Systems from Google, OpenAI, Anthropic, and xAI struggle with the Premier League.
See full article...
See full article...
AI models are terrible at betting on soccer—especially xAI Grok
- Thread starter JournalBot
- Start date
Interesting site, thanks.
The "average chess.com player" rating in that list is probably meaningless. Presumably dragged down by all the random Joe Sixpacks who barely know the rules of chess, sign up for chess.com, play a game or two, lose, and then never play again.
IIRC the median rating of a USCF club player is around 1200. These are the people I would call "chess players."
Upvote
-1
(3
/
-4)
There's nothing more tedious than the "soccer vs. rugby" debate that occurs whenever a Brit discovers that American English differs a little from British English. I must have stumbled on this same exact exchange a hundred times over my years on the Internet.
Yup, different dialects sometimes have different words or spellings for things. Incredible, I know. Can we move on?
Yup, different dialects sometimes have different words or spellings for things. Incredible, I know. Can we move on?
Upvote
2
(2
/
0)
You seem to be associating surprise to the the wrong person. "Brits" are only too aware of the linguistic foibles of USAns.
Upvote
1
(3
/
-2)
Did they try using anthropics newest scariest terrifying model Mythos on this. I was told its so powerful it cant be used by the public to audit its effectiveness. A data driven decision like betting surely should be dominated by highest end AI systems right? I guess if we spend a few more trillion on data centers and training they will be able to do better than chance.
Upvote
1
(1
/
0)
say, how do you pronounce "Cholmondeley" and "Quernmore" ?
Upvote
0
(0
/
0)
No, you need to prove that, and you need to provide evidence for how they are different.
Upvote
3
(3
/
0)
1000?
You gotta be joking.You can get stockfish 18 (3600+ ELO) to run on your phone. Pretty sure these big companies won't have a problem running and doing tool calling to that. So they definitely don't do it.
Upvote
-2
(1
/
-3)
1000?
You can get stockfish 18 (3600+ ELO) to run on your phone. Pretty sure these big companies won't have a problem running and doing tool calling to that. So they definitely don't do it.
just note that Stockfish is not an LLM. it is a NNUE
Upvote
4
(4
/
0)
I agree - I didn't say it was. The prior poster was saying that the reason LLMs are improving at chess is because they're calling engines as tools in the bg.
I was replying that if that were the case - they wouldn't be at 1400 ELO.
Upvote
-1
(0
/
-1)
great.
NM. brainfart.
Last edited:
Upvote
0
(0
/
0)
I have provided evidence:
https://blog.google/innovation-and-ai/models-and-research/google-deepmind/kaggle-game-arena-updates/
From the article:
"While traditional chess engines like Stockfish function as specialized super-calculators, evaluating millions of positions per second to find the optimal move, large language models do not approach the game through brute-force calculation. Instead, they rely on pattern recognition and ‘intuition’ to drastically reduce the search space — an approach that mirrors human play.
Gemini 3 Pro and Gemini 3 Flash currently have the top Elo ratings on the leaderboard. The models’ internal ‘thoughts’ reveal the use of strategic reasoning grounded in familiar chess concepts like piece mobility, pawn structure, and king safety. This significant performance increase over the Gemini 2.5 generation highlights the rapid pace of model progress and demonstrates Game Arena’s value in tracking these improvements over time."
Upvote
-4
(1
/
-5)
Huh? Are you not following the thread?
They're not being 'shown' (no tools are being called) if they were being called their ELO would be much greater than 1400.
Upvote
0
(1
/
-1)
... and you think Stockfish 18 is "the most basic chess engine"?1000?
You can get stockfish 18 (3600+ ELO) to run on your phone. Pretty sure these big companies won't have a problem running and doing tool calling to that. So they definitely don't do it.
Upvote
0
(1
/
-1)
Upvote
2
(2
/
0)
No you can take cull that list down to people who have played more than 100 games and it would still be in the 500s.
USCF club players need to pay for membership and official ratings - not to mention that you have play 20 odd games to get a non-provisional rating. You can't use a paid tier as a purity test for something as simple as a 'chess player'.
It can run on your phone. Exactly how small do you want to go without deliberately nerfing it? Do you think these big companies like Google or Anthropic will have any trouble running them and calling them as tools?
Upvote
-4
(0
/
-4)
Over the decades, people have created thousands of different chess engines.
They're all different from each other in terms of how "basic" they are.
Stockfish 18 is one of the most complicated, sophisticated chess engines ever made. Maybe THE most complicated and sophisticated.
Suggesting that it might be "basic" to any degree at all is catastrophically wrong.
Upvote
1
(3
/
-2)
I mean, this is getting to be a question of personal interpretation more than anything.
Personally, I wouldn't call somebody an "XYZ player" if they didn't play XYZ in some kind of organized fashion, usually being on a team or belonging to a league, paying dues, etc.
But if you think "XYZ player" just means somebody who knows how to play XYZ, or has done it a few times, fine.
Upvote
1
(2
/
-1)
Whether you want to call something that can run on a phone (vs something that needs a supercomputer to run) as 'basic' is a matter of interpretation.
But we have really deviated from the main point that these companies like Google will have no trouble calling Stockfish 18 as an MCP tool.
You seem to think that they might and I'm saying that there is absolutely no way they would choose a 1400 ELO engine that sometimes gives illegal moves (if such an engine exists).
