Olga Kolesnikova

Sentiment drift in RLHF can strip emotional nuance from summaries, but a new regularization technique can mitigate this effect without sacrificing quality.

Classifying political sentiment on social media is more complex than expected, with leading models achieving modest F1-scores that underscore the challenge.

Sentiment drift in RLHF-based summarization can suppress emotional expressiveness, revealing a critical trade-off in alignment methods that researchers must address.