Nvidia is facing a copyright lawsuit. The company is said to have used proprietary work to train LLMs for the NeMo platform.

Three authors file a lawsuit against Nvidia. Reuters knows this. They claim that their works were part of a dataset of 196,640 books that Nvidia used to teach an LLM language written in its NeMo framework. Nvidia has taken the LLM offline and the authors claim that the company has thereby admitted that mistakes were made. Now you want a so-called Class action lawsuitLawsuit in the USA, which can be joined by anyone who has seen how copyrighted works were misused for Nvidia. It is not clear what exact amount the complainants are demanding.

This lawsuit is not the first against companies behind LLMs. The most notable case is probably the New York Times v. OpenAI. The major American newspaper correctly noted that ChatGPT was trained on NYT data without ever giving permission for it.

Challenge for Gen AI

Copyright protection is a sensitive issue for generative AI. In fact, almost all large models today are trained using stolen data. Journalists, authors, graphic designers and artists experience how advanced LLMs can replicate their work by training their texts and images without ever being able to give (or withdraw) consent.

There is a gradual change in the way the training of future LLMs works. Redit and Stack Overflow have already developed an API for their data, which they make available for a fee to organizations that want to train models. For example, training of AI models can continue, but data owners receive compensation. Unfortunately, much damage has already been done. The three authors want this damage to be repaired now.