As A.I. Makes Strides in Mathematics, Mathematicians Urge Caution

[highplainsdem]

Source: NYT

-snip-

On Tuesday, a group of 16 mathematicians, in consultation with colleagues and math organizations worldwide, published the Leiden Declaration on Artificial Intelligence and Mathematics. It aims to “frame the conversation about future directions,” said Dame Ursula Martin, one of the authors, and a mathematician and computer scientist at Oxford.

This effort comes as A.I. models have been making headlines with successful results in research-level mathematics. In late May, OpenAI, the maker of ChatGPT, announced that one of its models had disproved a notable 80-year-old mathematics conjecture in the field of combinatorial geometry.

-snip-

Among the potential threats that the Leiden Declaration authors articulate are accuracy and reliability: Journal editors are already complaining about a flood of plausible seeming A.I.- generated papers and proofs that have turned out to be incorrect, and in ways that are difficult for mathematicians to discern.

-snip-

HARRIS I would add that a lot of the enthusiasm and excitement is artificially generated by the corporations. The declaration warns against that: “Don’t believe the hype.” It is important for the mathematical community to have the last word on what, mathematically, is, and what is not, exciting.

-snip-

Read more: https://www.nytimes.com/2026/06/02/science/ai-mathematics-leiden-declaration.html

---

This declaration was a good idea. The AI companies have been overstating and hyping their AI models' alleged abilities. There have been a number of hypefests that got AI companies a world of free publicity before the reality of what the AI had - or hadn't - accomplished became known, whether it was the AI simply finding answers online or the hallucination level making the supposed advance much less useful. And there was a study showing that LLMs can come up with correct answers possibly by simple guessing or hallucinating, since their explanation of how they reached that answer didn't actually lead to it.

A very pro-AI account on both Bluesky and X posted about a "disturbing" Stanford paper on LLMs' failures at reasoning
https://www.democraticunderground.com/100221009224

One of the most disturbing findings is how often models produce unfaithful reasoning. They give the correct final answer while providing explanations that are logically wrong, incomplete, or fabricated.

This is worse than being wrong, because it trains users to trust explanations that don’t correspond to the actual decision process.

That sort of thing is why it caught my attention when a Harvard mathematician mentioned in the NYT article pointed out a lack of related ideas.

Another of the mathematicians pointed out that OpenAI "got lucky with this one" and so far people being asked to praise this latest hyped AI accomplishment "are not told about the model’s failures."

Would Sam Altman lie to get more attention for his company?

Sam Altman May Control Our Future--Can He Be Trusted? (Ronan Farrow and Andrew Marantz, The New Yorker, 4/6)
https://www.democraticunderground.com/100221158209

[2naSalit]

1. ...

[Tetrachloride]

2. Mathematics and Physics is what AI is fundamentally

look in a mirror? Scary

[EYESORE 9001]

4. But it's built on *sound* principles

Not AI-generated hullabaloo-lations with no footing in reality.

[ancianita]

3. Yes, Sam Altman would pitch to get attention, but mostly to get money, even if what he said seemed true at the time.

They might be lies, but IIRC, the article says he believes them when he says them.

[ret5hd]

5. MAYBE altman/thiel/musk et al...

are themselves AI and are attempting to create a “Bride Of Frankenstein”.

c’mon guys…they’re just lonely.

[Happy Hoosier]

6. LLMs are bad at math... don't trust them.

Computers, of course, are fancy adding machines. But interestingly, LLM's cannot yet properly route mathematical questions to the resources it has access to in order to give you a correct answer. It uses the same models it uses to hallucinate about other subjects to predict what the answer to math question is. It's USUALLY right, but USUALLY right isn't good enough if people are trusting it.