Why LLMs Can't Play Chess(nicowesterdale.com)
nicowesterdale.com
Why LLMs Can't Play Chess
https://www.nicowesterdale.com/blog/why-llms-cant-play-chess
3 comments
One of my favorite things about the current generation of "AI" is that it puts a powerful analytical mirror up to what language actually is. Like, quantitatively - what is language.
Why do I take pleasure in this?
In document engineering and contentlandia in general, there have been so, so, so, so many generations of "We've Figured Out Content Architecture", for many decades, generations even. Sometimes bordering on outright ideology, with party faithful, catemites, and all the rest. Each Vision pushed by these Faithful have added hours, weeks, centuries to the labor devoted to the human work of communication, with gains that rarely if ever stack up in positive integers. So it has been deeply satisfying to watch these deep networks of hierophant priesthoods struck dumb-angry by the simple (but inexplicable!) mathematics underlying meaning.
Thing is, it was predictable, from a traditional computational linguistics perspective, that formalizing structure on top of - as a deep structure! - the corpus of all natural language was going to be a "squaring the circle" sort of endeavor. But these priests and evangelists didn't come from computational anything, so they floated on past the fundamentals like a defensebro who doesn't know what "Bernoulli" means.
Why do I take pleasure in this?
In document engineering and contentlandia in general, there have been so, so, so, so many generations of "We've Figured Out Content Architecture", for many decades, generations even. Sometimes bordering on outright ideology, with party faithful, catemites, and all the rest. Each Vision pushed by these Faithful have added hours, weeks, centuries to the labor devoted to the human work of communication, with gains that rarely if ever stack up in positive integers. So it has been deeply satisfying to watch these deep networks of hierophant priesthoods struck dumb-angry by the simple (but inexplicable!) mathematics underlying meaning.
Thing is, it was predictable, from a traditional computational linguistics perspective, that formalizing structure on top of - as a deep structure! - the corpus of all natural language was going to be a "squaring the circle" sort of endeavor. But these priests and evangelists didn't come from computational anything, so they floated on past the fundamentals like a defensebro who doesn't know what "Bernoulli" means.
LLMs can play chess, Gemini 3.1 Pro reaches an accuracy on 96.9 on chess.com and wins regularly against stockfish 18 set on elo 2000
No doubt that AI is useful, but not generally useful for everything.