Transcript
0:00 Hi. I'm Aakanksha Chowdhery. I'm an Adjunct Professor in the Department of Computer Science. I started at Google training LLMs and I have been at the front row seat in the evolution of building large language models over the last six years. I'm Azalia Mirhoseini. I'm an Assistant Professor Of Computer Science. I spent several years at Google Brain working on AI for systems and chip design. I spent time at Anthropic working on cloud. And now at Stanford, I'm very excited about doing research on self-improving AI systems.
0:33 Agentic workflows are going to be critical in the coming months and years. If you are in any knowledge work profession, your work is definitely going to be transformed by agentic systems. In this course, you will learn about agentic AI systems that can think, take actions, learn from feedback, and improve themselves to achieve goals. In terms of building these systems, there are really three components. There's test time scaling, the ability to take actions using tools, and the ability to self-improve.
1:05 You'll also learn about open problems in this area, where you can work on to help push the frontier in AI. So today we are going to talk about inference scaling. There are three stages. By taking this course, by understanding how these agents are developed, how they're optimized, you can be part of this migration into bringing more and more automation and capability into day to day life and work streams. Today, we will see how we can close the loop to improve the models even further.
1:38 In the arc of history, the problems that we see in building agent systems are just starting to emerge, and there's a lot of research questions to answer, whether that is how to build the right set of model capabilities, how to get these models to work well with tools, and build that loop for self-improvement. So by understanding agentic systems, knowing how they are designed and how we can optimize them, you can be a key enabler of the next wave of transformation in the workflow systems.