Kinda wish we got a round 2 of this but on a newer, smarter base.
#3
by BigBeavis - opened
Revisiting this gem. It's far from perfect, but it represents a very unique use case that's a hassle to replicate with other models without bloating context with extensions. The limited brain power of mistral is holding it back imo, but the premise is too good to leave it confined to a one-off experiment.