Moritz Pfau
Hi, I'm Moritz. I work on the Alignment Team at Arcadia Impact in London. Before that I did product management at Ramboll and a physics degree in Heidelberg.
InterestsAI safety · Effective giving · Being outdoors, bikepacking · Podcast addict · History of the last ~two centuries
Writing
I write Minimal Mind, where I write about whatever's on my mind. Mostly everything except AI safety, though lately that separation has been drifting a bit.
Projects
I built this in late 2025 as an AI tutor for working through academic papers. I received a Google Cloud startup compute grant for it, so it's entirely free to use :)
After listening to far too many episodes of a great Bach podcast, I made a little project to visualise the whole catalogue of Bach's works. Blog post on this coming (hopefully) soon.
Most of the books I've read over the last few years, with many of the quotes I noted down because they meant something to me.
A few things I believe in
Thought through to varying degrees, and mostly here as an invitation: if you'd like to discuss any of them, reach out.
Life
- I'm a hedonist: what ultimately matters is the quality of conscious experience, mine and everyone else's.
- I follow Peter Singer's argument. From any consistent moral framework I can support, I end up in the same place: given the coincidence of being born in a rich country, I should be helping others significantly.
- Giving works. I've pledged 10% of my income to effective charities, and it's one of the best decisions I've made.
- I'm trying to be attackable. With mixed results.
- Mountains make me unreasonably happy.
AI
- AI is the most consequential technology of our time, and it going well for humanity is not a given.
- I think there's a significant chance of catastrophic outcomes from transformative AI in the coming decade. The scenarios that seem plausible to me:
- Scenario 1: AI capabilities fall off the trend of the last eight years. Progress stalls, the AI investment bubble pops, and we get an economic crisis; afterwards the economy returns to something like the trend of the past decades.
- Scenario 2: Capabilities keep improving and we land in a world of recursive self-improvement, leading to superhuman intelligence (ASI). Either (a) we have solved alignment by then and ASI is aligned with humanity's goals, or (b) we can't control superhuman systems, with catastrophic outcomes.
- Scenario 3: Capabilities would keep improving, but we decide to regulate the development of ASI significantly and slow down globally. That buys much more time to sort out alignment.
- I'm highly uncertain about capability increases, but I find it reckless to bet on living in a scenario-1 world. So I believe that solving alignment and/or regulating the development of ASI is one of the most urgent challenges of our time.
Feedback
I believe candid feedback is one of the fastest ways to improve. If you have feedback for me, about anything, leave it here anonymously. I'd love to read it.