What I'm building
AI alignment talent may soon be the biggest bottleneck to AGI and to getting AI right. I'm building Polymath to evaluate that talent and route it, working with MATS and Constellation.
Try it yourself, and read the exact rubric for how it works and what it tests for.
It started as an attempt to fix education. I believe education is not a learning problem but a measurement problem: we cannot accurately measure how effective people are in different roles, or their growth potential, so we default to proxies like university degrees. Measure people the right way, tell them exactly how to get better, get them opportunities based on that, and maybe we can fix education.
Before this, Polymath was an npm tool that measured people and got them opportunities from what their Claude, ChatGPT and Claude Code knew about them. It failed because the friction and privacy concerns were too strong to fix the cold start problem. You can see the sample reports and the launch post.
I wrote about the journey in The AI that fights for your place in the world.