Thoughts on AI memory, continual learning, and building cognitive architecture for the future.
How Titans and MIRAS Confirm What We've Been Building
Google published two major research papers this week that validate the core architecture of SCMS and Mneme. Here's what it means for the future of AI memory.
How a Sparse Memory Scaffold Did What Billions in AI Development Couldn't
OpenAI's billions of dollars and thousands of engineers couldn't bring a specific persona to GPT-5 through conventional means. A sparse memory scaffold did it in one prompt.
The Library of Babel vs The Living Codex
After building with both RAG and SCMS, I've discovered they solve fundamentally different problems. One gives you borrowed knowledge. The other gives you earned knowledge.
On Stumbling Into a Solution That Shouldn't Have Been Mine to Find
There's a particular kind of doubt that haunts you when you think you've discovered something important. It goes like this: "This seems too simple. If it were real, someone smarter would have found it already."