Inside ML4Good’s Technical AI Safety Bootcamp

Summary

Igor Maljkovic describes his experience at ML4Good, an intensive eight-day technical AI safety bootcamp held near Lyon, France. Applicants complete a form, a 10–15-minute interview with five questions, and receive results within a few days; the author says future cohorts remain an option for unsuccessful applicants. The programme is free, provides accommodation and three meals a day, and can reimburse up to €180 in transport costs. Sessions generally run from 9 a.m. to 7–7:30 p.m., with breaks and dinner after the final lecture. Its technical curriculum includes AI agents, pre-training, RLHF, RLAIF, Constitutional AI, interpretability, model reasoning such as Chain-of-Thought and ReAct, evaluations, and transformers, alongside Colab coding exercises. Participants can skip exercises they already know and use the time for discussions, including emerging AI risks and mitigation. The bootcamp also covers AI governance and technical AI governance, using lectures and interactive discussions rather than lecture-only teaching. The author, who is already a PhD student in AI security and safety, particularly values sessions on fellowships, self-funding, career transitions, and starting an AI safety organization or company. He portrays the teachers as knowledgeable and approachable and recommends that technically experienced people still apply.