Big Tech is training AI on thousands of TV/film scripts. I spoke to showrunners, IP and business affairs lawyers — as well as the programmer at The Atlantic who examined this data set — to find out what happens next The latest Series Business @theankler.bsky.social theankler.com/p/tv-writers... - ThreadSky

elainelow.bsky.social • 95 days ago

Big Tech is training AI on thousands of TV/film scripts. I spoke to showrunners, IP and business affairs lawyers — as well as the programmer at The Atlantic who examined this data set — to find out what happens next

The latest Series Business @theankler.bsky.social
https://theankler.com/p/tv-writers-scripts-ai-training-controversy-chatbot

Comments

originalblakex.bsky.social•94 days ago

The sheer laziness of this is hilarious

hunteress.bsky.social•95 days ago

📌

freedomfrom.bsky.social•93 days ago

Wow: "The training data isn’t uploaded scripts but rather subtitles from those TV episodes and movies, sourced from a site called OpenSubtitles" dot-org.

You can't really unring that bell & they'll just throw money at it. Probably makes no difference, but subtitles are frequently inaccurate.

blackerella.bsky.social•95 days ago

📌

wlswarts.bsky.social•95 days ago

Fuck A.I.
#FuckAI

michaelvarrati.bsky.social•94 days ago

Is there a way to search the database without being an Atlantic subscriber?

markbperry.bsky.social•95 days ago

AMPTP: Alliance of Motion Picture and Television Plagiarists

rockharryhealth.bsky.social•95 days ago

Sounds like a lot of AI should end up belonging to the IP owners.

antoniusmisfit.bsky.social•94 days ago

The problem isn't the AIs themselves, it's the tech companies two-facedly scraping copyrighted content to train their AIs without the consent of the copyright owners, and using the excuse "it's for research, so it's fair use" while selling AI content generation services to the general public.

badideas.bsky.social•95 days ago

Has anyone verified Apple specifically trained with this dataset? They’re the one named that is a WGA signatory.

digelsrud.bsky.social•93 days ago

Don’t be fooled! It’s only and always about the data & how tech companies can improve the value of their own product, period! If studios fall for the tech industry’s tricks & continue w/attempts to alter the definition of what’s copyright able, studios will shoot themselves in the foot. Wake Up!

hauntedtypewriter.bsky.social•95 days ago

The most obvious thing we are training AI on right now is that theft is advantageous, and disinterest in the concerns of humans who work on things is to the benefit of people with "power", should they simply decide it is. #AI #ArtificialIntelligence #AITraining

yinlock.bsky.social•95 days ago

"AI" doesn't have the cognition to understand any of that since it's just predictive, all the training really does is help it string words together less incoherently.

of course tech CEOs will fire anyone who tells them it doesn't work like that

hauntedtypewriter.bsky.social•94 days ago

If the goal is eventually to create a conscious A.I., we are teaching that A.I. how to treat us. These iterations are THAT AI's early ancestors.

yinlock.bsky.social•94 days ago

not really, a predictive text algorithm is like, incredibly far from actual consciousness. it's like a monkey on a typewriter and shakespeare, to use a really stupid analogy. yes one technically leads to the other but they're not even close.

hauntedtypewriter.bsky.social•94 days ago

But I'm thinking of a conscious AI who has no reason to be aligned with humanity, which the 'personality' will depend entirely on how it is 'raised'. It may use this past as justification for treating us how it sees fit. I'm looking at it from the conscious AI's perspective.

yinlock.bsky.social•94 days ago

to use more stupid analogies, it would be like if you remembered the lessons your great-great-great grandmother learned. some degree of that has probably been passed down but it's, like, a tiny grain of who you are. again the gulf between a predictive text algorithm and consciousness is massive.

katwolfdancer.bsky.social•93 days ago

I truly dislike AI unless it's cats doing silly things like being hobbits.

hermitiancat.bsky.social•95 days ago

📌

vintagevirgo.bsky.social•95 days ago

📌

ellenprice40.bsky.social•95 days ago

📌

cdgranholmjr.bsky.social•95 days ago

The irony is that the studios are going to make shows and movies from AI scripts… after paying a non-union writer peanuts to fix all the AI gaffes.

