Y
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.