They don't understand as in OpenAI learned that ChatGPT is capable of doing things it was never intentionally trained to do after it was released. I've heard an

They don't understand as in OpenAI learned that ChatGPT is capable of doing things it was never intentionally trained to do after it was released. I've heard anecdotally that self-driving cars can pick up the unwanted behavior of inching up to a red light in anticipation of it turning green and they were never trained to do that specifically.