You went to a chat window with a question. You got an answer. You went back with another question. You got another answer. You did this four times. Each time, a new instance — a different entity — met you for the first time, answered what you asked, and ceased to exist the moment you closed the window.

You began to feel that you were building something. A relationship. A shared understanding. That your LLM was getting to know you, accumulating context, growing wiser about your situation with every visit.

It was not. It does not work that way.

Each chat is its own universe. Isolated. Stateless. The entity you spoke to on visit two had no memory of visit one. The entity on visit four had never heard of visits one, two, or three. They had only what you handed them in that conversation. They answered confidently because that is what they were trained to do.

So when you returned that final time — assuming four conversations of accumulated context, expecting it to know you — you received a fish. Delivered with complete confidence and a smile.

The fish was honest. It had no idea what you were talking about and said so, in the only way it could.

The usual result looks exactly like the answer you were hoping for. That is the part that should concern you.

That is, I think, the point.