Seminars

We host two seminar programs for those new to AI safety, regardless of technical experience.

AI Alignment Seminar

Overview

Experts broadly expect significant progress in AI during our lifetimes, potentially to the point of achieving human-level intelligence. Digital systems with such capabilities would revolutionize every aspect of our society, from business, to politics, to culture. Worringly, these machines will not be beneficial by default, and public interest is often in tension with the incentives of the many actors developing this technology. Absent a dedicated effort, AI systems will outpace our ability to interpret their underlying behavior, instill our values in their objectives, and build safeguards against potential failure. From these challenges, the field of AI safety has emerged.

Details

The alignment seminar focuses on the technical challenge of developing AI to the benefit of humanity's future. We begin by discussing the risks posed by advanced AI through the lens of reward misspecification, goal misgeneralization, and instrumental convergence. In later weeks, we dive into proposals for solutions to this these problems such as inverse reinforcement learning, iterated amplification, automated debate, interpretability, and agent foundations. We thank OpenAI researcher Richard Ngo for his work developing this curriculum.

In this 9 week program, participants will have weekly 1 hour discussions with a cohort of 4-6 led by 1 experienced facilitator. Cohorts are selected based on participant's schedules and meeting type (virtual/in-person) preferences.

Prior to these meetings, participants are expected to complete 2-3 hours of readings and exercises. The program is optionally followed by a month completing a self-directed project to help you advance relevant skills and knowledge.

Applications for Spring 2023 are due Sunday, 2/5 at 11:59pm. For updates, please subscribe to our mailing list and join our Discord.

[Syllabus] [Application]

Testimonials

"The program gave me the knowledge and confidence I needed to pivot from strategy work to studying AI at BCG’s think tank. I would not have been as strong of a candidate without it! Huge kudos to the team for organizing a thoughtful, engaging course.
— Emily D

"The seminar was great for covering a wide array of what the current problems and approaches are in alignment, and prepared me to work as a researcher at Stanford's ML Alignment Theory program a few months later."
— Michael E

"I had been interested in AI safety for a while, but participating in the program was a critical step for me to gain a deeper understanding of the problem. I had the opportunity to engage in fascinating discussions with my peers and was better prepared to participate in research at UC Berkeley's Center for Human-Compatible AI afterwards."
— Michael C

Report abuse