This looks like a start of a cascade. Capture data (memory) - too much data confuses context - selective memory based on situation - selection is a chore for hu

This looks like a start of a cascade. Capture data (memory) - too much data confuses context - selective memory based on situation - selection is a chore for humans - automate it with a “pre prompt” - that will select relevant memories for the conversation Now we have conversations that are 2 layers deep. Maybe there are going to be better solutions, but this feels like the solid step up from LLM as tools onto LLM as services.