AI Safety Fellowship
Exploring the fields of AI Alignment and Governance
Duke's FIRST AI Safety Fellowship, focusing on Technical Alignment, will be happening this fall, with the following topics:
Sept. 17/18: Intro to ML (optional)
Sept. 24/25: AGI
Oct 1-2: Goals and Misalignment
Oct. 15/16: Threat Models and Types of Solutions
Oct. 22/23: Learning from Humans
Oct. 29/30: Decomposing Tasks for Outer Alignment
Nov. 5/6: Towards a Principled Understanding of AI Cognition
Nov. 12/13: AI Governance and Careers
Nov 19/20: What now? Developing an Inside View and Starting Projects
Fellowship Structure
The Fellowship takes place over the course of the 2022 Fall semester. We expect to accept 6-8 Fellows. There will be a discussion series with 9 meetings (each approximately 90 mins), group socials with all Fellows as well as members of the broader Duke EA community, as well as academic/career advising. Meetings will take place once a week. Timings will be flexible and will be arranged once Fellows have been selected. Each discussion will consist of a set of study resources and an in-person discussion. The resources will take approximately 90 mins to go through and must be completed in preparation for the meeting.
Fellows will also gain the skills and knowledge to form ‘inside views’ on AI Alignment and begin work on a project to expand upon the knowledge and skills gained. While the completion of such projects is not expected during the Fall semester, we will reconvene in Spring to discuss and present everyone’s progress and any updates or questions Fellows have at that time.
The weekly time of the discussion meeting is TBD. We will have several different sessions at different times each weekend. We will decide times democratically, once Fellows have been selected.