Welcome to Reinforcement Learning from Human Feedback, or RLHF, built in partnership with Google Cloud. An LLM trained from public internet data would mirror the tone of the internet, so it can generate information that is harmful, false, or unhelpful. RLHF is an important tuning technique that has been critical to align an LLM's output with human preferences and values. This algorithm is, I think, a big deal and has been a central part to the rise of LLMs. And it turns out that ROHF can be useful to you, even if you're not training an LLM from scratch, but instead building an application whose values you want to set. While fine-tuning could be one way to do this, as you learn in this course, for many cases, RLHF can be more efficient. For example, there are many valid ways in which an LLM can respond to a prompt such as, what is the capital of France? It could reply with, Paris is the capital of France, or it could even more simply reply, Paris. Some of these responses were few more natural than others. And so, RROHF is a method for gathering human feedback on which responses they prefer in order to train the model to generate more responses that humans prefer. In this process, you start off with an LLM that's already been trained with instruction tuning, so it's already learned to follow instructions. You then gather a dataset that indicates a human label's preferences between multiple completions of the same prompt, and use this dataset as a reward signal, or to create a reward signal, to fine-tune an instruction an instruction tuned LLM. The result is a tuned large language model that generates completions or outputs that better aligns with the preferences of the human labelers. I am delighted to introduce the instructor, Nikita Namjishi, who is developer advocate for Gent of AI on Google Cloud. She is a regular speaker at Gen2AI developer events and has helped many people build Gen2AI applications. I look forward to her sharing her deep experience, her deep practical experience with Gen2AI and with ROHF with us here. Thank you, Andrew. I'm really excited to work with you and your team on this. In this course, you learn about the RLHF process and also gain hands-on practice exploring sample data sets for RLHF, tuning the LLAMA2 model using RLHF, and then also evaluating the newly tuned model. Nikita will go through these concepts using Google Cloud's Machine Learning Platform, Vertex AI. What really excites me about RLHF is that it helps us to improve an LLM's ability to solve tasks where the desired output is difficult to explain or describe. In other words, problems where there's no single correct answer. And in a lot of problems we naturally want to use LLMs for, there really is no one correct answer. It's such an interesting way of thinking about training machine learning models, and it's different from supervised fine-tuning, which you may already be familiar with. RLHF doesn't solve all of the problems of truthfulness and toxicity in large language models, but it's really been a key part of improving the quality of these models. And I think we're going to continue to see more techniques like this in the future as the field evolves. So, I'm really, really excited to share with you just how it works. And I'm happy to say you don't need to know any reinforcement learning to get started. Many people have worked to create this course. I'd like to thank, on the Google Cloud side, Bethany Wang, Mei Hu, and Jarek Kazmierczak. From DeepLighting.ai, Eddie Xu and Leslie Zerma had also contributed to this course. So with that, let's go on to the next video where Nikita will present an overview of RHF so you can see all the pieces of how it works and how they fit together. Let's go on to the next video.
Reinforcement Learning From Human Feedback