You can do this too. There's no reason an inner monologue can't serve as continuous input forever. That's a project that does exactly this. Letting cGPT think and continuosly driving other thoughts and actions and to also allowing other user input
Correct, and it's sad as to how few people, especially within the ML research community, ever considered how blatently easy it is to embody agents.
Literally go open ChatGPT now and ask it give you a time between X and Y where it will respond if not prompted again. Ask chatGPT to Write some kind of code that parses the tidbit at the end indicating when it will next respond, and have it blank prompt it if it is not responded to within the time limit. Boom, embodied agent with a few prompts and a tiny bit of extra code.
It's a result of the general intelligence of LLMs. Not a lot to be done about that.
But people are starting to see that. If you read the recent Microsoft paper, they essentially say, " wow it'd be cool to see what Gpt-4 can do with agency and motivation. We leave that for later research." Lmao
Literally go open ChatGPT now and ask it give you a time between X and Y where it will respond if not prompted again. Ask chatGPT to Write some kind of code that parses the tidbit at the end indicating when it will next respond, and have it blank prompt it if it is not responded to within the time limit. Boom, embodied agent with a few prompts and a tiny bit of extra code.