Osbert Bastani (Penn)- Unlocking the Potential of Coding LLMs
Abstract: One of the most striking capabilities of LLMs has been their proficiency at coding. This capability has unlocked a wealth of benefits beyond helping build software — it enables them to access vast capabilities implemented in software libraries, and even to take actions that affect the external world. I will present two projects on such “code actions”. First, I will introduce a series of projects leveraging code actions to design more effective robotics simulation environments, from generating shaped reward functions to creating interaction-ready simulation assets. Second, I will describe a new programming language called Quasar designed to improve the performance, security, and reliability of code actions.
Speakers
Osbert Bastani
Osbert Bastani is an associate professor at the Department of Computer and Information Science at the University of Pennsylvania. He is broadly interested in trustworthy machine learning, especially how to build reliable systems that incorporate (unreliable) machine learning models.
