“Aware of its surroundings” is a pretty general phrase though. You, presumably a human, can only be as aware as far as your senses enable you to be. We (humans) tend to assume that we have complete awareness of our surroundings, but how could we possibly know? If there was something out there we weren’t aware of, well we aren’t aware of it. What we know as our “surroundings” is a construct the brain invents to parse our own “raw sensor data”. To an LLM, it “senses” strings of tokens. That’s its whole environment, it’s all that it can comprehend. From its perspective, there’s nothing else. Basically all I’m saying is that you seem to be taking awareness-of-surroundings to mean awareness-of-surroundings-like-a-human, when it’s much more broad than that. Arguably uselessly broad, granted, but the intent of the phrase is to say that an AI should observe and react flexibly.

Really all “AI” is just a handwavy term for “the next step in flexible, reactive computing”. Today that happens to look like LLMs and diffusion models.