Denying Levent the Fields Medal because his proof of RH was AI-assisted, it's like refusing physicists a Nobel because they use telescopes and particle accelerators.
It would signal that mathematics is choosing to sit out the future of scientific discovery.
@p0@thinkymachines okay this might indeed be the thing that sells me not just on this model but on whatever the fuck the cursed pipeline that brought it to life was. This is *fucked*. How the hell did they decouple knowledge from intelligence so cleanly???
@teortaxesTex i admit this is 100% based on priors (fancy word for "vibes"), there may be literature (fancy word for "non-reproducible papers on arxiv made purely for hype or grant money") on the topic that I don't know of
@teortaxesTex it's not like you can skip layers and go straight to the lm_head on a normal transformer either
this also doesn't seem like something that can be fixed in post-training, i think variable looping is a very different mechanism from fixed looping and needs to be pretrained in
@nicbarkeragain okay i immediately realize what the problem here is but I'm more surprised that it even compiles. though i guess giving you a warning when you aim the shotgun to your foot but not stopping you is perfectly in the cpp ethos
@PJ11819211 ice is reflective, water isn't, you heat up the air temporarily but if the ice holds it slows down the rate at which sunlight heats it up, so yeah it's the albedo they want. like this is actually pretty smart in spite of how stupid it sounds
People who have been wrong constantly about every AI prediction going “haha see I was right! total vindication lol” every time there’s a 10% market downturn is one of the most annoying discourse cycles on here.
@teortaxesTex dawg I'm not even sure deepseek nano will be a thing
i know they're the kings of opd and whatever but i just don't think they'll do it because they have no reason to
qwen does it bc they need community goodwill
deepseek has that in droves. they're beloved by everyone