Claude solves two open problems in the physics of black holes (anthropic.com)
easterncalculus 35 minutes ago
LeoPanthera 29 minutes ago
easterncalculus 22 minutes ago
For example, the metaphor they give about walking from Penn Station to the High Line comes off as somewhat out of touch.
willyg123 19 minutes ago
numbsafari 13 minutes ago
pamcake 27 minutes ago
scrollaway 26 minutes ago
Most people here care more about defining what AI is and whether what it can do really is "physics", than about the advance of humanity.
I dunno, I'm not one to say "HN was better before", but I did get bored a couple of years ago with all the "akshually, it's not real AI until it can solve [this problem], anything else is just sparkling inference algorithms".
(And of course the goalposts get moved so often, they need an amazon prime subscription just for the free delivery)
easterncalculus 24 minutes ago
I would make a comparison to illuminate the phenomenon I'm seeing in the rest of this thread, but unfortunately there is a specific rule artificially preventing me from doing so.
badgersnake 8 minutes ago
Toynbeeidea 20 minutes ago
_s_a_m_ 8 minutes ago
simonw 3 hours ago
byteknight 3 hours ago
bananaflag 3 hours ago
Those theoretical physics papers look like the real thing (i.e. not mathematical physics like the OpenAI Yang-Mills result). It looks like AI is finally branching out of math.
pizza234 3 hours ago
Is it really, though? It's still a heavily mathematical domain, rather than one involving empirical research, where AI still struggles.
ModernMech 2 hours ago
Sharlin 2 hours ago
bananaflag 2 hours ago
"It's not much, but it's honest work."
suddenlybananas 2 hours ago
boothby 2 hours ago
If you attempt to formalize math involving such assumptions, one expects to encounter a proliferation of contradictions. So I’m curious to see how the contributions of AI authors is handled by the physics community. With pure math, you can ostensibly write a compact statement of a theorem in Lean, and if so, trust the process.
jakzurr an hour ago
layer8 2 hours ago
BeetleB 2 hours ago
For people not in the know, it's common - both now and historically - for physicists to use "mathematical techniques" that have not been proven valid.
I took grad level physics courses, and this drove me bonkers. I was accustomed to the comfort mathematics provided me. If something was unclear, I could always dig deeper and find a proof. But a lot of the stuff the physics professors would do - I couldn't, and nor could they. They just responded with "intuition". Some of them had dug deep earlier in their careers and stated that the technique was likely invalid mathematically, "but still useful in getting answers experiments can verify".
I won't even talk about relying on diverging series...
ape4 an hour ago
keeda 2 hours ago
It is very good at understanding research goals, identifying appropriate datasets to work with, doing data wrangling and data science, determining appropriate metrics, running simulations, analyzing results, writing up reports with charts and tables, and proposing follow ups or “high potential next steps” which show a surprising amount of judgment, and if instructed to, executing on those too.
I use Astra and Fable (who delegate to cheaper models for coding) and have them review each others work, and only spot check things when they are done calling each other out. I rarely find issues, and when I do, it’s usually a misunderstanding of intentions.
I’ve only explored a few areas though, including computer vision, vector search, and wireless signal processing, all in applications that, as far as I can tell, are genuinely novel (though of varying levels of novelty and economic value, as the models are fond of pointing out to me!)
I prefer being in the loop and setting the direction — because I do not have unlimited tokens, hah! —- and I regularly identify new directions they had not considered or had prematurely dismissed… so I am a bit more useful than just a financier of tokens! But I can totally see how they could grinding ahead by themselves and brute force useful results, especially if given very concrete goals like a theorem to prove.
I don’t think we’re at Recursive Self Improvement yet, but I don’t doubt for a minute that this is behind the insane pace of development at Anthropic and OpenAI.
glitchc an hour ago
keeda 42 minutes ago
My perception is that most data collection, once designed (which does take considerable thought, as you point out), is a slog of executing rather mechanical processes over a long time. (Which is what RA’s and lab assistants are for!) For sure, regular analysis is needed to ensure the right data is being collected and no confounding factors are sneaking in, but seems to me the majority of time goes into the relatively rote work of setting up equipment, running experiments and recording the results.
I do think that with access to MCP-enabled equipment, AI would be ideal for this too, given its infinite patience. The problem, as with all other fields, is this would rob novices the opportunity of building the necessary intuition that comes from spending time in the weeds.
tomrod an hour ago
Context is king.
vvilliamperez 2 hours ago
kragen 2 hours ago
ronnier an hour ago
kragen an hour ago
keeda 17 minutes ago
pyaamb 35 minutes ago
troupo 34 minutes ago
Because they want you to think that you will be able to do the same, too.
