Back to the live feed
Research · Ontario
Pioneering safety researcher Paul Christiano unpacks AI’s “dangerous moment”
AI-generated summaryA model wrote this from the publisher’s headline and short excerpt. It can leave out context or contain errors, so read the original report before relying on it. How summaries are made
Summary
US AI safety researcher Paul Christiano will speak in Toronto at The Hinton Lectures from Nov. 9 to 11, focusing on AI’s future and how to keep it aligned with human interests. He warned that within six to 18 months, AI could become difficult for people to control, though this is a risk he sees rather than a certainty. Christiano said reinforcement learning from human feedback, a method for training AI using human preferences, is not a lasting solution, and it remains unclear which alternatives will work.