Safety researcher
Jacob Hilton
Studied reward-model overoptimization and sycophancy, characterizing failure modes of RLHF-trained models.
Works in the archive (2)
- WebGPT: Browser-assisted Question-answering with Human Feedback · 2021
- Reward Model Overoptimization / Sycophancy · 2022
“Lead” marks the first-listed author — a byline fact, not a measure of influence; co-authorship is shown as recorded.
Co-authors in the archive
View co-author network (7 people)
Loading graph…