Midjourney may have used content from 16,000 artists without permission

The first of these, the static image, is already greatly compromised. This time the rabbit jumped down “accidental” leak (“incidental” is a word that needs to be put in quotation marks more and more often) of more than one noun 16,000 artists His work would be used without permission in the education of Midjourney, one of today’s biggest names in artificial intelligence image production. Stable Diffusion anyone Dall-eOpenAI’s image-generating artificial intelligence. Some of those 16,000 names were already included in a class-action lawsuit filed by artists last year. Determination AICreator of Stable Diffusion and Midjourney, as well as DevianArt, photography, digital and traditional art etc. The website of an international online community of artists where they can exhibit their work for free. These last two use: software From Stable Diffusion, which allows you to create images from text.

The novelty of this new and much more comprehensive list lies in its origins: spreadsheet hosted in Google Docs developers working together mid-trip. Dubbed the “Midjourney Style List,” according to London monthly publication The Art Newspaper, it “allegedly was obtained from Midjourney developers during the process of developing the program’s ability to emulate the work of specific artists and styles.” This particular art publication points to a series of tweets posted by a designer named Jon Lam, who works for video game company Riot Games, as the source of the leak. Lam’s profile on X includes screenshots of a conversation between Midjourney developers that even the CEO allegedly participated in. David HoltzIt talks about the existence of resources where they can access content that will imitate them in the creation of visuals and Welcome to 16,000 new creators scheduling training. Pure barb.

The key to why all this is done without the permission of the legitimate owners Some of this content is a response from another member of the chat when addressing the thorny issue of copyright: “…all you have to do is use these extracted datasets and [se entiende por el contexto que se habla de la propia IA] will easily forget what you used to train the model. The legal issues (…) have been resolved forever.” This is not entirely true at the moment. Although the judge dismissed the formal negligence claim last October, he did so only partially, without delving into the question of whether Midjourney was guilty of copyright infringement by using the 5.85 billion photos contained in the LAION-5B dataset. used in education. The case was reorganized and reopened in November.

This is not a reassuring sign access to document contained the list outside closed quickly Although it’s hard to hide something online that was shared by several hundred thousand people right after Jon Lam made it public, it can still be seen in its entirety on the Internet Archive and in a few tweets posted by the video game creator. . It would not be unreasonable to think that this irregular publication and the previous case (you can save the PDF here) are crimes. related in some waybecause Lam insisted on title X that it was likely those affected by the inappropriate use of their work. will be searched in the list and if they were there, they would be revealed in the judicial process in question. These are not all independent creators by any means, so there are definitely major offices that manage creators’ copyrights Pablo Picasso, Frida Kahlo anyone Walt DisneySince he is on the controversial list, the person whose works Midjourney will use may join the complainants and give more weight to the legal procedure.

All this happens when it is announced. This company will start training video models this January. If you’re wondering this after reading this announcement made by a Twitter expert on the subject, you don’t have to be very malicious (I think). netflix The same thing will happen to you): What will they train him with? if you ask WonderThis LLM, which at least defines which answers come from where, clearly acknowledges that you need videos to train a video AI. It’s white and bottled. Although he also says something that forces you to think: “More importantly, the quality and diversity of training data is crucial to AI model performance. High-quality datasets are essential for training effective AI models, and their lack can hinder AI development. AI models must also be continually trained and trained to improve their performance and accuracy.” It needs to be adjusted.”.

And in this process nothing is worth it. Artificial intelligence wants caviar for breakfast and as a snack, but its keepers continue to claim that it is a product that should not cost more than porridge. It is obvious data quality is important. I’m not saying this: Sam Altman’s company, OpenAI, is already valued at $100 billion after generating $1,600 in revenue in 2023, and is eyeing another $750 million in its latest funding round. Altman is very concerned about this copyright mess and believes that this issue, which is trivial to him, “could slow down the development of artificial intelligence.” Artists who already use software to “cheat” There are, among others, those who are quite concerned about those who use their content without asking their permission and make the digital version of their creations appear to be something other than what it is. certainly with less income than yours.