OpenAI has provided previews of its newest tool, "Sora".
Japanese for "Empty sky", Sora can generate video clips instantly from
short text prompts.
Perhaps Sora tells us a little about when we dream. Dreams appear to
be a way for us to organize and consolidate associative memories.
Sora's training would have been on a rich set of imagery. At the heart of
an AI system is a neural network, a form of associate memory, with
a multidimensional landscape of peaks and valleys. We consolidate
associated memories into the same or neighboring valleys, mathematical
local minima. It's why, for example, if we watch a movie about people
escaping danger, we may dream that night of being chased.
And dreams don't always follow the rules of the physically possible.
They often break the rules of time and space. Check out the leaping
wolf puppies generated by Sora that pop in and out of their siblings
in the link below. Human dreams are like that at times.
Speaking of dangers, everyone has become acutely aware of misusing
the technology for the creation of "deep fakes" for nefarious purposes.
What's curious though is that despite many depictions by science fiction
writers of intelligent computer and robots, there were no shortage
of talking and walking and even shape-shifting running robots but
I don't recollect many ever envisaging computers being able to create
such detailed imagery. The closest genre perhaps would the the brain-linked
depictions of artificial reality as in The Matrix. But computers as such
highly skilled artists? It seems to be largely missed, but someone
with a wide science fiction reading repertoire might have a prior-art
reference.
Story and several impressive short video clips here :-
https://www.abc.net.au/news/2024-02-...unch/103475830
