Millions of people are interested again in Chatgpt. The problem is that he has achieved it by violating copyright

Networks had long since They didn’t go so crazy with an artificial intelligence tool. Normally there is a certain bustle when something attracts more attention to the account, but what has happened with the generation of chatgpt images based on GPT-4O It does not make any meaning. The generative AI has achieved something that had not achieved: surprise the user on foot. And he has done so shows one of the greatest criticisms of this technology: the violation of copyright.

Content ©. In recent hours, social networks have been filled with memes, images and avatars edited by ChatgPT for look like Studio Ghibli drawings. The images are really spectacular, to César what is from Caesar, but it is not convenient to forget that an AI knows how to generate an image of a horse because, among other things, it has been trained with millions and millions of images of horses.

Click on the image to go to Tweet.

Otherwise. If an AI like chatgpt-4o is capable of converting or generating an image With the style of a specific author It is because it knows what the concrete author’s style is like. That is, ChatGPT-4O must have been trained with related content, based or generated by the study founded by Hayao Miyazaki. And what about that content? Which is beautiful, emotional and close, but not free or public domain. It is contained in copyright, an issue that has brought to ChatgPT and head OpenAi since its inception.

It is no secret. Of course not. Chatgpt was trained with a huge amount of data obtained from the Internet, websites, books, publications in social networks, academic articles, etc. Content that can be freely accessible, but not for that reason. An image that is “on the Internet” is not “on the Internet”, is housed on a server that can belong to a company and can have (and surely) copyright. That you can see and download it for free to your mobile to use the wallpaper does not mean that you can print it and sell it or illustrate the cover of your next novel with it.

Click on the image to go to Tweet.

“Live artists”. Openai claims to have opted for a “conservative approach” for the images that use the work of other artists and have “added a denial that is activated when a user tries to generate an image with the style of a living artist.” Like Miyazaki, for example. Before the flood of images generated with the style of the Japanese cartoonist, a company spokesman has told Business Insider that Openai will prevent “generations with the style of individual individual artists”, but will allow “broader studies styles.”

In other words, Hayao Miyazaki no style, Studio Ghibli style yes. Which has its ironic point, because in the year 2016after seeing a demo of an animation generated by AI, the teacher Miyazaki said “I would never want to incorporate this technology into my work. I firmly believe that it is an insult to life itself.”

My neighbor Totoro | Image: Studio Ghibli

The style. It should be noted that no one can prevent someone from doing works with the style of Miyazaki or Studio Ghibli. The style is not protected per se. Another story, and is where the quid of the matter is, is to use protected works to train an AI capable of replicating that style. That is the real problem. We could understand it as the fan art: You can make an illustration of Pikachu, print it and put it in your room, no problem. What you can’t do is sell that illustration.

OpenAi’s headache. This access and use of copyright content for commercial purposes has earned Openai some other complaintbeing the most important that of New York Times. Getty also denounced Stable Diffusion for having used their images to train models, Anthropic was denounced By a group of authors for having used their books to train Claude and a goal, apparently, downloaded 81.7 TB of books With copyright to train your models.

The conclusion is clear and we have addressed it on occasion: The price to be paid for having artificial intelligence is the looting of all the contents on the Internet, beyond that AI companies They support and hide in the Fair Use. With generative artificial intelligence it seems to have assumed that if it is on the Internet it is free, and the reality is that it is not always. All large AI companies have ignored Copyright laws And, for the moment, there is no consequences. The debate, however, is far from finishing and probably this is not the last time it is put on the table.

Cover image | @MDURBAR

In Xataka | The generative AI has a huge problem with the content without a license to train. Adobe is trying to solve it