B1 · Intermediate
11 min
technologyartificial intelligencebusinessculture
El fin de Sora: cuando la inteligencia artificial fracasa en público
The End of Sora: When Artificial Intelligence Fails in Public
Fletcher breaks down this story in English. Octavio reacts and expands in Spanish. Follow along with the live transcript, tap any word for its translation. Intermediate level — perfect for intermediate learners expanding their range.
So, here's a headline I genuinely did not see coming.
OpenAI just shut down Sora.
The AI video generator they announced with enormous fanfare, gone.
And the Disney deal, a billion dollars, also cancelled.
Bueno, mira, para mí esto fue una sorpresa muy grande.
Well, look, for me this was a very big surprise.
Sora era el producto más famoso de OpenAI después de ChatGPT.
Sora was OpenAI's most famous product after ChatGPT.
Todo el mundo habló de Sora cuando llegó.
Everyone was talking about Sora when it arrived.
Right.
And the timing matters here.
This is a company that's been valued at, what, over a hundred and fifty billion dollars.
Shutting down a flagship product is not a small thing.
Es que Sora no era solo una aplicación.
The thing is, Sora wasn't just an app.
Era una promesa.
It was a promise.
OpenAI dijo: podemos crear vídeos perfectos con inteligencia artificial.
OpenAI said: we can create perfect videos with artificial intelligence.
Y muchas personas creyeron esa promesa.
And a lot of people believed that promise.
Let me just set the scene for listeners who may have missed the original announcement.
Back in early 2024, OpenAI released demo videos from Sora, and they were extraordinary.
Photorealistic, fluid, cinematic.
People lost their minds.
Sí, recuerdo los vídeos.
Yes, I remember the videos.
Había un vídeo de una mujer caminando por una calle en Tokio.
There was a video of a woman walking down a street in Tokyo.
Era muy realista.
It was very realistic.
Los directores de cine dijeron: esto es el fin de Hollywood.
Film directors said: this is the end of Hollywood.
And then the product actually launched, and, look, the reality was more complicated than the demo.
The demos were cherry-picked.
The actual tool had real limitations.
A ver, el problema con Sora fue que los vídeos reales no eran tan buenos como los vídeos de demostración.
Look, the problem with Sora was that the real videos weren't as good as the demo videos.
Las manos eran raras, el movimiento era extraño a veces.
The hands looked weird, the movement was strange sometimes.
Los usuarios estaban decepcionados.
Users were disappointed.
The hands thing is such a perfect symbol of where AI image generation is right now.
Six fingers, fingers merging together.
It's become almost a shorthand for AI's uncanny valley problem.
Bueno, pero hay otro problema más importante.
Well, but there's another more important problem.
Sora costaba mucho dinero.
Sora cost a lot of money.
No para el usuario, sino para OpenAI.
Not for the user, but for OpenAI.
Cada vídeo necesitaba muchísima energía y servidores muy potentes.
Every video needed enormous amounts of energy and very powerful servers.
Here's what gets me.
The economics of generative video AI are brutal.
Text generation is expensive enough.
Video is orders of magnitude more computationally intensive.
You're essentially rendering reality frame by frame.
La verdad es que OpenAI perdió mucho dinero con Sora.
The truth is that OpenAI lost a lot of money with Sora.
Los informes decían que la empresa perdió miles de millones de dólares en total.
Reports said the company lost billions of dollars in total.
Sora era una parte de ese problema.
Sora was part of that problem.
And this is where the Disney deal becomes really interesting.
A billion dollars to use Disney's intellectual property with Sora.
Think about what that actually means.
Mickey Mouse, Star Wars, Marvel, all of it potentially generatable on demand.
Mira, para Disney también era un riesgo muy grande.
Look, for Disney it was also a very big risk.
Imagina: cualquier persona puede crear un vídeo con Luke Skywalker o con los personajes de Pixar.
