carodgers

They're notoriously, hilariously bad at chess, even today. So yes, agree, their shortcomings come much more forcefully and immediately to light when they attempt spatial tasks.

You might be interested in the con leche project. A subset of the Lean theorem prover which was sufficient to prove the entire contents of Lean mathlib has been proven consistent.

Without exaggeration, given a choice between models, I would pay for Mistral's model over Anthropic's based on the name alone, completely ignoring features or other technical considerations. The name is playful and is such a refreshing contrast to Anthropic's (and OpenAI's) doomsaying, scaremongering, and god-posturing.

EDIT: Can no longer edit my own comment. Lots of folks, on closer inspection, are finding errors in the board behavior. Shame on me for not validating the output before praising the author.

Yeah, seems the longer I look the more problems I find. Circuits on the PCB terminate at nonsensical points or overlap where they shouldn't. I'm disappointed. I'm also disappointed in myself because I knew this was LLM generated and did not pause to validate before I gave praise to the author.

If you have not tried panning the camera and zooming in to one of the chips, you are missing out.

Bravo to the author. This is inspiring work.

> Claude Fable would destroy any human at chess by coding a strong enough engine on the fly.

A bash script can clone and build stockfish, feed in human moves, and reply. By your standard, this bash script would "destroy any human at chess."

Are you interested in assessing the intelligence of the model, or the intelligence of the tools the model can use?

This April 2026 paper is a fun and related read.

https://arxiv.org/html/2509.24239v4

Researchers asked frontier models to play chess. Have a look at the MAR rates in Table 3. When not explicitly told which moves were legal, no model identified legal moves at a rate better than 80%. Many asked for more illegal moves than legal moves. And even when explicitly told which moves were legal, the models continued to ask for illegal moves. With illegal asks discarded, none of the bots could beat a chess model calibrated to 1100 ELO.

The author of the originating post says that "current frontier models need laborious oversight and guardrails on even the simplest tasks", and he's absolutely correct.

"It just can't play it on a master level."

Technically correct. Losing to random play is definitely not master level.

"Chess is not an interesting target for AI Research and not for AGI."

I don't care if it's interesting or not. A bot too stupid to play chess is too stupid to be a threat to me. So I reject Dario's fearmongering and that of anyone allied with him.

"Do you really think if you would get unlimited tokens and a compute cluster that something like a fable or astra frontier model setup couldn't build a better chess engine?"

Disconnect it from the internet, where it cannot copy from humans, and no, it certainly will not. If it can copy code from humans, then maybe, with guidance. But if a human is not there to offer guidance, no, I don't think it will improve on stockfish.

Worse, if it cannot copy code from humans, it will fail embarrassingly at playing chess on its own. This April 2026 paper found that frontier models which did not have specialized (human guided) tailoring could not beat an amateur-level bot (1100 ELO), and some were so bad they lost to random play.

https://arxiv.org/html/2509.24239v4

Look at the MAR rates in table 3. When not explicitly provided with a choice of legal moves, no bot could ask for legal moves at a rate better than 80%. Many of them asked for more illegal moves than legal moves. All of them continued to request illegal moves even when explicitly asked to choose from legal moves. That's not just less than perfect. That's abysmal.

So, recapping, if the most advanced bots on the planet, in April 2026, are not allowed to copy code written by humans, and are forced to play chess by themselves, they will embarrass themselves. And you and others are asking me to believe they are a threat to me somehow. That kind of talk is disconnected from reality.

What is common with all of the "impressive" examples you gave me? The bot had material to copy from, or a human standing nearby to tell it when it screwed up and suggest a path forward. They are not intelligent. They're so stupid, in fact, I could not trust them to balance my checkbook without knowing guard rails are in place. So spare me the hysterics.

"When do you draw the line of LLM is just some stupid thing vs. AI/AGI?" I'm not sure, but I do know the line is well beyond playing a game of chess without requesting illegal moves 20% of the time. There are 5-year-olds who can do that.

You need to ask yourself some hard questions.

1) Why have I been conditioned to expect that given enough compute, a bot can build anything, unguided?

2) Why have the embarrassing details about bot performance been hidden from me, or at least drowned out by hysterical claims about their wild successes? Why is the human role in these successes never highlighted?

3) What do the people training these bots stand to gain if they convince 51% of the voting population that bots are actually a meaningful threat?

I am happy. AI is great. I use it. I'm glad that the U.S. is building datacenters. I work for a company involved in siting and building datacenters. But AI is not as scary or dangerous as Dario claims.