Except, you don't have a dedicated cluster for a week with an unrestricted model.
dap an hour ago
What does it mean for Claude to solve it? What was the prompt? I didn't read it all, but the word "prompt" does not appear in the blog post, either abstract, or either paper.
tomrod an hour ago
When my accountant whips out a calculator, it is the accountant that defined the problem and set the solution in motion. Fantastic tools are still tools. Claude does not love me nor you no matter how its trained and encoded patterns express.
KoolKat23 an hour ago
tomrod 19 minutes ago
garciasn 17 minutes ago
2OEH8eoCRo0 43 minutes ago
jcmontx an hour ago
chunky1994 3 hours ago
The first result is cool (but basically a complex error-prone calculation), the second is much much cooler, Claude proved the upper limit for the mass of a black-hole which is consistent with a given central charge. (thereby making it a consistent theory of quantum gravity), any theory predicting above this is not a theory of quantum gravity of that universe. This is very exciting!
dguest 2 hours ago
chunky1994 2 hours ago
bananaflag 2 hours ago
https://nautil.us/what-does-any-of-this-have-to-do-with-phys...
cvoss 2 hours ago
Careful. Thereby making a not-inconsistent-via-bootstrap theory. Such a theory could be inconsistent for other reasons shown by other methods. The article expresses a desire to show inconsistency for M1 above c/12 for example, but it's now been shown that the boostrap method can't get you that far.
chunky1994 an hour ago
Merrill 3 hours ago
"Neither result describes our own universe directly, and neither has practical applications, but both contribute to a long-running theoretical effort to understand how quantum mechanics and relativity fit together."
nisegami 3 hours ago
eru 3 hours ago
curt15 2 hours ago
chunky1994 an hour ago
Sharlin 2 hours ago
bongofizzbuzz 3 hours ago
tiborsaas 3 hours ago
raverbashing 3 hours ago
4ndrewl 2 hours ago
steinwinde 3 hours ago
Ampersander 2 hours ago
pona-a an hour ago
Even through after all my skepticism about AI in science, Anthropic proved they can search for and find non-trivial discoveries with their tech, they've done very little to dissuade my fear they are actively hurting the scientific community for PR.
We could treat AI as just a helpful fuzzer for open-problems that occasionally opens new interesting paths for the actual scientists, but we know the end-goal of these people is a future where there are no qualified people left to review, let alone make new discoveries independently from the machine.
dodger-dog an hour ago
shagie an hour ago
scrollaway 38 minutes ago
:)
tzfld an hour ago
Madmallard 22 minutes ago
gr-mk 3 hours ago
Rochanhlua 29 minutes ago
pelagicAustral 3 hours ago
eru 3 hours ago
jcmontx an hour ago
Answering myself: yes, one of the problems was solved using Lean. The other was undisclosed. Nothing new under the sun, huh.
mentalgear 2 hours ago
nutjob2 an hour ago
themgt 2 hours ago
Representatives of AI labs and other organizations may email [email protected] to arrange a verified lab account.
https://hexagonmath.org/organizations/anthropic
Also: The main result was first obtained autonomously by an internal Claude research model. The publicly available Claude Opus 5.5 later obtained the same result independently. Claude then wrote this paper, following brief guidance from a physicist, who also checked every main result.
Wild.
jon128 44 minutes ago
That talk did make enough sense.
I'm still waiting to see a "picture" of a black hole that didn't involve a "pre-trained model" developed with "synthetic data" or something like that.
I'm not feeling sufficiently persuaded that I can feel confident that gravitons are a thing.
Nonetheless, I am remaining open minded to this black hole stuff.
nozzlegear an hour ago
physicsgoat an hour ago
Stopped reading right here. It took one. Maybe you don't know which one it took but things can't be in two places at once.
motbus3 3 hours ago
CaptWorld 2 hours ago
0xBA5ED 2 hours ago
CaptWorld an hour ago
0xBA5ED 28 minutes ago
CaptWorld 3 minutes ago
coliveira an hour ago
CaptWorld an hour ago
Qem an hour ago
dodger-dog an hour ago
AlecSchueler an hour ago
They do but we've seen enough solutions by now that surely one of these researchers would come forward to point it out?
bbor 3 hours ago
It was a comforting moment, for sure. But ephemeral, self-imposed ignorance is no bliss at all :(
aanet an hour ago
Next up, the Origins of the Universe... stay tuned for our Fable e^2. More on the 10 o'clock news...
/s
2OEH8eoCRo0 an hour ago
insane_dreamer an hour ago
techscruggs an hour ago
brokencode 42 minutes ago
totalforge 2 hours ago
claaams 2 hours ago
nutjob2 an hour ago
ironborn123 an hour ago
geraneum an hour ago
Any minute now
nutjob2 an hour ago
OK, but it's still an unfounded claim and...
> a model or system [...] can solve hard problems across math, physics, biology
...is all that has been proven. That may one thing that intelligent humans can do, but it doesn't mean it requires human level intelligence. All computers are superhuman in some narrow sense.