roenxi
a year ago
Cool but probably not that interesting to the development of AI in Hollywood over the longer term. As the tech improves, at current rates, I expect we'll see something like VTubers on a mass scale. Companies creating their own virtual people - where they control the IP - and putting all their efforts behind promoting them instead of humans. It'll be cheaper and easier in the long run.
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.
pjc50
a year ago
> I expect we'll see something like VTubers on a mass scale. Companies creating their own virtual people - where they control the IP - and putting all their efforts behind promoting them instead of 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.
roenxi
a year ago
> 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.
mpalmer
a year ago
The Turing test has a variable in the form of the human that is playing.
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.
lukev
a year ago
Yes. And another variable is what humans believe state of the art AI to be capable of.
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.
user
a year ago
scarface_74
a year ago
Even if you did it now and it would stay in the “voice” that you gave it, with careful prompt engineering you could fool most people
ffsm8
a year ago
The turing test is
> 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.
noobermin
a year ago
That isn't what happened to kizuna ai, she just wasn't as popular because she didn't stream like the others.
Eddy_Viscosity2
a year ago
Hollywood already had this with characters they own: James bond, mickey mouse, etc. The people who scrpt/act/draw/voice them are all swappable. An Ai actor isn't much different than a cartoon character but for live action.
erie
a year ago
They are already here with outfits such as Synthesia and heygen.
JumpCrisscross
a year ago
> Companies creating their own virtual people - where they control the IP - and putting all their efforts behind promoting them instead of humans
The 'killer app' will be a personal cast of virtual performers. Tailor made to appeal to you.
gedy
a year ago
I really want this - if you want a movie with a diverse cast, great. All black, no problem. K-pop star wars, etc.
I highly doubt Hollywood and actors would ever go for this, but I think people would enjoy it.
consteval
a year ago
The consequences of such personalized entertainment will be great:
- 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.
JohnFen
a year ago
I would hate that so very much for a number of reasons, starting with that it will increase overall dehumanization of everything.
DrSiemer
a year ago
Not a great movie, but one part of "the Congress" was interesting: if a famous actor or actress gets older, they could choose to sell the rights to their likeness to an AI company.
shafyy
a year ago
> As the tech improves, at current rates
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.
roenxi
a year ago
I'm extrapolating exponentially.
votepaunchy
a year ago
What is either of yours dependent variable?
roenxi
a year ago
$/FLOP, FLOP/Watt or something like that. Eg, https://epochai.org/blog/trends-in-gpu-price-performance
shafyy
a year ago
Even worse