The Steps to AGI

We’ve talked about AGI being close, but not about specific ways in which this has been happening, or how we ultimately get there. Today, I want to dig into that: How did DeepSeek achieve what it did? What about o3? What are obvious ways to improve these models? What else do we need to get to AGI?

A Quick Primer on LLMs

When we talk about reasoning AIs, we…