))
Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback
Manage episode 429711880 series 3498845
Content provided by BlueDot Impact. All podcast content including episodes, graphics, and podcast descriptions are uploaded and provided directly by BlueDot Impact or their podcast platform partner. If you believe someone is using your copyrighted work without your permission, you can follow the process outlined here https://player.fm/legal.
This paper explains Anthropic’s constitutional AI approach, which is largely an extension on RLHF but with AIs replacing human demonstrators and human evaluators.
Everything in this paper is relevant to this week's learning objectives, and we recommend you read it in its entirety. It summarises limitations with conventional RLHF, explains the constitutional AI approach, shows how it performs, and where future research might be directed.
If you are in a rush, focus on sections 1.2, 3.1, 3.4, 4.1, 6.1, 6.2.
A podcast by BlueDot Impact.
Learn more on the AI Safety Fundamentals website.
Chapters
1. Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback (00:00:00)
2. Abstract (00:00:30)
3. 3 Open Problems and Limitations of RLHF (00:01:23)
4. 3.1 Challenges with Obtaining Human Feedback (00:03:17)
5. 3.1.1 Misaligned Humans: Evaluators may Pursue the Wrong Goals (00:03:38)
6. 3.1.2 Good Oversight is Difficult (00:06:51)
7. 3.1.3 Data Quality (00:11:08)
8. 3.1.4 Limitations of Feedback Types (00:12:59)
9. 3.2 Challenges with the Reward Model (00:17:03)
10. 3.2.1 Problem Misspecification (00:17:27)
11. 3.2.2 Reward Misgeneralization and Hacking (00:20:24)
12. 3.2.3 Evaluating Reward Models (00:22:30)
13. 3.3 Challenges with the Policy (00:23:49)
14. 3.3.1 Robust Reinforcement Learning is Difficul (00:24:13)
15. 3.3.2 Policy Misgeneralization (00:26:23)
16. 3.3.3 Distributional Challenges (00:27:35)
17. 3.4 Challenges with Jointly Training the Reward Model and Policy (00:29:54)
83 episodes
Share