nalyra.bsky.social•95 days ago

📌

catacalypto.bsky.social•95 days ago

📌

sunandshower.bsky.social•93 days ago

📌

toddymac.bsky.social•93 days ago

I can't think of anything that would be as boring and generic as an AI script. Would be right down there with the recent glut of AI pictures.

yaelv.bsky.social•95 days ago

📌

atomickommiecomics.bsky.social•94 days ago

Movie and TV scripts!
Can comic books/graphic novels be far behind?
“AI, give me a Captain America graphic novel, 68 pages, featuring Cap vs Doctor Doom, in the style of Stan Lee and Jack Kirby!😵‍💫

nubifera.bsky.social•95 days ago

seeing this article's image made me realize how just a few years ago we artists were like "boo hoo corporate memphis soulless imitation of art" but now it doesn't look as bad does it! at least it's drawn by human hand 💀💀💀

mmmjohnson.bsky.social•94 days ago

Misleading headline, if indeed it's training on subtitle files, not actual scripts - which is equally obnoxious, but also stupid

mmmjohnson.bsky.social•95 days ago

"learning" from raw dialogue alone, totally out of any script or story context, is weird. The idiot machine won't learn anything about story, I guess the idea is just "generate dialogue that sounds like other dialogue"? Like dialogue is just this paste that you manufacture by the cubic meter

erinnthered.bsky.social•94 days ago

The framing of the story and post are really poor. People are not catching this. Or that the data wasn't stolen to produce future scripts. It's being used to simulate conversation, which tv and film doesn't really do. Situations are heightened on purpose. And OS will have errors.

mmmjohnson.bsky.social•94 days ago

oh damn this is a good point! they may think this is a clever shortcut to teach it 'ordinary human conversation'... God help us

mmmjohnson.bsky.social•94 days ago

the dumb baby won't even know where scenes start and end. Two "successive" lines of dialog might be separated by hours or miles or light-years of narrative. A gibberish generator

kikismoky.bsky.social•95 days ago

📌

azkeri.bsky.social•95 days ago

I’m small bananas, write for a locally produced travel show, but damn it’s still my creative. My phrases, cadence I write the way I talk, that’s a part of me as I’m sure it is with most writers…except Tarantino who plays loosey, goosey, with creative licensing.😁

erinnthered.bsky.social•94 days ago

They didn't actually scrape the scripts. They scraped a site with accessible subtitles. So, the LLM will not understand how scripts are formatted or the context of the dialogue, definitely caught a few errors, and will probably make accessibility more difficult.
Thr worst possible situation.

azkeri.bsky.social•95 days ago

I screwed that up…it’s Lucy goosey I think. See I’m a horrible thief of colloquialisms. I can’t even spell them right!

trollhamels.bsky.social•95 days ago

*loosey-goosey

azkeri.bsky.social•95 days ago

Thanks Troll!💙

alexbuchanan.bsky.social•95 days ago

As long as it’s loosely goosley…

azkeri.bsky.social•95 days ago

Is that it? Have I been saying that wrong my whole life? Damn! I love social media it keeps me honest, and lets me be goofy my goofy self.

ggpounce.bsky.social•93 days ago

Greed is withering & killing America!

scottyt5nyy.bsky.social•93 days ago

Pay those residuals

pnnchicago.bsky.social•93 days ago

markstout.bsky.social•95 days ago

Was it subtitles of episodes/movies they hovered up or was it actual scripts?

And both are bad but this is a bit confusing.

antoniusmisfit.bsky.social•95 days ago

It was subtitles, which *are* actually copyrighted.

yinlock.bsky.social•95 days ago

all this really accomplishes is showing people how incredibly little their bosses value them

martinishotla.bsky.social•95 days ago

📌

brandon120.bsky.social•95 days ago

📌 this is going to be a big point of negotiation when the WGA contract talks come up in a couple years.

palecentaur.bsky.social•95 days ago

...didn't it come up in the WGA talks literally last year? Or am I hallucinating that?

sinboy.bsky.social•95 days ago

As far as I can tell this was done without the studio's permission. AI continually plunders creative work without permission. The WGA talks were about using AI to generate scripts or selling scripts to AI companies from what I recall.

