eru

7 ms·

151.
by eru 1mo ago
Offshoring worked really well. So I don't see the problem in repeating something like that success story.
152.
by eru 1mo ago
Exponential growth in hardware's capabilities, like we are used to, means that several orders of magnitude are a matter of a relatively short time.
153.
by eru 1mo ago
So far we haven't seen much of that consolidation. And a lot of people are using trailing edge models just fine already. > [...] but the level of quality that is considered essential for work will always stay pinned at "near-fr
154.
by eru 1mo ago
> What I feel a bit annoyed by, and what I feel obviously LLM-run ablations like this fail to capture, is any kind of reflection around previous research or any kind of proof that this is the best you can do. At least the former is easy:
155.
by eru 1mo ago
It was just the first example that stuck out to me. I didn't do an exhaustive check to look for the worst offender.
156.
by eru 1mo ago
> In my experience, AI tends to avoid the first person, instead writing from a passive observer or even using sentences without verbs. Not my experience with Claude at all. Which AIs are you talking about?
157.
by eru 1mo ago
> It didn't work, and why it didn't work turned out to be more interesting than the policy would have been. Spoken like a true Claude. Snarking aside, I am glad that our AI agents make it cheap enough to do these experiments an
158.
by eru 1mo ago
As I said, these people exist. And math is fun! But it's far from the only reason people in general do and have done math.
159.
by eru 1mo ago
Huh? The resolution of Navier Stokes won't have much of an effect, yes. There's lots of problems like that. Eg if we prove P != NP, that won't have much of an immediate effect either. However, there's also plenty of pr
160.
by eru 1mo ago
Huh? Where do I say that AI needs to solve every intellectual problem? For all we know, P vs NP will never be resolved neither by human nor AI, and the world will just keep turning.
161.
by eru 1mo ago
Yes. And for most applications it's more important that the open models get better in absolute terms and perhaps that they are competitive on a per-Watt basis.
162.
by eru 1mo ago
You can enjoy math, but that was never the only reason people in general did math. (It was the only reason for some people, sure.)
163.
by eru 1mo ago
Applications don't care about Navier Stokes, yes. But they care about eg proving crytographics secure, or proving that your algorithm doesn't blow up under adversarial input.
164.
by eru 1mo ago
> When cars made horses obsolete, it didn't go so well for horses. I'd guess horses today have on average a better life than back then.
165.
by eru 1mo ago
Hasn't happened so far.
166.
by eru 1mo ago
I grew up in Germany, but I got frames in UK and Singapore and Australia, too. Fielmann in Germany has been offering very reasonably priced glasses and frames for ages.
167.
by eru 1mo ago
Well, Lean needs to get its act together to fix the bugs.
168.
by eru 1mo ago
Now, I'm saying that if you found a bug that lets you prove nonsense stuff (from true premises), you can probably prove whatever you want very quickly.
169.
by eru 1mo ago
Someone made a 'proof' of the Collatz conjecture that way in Lean. However if the prove relies on a bug like that, you'll be able to 'simplify' the proof a lot and you'll be able to proof contradictions.
170.
by eru 1mo ago
You don't need to read the lean proof for that, only the statement.
171.
by eru 1mo ago
I'm not sure it actually makes a difference. OpenAI doesn't care about the million dollars in any case. And the judgement that they did it is independent of whether the Clay people agree: you can make up your own mind and so can
172.
by eru 1mo ago
Then instead of publishing a paper, just upload your AI result to github. Same advancement of the state of the art. Papers are overrated here.
173.
by eru 1mo ago
Even aggregate employment of translators has held up well in the US. Even though machines do a much better job of what used to be the most basic job of a translator.
174.
by eru 1mo ago
Applications don't care whether the math was proven and understood by humans or computers. Your algorithm will get faster no matter where the insight came from.
175.
by eru 1mo ago
Why do you need AI to ask the questions? You give your AI the problem, like 'make this computer program faster and more reliable', and it'll go off and do the math necessary to make it so.
176.
by eru 1mo ago
You can have a look at eg improvements in solving linear programming problems or in mixed integer linear programming or SAT solvers.
177.
by eru 1mo ago
> Right now, we don't know if AI will be able to solve bigger problems, or if they'll be so efficients that you no longer need a whole datacenter for running them. For the latter: I assume that having a whole data centre will a
178.
by eru 1mo ago
Math also has applications, and they don't rely on human mathematicians doing the math.
179.
by eru 1mo ago
> Though I think even that's probably too socially utilitarian. I think ultimately the 'real' drive is the same in both fields - it's fun and personally rewarding. That's fine, and no amount of machine excellence
180.
by eru 1mo ago
Huh? I've been able to buy reasonably priced glasses for all my life. (However, I've never lived in the US nor China.)

More ›