Hacker News Reader

Have you trained models? If not then you can't really speak to how effective distillation is (answer: it is very effective)

View on HN

Why don't we see more American companies doing distillation? If you could make an opus 5.5 cheaply by distillation in china, you can do so even easier in America where you have more gpus.

Why don't we see the exact car clones you used to see in China in the US. I imagine labs were want to try and do it themselves -> or aren't willing to steal credit cards/mass create accounts/etc.... You haven't answered the question, either

lol. Have you? What kind of distill you are talking about? It’s only effective in the surface, small scale model or output preference. Not useful for long reasoning chain.

Why lol? Yes. All major labs use SFT... (look at job postings)

> Not useful for long reasoning chain.

I have a feeling you don't actually know anything about ML

“All major labs use SFT” <> it’s effectively for reasoning. And SFT <> distilling.

lol looks like you know about model trainings from water cooler chats mostly.

Reply on news.ycombinator.com