I knew before clicking it will be one of the things they say. people are sheeps. they will say whatever they think is right up to a limit. then they will break this limit and punish everyone that is deviating from it. this is what happened in nazi germany, soviet union, china and in the US under biden and I can already see the premises of the next term of US presidency through the lindsay clancy reaction from millions of women. to me it feels like we broke this limit in the modern developed world now. sad that as humans we cant use our brains for better than following what our hiveminds say is truth
Its a s3 wrapper. Can be replaced by a wordpress with torrents too. But they built an ecosystem around it and some libs/content. i guess thats what nvidia value
Unfortunately not all deals go through. Like Wiz + Google, or Figma + Adobe. A lot can go wrong between “in talks to acquire” and “acquired.” The official announcement is less surprising but still newsworthy.
its always the most virtue signaling people (sam on "openess", dario on pretty much every word coming out of his mouth, HF) that ends up doing the worst things because they always tend to extreme. Funny to see them joining the biggest corporation on earth by market cap after the every day virtue signaling of pretty much every employee working for them.
very few people can lock a vision and fight billions dollars with integrity. In tech they mostly end up shadowed. People like JB Kempf with VLC or Bellard. Maybe even Torvald or Jeremy Howard belong to this list of people who could have been way richer than they are now if they wanted to go against their visions and beliefs.
I did that on SVG mostly to teach it to draw pelicans but also to generalize it. most of the behavior is from SFT on the base model tho. RL is very ineficient at style or at least at generating novelty out of distrib.
Nothing against what I said in there. Distillation is far from enough to get to the frontier. Its at best a ramp (the most efficient one used by everyone ) that shortcut and saves millions of rl runs before a model moves.
the initial post infers that distillation is all you need. it is not. in 2026 you need large scale distributed inference of gigantic models, rl envs and millions of dollars runs to get to something decent. if you think glm just has to sft on traces of claude to edge Mythos on some cyber benchmark you are fooled.
I said there is no evidence distillation is all you need as per the initial comment "The fact that AI models can be so easily distilled and replicated is such a stroke of luck." it is far from trivial to reproduce the capabilities of anthropic models for instance
reply