← Back to context

Comment by mountainriver

2 months ago

It’s a problem specific to autoregressive LLMs, the early tokens bias the output

0 comments

mountainriver

Reply

No comments yet

Contribute on Hacker News ↗