is LLM a language specific AI?
Chess will never be solved, here's why
Sort:
no. three brown one blue does a pretty good job w/this umbrella clip. here:
https://www.youtube.com/watch?v=LPZh9BOjkQs
and justa huge warning ?... ddue barely knows a/t abt llms. hes just tryn2impress u. and trust me. hes gonna watch this. AND he'll learn lots & gobs.
Jun 23, 2025
0
#16764
I theory, chess could be solved. Stockfish, if we made him play over 1000 games against himself over the course of a decade, then chess would be theoretically solved. When he finished one game, he would go to another immediately. New 'better' AIs wouldn't be a problem because Hikaru, for example, would play the exact same moves as magnus would do, or ethereal, or gukesh,or even Stockfish, given unlimited time. Chess could be fundamentally solved, but I'm sure there's a few problems. Please point them out.
Jun 23, 2025
0
#16765
If the current iteration of Stockfish played a 5 year old version, the latest program would win. In 10 years there will be a version that can defeat today's Stockfish. This will not solve chess
Micloi wrote:
I theory, chess could be solved. Stockfish, if we made him play over 1000 games against himself over the course of a decade, then chess would be theoretically solved. When he finished one game, he would go to another immediately. New 'better' AIs wouldn't be a problem because Hikaru, for example, would play the exact same moves as magnus would do, or ethereal, or gukesh,or even Stockfish, given unlimited time. Chess could be fundamentally solved, but I'm sure there's a few problems. Please point them out.
Not even close. Hint: Engines are not perfect chess players. If they were, they would not need opening books, nor endgame tablebases. Everything else falls apart for your premise after that simple fact. Whether human beings can beat them has nothing to with chess being actually solved.
Jun 23, 2025
0
#16768
Let's say stockfish versus stockfish,10000 games. Every game, it will choose a 1 out of 50 most popular and optimised openings. It cannot repeat until it has done all 50. If it changes it's technique, stockfish has found a better strategy. As well as engines are not perfect, it could gives some insight for the question we are trying to solve.
Thee_Ghostess_Lola wrote:
no. three brown one blue does a pretty good job w/this umbrella clip. here:
https://www.youtube.com/watch?v=LPZh9BOjkQs
and justa huge warning ?... ddue barely knows a/t abt llms. hes just tryn2impress u. and trust me. hes gonna watch this. AND he'll learn lots & gobs.
I'm glad you found a 7 minute video that could teach you about LLMs.
If you consider yourself any kind of developer and everything in that video was not readily apparent to you the moment you first heard that LLMs train on massive amounts of text, I'm not sure what to tell you...probably "find another career, you have a low ceiling in this one".
The training datasets are self-evident, the RLHF "tweaking" is self-evident, the use of GPUs for parallel processing is self-evident. The first two dovetail with exactly what I said...LLMs are not a resource you can go to for advances/breakthroughs. They can assist you in sifting through everything so *you* can make a breakthrough...were you capable of one.
The only section that explains anything reasonably informative is the "attention" section, but even that is just going over the weighting tables and iterating, which, again, are pretty self-evident mechanisms for how LLMs work. The specific algorithms and parameters for the weighting and so how they will connect words contextually on the fly are the only real "beef" here, and this video does not cover that.
Micloi wrote:
Let's say stockfish versus stockfish,10000 games. Every game, it will choose a 1 out of 50 most popular and optimised openings. It cannot repeat until it has done all 50. If it changes it's technique, stockfish has found a better strategy. As well as engines are not perfect, it could gives some insight for the question we are trying to solve.
Insight is irrelevant here. You cannot extrapolate perfect play by using imperfect play to try and prove it.
What are examples of insights that are relevant to everyday chess?
watchout w/him 888. hes on a midnite picknick. with imperfect play chess is still he!!ahard solved. for ex. K+R vs K. thiss 'diamond' solved yet one doesnt need perfect play to solve it right ? iows u can skroo up along the way a still quackmate s/o.
so member to consider the source. epsilon doo is a international patzer AND takes hisself way too seriously lol !
sahrul888 wrote:
What are examples of insights that are relevant to everyday chess?
I trust that reading Lola's replies will help make it clear to you that I am correct.
Jun 27, 2025
0
#16776
There are 64 squares on the board. There are 32 pieces at most on the board at most. 64 times 32 is 204
I would not call LLMs "true" AI in the sense that most sci-fi authors over the decades have (that is, an artificial intelligence that thinks entirely on its own and possibly achieves self awareness), and LLMs are certainly not really applicable to solving chess. They are constrained by the information they have to work with, so the LLMs' conclusions are not going to be much better in the end than what we already know. Where an LLM comes into play is being able to sift through and refine consensus in a way that *assists* experts in taking incremental steps forward, but an LLM will not come up with a completely new breakthrough on its own.