OpenAI Sora research engineer Gabriel Petersson predicts a surge of AI-generated fan episodes for every TV show within 12 months

making good shows is hard because you can only make bets on 1000 people in the world to direct it, because each show cost tens of millions to produce

soon millions of people will have the same opportunity because it will cost 100$

@gabriel1 sounds really lame tbh

@gabriel1 Overestimating the agency of the typical person. I’ll fade this. Majority of TV/movies will be made by people who make TVs / movies.

Kind of like Sora, the idea of “everyone making personalized everything” sounds good but most people are consumers not creators.

@gabriel1 @Indian_Bronson Why has nobody shipped longform yet? Is it too expensive or does it come out too mid?

@gabriel1 True that 💯

Though disagree on the timelines.

@just_nikhil01 @akankshaluvcats @sa_vatsa will make it mainstream in 1 month 🔥🚀

@gabriel1 We need a reboot of Silicon Valley the tv show but around AI startups now lol

it will be very hard for IP holders to control their IP, but also hard for creators to make money from fan creations

i really like what @wideworlds_ is doing. they make IP holders feel in control while people can create content and make money from it. which is inevitable

@nicdunz @boneGPT yes it will look indistinguishable to real life at some point. that will take some time though, likely we'll just have extremely advanced video editing software for a while. will be similar to photo editing, you can half ass something and it looks fine, you can also perfect it

@boneGPT @sponkostonko @Indian_Bronson oh damn yeah quality is pretty crazy

@gabriel1 @boneGPT I’ve been trying to preach when people can make their own “what ifs” at 1:1 quality it’s over

@gabriel1 @boneGPT but will it ever not be cringe ai slop at its core that you can feel in your bones as you watch it?

@boneGPT @Indian_Bronson people don't know how to do consistency perfectly (i do) + very expensive + ai video editing tools are kinda clunky so very time consuming. i think we are still one model generation from models being good enough. like look at karpathy, not good at all

@Jack_Raines nope sora model was just too bad. would have looked very differently if it was cheap and fast and perfectly consistent

i agree that current directors wont lose out cause they'll be the first to get good at this but obviously the number of directors will 100x

@gabriel1 Can’t wait for Silicone Valley reboot.

@boneGPT @gabriel1 @Indian_Bronson cost + time expense is real. iterative processes are brutal. access to compute is a steep curve. and on. you can spend a lot of money on a lot of subpar outputs. you can spend a lot of time getting to subpar. when you get to what you want, finally, there is a mountain of work.

Yeah. One way to think about this (although it is not precisely the same) is to think of ai generated video as being a collection of ai generated images, as though it were like old timey frame capture being spliced together

You probably could use old GAN generated images to create a series of frames that are then interpolated to produce “ai generated video” and it would be absurdly difficult because those old models were totally superseded by diffusion and transformer architecture. But someone could easily have looked at that old then-state-of-the-art stuff and doubted we’d ever get anywhere near where we are — and people did!

@gabriel1 people are heavily misunderstanding this

you aren’t saying all of the fans of popular shows will be making episodes (10s of millions of people)

you’re saying that 10,000 people will be making hundreds of episodes each — people who want to make tv but dont have $/opportunity

@_Abides_ @boneGPT @gabriel1 I would not be surprised if by 2030, albeit expensively (think, paying $1000 a run) people are in fact producing entire longform custom works at the enterprise tooling level and this is how studios make stuff for a massively fragmented consumer base.

@sponkostonko @gabriel1 @Indian_Bronson quality is fire tho