We are really not ready for paying the real price of tokens….
And I’m not only talking about API price. Just check how much it costs to rent GPUs to run shitty 100B models
@joshwhiton Both sides can be defendable. But when a smaller team with less budget, make a better, cheaper version of your product in less than 2w after release, this IS core innovation, distillation or not.
@Capetlevrai Même avec Max x5, Fable est presque inutilisable. Le x20 c’est pas mal. Codex est dur à battre niveau rapport qualité prix en attendant que moonshot s’équipe en compute