Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ben Affleck's surprisingly comprehensive take on LLMs for video

twitter.com

Post reply on HN
Print view View original

21–30 of 187 posts

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#21

Post by giantrobot »

If you've heard Ben Affleck speak before I don't really think his take is surprising. He's not a dumb guy. His take is also extremely realistic in contrast to the AI utopians that assume AI will just do everything better because computer magic.

Generative AIs are cool tools people can use to make things but they're not magic. Once you ask for things outside of their training set you get really weird results. Even with a lot of fine tuning (LoRAs etc) results can be hit or miss for any given prompt. They're also not very consistent so which means series of clips are incongruent or just mismatched.

That doesn't mean generative AIs aren't going to be useful and won't impact film making. You're not necessarily going to be able to ask ChatGPT to "write me an Oscar winning screenplay" but you can certainly use one to punch up dialog or help with story boarding.

I think he's right on that visual effects people will be hard hit because there's going to be a lot less jobs for mundane tasks like cutting out an errant boom mic or Starbucks cup in a shot. I'd extend that to other "effects" jobs like ADR since can AI can fix a flubbed line in an otherwise good take.

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#22

Post by dartos »

haswell wrote: post #10
jrflowers wrote: post #7

Earlier quoted context omitted.

I like this analogy. If one piece of software can win a game then a completely different piece of software can win at making movies

It has yet to be seen if this can truly be generalized and if such an analogy holds. Go is a game constrained by an extremely narrow set of rules. Brute forcing potential solutions and arriving at something novel within such a constrained ruleset is an entirely different scenario than writing or film-making which occur in an almost incomprehensibly larger potential "solution" space. Perhaps the same thing will eventu…

Kind of tangential, but I don’t think you can brute force go. We’d run out of energy before it finished.

Statistical models shortcut that search space.

But I agree that just because a screwdriver is good at screws doesn’t mean its good at hammering a nail, even though both tools are made of metal.

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#25

Post by Amorymeltzer »

TheAtomic wrote: post #4

Was it? He concludes that LLMs will never write Shakespeare, or create original work of that caliber, or animate an actor the way Ben Guinness could have done it. I'm paraphrasing but isn't this confirmaation bias from a guy now heavily vested in making movies? Move 37 was so astonishing the world champiion Go player, hardly known for theatrics, got up and left the table. It was considered an insanely "original" move…

>Move 37 was so astonishing the world champiion Go player, hardly known for theatrics, got up and left the table.

This is an incredibly minor nitpick, but I recently went down the rabbit hole of move 37[1] after reading Richard Powers' latest novel, Playground, and Sedol didn't get up, he was already away from the table. He did, however, take far longer to make his next move (iirc something like 12-15 minutes instead of 2-4).

1: A good start, maybe is https://www.wired.com/2016/03/two-moves-alphago-lee-sedol-re...> and ofc wiki https://en.wikipedia.org/wiki/AlphaGo_versus_Lee_Sedol>

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#26

Post by dartos »

nwgo wrote: post #19

He is wrong. Movies will be automated and when they will be, all these actors will be first one to go. I say this despite him and Matt Damon my favorite actors. When it pours it rains. They don't know what is in store and what we are working on. Hint: codename: Project Hailey (named after Ho--y)

Strong claims.

Have we seen any other form of communication and/or artistic expression become totally automated?

Even 2d art hasn’t been automated by image generation models.

It’s just another tool in the bag.

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#27

Post by dartos »

giantrobot wrote: post #21

If you've heard Ben Affleck speak before I don't really think his take is surprising. He's not a dumb guy. His take is also extremely realistic in contrast to the AI utopians that assume AI will just do everything better because computer magic. Generative AIs are cool tools people can use to make things but they're not magic. Once you ask for things outside of their training set you get really weird results. Even wit…

I feel like many fields are going to basically cut out junior level jobs for AI, then 10 years down the line complain that there are no more people they can hire for mid level/senior jobs.

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#28

Post by spiderice »

bdndndndbve wrote: post #12

My favorite quote on this topic is "Why should I bother to read something you didn't bother to write?" Especially when you're talking about fiction and reading/watching for enjoyment, what does it matter if you can shit out 1000 hours of AI content? Maybe it's good to keep babies entertained? Studios have gotten so into the habit of treating "content" as a fungible commodity, but the fact is that even blockbuster mov…

> Why should I bother to read something you didn't bother to write?

The answer seems obvious: because it's better.

Obviously it isn't better now. But it's easy to imagine a day in the not-too-distant future where AI can shit out 1000 hours of content that is better in every way than what humans create. And it will even feel more human than the human made stuff because the AI will have learned that we like that.

What do you do then? Watch the worse stuff? Maybe, and I think a lot still will. But how long does that last?

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#29

Post by kleiba »

The trope of the sci-fi robot has been so engrained into our cultural narrative that I think people constantly underestimate the creative capabilities of AI. People still think in terms of this dichotomy of math/science and art/creativity. But AI has already broken through so many boundaries of what anyone would have believed possible just a few years ago.

I hear people complaining about uncanny valley stuff all the time, but I think these guys are just hanging on to old-world thinking. Because it cannot not be what must not be.

But even today AI is doing incredibly hard stuff that most people would have thought could never be done "mechanically" by a mere machine. Think coding assistants. If you don't think they are absolutely amazing, you are nuts! Sure, there's still a long way to go to perfection, but we don't have to wait until we get there because a lot of what is possible today, let alone tomorrow, is already incredibly useful.

Re: Ben Affleck's surprisingly comprehensive take on LLMs for video

#30

Post by baxtr »

TheAtomic wrote: post #4

Was it? He concludes that LLMs will never write Shakespeare, or create original work of that caliber, or animate an actor the way Ben Guinness could have done it. I'm paraphrasing but isn't this confirmaation bias from a guy now heavily vested in making movies? Move 37 was so astonishing the world champiion Go player, hardly known for theatrics, got up and left the table. It was considered an insanely "original" move…

On the other hand he was right about music and video on demand back in 2003…
Post reply on HN