The hardest six months of a young African's life. A full-time, fully funded residency for people transitioning into AI and AI safety.
Everyone builds the same foundations. From there you challenge yourself to go deep in one track.
Training, optimisation and the mathematics underneath them, built up until the abstractions stop being abstract.
Open the model. Find the features, circuits and representations that explain what it is doing.
How systems learn from feedback, where that goes wrong, and what specification failures look like in practice.
Design and run evaluations that hold up, including in languages the frontier labs have barely tested.
Adversarial behaviour, misuse, model and infrastructure security, and the failure modes that arrive on deployment.
We are not looking for people who already do AI safety. Most residents will be new to it. We are looking for people who will challenge themselves to go quantitative, and who will burn the midnight oil to get there.
Background, motivation and a short technical exercise.
A conversation about how you think, not a quiz on what you memorised.
A funded place in the cohort, with a stipend for the duration.
Register your interest and we will write to you when the first cohort opens.