Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
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-layered approach to the development of safer AI systems.
Cite
Text
Casper et al. "Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback." Transactions on Machine Learning Research, 2023.Markdown
[Casper et al. "Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback." Transactions on Machine Learning Research, 2023.](https://mlanthology.org/tmlr/2023/casper2023tmlr-open/)BibTeX
@article{casper2023tmlr-open,
title = {{Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback}},
author = {Casper, Stephen and Davies, Xander and Shi, Claudia and Gilbert, Thomas Krendl and Scheurer, Jérémy and Rando, Javier and Freedman, Rachel and Korbak, Tomek and Lindner, David and Freire, Pedro and Wang, Tony Tong and Marks, Samuel and Segerie, Charbel-Raphael and Carroll, Micah and Peng, Andi and Christoffersen, Phillip J.K. and Damani, Mehul and Slocum, Stewart and Anwar, Usman and Siththaranjan, Anand and Nadeau, Max and Michaud, Eric J and Pfau, Jacob and Krasheninnikov, Dmitrii and Chen, Xin and Langosco, Lauro and Hase, Peter and Biyik, Erdem and Dragan, Anca and Krueger, David and Sadigh, Dorsa and Hadfield-Menell, Dylan},
journal = {Transactions on Machine Learning Research},
year = {2023},
url = {https://mlanthology.org/tmlr/2023/casper2023tmlr-open/}
}