The MCO Project

Today I’m sharing a project I’ve been working on called MCO (Massive Chat Training), which explores how we might get language models to deeply understand individual thought patterns through personal chat data.

The core idea is simple but powerful: take a large pretrained language model and train it on your own conversations and communication patterns. It’s about moving beyond generic, one-size-fits-all AI to something that genuinely understands your specific way of thinking and reasoning.

What makes this particularly interesting is the focus on simplification. Instead of building increasingly complex architectures to understand human thought patterns, MCO strips things down to the essential: your actual communication data. The hypothesis is that the best way to understand someone’s thinking is to learn from how they naturally express themselves.

This is very much a personal project—something I’m exploring because I’m genuinely curious about whether this approach could work. Like many side projects, it started with a simple question: why do we try to build complex systems to understand human thinking when we have such rich data in our everyday conversations?

The technical details are still evolving, but the core approach is straightforward:

• Take a strong base model
• Train it on personal chat histories and communication patterns
• Focus on learning individual reasoning chains and thought processes

I’m sharing this early because I think there’s something important here about the direction of model development. We spend a lot of time trying to make AI systems that understand humans in general, but maybe the path forward involves systems that deeply understand specific humans first.

This is still very early—there are a lot of open questions about privacy, data quality, and whether this approach will actually work. But I’m excited about the potential, and I think it’s worth exploring.

If you’re interested in this kind of work, I’d love to hear your thoughts. The best ideas often come from sharing early and getting feedback from people who think deeply about these problems.