Reinforcement Learning (RL) from Human Feedback (RLHF)

Jump to: navigation, search

YouTube ... Quora ...Google search ...Google News ...Bing News

Deep reinforcement learning from human preferences | P. Christiano, J. Leike, T. B. Brown, M. Martic, S. Legg, and D. Amodei

Reinforcement Learning from Human Feedback (RLHF) - a simplified explanation | Joao Lages

Illustrating Reinforcement Learning from Human Feedback (RLHF) | N. Lambert, L. Castricato, L. von Werra, and A. Havrilla - [[Hugging Face]

Reinforcement Learning from Human Feedback: From Zero to ChatGPT
In this talk, we will cover the basics of Reinforcement Learning from Human Feedback (RLHF) and how this technology is being used to enable state-of-the-art ML tools like ChatGPT. Most of the talk will be an overview of the interconnected ML models and cover the basics of Natural Language Processing and Reinforcement Learning (RL) that one needs to understand how Reinforcement Learning (RL) from Human Feedback (RLHF) is used on large language models. It will conclude with open question in RLHF.

Nathan Lambert is a Research Scientist at HuggingFace. He received his PhD from the University of California, Berkeley working at the intersection of machine learning and robotics. He was advised by Professor Kristofer Pister in the Berkeley Autonomous Microsystems Lab and Roberto Calandra at Meta AI Research. He was lucky to intern at Facebook AI and DeepMind during his Ph.D. Nathan was was awarded the UC Berkeley EECS Demetri Angelakos Memorial Achievement Award for Altruism for his efforts to better community norms.

How ChatGPT works - From Transformers to Reinforcement Learning with Human Feedback (RLHF)
ChatGPT has recently been released by OpenAI, and it is fundamentally a next token/word prediction model. Given the prompt, predict the next token/word(s). When trained on a massive internet corpus, it manages to be very powerful and can do many tasks like summarization, code completion, question and answer zero-shot.

Amidst the hype of ChatGPT, it can be easy to assume that the model can reason and think for itself. Here, we try to demystify how the model works, first starting with a basic introduction of Transformers, and then how we can improve the model's output using Reinforcement Learning with Human Feedback (RLHF).

Slides and code here

Transformer Introduction here


  • 0:00 Introduction
  • 3:09 Embedding Space
  • 15:35 Overall Transformer Architecture
  • 36:06 Transformer (Details)
  • 49:28 GPT Architecture
  • 56:38 GPT Training and Loss Function
  • 1:05:25 Live Demo of GPT Next Token Generation and Attention Visualisation
  • 1:16:55 Conversational AI
  • 1:19:00 Reinforcement Learning from Human Feedback (RLHF)
  • 1:45:15 Discussion

AI and ML enthusiast. Likes to think about the essences behind breakthroughs of AI and explain it in a simple and relatable way. Also, I am an avid game creator.