Joan Velja
LightNews LightNews
banner
joanvelja.bsky.social
Joan Velja
@joanvelja.bsky.social
13 followers 150 following 6 posts
Container of multitudes | MS AI @UvA Amsterdam | Prev @LondonSafeAI

joanvelja.vercel.app/about
Posts Replies Media Videos
joanvelja.bsky.social
This one is also interesting:

arxiv.org/abs/2409.12822
Language Models Learn to Mislead Humans via RLHF
Language models (LMs) can produce errors that are hard to detect for humans, especially when the task is complex. RLHF, the most popular post-training method, may exacerbate this problem: to achieve h...
arxiv.org
January 13, 2025 at 6:38 PM
joanvelja.bsky.social
I’d be happy to do so :)
December 11, 2024 at 1:41 PM
joanvelja.bsky.social
Do you reckon it is more due to backprop or more downstream (e.g. autoregression)?
November 22, 2024 at 12:40 PM