I am an American taxpayer. The money to pay the regulators that Dario wants to permanently hire will come out of my paycheck and will take from me time I could spend interacting with family, traveling, or pursuing hobbies. Dario and others demanding this can get bent.

"You picked a piss poor metric and you are getting piss poor results."

Well, we agree that the results are piss poor, at least.

Chess ELO is not a proxy for general intelligence because many intelligent people choose to learn things other than chess. But any intelligent person can, if they choose, learn chess and play it well.

LLMs cannot do this. They still play very poorly relative to human pros, and even modern LLMs still understand the rules so poorly that they request illegal moves.

Anyone claiming that we should be afraid of a text generator that can't understand the rules of chess is being ridiculous.

"And an LLM can easily, today, just rebuild a chess engine and beat whatever they need to beat."

Always claims and promises about what they "can" do. This pattern is constant across AI hype. No one can prove that an LLM will forever fail to solve problem X, so the claim goes uncontested, and humans are biased to believe unrebutted claims.

I reject this fallacious line of reasoning. I will instead require evidence of what they "have" done. No LLM to date has written a chess engine that beats stockfish. More importantly, no LLM has ever written any chess engine without heavily copying code and patterns from humans.

There's been substantial effort put into designing and running benchmarks for these models. You know which one they don't report any more? Chess ELO.

Watt for watt, every one of these models loses to stockfish. There is no possible scenario where we need to be afraid of "rogue superintelligence" when the intelligence in question cannot reason or plan well enough to play chess.

Lecun has the correct approach. Mock these people relentlessly for their attention-seeking doomerism.

"Delivers something they wanted" is the most bizarre way possible to phrase "followed through with the majority of a reprehensible agenda and were forced to carve out an exception due to technical constraints and massive blowback."

Also the screen sizes they are showing have got to be exceedingly heavy.

Even before LLMs, proving that a specific human wrote some piece of text was difficult or imprecise due to coauthoring, editing, plagiarism, etc. The important question then, as now, is rather to determine whether a specific human approved some piece of text for publication under their name.

I would just like to say that with this page open I have ~15% utilization on my GPU. XD

This is wrong, given only the text of the article. The two books explicitly listed are not banned in Portugal or anywhere else. They are simply not subsidized with public money in some libraries.

Only two books are explicitly listed by the article, neither of which have been banned anywhere.

Unsubsidized is the more accurate word here. Some governments have chosen not to pay public money to stock these books in libraries, but no government has created criminal penalties for ownership.

It may be the case that her library includes some books that genuinely carry criminal penalties, but the article does not provide enough info to assess that.

It behooves every company working on some form of actual AI to immediately switch branding to distance themselves from the slop bots. Call it image recognition, automated translate, inventory automation, what have you.

I tried to use claude for a completely menial task yesterday, adding some resistant-starch foods to my diet, and it was worse than useless. Actively wasting my time with claims that buckled as soon as I questioned them and completely falsified citations from product pages that it recommended.

If you are branding your product as "AI" when these word salad spitters are still dominating headlines, you deserve all the headwinds coming to you.

Another day, another breathless claim about the abilities of LLMs based on output that looks valid to underinformed user but is actually defunct.

On other topics I imagine the conversation would flow better and, I hope, be just as lively and engaging. But to be direct, no, I'm not eager to continue when universes that admit contradictions are on the table. A conversation that permits contradictions is not likely to be productive. So I'll be content to just restate my point. Cheers!

What does it mean to admit a contradiction? Give an example.

Modus ponens holds regardless of whether it is assumed correct or not. By that, I mean that excepting contradictions, not only are no counterexamples known, but it is known that none will ever be found.

So I decline to adopt your convention of classifying it as an assumption. It might be more accurate to call it an observation.

"We’ve established that the truth or falsity of a statement depends on its context — that is, on the assumptions we take as true or false in order to justify it."

I am inexplicably and disproportionately irritated by people like this. They've an a priori commitment to epistemic relativism. Unnaturally allergic to any claim that certainty is possible, and dedicated to undoing the careful work of those who are building up what may be known.

No, not every statement requires assumptions in order to hold. "The assumptions A implies B and B implies C, taken together, yield A implies C." This statement contains assumptions and makes observations about them, but it is true regardless of whether the assumptions it describes are true. The statement as a whole is "true" in the exact sense that the no counterexample to it can ever be given in any universe, under any set of assumptions.

I cannot wait for the day that Lean and other proof systems become accessible enough, conversant enough, and interdisciplinary enough to put these eternal confusion peddlers out of business.