Hacker News new | ask | show | jobs
by astrange 136 days ago
The GPT3 base model was pretty funny if you like nonsense. Instruct tuning and RLHF seem to destroy it when they recalibrate everything.