Y
Hacker News
new
|
ask
|
show
|
jobs
by
sumo43
810 days ago
Maybe true for instruct, but pretraining datasets do not usually contain GPT-4 outputs. So the base model does not rely on GPT-4 in any way.