## Who is BlueDot Impact?

We're a philanthropic startup **building the workforce that protects humanity**.

Since 2022, we've trained over 10,000 people in AI safety and biosecurity. Our alumni work at Anthropic, DeepMind, the UK AI Safety Institute, and dozens of other organisations shaping how transformative technologies affect the world.

We now train over 1,000 people a month, and are deploying over $1M per month in grants to accelerate the most exceptional talent coming through our pipelines. We're a small team headquartered in San Francisco. We've raised $34M from Coefficient Giving.

## Why this role matters

There are a huge number of open technical problems we must solve to make AI go well, and the field doesn't have enough exceptional people working on them. Worse still, public writing on these open problems is largely out of date, and is missing many of the most important nuances that are locked in experts' heads and private Google docs. Talented newcomers spend their first months building toy projects the field doesn't need and experts drown in open problems without the capacity to find the people who could take them on.

Our [Technical AI Safety Project Sprint](https://bluedot.org/courses/technical-ai-safety-project) is where strong technical researchers and engineers do their first real technical AI safety project. Each monthly round gets ~250 applications, we accept ~110, and invest additional resources into ~10 (each of these figures has ~tripled since January).

Currently, it's nobody's whole job to make the program excellent. Most projects don't tackle the problems experts actually care about, leaving many participants struggling to land next steps after the program ends.

You'll own the sprint end to end: which problems participants work on, who gets in, how the work is judged, how the program itself runs, and who we accelerate afterwards with advising and funding.

## What you'll do

**Decide what the field's newcomers build**

- Sit down with the field's best researchers and extract the problems they wish someone was working on, including the nuances that are often missed.
- Translate these open problems so that a strong newcomer can quickly understand them and start solving them.
- Every participant should be able to defend why their problem matters and how solving it contributes to AI safety before they start executing.

**Build the system to spot great work**

- Build the scalable end-of-sprint review system: every submitter gets substantive feedback and a calibrated quality rating that experts would agree with.
- Use it to catch exceptional people who slipped through our sourcing net, and to track whether your changes to the program actually improve the work.

**Accelerate the best participants**

- Every "strong yes" participant gets a 1-1 advising call with you: understand their blockers, give them opinionated advice on their next step, and unblock them with career transition grants when funding is what's in the way.

**Fill the sprint with the people the field needs**

- Triple the number of "strong yes" applications, from ~10 to 30 a month.
- Identify the domain expertise the most important open problems require, and source the people who have it.

**Run an excellent sprint**

- Ensure the top participants finish with work that experts recognise, allowing them to take impactful next steps in their AI safety journeys.
- Build a bench of excellent facilitators who are already contributing to technical AI safety and push participants to be more ambitious.
- You won't do this alone. Two course ops staff work with you on the inbox, onboarding, and logistics, so your time goes to strategy, talking to experts, and supporting the top participants.

## About you

You might be a great fit if you:

- **Have supervised researchers.** You've been a research manager, mentor, or project supervisor, and your mentees produced better work than they would have alone.
- **Can scope.** You've turned messy, ambitious research questions into concrete projects that other people actually executed (fellowship proposals, hackathon challenges, open-problem write-ups that strangers picked up and ran with) and your write-ups make hard problems legible to people new to the field.
- **Have technical AI safety research experience.** You've done research through a fellowship (MATS, LASR, Astra, ERA, Anthropic Fellows, Pivotal) or at a safety org, and you've gone deep enough on at least one research agenda to advise someone else on how to do the same.
- **Write in public.** You have an opinionated blog or have written explainers or given talks on hard technical concepts that smart newcomers can understand.
- **Have takes.** You hold specific, falsifiable opinions about which agendas, orgs, and interventions are over / underinvested, and you can make the strongest case against your own view.
- **Spot exceptional people fast.** In a 30-minute conversation you can read how mission-aligned, agentic, and technically sharp someone is.
- **Default to action.** Your instinct is to launch a two-week experiment rather than spending months writing a doc. You fix and improve things fast.

Bonus points if you've been a BlueDot facilitator or alum, run your own fellowship or teaching program, or evaluated talent for grants or admissions.

We encourage speculative applications: most strong candidates won't meet all these criteria!

## What we offer

- **Influence 100s of exceptional technical researchers:** Your work will directly influence a large proportion of the people who enter this field. You decide what hundreds of talented people build when they first enter technical AI safety, and the field gets real work on the problems you think matter most.
- **Real autonomy:** The sprint is yours. Our expense policy is "[act in BlueDot's best interest](https://blog.bluedot.org/p/expenses)", unlimited PTO, minimal bureaucracy.
- **Location:** San Francisco strongly preferred, working in person with the team. We're open to London-based or remote for exceptional candidates. US visa sponsorship available.
- **$160-250k salary** depending on experience, 10% employer 401(k) contribution (no match required), and comprehensive health insurance.

## Apply today

Applying takes 20-30 minutes, and we encourage you to **apply as soon as possible**.

We're evaluating candidates on a rolling basis and want to make an offer quickly.

## Application process

- Initial application
- 3-hour work test (paid)
- 45-minute interview
- In-person work trial in San Francisco

If you have any questions about the role, email [will@bluedot.org](mailto:will@bluedot.org)


Scale the project sprint program to get exceptional technical researchers and engineers doing their first work on real problems, so they quickly start contributing to the field.

Program Lead, Technical AI Safety Project Sprint