AI safety

Why

We are living through the strangest period in human history, and the line between reality and science fiction gets thinner every quarter. We are racing towards systems more capable than us without proper guardrails and without any deep understanding of what these alien minds want or think. That is a safety problem of a size we have not faced before.

I did not start my PhD worried about this. My research was not about AI safety and I won’t pretend it was. But I watched the field’s capabilities grow at close range while training networks to emulate cellular automata, and I made deliberate choices to stay close to the state of the art. The actual pull came from two good friends who had read MacAskill and could not stop radiating effective altruism at me. I have a strong sense of justice, and the movement’s habit of asking “does this actually help?” resonated. Once I noticed that AI safety was both the problem that world considers most pressing and the one my background lets me contribute to, the choice was more or less made for me.

So I joined formally, helped build EA Ghent, went to the conferences, read the books. Near the end of my PhD I took the next steps seriously: the BlueDot Technical AI Safety course, and a place in the Iliad Intensive in San Francisco. I’m excited about what’s next.

Interests

Interpretability and control, mainly. Field building locally. Not governance as a career, but I want to keep strong ties with that side: Brussels is half an hour away by train and that proximity is worth using.

So far

• Accepted into the Iliad Intensive, a full-time taught course on foundational alignment research, in October in San Francisco.
• Board member of Effective Altruism Ghent since 2025, where I host and organise events including Powerful AI is coming. How do we get it right? and Your Career as a Force for Good.
• Cohort student in the BlueDot Technical AI Safety course, August to September 2026.
• EA Summit Brussels 2026 and EAGxAmsterdam 2025.

What I bring

• Dynamical systems on networks: stability, Lyapunov spectra, damage response, mean-field approximations.
• Experience training small neural networks to emulate and classify discrete systems, and an honest record of when the learned model lost to plain simulation.
• Reproducible research as a habit: pinned environments, one script per figure, CI that checks it.
• Five years of teaching mathematics to engineers, and the organisational side of running a conference and a community.

Looking for

Right now, a series of short, strong experiences: courses, short fellowships, research sprints. Longer term, joining an existing organisation, remote or in Belgium, or, better, founding one here in Ghent.