[jmward01]

This is definitely the right problem to focus on. I think the answer is a different LLM structure that has unlimited context. The transformer with causal masks for training block got us here but they are now limiting us in many massive ways.