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.