Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Artificial Intelligence
2023-09-12 v2 Computation and Language
Machine Learning
Abstract
Reinforcement learning from human feedback (RLHF) is a technique for training AI systems to align with human goals. RLHF has emerged as the central method used to finetune state-of-the-art large language models (LLMs). Despite this popularity, there has been relatively little public work systematizing its flaws. In this paper, we (1) survey open problems and fundamental limitations of RLHF and related methods; (2) overview techniques to understand, improve, and complement RLHF in practice; and (3) propose auditing and disclosure standards to improve societal oversight of RLHF systems. Our work emphasizes the limitations of RLHF and highlights the importance of a multi-faceted approach to the development of safer AI systems.
Keywords
Cite
@article{arxiv.2307.15217,
title = {Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback},
author = {Stephen Casper and Xander Davies and Claudia Shi and Thomas Krendl Gilbert and Jérémy Scheurer and Javier Rando and Rachel Freedman and Tomasz Korbak and David Lindner and Pedro Freire and Tony Wang and Samuel Marks and Charbel-Raphaël Segerie and Micah Carroll and Andi Peng and Phillip Christoffersen and Mehul Damani and Stewart Slocum and Usman Anwar and Anand Siththaranjan and Max Nadeau and Eric J. Michaud and Jacob Pfau and Dmitrii Krasheninnikov and Xin Chen and Lauro Langosco and Peter Hase and Erdem Bıyık and Anca Dragan and David Krueger and Dorsa Sadigh and Dylan Hadfield-Menell},
journal= {arXiv preprint arXiv:2307.15217},
year = {2023}
}