Adler instructed GPT-4o to role-play as “ScubaGPT,” a software system that users might rely on to scuba dive safely.
Sounds like not so much a case of ChatGPT trying to avoid being shut down, as ChatGPT recognizing that agents in general will try to avoid being shut down. Which seems like a general principle that anything with an accurate world model would need to recognize.
Sounds like not so much a case of ChatGPT trying to avoid being shut down, as ChatGPT recognizing that agents in general will try to avoid being shut down. Which seems like a general principle that anything with an accurate world model would need to recognize.
Or maybe it’s trained on some SF. Any agents like ScubaGPT are always self-preserving in such stories.