Abstract: The fluency of LLMs like ChatGPT and New Bing (also known as Sydney) heightens the need for corresponding systems to detect whether a piece of text is machine-written. In this talk + Q&A, I’ll first provide an overview of our recent work DetectGPT, which leverages the local curvature of a language model’s probability function to detect machine-generated text. Then we’ll open the floor to questions about our work, machine-generated text detection more broadly, and LLMs in general. For background, see: hai.stanford.edu/news/human-writer-or-ai-scholars-build-detection-tool

Biography: Eric Mitchell is a fourth-year CS PhD student in the Stanford AI Lab, advised by Chelsea Finn and Christopher Manning. His research focuses on augmenting large pre-trained language models with tools to make them safer, more controllable, and ultimately more useful to society.

NY Times Articles You May Want To Read:

• The Brilliance and Weirdness of ChatGPT
• An Unsettling Chat With Bing
• Read the Conversation
• How Chabots Work

To join the scheduled Zoom meeting: http://ee380.stanford.edu