Vierailuluento: Dan Brown "Truth, lies, fiction, and creativity in AI"
Milloin
Missä
Tapahtuman kieli
Truth, lies, fiction, and creativity in AI
Dan Brown
University of Waterloo
Abstract: A baseline need for language models and visual models to make compelling art is a sense of what is true, what is false, and what is a fictional worldview. We discuss a number of recent experiments and theoretical ideas connected to these ideas: do LLMs understand truth and fiction? Can they even consistently maintain a worldview across multiple prompts, let alone a whole story? Can they understand minority groups outside stereotypes? And what are we hoping for them to do in the first place? What creative tasks can AI models even succeed at, and what does it mean for them, or anyone, to be creative? Joint work with a lot of people, including Anna Jordanous, Max Peeperkorn, Piotr Sawicki, Marek Grzes, Aisha Khatun, Maya Ackerman, and others.
This guest talk is organised by Nicola Dainese, and hosted by Assistant Professor Christian Guckelsberger, Department of Computer Science.
Bio: Dan Brown is Professor of Computer Science at the University of Waterloo, where he has been a faculty member since 2000. His research is alarmingly interdisciplinary, ranging from finding rhyme in music lyrics to identifying genes in genome sequences to preventing problem gambling to theorizing about what makes for a high-quality review to teaching computers to write poetry and tell stories. Right now, he’d be most excited if he could get an LLM to tell a good story about two men in love that doesn’t resort to cliche and stereotype. No luck so far.