← Back to context

Comment by bonzini

1 year ago

Apparently the biases in the output tend to be stronger than what is in the training set. Or so I read.

0 comments

bonzini

Reply

No comments yet

Contribute on Hacker News ↗