kromem, 6 months ago

polishing each approach gets you slight improvements

Without the base model changing at all, research into better use of the models has, depending on the measurement, gone from around 35% success rates to 85% success rates.

I wouldn’t define that as slight.