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

841–850 of 1000 posts

Re: Ten advances in mathematics and theoretical computer science

#841

Post by DiscourseFan »

andai wrote: post #807
MostlyStable wrote: post #644

Earlier quoted context omitted.

From a mathematician who was intimately familiar with some of these problems [0] >I don’t understand it yet. Maybe it’ll take me an afternoon to check all the calculations, but what would still be missing is why this was an approach that would’ve made sense in the first place. Is there some broader context or theory within which this would’ve been the obvious thing to do? What other results can be proven using these…

I think it might be like waveform collapse, but very high dimensional.

Can you expand on this?

Re: Ten advances in mathematics and theoretical computer science

#842

Post by vasco »

mitthrowaway2 wrote: post #757
annzabelle wrote: post #692

Earlier quoted context omitted.

The author of that post is a prominent Bay Area "rationalist," who have had a quasi-theistic relationship with the concept of all-powerful AIs for a couple decades now.

So are they right or wrong about the sigmoids?

All of the point of that article is that most people that think something is a sigmoid think it'll bend just as they are publishing their analysis. And the article says, don't do that, assume it'll be related to how long we've been on the "goes up" part.

Nothing in that article says it's not a sigmoid.

Re: Ten advances in mathematics and theoretical computer science

#843

Post by sillysaurusx »

avaer wrote: post #37

What happens when OpenAI et al stop being open about these things, and just pack it into the training?

Hey man, just wanted to say hi and catch up a bit. I tried DMing you on Twitter. If that sounds interesting then shoot me a message sometime. Hope you’ve been well :)

Re: Ten advances in mathematics and theoretical computer science

#844

Post by armchairhacker »

tyre wrote: post #749
lettergram wrote: post #553

Earlier quoted context omitted.

> My guess is that, in the US, the right will cynically adopt manipulation to great effect and the left will take a moral stand against shady practices and lose elections. I think that statement may itself highlight how prevalent manipulation is. I fully anticipate all groups to continue maximal manipulation they can. One thing with LLMs is that it'll be a far less unified view, so a "divide and conquer" strategy is…

In US politics, the right is far, far better at winning elections than the left. This isn’t about personal preference. It’s objective political science. Look at the most contentious issues in the US: abortion, climate change, taxing the wealthy, gun control, Affordable Healthcare Act. The Democratic Party platform is aligned with national polling on every one. Every one of those issues has >60% support with voters an…

> Look at the most contentious issues in the US: abortion, climate change, taxing the wealthy, gun control, Affordable Healthcare Act.

You left out immigration, crime, and “moral values”: https://www.pewresearch.org/politics/2024/05/23/top-problems...

Re: Ten advances in mathematics and theoretical computer science

#845

Post by hnlmorg »

tyre wrote: post #744
hnlmorg wrote: post #720

Earlier quoted context omitted.

It’s unfortunate that you’re being downvoted because this comment is true. And it just goes to show how far we’ve sunk that people have forgotten the problems of ad tech in the age of LLMs.

You’re making a weird number of assumptions about people’s voting and extrapolating to assertions about Society. Ads are not really the same. They can’t be as tightly targeted to what resonates with someone. Programmatic ads are certainly much better and closer than, say, television advertising, but users can’t _engage_ with them. Like actually chat with them. That’s where this is headed and people are not ready. I d…

I’m not making any assumptions.

> Ads are not really the same. They can’t be as tightly targeted to what resonates with someone.

The EU referendum in the UK proved your point false.

> Programmatic ads are certainly much better and closer than, say, television advertising, but users can’t _engage_ with them. Like actually chat with them.

When people talk about “ad tech”, they’re not talking about TV ;)

And yes, people can and do engage with them. That’s how ads on social media works.

During the EU referendum, people were even resharing ads on Facebook without even realising they were ads.

Re: Ten advances in mathematics and theoretical computer science

#846

Post by vasco »

andsoitis wrote: post #826
akoboldfrying wrote: post #766

Earlier quoted context omitted.

> A 0.001% optimization on a packing problem just isn't interesting for the amount of investment. I think you have completely misunderstood what OpenAI have accomplished here. Almost certainly no one cares about the specific concrete results achieved; they only care about (a) how difficult it would be for an intelligent human to achieve the same feat (ETA: the feat is the proof), which can be estimated by the amount…

> It's as if I showed you a dog that I had taught to speak German fluently And you’re thinking this is an accurate comparison?