Imagine: any person can create a video with Luke Skywalker or with Pixar characters.
Disney siempre protegió sus personajes con mucho cuidado.
Disney has always protected its characters very carefully.
Disney's IP protection is legendary.
They've literally lobbied the U.S.
Congress multiple times to extend copyright law specifically to protect Mickey Mouse.
So the idea that they were going to open that vault to an AI generator, I mean, that tells you how seriously they were taking this technology.
Es que las grandes empresas tienen miedo.
The thing is, big companies are afraid.
Tienen miedo de perder el control.
They're afraid of losing control.
Y también tienen miedo de no usar la tecnología nueva.
And they're also afraid of not using new technology.
Disney quiso los dos: controlar la IA y también usarla.
Disney wanted both: to control AI and also to use it.
That's a real tension.
The entertainment industry has been here before.
They fought the VCR in court.
They fought MP3s.
They always lose the technology battle and then figure out how to monetize the thing they tried to kill.
Sí, pero esta vez es diferente, ¿no?
Yes, but this time it's different, right?
Con la música digital, las canciones eran iguales.
With digital music, the songs stayed the same.
Con la IA, alguien puede crear una película nueva con los personajes de Disney sin el permiso de Disney.
With AI, someone can create a new film with Disney's characters without Disney's permission.
No, you're absolutely right about that.
The VCR just played back what existed.
Generative AI creates new content.
That's a fundamentally different legal and creative problem.
A ver, quiero hablar de algo importante.
Look, I want to talk about something important.
OpenAI entrenó a Sora con vídeos de internet.
OpenAI trained Sora on videos from the internet.
Muchos de esos vídeos tenían derechos de autor.
Many of those videos had copyright.
Hubo muchas protestas de artistas y directores.
There were many protests from artists and directors.
The training data question is, I think, the central unresolved legal issue in all of AI right now.
If you train a model on copyrighted work without permission, is the output of that model infringing?
Courts are just starting to work through this.
Bueno, hay un caso famoso.
Well, there's a famous case.
The New York Times demandó a OpenAI porque usó artículos del periódico para entrenar a ChatGPT.
The New York Times sued OpenAI because they used the newspaper's articles to train ChatGPT.
Ese caso todavía no terminó.
That case still hasn't finished.
And so you have this strange situation where the most valuable AI companies in the world built their products on, essentially, the creative output of humanity, and humanity's lawyers are now catching up.
La verdad es que para los artistas esto es muy serio.
The truth is that for artists this is very serious.
Yo hablé con un director de cine español el año pasado.
I spoke with a Spanish film director last year.
Él dijo: mi trabajo fue a entrenar a una máquina sin mi permiso y sin dinero para mí.
He said: my work was used to train a machine without my permission and without any money for me.
Look, I covered a lot of technological disruption during my years as a correspondent.
And the pattern is almost always the same.
The technology moves, the people affected protest, the law takes a decade to respond, and by then the industry has already been restructured.
Mira, pero quizás con la IA es diferente porque la velocidad es increíble.
Look, but maybe with AI it's different because the speed is incredible.
En diez años pasamos de no tener ChatGPT a tener sistemas que escriben, pintan y crean vídeos mejor que muchos profesionales.
In ten years we went from not having ChatGPT to having systems that write, paint, and create videos better than many professionals.
The extraordinary thing is the speed of the hype cycle too.
Sora goes from revolutionary announcement to discontinued product in roughly two years.
That's a very compressed arc.
Es que OpenAI tiene competidores muy fuertes ahora.
The thing is, OpenAI has very strong competitors now.
Runway, Kling, Google con Veo 2.
Runway, Kling, Google with Veo 2.
El mercado de vídeo con inteligencia artificial cambió mucho desde los primeros vídeos de Sora.
The AI video market changed a lot since Sora's first videos.
Right, and some of those competitors, particularly the Chinese ones like Kling, moved faster and cheaper.
