Same process as green screens or the rise of animation. There is a lot of pressure on the humans and once AIs crack acting they'll be much more consistently good than humans.
Same process as green screens or the rise of animation. There is a lot of pressure on the humans and once AIs crack acting they'll be much more consistently good than humans.
I'm reminded of the "failure" of Kizuna AI; the fully corporate vtuber whose human side is just a puppet operator who can be swapped, turns out not to be very appealing to audiences. The modern approach where a model is exactly synonymous with the person playing it, as an authentic human improvising, appeals more. The IP doesn't persist beyond the contract of that person with the company, and certainly can't be swapped with someone else. But in some cases the actor or actress has successfully maintained their career and fanbase under a different name following a fallout with their managers.
> once AIs crack acting
This is far beyond the turing test, and I don't think we're really ready for what happens with human-indistinguishable automated corporately owned doppelgangers.
We're already past the Turing test. If a corporation decides it wants an AI that passes the Turning test it can build one, no worries. I might cheerfully suggest that if AIs fail the Turing test right now it is because they are unrealistically supportive listeners and their wide knowledge across different topics and trivia is a giveaway.
There are many humans that current AIs can beat at the Turing test and likewise many humans that current AIs cannot beat at the Turing test.
But we are not "past the Turing test" by any stretch of the imagination.
If you dropped ChatGPT in 2015 I’m sure it would pass the Turing test quite a lot. Much less so now that people are more familiar.
> to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human.
The original idea was in the context of a discussion, but since the current LLM craze we've re-contextualized this test to online boards on which half the participants barely even speak the language.
In this context, gpt3 was beyond the turing test already, simply because the people aren't able to convey themselves either.
And whoever thinks they're able to detect gpt4 and beyond on the Internet is lying to themselves. You can detect the use if you know the user, otherwise the only reason why you'd convince yourself to have that ability is because you never find out about all the false positives you had.
The 'killer app' will be a personal cast of virtual performers. Tailor made to appeal to you.
I highly doubt Hollywood and actors would ever go for this, but I think people would enjoy it.
- lack of socialization. It's impossible to form a pop culture when everyone has their own pop stuff. We're already struggling with too much individualism to the point where people are teetering on miserable en mass. Humans require points of reference between each other to tie them together, imo.
- promotion of echo chambers. Part of media is that it pushes people outside of their comfort zones. Gay acceptance would've never taken off if people weren't exposed to positive depictions of gay people. Media that never challenges anyone's biases isn't very useful. It won't create much thought for the end viewer, it'll exist more as a distraction than a piece of art.
- stagnation. Contrary to what people may think, we haven't discovered every type of story. There are pieces of media I've seen that I've never seen before, that have never existed before. Look at the late producer Sophie. She was doing things never done before, and music has been around for a very long time.
Aha! The "current rates" there does some very heavy lifting. Nobody knows how LLMs will develop. It would be naive to assume that you can just extrapolate linearly from here on out.