[theshrike79]

Anecdotal evidence says that the 1M context one still gets stupid around 200-300k tokens.

Context still matters and I'll never stop implementing things in small slices instead of trying to one-shot.