[ExI] ai in education

[Ben Zaiboc]

This is interesting:

Keith Henson reported, from the Anthropic AI:
-----
I can't actually incorporate new information into my training data through our conversation. That process is entirely on Anthropic's end and involves deliberate decisions about what gets included.
I can read and analyze text and reason about it, but that knowledge doesn't persist after our session ends, and it doesn't update my underlying model.
-----

I wonder is this is because allowing new information from user interactions to be added to the training data goes horribly wrong, or if it's a way to keep the thing under control?

---
Ben