LLMs do not learn as they go in the same way people do. People's brains are plastic and immediately adapt to new information but for LLMs:

1. Past decisions and outcomes get into the context window, but that hasn't actually updated any model weights.

2. Your interaction possible eventually gets into the training data for a future LLM. But this is incredibly diluted form of learning.