@MiTypeScript Which paper? Either the quality of the context from the ocr is lower, or the context is the same quality and therefore images are mispriced, no?
@MiTypeScript@EthosVentures yeah the distillation angle is the part people undersell. cheap inference that's actually good changes who can ship, not just what gets built
@MiTypeScript@EthosVentures Their paper was published like a year ago, right? I recall seeing it a long time ago but it ended up not being implemented in the end at that time
@MiTypeScript@EthosVentures@grok what deepseek paper, what more than a pricing hack is he prbly referering to, any other fantastics in this paper?, what other papers are worth people's time?