This is an artifact of them being diffusion models, generating by basically refining random noise in multiple passes until patterns emerge, usually leaving a residual watery or static sound. Same reason basic image models are bad at straight, fine lines and put random noisy detail everywhere
![philipswood 12 hours ago | prev | next [–]
Honest question: how is this different from some unknown mathematician having a breakthrough?
I mean: if some reclusive Japanese genius had a breakthrough on your problem and published it, would you have felt the same?
And if not, why not?
reply
jboggan 12 hours ago | parent | next [–]
If that had happened I would be overjoyed, maybe a hair chagrined that I didn't get it myself, but truly happy that someone got it and that I could go and talk to that person. Because it's the kind of problem I don't think would have fallen to a human after a few hours of thought, and I would have so much to talk about with that person. I would fly to Japan and hope to have tea with them, I would learn some Japanese to make the conversations easier. I would learn some interesting things hearing about their struggles and their false starts. I would make friends with that reclusive Japanese genius and my life would be far richer for it.
I will never meet that person and I will never hold a real conversation with the "creator" of that proof. They will never tell me how they came up with the cancelling exponential summation that cracked the constructio](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:jdum3zo5iqmx4gcnt5vuy7s4/bafkreih7e22vinqkztbymohgv4ldy4qojed452h2xfddb25vvpi2i2tlqi)
