[humblefactory]

Also, remember that, like an LLM, his conception of the world is flawed, because somebody fed him polluted facts in his original training data set (I think I remember that right, I haven't watched that movie in decades).

[neverokay]

Mind fuck time, he doesn’t have a conception of the world. His conception is entirely defined by his immediate memory (effectively the most recent prompt). The first tattoo or photograph he sees in the morning is the first prompt every morning, and it overwhelms his context length.

Lack of … useful memory (what’s useful? We consider big memory useful right now) is what handicaps Leonard. It’s the first thing Gemini looked to solve.

Anyways. It’s his system prompt that’s in the cmos that can’t be overridden. He wakes up every morning with the same plan. People are overly concerned about sentience, but in reality it’s the dogmatic machine code that can’t be changed that makes a robot scary.

Ask Leonard, he can’t forget his system prompt.