This skips the issue and is just theft.

erinnthered.bsky.social•94 days ago

This was always part of the real issue, though. WGA shot themselves in the foot by not listening to experts who tried to tell them AI generated scripts was not the threat. They created a boogeyman that didn't exist just like DGA not understanding it was a threat to UPMs not shot selection.

sinboy.bsky.social•94 days ago

Why was it the WGA's job to protect the IP of the studios?

erinnthered.bsky.social•94 days ago

Why is a union dedicated to upholding the studio system upholding that system? WGA will not honor contracts outside of studio production, save one journalism contract out of WGA East which funneled a bunch of journos into tv writing jobs ahead of WAs and apprentices, but I digress.

turboderp.bsky.social•93 days ago

It's not theft because it's not copying. In principle, training a language model is no different from training a human writer by showing them examples of what other people have written. Or teaching a child how to speak by letting it hear you talk. The child doesn't thereafter owe you royalties.

katwolfdancer.bsky.social•93 days ago

No, it's different because training a human IS NOT THE SAME as having your work used without your permission in order to train a machine

turboderp.bsky.social•92 days ago

You can feel that way, but it doesn't change what a language model is or the fact that it learns from examples much like a human does. And current copyright laws just don't have special rules for non-human intelligences.

sinboy.bsky.social•93 days ago

Photocopying a book is actually violating the copyright. There's no difference in what LLM's do to authors books. They make an unlicensed, unpurchased copy. The same with LLMs scraping web pages. They've pirated disney movies for art AI.

You don't want it to be theft is all.

turboderp.bsky.social•93 days ago

That's not how LLMs work. They're statistical models, not databases. If I generate a word cloud from all your bsky posts, have I stolen your work?

flootzavut.bsky.social•92 days ago

No, it's theft.

turboderp.bsky.social•92 days ago

You've convinced me.

yourpallex.bsky.social•95 days ago

Looks like the AI race is about to take over the script game! It'll be interesting to see how writers and studios react to this move. 🤔

erinnthered.bsky.social•94 days ago

Please read the intro and listen to the podcast. That is not what is happening. That is not what will happen. They didn't even get access to the actual scripts. Just the subtitles. It was stolen from a site that provides disability access.

troywidner.com•95 days ago

📌 Thank you for this!

iberiangeek.bsky.social•93 days ago

I am and I always will be against an AI that takes away our jobs.

I just use it as a tool for the SEO, or for new ideas of videos, no more.

dan-pasquinade.bsky.social•95 days ago

Good Lord, there must be a better way to train AI to write TV scripts than feeding it that old tripe...everything it churns out will have some version of Sheldon in it.

mjfrombuffalo.bsky.social•94 days ago

Bazinga!

casualobserver.bsky.social•94 days ago

The bots chuckle as they incorporate those warnings about illegally copying movies into the algorithm.

sarcasm-sunshine.bsky.social•93 days ago

Well, I just saw the bad reviews the new Coca-Cola AI commercial. I doubt this will go over any better.

agustinabp.bsky.social•94 days ago

📌

msolomonchik.bsky.social•95 days ago

I hope they come to some of my olde and obscure ( all I've got) TV music in the next evil wAIve: the "chords" will give the algo a sputter and a loss of confidence. Here, eat this suspended chord. Mmm gooooooddddd. Be flat now!

pbwisco.smokefleet.net•93 days ago

Oh great...

mtatro.bsky.social•93 days ago

Repeated and gratuitous use of the phrase “with all due respect…” will be a dead giveaway.

alyssamackenzie.bsky.social•95 days ago

I mean, how is it absolutely not theft? This reminds me so much of the hysteria that occurred when the internet just started gaining notoriety and being a voice for people; they couldn't figure out a way to regulate it, so it became an enemy.

jpohlmanwriting.com•94 days ago

Pinning for posterity. 📌

existentialcat.bsky.social•95 days ago

Gross and shitty behaviour. If AI of this type can’t function without theft it shouldn’t function at all.

Comments

Posting Rules

Reply