Bio
Max started researching alignment circa 2009, and is perhaps most famous for being the author of Crystal Society, a 2016 novel about a collective of rogue AGIs that bears an uncomfortable similarity to some recent events. He professionally transitioned from the software engineering industry to being a MIRI researcher in 2017. There, as part of MIRI's general policy, he did closed research. In 2024 he published the CAST - Corrigibility As Singular Target - research agenda. In 2025 he published the novel Red Heart, focusing on exploring ideas of AI x-risk, corrigibility, China, and digital personhood. In 2026 he helped create (and is the sole manager of) the Corrigibility Research Fund, and is more broadly involved in field-building.
Projects
Research Direction: Advancing scientific understanding of corrigibility
- Pure corrigibility is a particularly attractive alignment target for transformative AIs. (See this essay for details on why.)
- The field/world remains largely unaware of corrigibility and/or misunderstands it.
- Open work to advance our understanding of corrigibility ranges from abstract philosophy to building formal models to interp to experimental training, depending on the skills and interests of the researcher.
- Skimming this post and this post about research that I've selected to get funding will give a rough picture of the space of possible work.
What I'm looking for in a Mentee
I think the field of corrigibility is ripe for a passionate, bright person to think hard about the topic for a couple months and potentially become the most active, leading expert. There are tons of low-hanging fruit with a low barrier to entry, and people like me and Eliezer and others are dropping the ball, for a variety of reasons. I'm hoping for someone with some aptitude for philosophy/math/ML, and an ability to doggedly pursue murky questions. Experience/background/skills are less important than brains and passion, according to me.
What I'm Like as a Mentor
I'm very informal, and interested in helping others cultivate their scientific taste. I live out in the California mountains and like a mixture of async communication emails/drafts/posts and weekly video calls. I run a modest corrigibility research group on Zulip that I would hope fellows engage with as part of building broader connections.