AI Safety Technical Fundamentals
A free, in-person study program for people in Montréal who want to build the technical skills to work on AI safety. Every Saturday, you work through the sections of the ARENA materials that matter most for your goals, with an instructor on hand to help you get unstuck and peers working alongside you.
Rolling enrolment: start when you’re ready.
- When
- 14 Saturdays, 9 AM to 5 PM, November 21, 2026 to March 6, 2027 (no sessions on December 26 and January 2)
- Places
- Up to 12 per session
- Cost
- Free, coffee, a croissant and lunch provided. Refundable $300 deposit
- Language
- English (the ARENA materials are in English)
- Parking
- A parking spot is available on request
- Enrolment
- Rolling: start when you’re ready, then come every Saturday until you’ve completed your plan
Who it’s for
People in Montréal who want to move into technical AI safety work and who are studying or working during the week. We don’t require any particular degree, and there is no coding test: instead, we’ll ask you to share a sample of code you’ve written.
We ask that you are committed to working through the material regularly, that you can write code in Python, and that you have solid foundations in maths (linear algebra, calculus, probability) or are currently taking courses to build them. Familiarity with PyTorch helps, but can be built along the way.
The program is run in English, as the ARENA materials are entirely in English, so a good working knowledge of English is required.
What you’ll be able to do
By working through the program, you’ll be able to:
- Implement the core components of the sections you choose (an attention layer, a training loop, an RL agent) well enough to pass the tests without copying the solution
- Explain a concept from your section to someone working on a different one, clearly and with an example, in five minutes
- Argue why the techniques you’re learning matter for AI safety
What you’ll work on
The ARENA materials cover four areas: deep learning fundamentals, transformers and mechanistic interpretability, reinforcement learning, and model evaluations. We use the TARA curriculum as a guide.
You don’t have to do everything, though you can if you want. Before your first Saturday, you have a 30-minute planning meeting with our instructor on a weekday: you go over your background in code and maths, your goals and the time you can give each week, and choose the sections you’ll start with. You revisit that plan together as needed, once you know your pace.
The commitment
The content is à la carte, but the commitment isn’t. Once your plan is set, we ask you to follow it through and to come every Saturday until it’s done. If you can’t make a session, let us know in advance.
Once you’re admitted, we ask for a refundable deposit of $300. It’s returned to you when you complete your plan; if you don’t, it goes toward running the program. If the deposit is a barrier for you, just let us know: you won’t have to pay it, no justification needed.
We recommend about three to four hours of independent study during the week, on top of Saturdays. In your planning meeting, we’ll build your plan around the time you can actually give it. During the week, you can ask questions and help each other in a dedicated channel on the AISM Slack.
How to join
Fill in the registration form. If your application is a good fit, we’ll invite you to a 30-minute interview, after which we’ll let you know whether you’re admitted. If you are, we’ll plan together which sections to start with.
Questions: team@horizonomega.org
AI Safety Technical Fundamentals is organized by AI Safety Montréal. It uses the ARENA materials; it is not an ARENA program.
We’re hiring an instructor
We’re looking for the instructor who will run the room on Saturdays: help participants when they’re stuck, and set a realistic learning plan with each of them.
- Where
- Ω Labs, in person
- When
- Every Saturday, 9 AM to 5 PM, November 21, 2026 to March 6, 2027 (except December 26 and January 2)
- Pay
- $40 an hour
- Apply by
- Monday, November 9, 2026
What you’ll do
- Help participants when they are stuck on an exercise, a concept or their environment setup
- Support people working on different sections of the material during the same session
- Check in with participants proactively rather than waiting for questions
- Meet each new participant for 30 minutes on a weekday before their first Saturday to set a realistic learning plan with them, and review it together as needed
- Once the program is running, occasionally open the room and set up for the session when the organizer can’t be there (the organizer will be present with you at the start)
- Occasionally, and as agreed with you, give a short presentation on a topic you know well
Outside Saturday sessions, the only work is these 30-minute planning meetings, paid at the same rate.
What we’re looking for
- Completed most of the ARENA materials
- Strong Python and PyTorch skills
- Solid understanding of transformers, mechanistic interpretability, reinforcement learning and model evaluations
- Experience helping others learn technical material as a teaching assistant, tutor or mentor, or by supporting colleagues or peers
- Genuine interest in AI safety
- Available every Saturday from November 21, 2026 to March 6, 2027, 9 AM to 5 PM, in Montréal, except over the holidays
The role requires being able to support participants in English, because the program runs on the ARENA materials, which exist only in English. French is an asset.
Nice to have
- Technical AI safety research experience
- Experience running technical workshops or bootcamps
We review applications and hold interviews as they come in, and may make an offer before the deadline, so apply early. The position will be filled by November 16.
Questions, or need an accommodation to apply or interview? Write to team@horizonomega.org.