It's more funny than the dog doing differential equations I'll tell you that. Specially when it gets mad.

Re: Ten advances in mathematics and theoretical computer science

#847

Post by energy123 »

casey2 wrote: post #709
sothatsit wrote: post #535

People argue whether we are at y-5, y, or y+5, meanwhile we seem to be on a y=2^x exponential that keeps delivering more and more impressive results. The most interesting question to me is what will be consumed by the exponential like math seems to be undergoing, and what won’t. Writing has been quite stubborn, but I’ve noticed Fable to be quite a big step up there. How about politics? Will we develop new ways to let…

We definitely are not on an exponential. Don't say we are because this isn't up for debate. AI progress is logarithmic the million dollar question is 2x or 10x for linear improvement. The nearest qualitative shift would be very fast inference so people could start writing real software on top of LLMs. A 0.001% optimization on a packing problem just isn't interesting for the amount of investment.

What is the y-axis in this claim about logarithmic improvements? Any exponential curve can be trivially turned into a logarithmic and vice-versa, and the y-axis redefined as "progress", with no loss of accuracy.

One example that always bugs me is when people point to "exponential" or "sigmoidal" progress on benchmarks. Benchmarks are artificial constructions (saturation at 100% by definition) and benchmark scores should not be mapped to these words when talking about overall progress.

Example - progress on ARC-AGI-3 at the moment is exponential, steeper than 2^t and e^t. Does that mean AI is progressing "exponentially" in the colloquial sense? No, it doesn't support or refute that colloquialism.

Likewise with MMLU saturation. We can't go above 100% by construction. Therefore we have a "sigmoid". Gah.

The colloquialism is not helpful to begin with.

Re: Ten advances in mathematics and theoretical computer science

#848

Post by tossandthrow »

AgentMatt wrote: post #723
thisisnotauser wrote: post #722

Earlier quoted context omitted.

My wife is working on her PhD in microbiology now, using OpenAI to implement her research ideas. Genetics is just too much data, and she eats through tokens like nobody's business. I thought I was careless with them, but she barely lasts a full day before exhausting her quota. There's definitely a lot of value there, but dealing with the data problem is a big obstacle in biology. I can only imagine what she could get…

What makes it so token hungry? Is she directly using the LLM for genome analysis rather then having it write the data analysis algos?

We so don't know what plan she is using.

Eg.a 20usd/m plan usually don't cut it for professional work.

Re: Ten advances in mathematics and theoretical computer science

#849

Post by aswegs8 »

wombatpm wrote: post #737
tyre wrote: post #548

Earlier quoted context omitted.

We will get much better at manipulation and better at people “writing” things to justify their own feelings. What’s new about LLMs is that you can scalably manipulate people individually. It used to be that you could either have scale (speeches, tweets, interviews, website, etc.) or individual engagement (replying to mail/tweets/town hall questions.) Now you can pull the history and preferences of an individual, then…

Biology would greatly benefit. We barely understand transcription and protein structure. And the straightforward systems that we know like insulin have complex post translational modifications. So while we have a map of the partial proteonome, we have barely scratched the surface on networks regulation and interactions.

I think the question is less about where it's most beneficial but which knowledge structure lends itself to LLMs most. Since biology's "language" is way more complex and irregular than maths or natural language, it isn't particularly accessible.

Re: Ten advances in mathematics and theoretical computer science

#850

Post by aswegs8 »

tyre wrote: post #745
attila-lendvai wrote: post #733

Earlier quoted context omitted.

the fact that the useless left/right divide is still so widely used shows that manipulation is working well even pre LLMs... when it comes to the important question, then both "sides" are the same team. or if you want it with a pinch of humor: when a boot is on your face, it makes precious little difference whether it's the left or the right boot. (i lived the first 10 years of my life in communism)

If you think that the American Left is anything like communism, I’m sorry, but, respectfully, I can’t take your comments seriously. The left in the US would be center-right in Europe, who are certainly not communist (they have separate parties that are communists!) Even the socialist strain of the US has nothing to do with socialism scaremongering about Venezuela, etc.

Not at all. Economically, yes. Socially they are far left of the European left.
Post reply on HN