When you give Gemini a FEN/PGN (chess game notation), its suggested move is all pure LLM next token prediction.
Upvote
0
(0
/
0)
I feel like we're speaking different languages. What does hardware have to do with how "basic" a chess engine is or isn't?
Yes, you can run Stockfish on a phone. What does that matter? How is that relevant to anything? You can run any chess engine on a phone. That has nothing to do with how "basic" a chess engine is or isn't. There is no chess engine that requires a supercomputer to run.
Huh? No I don't. I had no idea that was your main point, nor do I care.
I took issue with this sentence that you wrote:
"If they did a tool call to even the most basic chess engine, it would be at 3400+ elo not ~2000 elo."
Because the most basic chess engine is dramatically weaker than 3400+ Elo.
Here's a list of 602 chess engines, with ratings:
https://www.computerchess.org/
89% of them are rated under 3400. So I don't know why you think "even the most basic chess engine" would be rated 3400+.
Upvote
2
(3
/
-1)
Play CHess Against ChatGPT
it is extremely bad, and absolutely loves illegal moves.
ex. it's playing black. and it either moved its white bishop from c8 to b4, leaping over pawns and landing on a black square. or it moved its black bishop from g8 to b4, also jumping a pawn to get there.
(the site that's drawing the board has no way to know which illegal and nonsensical move it was. 'Bb4' is an illegal move either way)
it is extremely bad, and absolutely loves illegal moves.
ex. it's playing black. and it either moved its white bishop from c8 to b4, leaping over pawns and landing on a black square. or it moved its black bishop from g8 to b4, also jumping a pawn to get there.
(the site that's drawing the board has no way to know which illegal and nonsensical move it was. 'Bb4' is an illegal move either way)
Last edited:
Upvote
2
(3
/
-1)
Upvote
0
(1
/
-1)
Back in the day the head of my father's firm of solicitors relied on local chauvinism for betting on football. Bookmakers lay off the odds based on their exposure.
So, London matches. E.g. Chelsea versus Tottenham. Quick trip to Chelsea where locals are putting money on Chelsea to win, so the odds for Tottenham are better. Reverse process in Tottenham. Based on who wins, collect profit. This got harder when gambling tax came in, but worked surprisingly well as a strategy till computers came in.
As I mentioned above, it was the British Prime Minister and brilliant statistician Harold Wilson who demonstrated in a paper that the results of First Division football matches were indistinguishable from randomness. Not that it mattered. The "Football pools" were as big a scam as the Lottery, with often less than a third of money bet being returned to winners.
Upvote
0
(0
/
0)
Fine if you have a problem with my usage of the term 'most basic engine' - I will concede that. It doesn't detract from my main point about LLMs not calling engines as tools (they'd be stronger than 1400 ELO).
Upvote
0
(0
/
0)
Again what does “intelligence” ? I’m willing to bet you can’t define it.
Upvote
-1
(2
/
-3)
The takeaway from this may be that "general purpose AI models fail in a highly competitive market where bespoke AI is likely a significant factor."
Given the size of the sport betting market, there is no way that we're not already seeing a good number of bets made with AI assistance, and those would be refined models for this specific task.
It would be interesting to simulate how well the betting would go against games and odds from a decade ago, where there would have been a fair amount of computer assistance but not modern AI.
Given the size of the sport betting market, there is no way that we're not already seeing a good number of bets made with AI assistance, and those would be refined models for this specific task.
It would be interesting to simulate how well the betting would go against games and odds from a decade ago, where there would have been a fair amount of computer assistance but not modern AI.
Upvote
1
(1
/
0)
Anyone with a rating above 400 knows what they're doing. People at that level are setting traps, understand and can avoid all of the early mating scenarios, and are thinking three or four moves ahead. Indian chess hustlers have taken over that middle area ever since Iran lost its internet. It's interesting how the ratings stratify like that. I miss chewing through the IRGC players.1000?
You can get stockfish 18 (3600+ ELO) to run on your phone. Pretty sure these big companies won't have a problem running and doing tool calling to that. So they definitely don't do it.
I know you guys are enthralled in your mechanical dick measuring contest, but I actually enjoy chess. With my brain.
Last edited:
Upvote
3
(3
/
0)
Hey don't blame me. I'm not the one putting random purity tests like 1200 elo and uscf membership.
Upvote
1
(1
/
0)
I read the article expecting to hear that ChatGPT lost badly at poker, which would have been understandable—almost all humans suck at poker, too, and one shouldn’t expect a general chat engine to bluff with perfect precision. What I didn’t expect to hear was that it couldn’t even declare the correct winners, or keep accurate account balances of the players.
Upvote
3
(3
/
0)
Yeah, "sorry" for trying to nail down what people mean when they say XYZ is better than 99% of "chess players."
Heaven forbid we have a common understanding of what the term "chess player" means. We might end up talking about the same thing.
Upvote
0
(2
/
-2)
For Chess, I know there have been chess engines since 1979. See https://en.wikipedia.org/wiki/Video_Chess . The question becomes, why reinvent the wheel? Clearly Chess engines are well known, and serve a purpose. Why reinvent the wheel?
Upvote
0
(0
/
0)
This is a pretty superficial take on things, akin to asking why we needed to invent the automobile when wheels have been around for thousands of years. Chess engines from 1979 bear little resemblance to the ones we have today, especially after the introduction of AlphaZero in 2017 made neural networks pretty much a required feature.
Upvote
0
(0
/
0)