
Deep Field Fellowship
Years of AI safety context in six weeks. Full-time and in person in San Francisco, with flights, housing and food covered.
Takes 5 minutes.
- Next round
- 1 Feb – 12 Mar 2027
- Where
- BlueDot Campus, San Francisco
- Cohort
- 20 to 30 people
- Covered
- Flights, housing and food
About the program
The people having the biggest impact on making AI go well have a deep understanding of the whole field. They've thought hard about the different risks, the plans for mitigating them, who's working on what, and what the biggest disagreements are. Based on that understanding, they're able to make their own informed decisions.
Most of them built it slowly, over years, with no one guiding them. We think it can be done in weeks.
Online courses give you tens of hours of learning. Research fellowships give you hundreds of hours focused on one narrow research agenda. The Deep Field Fellowship gives you hundreds of hours across the whole field: six weeks, full-time, in person, alongside a small group of exceptional people who are dedicated to making AI go well.
What you'll study
How AI works and where it is today
How modern AI systems are built and trained, what they can do now, and the race to build more powerful ones, both between AI companies and between the US and China.Where AI is going
How to forecast AI progress, the case for short and long timelines, and what a world with superintelligent AI could look like.How AI could go wrong
Three ways AI could lead to catastrophe: humans lose control of it, a small group uses it to seize power, or civilization collapses. For each, you'll study how it could happen, the evidence so far and the strongest objections.What can be done about it
Four kinds of work: preventing dangerous AI from being built, constraining AI and the people who direct it, defending the world if that fails, and redesigning institutions so people keep power.Plans for making AI go well
What a good future could look like, and three bets for getting there: don't build superintelligence, build it carefully, or spread it widely and accelerate defenses. You'll read the leading proposals for each.The open questions
The most important questions for evaluating which bet to take, including how difficult it'll be to make AI safe, whether defenders can keep up with attackers, and what it would take for governments to act. You'll investigate them yourself and you'll build your own view.
Alongside these, you'll learn tools for thinking clearly, including forecasting and game theory. You'll also learn the history of the field, including the many mistakes it has made.
How it works
- Lectures each week from people working on these problems.
- Time to read and deeply engage with the most important pieces of literature.
- Small discussion groups of about four people, each led by an instructor who's deep in the field.
- You'll publish a blog post every week and get feedback on your ideas.
We'll show you a wide range of views, including the strongest criticisms of standard thinking in AI safety. We'll also tell you what we think. The goal is for you to reach your own conclusions.
What you'll leave with
A deep understanding of the field
You'll understand the threats, the work that reduces them, the main plans and organizations, and the open disagreements. When a new proposal appears, you'll be able to analyze what assumptions it's making, what threats it might mitigate, and where it falls short.Views you can defend
You'll have formed your own opinions, had them challenged over and over, and changed your mind where you were wrong.Public writing
You'll have six published posts that demonstrate to the field how and what you think.Close peers and a wide network
You'll have strong relationships with the other exceptional fellows who've gone through this journey with you. You'll also meet many people in the field in San Francisco, including funders, nonprofit staff and people at AI companies.
About you
- You've decided to work on making AI go well, and you plan to do it full-time
- You're new to the field, or you know one part of it and want the whole picture
- You can be in San Francisco full-time for six weeks
- You don't need a technical background
- You don't need to know yet what your role will be. That is what the six weeks are for
About BlueDot
BlueDot builds the workforce that protects humanity. Since 2022 we've trained more than 10,000 people through our courses, and our alumni work at the major AI companies, in governments and at AI safety organizations.
We're running a five-week online pilot of an early version of the curriculum, and using what we learn to improve it for this first in-person round.
Questions
Do I need a technical background?
What does it cost?
Can I take part remotely or part-time?
What happens after the program?
How will you choose people?
When will I hear back?
I'm not based in the US. Can I come?

Spend six weeks getting the whole picture.
We'll invite a small cohort to apply.