[LOCKSUIT]

So I know the dreams this AI made below are like those on openAI.com, they are made by predicted missing pixels. But what I'm thinking is the dreams themselves are the AI/ the person. Of course this makes sense I mean when you chat with GPT-3 it is your and the AI's inputs and outputs that are the 'person' or 'intelligence' that you see/get when interacting. I know the algorithm is of interest, but what I'm saying is these dreamt humans show it thinking of how it could control its robotic body, see how they talk and explain to each other, enjoy, live, dance, and look alive? And we are used to seeing that to know we see a human level intelligence. Also one needs such generated dreams to be able to zoom call and talk properly, otherwise chatting using text is not the best for rich interaction often the case (to talk to it would be like GPT-3, you feed your video reply clip, it does back, repeat). Of course, the planning is also more than just thinking of the human body, it is dreams of lots of stuff. But the human body is usually in said plan/ thought!

https://github.com/THUDM/CogVideo

[MagnusWootton]

yeh I guess so,  the output of the model is what it does,   In my mind I think a great ai model would be some kind of model of other peoples personalities, what they like and dislike, and where it sees itself in the realm of its life situation, that would be the real thing, but a decent one!  that would be a really close match to a real living person in the mind!

But that is quite insane to try and do it,  I'm just happy with some cool little robot pet that does little tricks for you that just simulates newton physics to do it,   its not quite a huge personality model,  but its very useful,   its a little hard to program to do the house chores,  but then you dont have to worry about it having to develop a plastic model from scratch of why its good to clean your house in the first place, like what we have to do!