Current research question
What can models learn from signals we barely notice?
My PhD currently looks at subliminal learning: how apparently innocuous patterns in data can transmit information and shift model behavior.
AI safety researcher · competitive runner
I study intelligent systems. I race over barriers.
I’m Tibo, a PhD researcher at Vrije Universiteit Brussel working on AI safety, alignment, and interpretability. Outside the lab, I race the 3000m steeplechase.
Brussels, Belgium
VUB
PhD researcher in artificial intelligence
39 notes
Public memos on language-model research
9:23.84
3000m steeplechase personal best
5 languages
Dutch, English, French, Spanish, German
Start anywhere
One person, three ways in.
If you arrived after a conference, a recruiter search, a race result, or one of the more surprising corners of the internet: hello. This is the concise version of what I work on, what I write, and what keeps me moving.
01
Research
Hidden learning signals, mechanistic evidence, and AI safety that stays useful outside ideal lab conditions.
02Notes
One precise question at a time, from attention sinks and truth directions to model merging and long context.
03Running
Official results, personal bests, and the less polished side of getting over barriers.
Beyond the bio
A life needs more than one lane.
My work is serious, but I do not want the website to pretend every worthwhile hour happens at a desk or on a track. I like cities explored slowly, meals that stretch into conversation, and the useful quiet of a book.
The best parts of a week rarely belong to one category: a hard research question, an honest discussion, a long ride, or something worth ordering twice.
- Home base Brussels, Belgium
- Away from work Books, long meals, cycling, and somewhere new
- Always welcome A recommendation with a reason behind it
- Best way to connect A thoughtful message beats a generic pitch
Reading / recommendations welcome
What stays open after the last page?
Reading is where I let questions remain unresolved for longer than a paper deadline usually allows. I am drawn to books that offer a framework rather than a verdict—and to people with a recommendation they are willing to defend.
A useful book does not have to win the argument. It has to leave a better one behind.
The shelf below is deliberately short: books I can actually say something about, not a decorative list of everything I have finished.
Six books / no ranking
A shelf I would happily argue about.
Not necessarily the six “best” books I have read. These are the ones that gave me a durable question, a useful method, or a line of thought I still return to.
Gödel, Escher, Bach
Formal systems, strange loops, minds, and the rare kind of ambition that changes what a book can be.
Antifragile
The one Taleb entry: a vocabulary for uncertainty, fragility, and systems that improve through disorder.
How to Solve It
A practical bridge between mathematical problem-solving and the craft of helping someone else learn it.
The Stranger
Spare, unsettling, and much harder to forget than its length suggests.
Bullshit Jobs
A provocative lens on work, meaning, and the strange institutions people learn to treat as inevitable.
Parkinson’s Law
Short, mischievous, and uncomfortably observant about organizations.