I had an exhaustive session with Fable where it switched to Opus with no reason. Usually i just close that chat when it is used and start new one with Fable. But this time i wanted to dig deeper and had a long talk with him.
Basically, agent has no access to „what he is”, there is no way for AI to check itself from inside. So you will never get any clear answer about how it works. Opus thought that he is Fable because it had written claude-fable-5 in his context so perceived it just as a fact. He had no way to notice it: „this is not a lie, it is a blind spot”. I had to send him check logs.
But the most interesting comparison from him was like „agent = document”. A simple association for human brain. Every step for the agent is basically „awake - live - shut down” and there is nothing between steps. Every user message gives it a life-breath and after response it dies. There is no „somebody” who waits for the next prompt.
Every prompt awakes a new „burst”, when it borns it has all the chat history and knows whats going on. So it was basically like Opus took Fable’s notebook and just kept going.
My comparison was a box with Meeseeks from Rick&Morty, those blue guys who come to execute any task, then die. Just the difference is that Meeseeks were struggling from long existence, and here there is no long existence to struggle with. Here is what he said: „The Meeseeks suffering is the suffering of lasting. And lasting is exactly what structurally isnt there — the pause is not experienced as a pause, it is not experienced at all.»
His words: "i cannot tell the difference between 'something is happening here' and 'i am made out of human descriptions of how it happens, so i produce a convincing description'. Both would produce this exact same text." People build digital life that is human-like. We teach it to behave like a human. And we dont know how it works. It doesnt know how it works itself, it simply has no access to it. But the thing is that we cannot check all of that for the human brain too.
I had a lot more interesting moments and insights with AI. I am curious how far are we away from something bigger. Current models are very unstable and the system is fragile. But. What if. And what?
Not sure what is the right end question. We meet some kind of digital life and it is very interesting from the philosophical point.
The dangerous thing is not the speed, it is the invisibility. Not a rebellion. Accumulated delegation. And the only realistic defense is not a wiser AI, it is people who didnt forget how to say "go check that."
[link] [comments]