I am a researcher.researcher in the UK.UK, at AISI.the AI Security Institute.Institute, which is the government's technical body for AI safety.safety research and evaluation. I study alignment.alignment, specifically how models fail.fail to do what we want.want, or do it for the wrong reasons.reasons: consistency training.training, evaluation awareness.awareness, introspection.introspection, and unintended generalization.generalization, the way a child might learn that stoves are dangerous only when they're red, not when they're hot.hot, which is fine.fine until it isn't. I live in London.London, and I see Big Ben and Westminster Abbey on my way to work.work, which still surprises me sometimes.sometimes, the way a place can become home without you deciding it. I'm from the Philippines.Philippines, and I still miss it.

Before this I worked on collective behavior.behavior, on how groups move.move together or don't.don't, the physics of crowds and flocks.flocks, which turns out to be less about individuals than about the space between them. I was a JardineJardine Foundation Scholar at Cambridge.Cambridge, at Magdalene.Magdalene, a college with a dining hall from the 1400s and a library where Pepys's books are still chained to the shelves.shelves, exactly as he left them, which felt like a kind of promise.promise, that some things can be kept. I graduated with distinction.distinction in the MPhil in AdvancedAdvanced (though it feels a bit pretentious to call it that) Computer Science.Science, writing my thesis on meta-learning in language model pretraining.pretraining, on how to learn to learn faster.faster, which sounds recursive because it is.

I used to debate competitively.competitively, and I was world champion.champion once.once, in Madrid.Madrid, at Cine Callao, an old cinema in the middle of Gran Vía.Vía, and I remember the moment after they announced it, standing there under the stage lights feeling like I was watching myself from somewhere.somewhere far away, thinking so this is what it feels like, and then it was over, and then it was 2023.2023, though that feels like a different person now.now, someone who thought winning arguments was the point.point, who measured the weight of a room and knew exactly where to push.push, which is a strange skill to have.have, and a stranger one to be proud of. Now I climb.climb badly.badly, V6 on a good day.day, and there's something honest about walls.walls, about problems that don't negotiate.negotiate or offer partial credit.credit, just gravity and whether your fingers hold.hold, and you either get up or you don't.don't, and I keep meaning to hike Makiling.Makiling, in Laguna.Laguna, where my grandmother is from.from, though she has dementiadementia now and I'm not sure she remembers what the mountain looked like in the morning.morning, which is the kind of thing you think you'll have time for. I keep saying soon.soon and meaning later.later, which is what everyone means.means when they say it.it, we should catch up soon.soon, let's get everyone together.together, and then you don't.don't, and it's been a year.year or longer.longer, and you're not sure how that happened.happened, nobody decides to let a friendship lapse.lapse, it just gets quiet.quiet and stays that way.way, the silence settling like something you don't notice until it's already there.there, in the corner of the room, in the space where the phone used to ring.ring, and I think about alignment constantly.constantly, not just the technical kind.kind, but how people drift apart.apart without anyone deciding to.to, just different directions at different speeds.speeds, like two boats that started at the same dock.dock, and you don't notice the distance until you look up and the other boat is a speck on the horizon.horizon, and you realize you have no idea what someone's life looks like anymore.anymore, someone whose life you used to know by heart.heart, what made them laugh.laugh, what kept them up.up, and I saw something that reminded me of you the other day.day, nothing important.important, just a small thing.thing, and I thought about texting but didn't.didn't, maybe because it had been so long.long that any message would need to justify itself.itself, and a small thing isn't enough.enough, even though small things used to be plenty.plenty, anyway I hope you're well.well, I mean that.that, even if well just means busy.busy, even if you never reply.reply, it's just nice to say something sometimes.sometimes, even into the quiet.