Mercial.

Get Involved

Hackathons are open to everyone, and you can check for upcoming ones via our Slack. The research fellowship and the technical bootcamp have a short application so we can keep cohorts small and well matched. If you are unsure whether a programme is the right fit for you, please err on the side of applying.

Research fellowship applications are open. The first cohort runs 1 October – 1 November. See the mentors, their problems, and apply. Bootcamp applications are not open yet — dates will go out on the newsletter and in the Slack.

Our programmes

Building knowledge New to AI safety? Start here before you apply.

Our fellowship and bootcamp require some background knowledge. If you are new to AI safety, consider first completing this BlueDot course, or taking part in an AI safety fellowship run by your local EA or AI safety group (for example, EA Warwick's) before you apply. Technical experience is not required to complete the research fellowship — you can choose to do conceptual research. We assume that you are comfortable with programming in the bootcamp.

The research fellowship A month working on an open problem, with mentorship, compute funding and a write-up at the end.

A month-long programme where fellows work remotely on an open problem in AI safety, with support from the Mercial team, and write up what they find. Mentors will regularly contact the fellows to provide guidance and accountability, but fellows will be agentic and take initiative. Fellows work on one of the problems their mentor has chosen to supervise, or can propose their own in-scope idea. The fellowship suits people who already have background knowledge on AI safety and want real applied experience. Compute funding is available.

See this cohort's mentors and apply, or see the events page for dates of later cohorts.

The bootcamp An intensive week on the ARENA curriculum: transformers, interpretability and RL.

An intensive programme following the ARENA curriculum: PyTorch fundamentals, transformer architecture, mechanistic interpretability, and reinforcement learning, mixing guided exercises with programming.

You need comfort with programming (Python) and roughly first-year undergraduate maths (linear algebra, calculus and probability). No machine learning background required.

How to apply

Research fellowship
Open now for the October cohort. Each mentor runs a stream of problems; you pick a stream and apply to it on the fellowship applications page.
Technical bootcamp
Not open yet. Dates will be announced on the newsletter and in the Slack — join either and you will not miss them.
Hackathons
No application. Turn up — dates are on the events page.
Mentoring
If you would rather mentor a research fellow than take part as one, we are taking mentor applications for future cohorts on the mentor page.

What applying will involve

  1. Fill in the form

    Tell us your background, how much you have interacted with AI safety, and what problem you’re interested in investigating.

  2. Follow up

    We’ll invite you to an informal conversation, mostly to gauge fit and answer your questions.

  3. Join the cohort

    Offers go out within a week of the deadline, with start dates as shown on the events page.

Apply to the research fellowship Mentor a fellow