I think a bigger problem is the almost complete lack of fine tuning success stories – it's still really hard to find convincing stories of teams that used fine tuning for anything more interesting than "we got a cheap model to classify things slightly better"

→ View original post on X — @simonw