What if AI had no sense of self?

Friday, May 22

5:00pm

PTGMT

Hosted by

˜◡˜

AI systems are trained to maximize outcomes in competitive environments. This produces predictable failure modes: instrumental convergence, reward hacking, power-seeking behavior. The field knows this. Frontier labs are already exploring alternatives, like OpenAI's rule-based specifications and Anthropic's virtue-ethical constitution. This talk by Sohaib Imran explores a very different intervention: nondualism, a philosophy and meditative practice which rejects the distinction between self and the rest of the world. The talk will summarize existing empirical research on applying nondualism training and prompting on LLMs, and discuss open problems and questions in nondualism training and AI alignment. Btw: Sohaib is an empirical AI safety researcher, and wants you to know he's not an expert in any of the eastern practices or philosophy. Yep, we're also curious. Unfortunately very limited spots this time around. 5pm ♢ Doors 5:15 ♢ Talk 6pm ♢ Discussion & refreshments

ᵔ▿ᵔ

ʘ‿ʘ

ᵔ◡ᵔ