this post was submitted on 14 Sep 2026
280 points (97.9% liked)

science

28975 readers
233 users here now

A community to post scientific articles, news, and civil discussion.

dart board;; science bs

rule #1: be kind

lemmy.world rules

founded 3 years ago
MODERATORS
all 29 comments
sorted by: hot top controversial new old
[–] nonentity@sh.itjust.works 89 points 6 days ago (4 children)

The LLM didn’t ‘solve’ anything, it regurgitated a guess.

Stop attributing comprehension and agency to something which is intrinsically incapable of possessing it.

[–] Hegar@fedia.io 44 points 6 days ago (1 children)

Producing a correct answer doesn't imply comprehension, using the word "solve" doesn't imply comprehension. Computers have been "solving" math problems for decades.

There is no attribution of comprehension implied in that phrase.

[–] BooBees@fedinsfw.app 4 points 6 days ago

Yeah there wasn’t a pre written or rewritten guess, it’s combining other ideas (processes) to get to a desired point. It’s novel, but it’s not conscious and derived from creativity on its own. And that’s fine. It’s a tool. It’s the most complex computational tool developed outside of quantum computing, and that’s pretty neat. It might also turn out to be really useful. The problem is the tech bros won’t use it long term for good, because good isn’t as profitable as using it for war, for mass surveillance, for oppression.

[–] TheTechnician27@lemmy.world 31 points 6 days ago* (last edited 6 days ago) (2 children)

Stop attributing comprehension and agency

That's not what this is doing. At all. It's saying it produced a solution.


Edit: Y'all, mathematicians literally use deterministic algorithms called SAT solvers that solve the Boolean satisfiability problem. Saying an algorithm "solved" a math problem is not ascribing it comprehension or agency.

[–] 8baanknexer@lemmy.world 4 points 6 days ago

Some more context on sat and first order solvers in math: the first open math problem solved autonomously by a computer was in 1997, well before the advent of generative ai.

[–] endless_nameless@lemmy.world 2 points 6 days ago* (last edited 6 days ago) (2 children)

It's the age of AI, words don't mean anything anymore. They only point to other words. The word "solve" may have a "definition" but that's irrelevant compared to how the word makes me feel.

[–] SippyCup@lemmy.world 9 points 6 days ago

That's literally always been true. Humans are emotional speakers and use language that feels right more than we use language that factually accurate.

The spicies have been trying to tell you this for decades.

[–] Not_mikey@lemmy.dbzer0.com 2 points 6 days ago

Using non anthropomorphic language for AI just makes discussing it more obtuse and complex for no reason.

[–] Zephorah@discuss.online 3 points 6 days ago

Search bots that have ingested speech patterns. Psycholinguistics is traceable down to codons, sure, but the meaning and nuance is lost. Lewis Carroll is probably LLM kryptonite, for example.

AI is a PR pick for a name. A name to lure in investors. It attributes so much more to the bots and invokes curiosity in everyone around it. Versus language bot or some other not so sexy name.

[–] Funkt4st1c@lemmy.world 2 points 6 days ago

Also answers are much much much less interesting than the steps to the solution and reasoning, which can be applied to future problems

[–] qevlarr@lemmy.world 41 points 6 days ago (2 children)
[–] cloudshouter@lemmy.zip 18 points 6 days ago

The goal, fully articulated or not, is to make it so easy to bypass the journey and the education that people loose the ability to do that part. Kill the need for educated and skilled people so everyone switches to AI and can’t switch back because the people that can do the work no longer exist. Always remember too that these tech guys look at the worst parts of sci-fi, fail to understand the moral of the story, and unironically think it’s the best goal.

[–] Mondez 1 points 6 days ago

LLM tools are like taking a helicopter that can only drop you off at the site because somone put down the trail markers and made the maps. It can't figure out how to go anywhere using a method a human didn't already do most of the important work finding out.

[–] daniskarma@lemmy.dbzer0.com 6 points 5 days ago* (last edited 5 days ago)

AI didn't solve it. Mathematicians working for OpenAI who signed NDA agreements took these other mathematicians idea and developed into a more complete prove to market the company share value.

These OpenAI mathematicians might or might not used a LLM to more or less extension to develop this prove. There is no evidence on how helpful was the LLM to the whole process.

[–] fubarx@lemmy.world 6 points 6 days ago
[–] courval@lemmy.world 4 points 5 days ago (1 children)

Sorry didn't read the article but they did a 20M marketing campaign involving maths a couple of weeks ago and now this 15M euros one.. Where are they getting the money for this?

[–] Goodlucksil@lemmy.dbzer0.com 1 points 5 days ago

Probably from the university (CUNEF) themselves, given that they're private.

[–] Zephorah@discuss.online 4 points 6 days ago

I’m still convinced this started out as a yacht race. And now it’s rolling downhill with a fuckton of force such that no one is stepping in the way.