Hacker News new | ask | show | jobs
by userbinator 883 days ago
It naturally assumes the average of its training data.
1 comments

I'm not sure this is necessarily the case, especially considering techniques like RLHF that can improve a model's performance.