AI Safety Founder Shares 7 Key Lessons from Starting Formation Research
Lessons on in-person work, niche conversations, and public career capital from a new AI safety org.
Alfie Lamerton, after completing an AI safety fundamentals course by BlueDot Impact and an AI MSc, founded Formation Research to tackle the neglected problem of AI-enabled lock-in and secret loyalties. Drawing from his experience in an incubator at LISA, he distills seven key lessons for aspiring AI safety entrepreneurs. First, he stresses the power of in-person work: being physically present at LISA allowed him to constantly rehearse his story and receive immediate feedback, accelerating progress far beyond remote work. Second, he advocates for targeted conversations with subject-matter experts—talking to ML researchers and Forethought members helped him refine his research agenda on lock-in and measurement techniques, using frameworks like the VFWPA template from The Mom Test to structure outreach and secure small commitments.
Third, Lamerton highlights the importance of building and publicizing career capital—demonstrating competence through visible work to attract collaborators and funding. Though he only details three of the seven lessons, the post underscores the need for systematic networking, iterative thinking, and leveraging community resources. Formation Research now seeks a founding team to empirically study secret loyalties in deep learning, aiming to develop defenses for AI labs. The lessons serve as a practical guide for researchers transitioning from coursework to real-world impact in the AI safety ecosystem.
- In-person collaboration at LISA (London) significantly outperformed remote work for refining the org’s narrative and accelerating feedback loops.
- Using the VFWPA template and Mom Test questions enabled Lamerton to secure targeted advice and small commitments from ML researchers and Forethought members.
- Public career capital (via 80,000 Hours principles) was crucial for demonstrating competence and attracting collaborators in the AI safety community.
Why It Matters
Actionable playbook for researchers scaling AI safety ideas into funded, team-driven organizations addressing high-risk technical problems.