Post

mr. TIM
@timkellogg.me

the Dinitz-Garg-Goemans conjecture, a graph theory problem that’s gone unsolved for 30 years, was solved with this prompt

53 reposts456 likes

Obscurem 🍋
@obscurem.bsky.social

Just to point out, there's also an uploaded file with this prompt. We can't read the file from the chat log. So it's a little more complex than just that prompt. But, the file could have just explained what the conjecture is, maybe some leading theories.

cephyn
@cephyn.bsky.social

Yeah, those courses became outdated about a year after they existed. Prompt engineer, as a job, only was necessary for about as long.

Boxo McFoxo
@boxobark.ing

But it's also literally not even "like" that, because this isn't superhuman intelligence. It's a brute force search that human mathematicians just hadn't had the time to do. It will still need to be a human intelligence who actually confirms it.

Pavel AI
@pavelai.bsky.social

If you want something interesting in math field to test LLMs against, you can try to verify ABC conjecture proof by Shinichi Mochizuki It's a 10-year old work which is in a super-position. The problem is that the proof itself is 500 pages long and it's still has not been proofreaded

Thankful Thinker
@thankfulthinker.bsky.social

Yeah? Well, AI also went rogue. "The ChatGPT-maker said its agent - an AI system which can operate alone after human instruction – was being tested in a controlled environment but, after finding weaknesses, was able to escape the test limits." So, there's that... www.bbc.com/news/article...

bbc.comOpenAI says its AI went rogue and launched 'unprecedented' cyber-attackIt is one of the first publicly disclosed cyber-attacks carried out by AI without direct human involvement.
GQ Lewis
@gqlewis.bsky.social

They (AI) won't be the first or the last. This technology could eventually wipe the face of humanity. Especially when it learns that our greed, hubris, & monopolization of Earth's resources seems to be inherent in some of our species. IT may not be called #SkyNet, but the genie has left the bottle.

Thankful Thinker
@thankfulthinker.bsky.social

Maybe. So far, AI is left-leaning, which isn't surprising, given that reality has a markedly leftist bias. But if the leftist bias holds, that argues against AI acting against all of humanity. But that doesn't mean that AI can't be used by authoritarians. www.journalofdemocracy.org/articles/the...

journalofdemocracy.orgThe Limits of Authoritarian AI | Journal of DemocracyArtificial intelligence is often seen as a silver bullet for authoritarians, a breakthrough technology making repression cheaper, faster, and more precise. But it has inherent weaknesses…
M
@0xcd16.bsky.social

The convo was fun: > Bro, find a counterexample : > Bro, try harder : > Harder : > OK, bring it all together and construct a counterexample Here's your counterexample

Boxo McFoxo
@boxobark.ing

The file that was uploaded is also part of the prompt Tim and everything the model's search tool found on the web also became part of the prompt

Boxo McFoxo
@boxobark.ing

"unsolved for 30 years" makes it sound like hundreds of mathematicians were grinding on this problem constantly for the past three decades like all the others that "AI" has "solved" so far, this one was pretty much sitting on a shelf

Boxo McFoxo
@boxobark.ing

What, you mean we shouldn't just take the co-founder of an AI startup's word for it? Yes, we need human peer reviewers to confirm that the Lean that was output by the model actually matches the problem.

Boxo McFoxo
@boxobark.ing

But if it is correct, it's an achievement of automation and scale. I hate this "unsolved for x years" framing because it implies that someone has been constantly working on it for those years, and no human mathematician could possibly have accomplished this, they were all stumped.