this makes me wonder about one of the big differences between these current AI ‘it just predicts the next word / denoises noise to make a picture’ systems, and humans.
Maybe our brains do some of these techniques to generate language/images in our heads. However, ChatGPT and Stable Diffusion (picking on them) just sit on their butts until someone makes them do an action.
Human brains, on the other hand, are continual thought factories. You have an enormous number of thoughts a day - they are happening constantly, and you have little control over them beyond how you might react to them. You also have the ability to observe and react to your thoughts intentionally, which is probably a part of what we consider human consciousness. (I’m really into this topic lately as I learn about mindfulness, acceptance, the default mode network in the human brain, etc.) TLDR, these content generation AI things have no ‘default mode network’ and their intent is only provided externally by humans providing triggering input.
What would it take to add that ‘functionality’ to an LLM? What would happen if we did?
