The PRISM Alignment Project – July 18, 2024

Session Description

Introducing the Toronto Data Workshop (TDW) Summer Series, bringing together academia and industry to share data science best practice. TDW is a joint initiative between the Faculty of Information and the Department of Statistical Sciences and is organized by Assistant Professor Rohan Alexander, Professor Kelly Lyons and MI student Michaela Drouillard. All workshops are held virtually on Zoom and are free to attend. Everyone is welcome.

Sign up to receive weekly invitations to Toronto Data Workshops and visit Professor Rohan Alexander’s website for full event details. Stay tuned as we update this page as more information becomes available.

Alignment has become a buzzword that frequently surfaces throughout the AI ecosystem. But what does it mean to align LLMs to the hundreds of millions of people who now interact with their outputs, each with different preferences for language, conversational norms, value systems, and political beliefs? This talk will explore the gnarly concept of alignment, specifically for large language models (LLMs). Using insights from the recently released PRISM Alignment Dataset, we’ll cover key questions around empirical alignment efforts, such as how to collect human feedback using various scales and signals, where to focus human labour, who to ask for feedback, and whether to target objectives of personalised or collective alignment. Through more comprehensive and representative data, PRISM highlights the challenges of aligning LLMs to meet the diverse expectations and beliefs of a global audience.