Thought experiment: what if we've been working with LLMs in the wrong shape?

Thought experiment: what if we've been working with LLMs in the wrong shape? If you use Claude or GPT every day, you've felt the same failure. The chat starts sharp. An hour later it's forgetting things, repeating itself, losing the thread. We call it