and trusted.

    Their wordsWhat the Apple paper shows, most fundamentally, regardless of how you define AGI, is that LLMs are no substitute for good well-specified conventional algorithms.

    ↗A knockout blow for LLMs?garymarcus.substack.com 3rd of 6 in this piece

    AGILLMsApple

    Interesting

  11. Our reading

    Success on the easy version of a task is evidence of nothing, because it seduces you into believing the model found a general solution when it did not.

    Their wordsWorse, as the latest Apple papers shows, LLMs may well work on your easy test set (like Hanoi with 4 discs) and seduce you into thinking it has built a proper, generalizable solution when it does not.

    ↗A knockout blow for LLMs?garymarcus.substack.com 4th of 6 in this piece

    LLMsApple

    Interesting

  13. Our reading

    Anyone who thinks large language models are a direct route to transformative AGI is kidding themselves.

    Their wordsBut anybody who thinks LLMs are a direct route to the sort AGI that could fundamentally transform society for the good is kidding themselves.

    ↗A knockout blow for LLMs?garymarcus.substack.com 5th of 6 in this piece

    AGILLMs

    Interesting

  15. Our reading

    The LLM line of research will reach a capability plateau.

    Their wordsBut this particular approach has limits that are clearer by the day.

    ↗A knockout blow for LLMs?garymarcus.substack.com 6th of 6 in this piece

    LLMsscaling laws

    Interesting
  16. 8 months earlier

  17. Our reading

    Large language models do not reason formally: their performance collapses as a problem is made bigger, in the way a calculator's never does.

    Their wordsAnother manifestation of the lack of sufficiently abstract, formal reasoning in LLMs is the way in which performance often fall apart as problems are made bigger.

    ↗LLMs don’t do formal reasoning - and that is a HUGE problemgarymarcus.substack.com 1st of 2 in this piece

    LLMs

    Interesting

  19. Our reading

    Scaling the 2019-2023 recipe — same architecture, bigger model, more data — is not enough; further progress depends on new architectural ideas.

    Their wordsNeurosymbolic AI — combining such machinery with neural networks – is likely a necessary condition for going forward.

    ↗LLMs don’t do formal reasoning - and that is a HUGE problemgarymarcus.substack.com 2nd of 2 in this piece

    OpenAIscaling laws

    Interesting