Speaker:
Mark Chen, Research Scientist, Open AI

When: 11:00 AM – 1:00 PM, April 7th, 2023

Abstract:

This talk will provide an overview of large language models (LLMs) and surrounding generative technologies, their development, and applications across various domains. It is tailored for a technical audience with limited exposure to Artificial Intelligence (AI). I will begin with a brief history of language modeling, outlining key developments that have culminated in the emergence of advanced LLMs such as ChatGPT. Subsequently, I will introduce the fundamental principles that drive the efficacy of LLMs, focusing on their ability to lead and represent complex language patterns. I will discuss the technical challenges involved in scaling LLMs as well as the remarkable properties that have emerged at scale, such as the ability to tackle new tasks with no or minimal new training data. I will further introduce representation learning through language modeling and its role across a range of applications. Beneficial AI systems must be both useful and harmless. ChatGPT, a cutting-edge language model developed at OpenAI, was trained using Reinforcement Learning from Human Feedback (RLHF) and other alignment techniques to optimize its behavior and safety. I will give a high-level overview of RLHF and its importance in creating safe and beneficial AI. I will conclude with LLMs applications across other domains, such as DNA and protein sequence modeling.         

Flyer for Event: Link