🧵 Untitled Thread
Anonymous at Fri, 19 Apr 2024 19:54:15 UTC No. 16136342
Why can't we use AI to find candidates for a superconductor?
Anonymous at Fri, 19 Apr 2024 19:55:23 UTC No. 16136343
>>16136342
Explain how you'd set up your training data, please.
Anonymous at Fri, 19 Apr 2024 19:56:55 UTC No. 16136345
>>16136343
we could use a simulation and just brute force shit.
Anonymous at Fri, 19 Apr 2024 19:58:08 UTC No. 16136348
We can, there is probably already people working on it. But ChatGPT likely struggles with the intricacies of calculating transport properties of complex n-body quantum systems. Come back in a few decades and we might be a bit closer to having a working model.
Anonymous at Fri, 19 Apr 2024 20:00:59 UTC No. 16136353
>>16136345
That is hard, incredibly hard. Such calculations scale exponentially and we don't have the computing power to simulated even existing superconductors we know about. That is why only inaccurate approximations are currently used.
Anonymous at Fri, 19 Apr 2024 20:03:32 UTC No. 16136354
>>16136342
1. There is no AI
2. ML isn't AI
>>16136348
ChatGPT is doesn't know, it's nothing but a digital parrot
Anonymous at Fri, 19 Apr 2024 20:05:37 UTC No. 16136356
>>16136354
> 2. ML isn't AI
In modern terminology it is. I agree, it's marketing bullshit. What was once called AI is now termed AGI, and yeah, we don't have it.
> ChatGPT is doesn't know, it's nothing but a digital parrot
That's the joke anon.
Anonymous at Fri, 19 Apr 2024 20:11:53 UTC No. 16136360
>>16136342
Is what people do
Also use this parameters
Where does light go in the darkness?
Anonymous at Fri, 19 Apr 2024 20:42:25 UTC No. 16136391
>>16136342
Because we don't have a perfect subatomic-to-macro-properties physics model yet.
And if we did, it would be too computationally intensive for our current hardware.
Anonymous at Fri, 19 Apr 2024 20:48:26 UTC No. 16136398
>>16136342
Just because you can't do something doesn't mean no one can do that. AI is used regularly in things like this.
Anonymous at Fri, 19 Apr 2024 20:56:41 UTC No. 16136403
>>16136391
Well you could just numerically solve the Time Dependant Schrödinger Equation for a large enough lattice. Might take a million years or two, but what's a few cpu cycles here or there.
Anonymous at Fri, 19 Apr 2024 21:04:25 UTC No. 16136416
>>16136342
Because you have to actually fabricate and test it. Look at LK-99, the argument wasn't over what the candidate was, it was over whether it was created according to specs and tested correctly.
Anonymous at Fri, 19 Apr 2024 21:10:41 UTC No. 16136430
>>16136354
Maybe you're confusing intelligence with sentience?