Gary Marcus

Rank 10 of 47

Score 110

@RobHafernik LLMs get trained on millions of games of chess, and the explicit rules, and still make illegal moves.

humans with a thousandth of that experience don’t make illegal moves, and many of them play much better.

6/15/2025, 3:10:06 PM

In reply to:

Rob Hafernik

@RobHafernik

41d

@GaryMarcus In other words, humans have to specifically train their neural nets to play chess and even then most of them are bad at it. Why would LLMs be any different?

Rob Hafernik

@RobHafernik

41d

@GaryMarcus I wonder how many HUMANS that Atari could beat? More than half of them, I would bet. It could probably beat any human who wasn't a serious chess player. I wonder what percentage of the population plays chess?

Gary Marcus

@GaryMarcus

41d

for $55 you can buy an Atari 2600 that can beat a billion dollar LLM in chess.

The statement critiques the performance of language models (LLMs) in playing chess compared to humans, highlighting that despite extensive training, LLMs still make errors that humans with less experience do not. The tone is analytical and somewhat critical, aiming to point out the limitations of LLMs in specific tasks like chess.

Principle 1:
I will strive to do no harm with my words and actions.
The statement does not cause harm; it is a critique of technology, not individuals.
Principle 3:
I will use my words and actions to promote understanding, empathy, and compassion.
The statement promotes understanding by discussing the limitations of LLMs in a specific context. [+1]
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 of LLMs without personal attacks. [+1]