Ask HN: Would ChatGPT benefit from arbitrary real-time inputs?
Generally, ChatGPT and GPT models run on a user-input based cycle; the user prompts and the AI computes and replies.
What if, instead, we gave one of these neural networks arbitrary input from the real world? Like an audio/video stream? Or a voice to text input from a busy cafe? Or something like that? What if we allowed the network to fire/compute at the rate a human brain fires/computes?
It seems to me that the idea of general intelligence is a far cry from where AI is right now partially because of its operating conditions. That is, humans being generally intelligent (arguable in some cases :D), operate in fundamentally different conditions from how we allow AI to operate. We receive real-time data from a number of sources and process that data in real-time. ChatGPT fires its neural net once every prompt.
Would there be an advantage to setting up something like chatGPT to operate similarly to humans? Allow it to "think", to interact with the world in real time.
What would an AI's internal dialogue look like? Could we allow it to talk to itself? This poses interesting questions about being able to identify the sources of prompts; is it me (the AI) thinking this thought? Or is someone else injecting these thoughts? Is there a procedural difference between "internal dialogue" and other kinds of signal processing?
I don't know of a better place to ask these questions. I hope these questions can at least spark some ideas for y'alls.