Learning P-Bits

About

I'm Dan Barrett, an independent machine learning researcher.

In 2021, during my first paternity leave, I spent my daughter's nap times learning GPT-2. That decision changed my career trajectory. What started as curiosity about a nascent technology turned into real opportunities—new projects, new connections, new ways of thinking about AI.

I documented that journey in Artificially Interesting, where I explored what happens when you let machines co-author your writing. The experiment taught me something important: the best way to understand emerging technology is to use it, not just read about it.

Now I'm doing it again with probabilistic computing and the thrml library. P-bits represent a fundamentally different approach to computation—one that could matter as much as transformers did. Whether that's true or not, I'll find out by building with it.

This blog documents that process. I'm not an expert. I'm figuring it out as I go. But that's the point—watching someone learn in public means you see the mistakes, the dead ends, and the moments when things finally click.

If you're curious about probabilistic computing, or just interested in how people learn hard things, follow along.