Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ten advances in mathematics and theoretical computer science

openai.com

Post reply on HN
Print view View original

121–130 of 1001 posts

Re: Ten advances in mathematics and theoretical computer science

#121

Post by robinhouston »

defrost wrote: post #108
pistoriusp wrote: post #106

Earlier quoted context omitted.

Interesting. I had no idea that a person could see what is flagged?

If you page through the /newest listings you can see [flagged] and [flagged][dead] submissions. eg. this: [flagged] A migrant surge tests Spain's open policies (economist.com) - https://news.ycombinator.com/item?id=49131860 is clearly marked as flagged. Unlike the current submission: Ten advances in mathematics and theoretical computer science (openai.com) which isn't [flagged]. * https://news.ycombinator.com/newest

That's true, but submissions are only killed in that way if they receive a ‘fatal’ number of flags. However, flags lower the rank of a story even at non-fatal levels. What antirez is suggesting here is that the rank of this story has been lowered by flags – and that seems plausible, if you compare its rank to that of other stories with a similar age and number of points.

Re: Ten advances in mathematics and theoretical computer science

#122

Post by revetkn »

antirez wrote: post #100
robinhouston wrote: post #87

In a way the most remarkable thing about this is that it isn't even at the top of the HN homepage. Even if this is a step up from what we've seen before, we're no longer astonished by the idea that AI can make significant advances in mathematics and computer science.

This is not at the top as it is actively flagged by people that can't psychologically cope with the advances of AI. Hacker News is no longer a web site of an elite.

[flagged]

Re: Ten advances in mathematics and theoretical computer science

#123

Post by defrost »

robinhouston wrote: post #121
defrost wrote: post #108

Earlier quoted context omitted.

If you page through the /newest listings you can see [flagged] and [flagged][dead] submissions. eg. this: [flagged] A migrant surge tests Spain's open policies (economist.com) - https://news.ycombinator.com/item?id=49131860 is clearly marked as flagged. Unlike the current submission: Ten advances in mathematics and theoretical computer science (openai.com) which isn't [flagged]. * https://news.ycombinator.com/newest

That's true, but submissions are only killed in that way if they receive a ‘fatal’ number of flags. However, flags lower the rank of a story even at non-fatal levels. What antirez is suggesting here is that the rank of this story has been lowered by flags – and that seems plausible, if you compare its rank to that of other stories with a similar age and number of points.

[flagged] submissions aren't [dead] (killed), they are still active and can be upvoted and commented upon.

> if you compare its rank to that of other stories with a similar age and number of points.

Ranking is complicated enough here even before weighting, speed of initial upvotes can play against ranking, number of comments and the shape of the comment tree also affect ranking. And yes, various subjects and submission sources do get weightings that impact ranking.

What's funny, to myself at least, is that any attention at all is paid to "HN front page ranking" - I've been on again off again active here since 2008 .. and can't recall ever really looking at a default HN "front page" ever.

( There's /newest /newcomments /active etc to browse and sites such as https://hckrnews.com/ )

Re: Ten advances in mathematics and theoretical computer science

#125

Post by Davidzheng »

artninja1988 wrote: post #115

Now that we've seen AI produce a fair number of proofs (and disproofs), I'm curious when we'll start seeing it build genuinely novel theory. Does anyone have predictions on when and how we'll get there and will it take new architectures/ training paradigms, or is the current approach enough?

There's no clean line between a collection of theorems and a theory.

Re: Ten advances in mathematics and theoretical computer science

#126

Post by Davidzheng »

lifeisstillgood wrote: post #47

On the token limits etc - one assumes that OpenAI et al are able to “hire expert in field, and let them spend the equivalent of a million dollars of tokens” because they are not actually selling their complete compute 24 hrs a day, so the cost internally is a negligible (ish) electricity bill. Which is very suggestive - if after everything they are not fully loaded then the next gazillion data centres being built loo…

RL training can use all of them - idk what needed means.

Re: Ten advances in mathematics and theoretical computer science

#127

Post by esperent »

dist-epoch wrote: post #98
esperent wrote: post #70

Earlier quoted context omitted.

> deliberate misleading “lack of transparency” etc It's a marketing post from a huge company. Only the naive would view it uncritically without assuming it's been written carefully to present the results in the best possible light while skirting the boundaries of outright lying.

The results speak for themselves. Imagine 2 years from now: "yes, GPT solved the Riemann Hypothesis, but cmon, it's just a marketing stunt to hype their stuff, it was probably Terence Tao doing the work but he's so obsessed with hyping AI that he doesn't want to take credit"

Nobody is claiming the results are false.

We're saying look critically at the claims for how it was done, that it only cost $2000, etc. it would be extremely easy to run 100 sessions that failed, each costing ~$2000, and then just publishing an article about the one that succeeded, for example.

This goes double since it's an internal secret model (Astra) so nobody else can verify the results.

Re: Ten advances in mathematics and theoretical computer science

#128

Post by azan_ »

irthomasthomas wrote: post #102
traes wrote: post #27

Earlier quoted context omitted.

> Also, have there been examples of researchers not affiliated with OpenAI (or another LLM creator), who have done something similar? A couple small ones that I've seen (example here [0]), but not anything of the magnitude that OpenAI and Anthropic have put out. Likely just related to token limits. > Another question I have is whether or not OpenAI 'simply' hired capable combinatorics researchers to work on problems,…

Why you think that?

I guess that's because there are serious problems on which many professional mathematicians worked on years. If it was just a matter of hiring an expert, they would've been solved long time ago.

Re: Ten advances in mathematics and theoretical computer science

#129

Post by artninja1988 »

Davidzheng wrote: post #125
artninja1988 wrote: post #115

Now that we've seen AI produce a fair number of proofs (and disproofs), I'm curious when we'll start seeing it build genuinely novel theory. Does anyone have predictions on when and how we'll get there and will it take new architectures/ training paradigms, or is the current approach enough?

There's no clean line between a collection of theorems and a theory.

I mean doing something like Grothendieck when he redeemed algebraic geometry or Galois when he invented group theory. We haven't seen that at all from LLMs.

Re: Ten advances in mathematics and theoretical computer science

#130

Post by utopiah »

xyzsparetimexyz wrote: post #104

Any implication of any of these findings? They seem like unimportant nerd snipes to me. If you want to do something actually relevant, get chatgpt to write a simulation of graphene nanotube construction and figure out how to do it at scale.

Very marketable nerd snipes indeed.
Post reply on HN