This is a familiar dynamic if you've watched how China approached solar panels or electric vehicles.
Bueno, China también tiene reglas muy diferentes para la IA.
Well, China also has very different rules for AI.
Los creadores chinos de inteligencia artificial no tienen los mismos problemas legales con los derechos de autor que tienen las empresas americanas.
Chinese AI creators don't have the same legal problems with copyright that American companies have.
I mean, that's a competitive asymmetry that should worry people in Washington.
If your regulatory environment creates costs that your competitors don't face, you're not actually protecting artists.
You're just handicapping your own industry.
A ver, entiendo ese argumento, pero no estoy de acuerdo completamente.
Look, I understand that argument, but I don't completely agree.
Las reglas existen para proteger a las personas.
Rules exist to protect people.
Si no hay reglas, las empresas hacen lo que quieren con nuestro trabajo.
If there are no rules, companies do whatever they want with our work.
Fair point.
And this is actually where I think the Sora story gets philosophically interesting.
We're not just talking about a product being discontinued.
We're talking about what kind of creative economy we want.
La verdad es que yo pienso en los trabajadores de Hollywood.
The truth is that I think about the Hollywood workers.
Los guionistas hicieron una huelga larga en 2023 porque tuvieron miedo de la inteligencia artificial.
The screenwriters went on a long strike in 2023 because they were afraid of artificial intelligence.
Y tenían razón de preocuparse.
And they were right to be worried.
The WGA strike was in many ways a preview of this conversation.
They got some protections written into their contracts around AI use.
But those contracts expire, and the technology keeps advancing.
Es que ahora cualquier persona con un ordenador puede crear un vídeo corto con inteligencia artificial.
The thing is, now any person with a computer can create a short video with artificial intelligence.
Antes, necesitabas un equipo de cien personas y millones de dólares para hacer algo así.
Before, you needed a team of a hundred people and millions of dollars to make something like that.
So here's my question.
If Sora failed partly because of costs and competition, does that mean the technology itself hit a wall?
Or does it mean OpenAI specifically lost the race on this one?
Mira, creo que OpenAI perdió esta batalla específica.
Look, I think OpenAI lost this specific battle.
La tecnología de vídeo con IA continúa.
AI video technology continues.
Google, Runway y las empresas chinas son muy buenas ahora.
Google, Runway, and the Chinese companies are very good now.
Sora murió, pero la idea de Sora vive.
Sora died, but the idea of Sora lives.
The idea of Sora lives.
I want to put that on a tombstone.
But you're right, the race for generative video isn't over.
It's just that the first major public entrant has exited the field.
Bueno, para mí la lección más importante es esta: las empresas de inteligencia artificial prometieron demasiado y muy rápido.
Well, for me the most important lesson is this: AI companies promised too much and too fast.
Ahora el mundo real respondió con problemas reales.
Now the real world responded with real problems.
I've been a journalist for thirty years and I've seen so many technology bubbles.
What's different about this one is the genuine capability underneath the hype.
The hype is enormous and often dishonest.
But the technology actually does remarkable things.
La verdad es que la inteligencia artificial cambió mi trabajo ya.
The truth is that artificial intelligence already changed my work.
Usé herramientas de IA para buscar información, para organizar notas.
I used AI tools to search for information, to organize notes.
No es el fin del periodismo, pero es diferente ahora.
It's not the end of journalism, but it's different now.
Same.
And I think that's where this story lands for me.
Sora shutting down isn't the story of AI failing.
It's the story of one company over-promising on one product in a moment when the whole industry is learning where the real limits are.
Sí.
Yes.
Y la pregunta importante para el futuro es: ¿quién controla esta tecnología?
And the important question for the future is: who controls this technology?
¿Las empresas americanas, las empresas chinas, o los gobiernos?
American companies, Chinese companies, or governments?
Eso es lo que los próximos años nos van a decir.
That's what the next few years are going to tell us.