There's another phenomenon. It's called pedantic denialism. Deriving conclusions from more data is the same thing as learning. You learned something from the new data hence the new conclusion. As long as that context window survives the LLM has learned.