Founding cohort · Applications open
A 12-week online program where small teams of students design, run and write up an original language-model research project, mentored by a former Meta AI researcher. GPU compute is included.
How it works
Weeks 1–4
Lectures on modern language models and how research works. Read the literature, pick a question your team can actually answer, and lock a proposal with datasets and metrics.
Weeks 5–9
Build baselines, run experiments on real GPUs, and iterate on what you find. Every week includes code review and a hard look at whether the results hold up.
Weeks 10–12
Draft the paper, revise it through internal review, release your code, and submit to a target workshop at a major AI conference.
Research areas
Probe what models represent internally, and use steering vectors to change their behavior in targeted ways.
Measure how language models treat different groups of people, and test methods that make them fairer.
Evaluate when AI agents follow, bend or break their constraints, and design ways to keep them in bounds.
Your mentor
AI/ML researcher working on interpretability, fairness, and safety in language models. Formerly at Meta AI. You'll work directly with Keren every week.
Pricing
Most students
[YOUR PRICE]
A team of 3–4 students and one mentor. Compute included.
Apply for the team trackIndividual
[YOUR PRICE]
Your own project with a dedicated mentor. Compute included.
Apply for the 1:1 trackNeed-based scholarships are available for every cohort; ask about one in your application. Refund policy: [YOUR REFUND POLICY].
FAQ
High school students, undergraduates and working professionals who want to do original AI research. Most participants have never written a paper before.
You should be comfortable writing and running Python. We teach the machine learning and research skills during the program.
Plan on 5-10 hours a week, including two live sessions.
No. Every team finishes with a submission-ready paper and submits it to a target venue, but acceptance is decided by peer review.
No. Each team gets a cloud GPU budget for the full twelve weeks.
Applications close [DEADLINE]. Seats are limited to [NUMBER].