Intro to AI Safety Fellowship

About

The Intro to AI Safety Fellowship is a 6-8 week, quarter-long course which explores the impact of increasingly powerful AI systems on society, explains the problems which the AI safety field is currently working on, and goes over active areas of research.

The program consists of a series of video and reading lessons which are accompanied by weekly, collective meetings that expand on the topics covered in the lessons. You can expect to learn from many prevailing sources in the field which will help you develop a solid research agenda that you may work towards in the following quarters.

In the 6-week technical fellowships, participants will complete about 3 hours of core content outside of hour-long weekly meetings, split between conceptual readings and technical content. In the 8-week governance fellowship, there is no mandatory time commitment outside of the two-hour weekly meetings, each of which will be equally split between readings and discussion. We encourage participants to do some or all of the optional readings before each meeting to dive deeper on topics that interest them!

Curriculum Options

Since Fall 2023, we have started offering multiple tracks for our intro fellowshups. Click on the links below to explore each tracks’ associated curricula:

  • Reinforcement Learning: Create an agent that learns to play connect four.
  • Transformers: Understand language models and build GPT-2 from the ground up.
  • Diffusion: Build your own text-to-image generation model like Stable Diffusion.
  • Computer Vision: Design models to analyze images for object recognition and classification, recommended for those with little-to-no machine learning experience.
  • Governance: Understand policy approaches to mitigating AI risk and propose ideas of your own.

Eligibility

For our transformers fellowship, we’re looking for students who meet any of the following requirements:

  • Have taken CS 31 or an equivalent class OR
  • Have worked on a Python project OR
  • Have strong skills in another programming language
  • Additionally, a basic understanding of machine learning fundamentals and familiarity with contemporary AI research would provide a good baseline for the course’s content.

For our governance fellowship, we do not enforce any particular requirements. However, we are very excited to see participants who:

  • Have strong writing and communication ability
  • Are familiar with US policy (e.g., have engaged in policy research or an internship)
  • Have some familiarity with computer science or another technical field.
  • Once again, these are not enforced! We encourage all participants to apply, because interest and motivation are the two traits we care most about.

Apply

Applications are open for Fall Quarter 2024!

The deadline to apply is Friday, October 4 at 11:59 PM.

Post Fellowship Opportunities

For those who have completed the Intro Fellowship, we will be offering various other additional programs. Please fill out this form if you are interested in any (or all) of these programs.

Reading Group

Come to the AI Safety weekly reading group to learn more about the latest topics in AI and alignment research. Discuss papers with your peers and deep dive into them! An understanding of machine learning at the C147+M146/Math 165 level is encouraged but not required.

Upskilling Programs

This quarter, we will be running various upskilling programs open to those who have completed the Intro Fellowship. This will include weekly meetings with content inspired by ARENA, but additionally, dependant on interest, deeper dives into the technology behind machine learning (ie. CUDA programming and autograd engines).

Alignment Research Team

The Alignment Research Team (ART) is our main post-fellowship program. This quarter, our ART projects will be predominantly structured around getting club-members access to our GPU resources for their projects. As such, ART will be much less structured so that organizer time can be spent on other programs. Additionally, we will be running weekly coworking sessions, similar to those run last year.

Newsletter

Please sign up for our weekly newsletter to stay up to date with events and important announcements. Click here to join.

Discord

We have a Discord server where you can discuss with us. Click here to join.