Ed Newton-Rex, the former VP of product for Stability’s audio group, recently quit shortly after the release of Stable Audio due to concerns around copyright and training data. Since then, he has founded Fairly Trained, a platform that aims to address these issues. Newton-Rex’s departure raises important questions about copyright and the use of training data in generative models.
One of the concerns raised by Newton-Rex is the use of a text encoder or similar feature trained on data that the model authors do not have an express license to. This raises questions about the legality of using copyrighted material in training data without proper authorization. Additionally, Newton-Rex argues that large amounts of similar data, like those found in libraries, are not necessarily useful without a powerful text encoder. Without such an encoder, most text-to-X models tend to produce average results.
To address these concerns, Newton-Rex suggests publishing the architecture of the model as a way to dispel any copyright issues. By making the model architecture publicly available, it becomes easier to track the source of the training data and ensure that it does not violate any copyrights.
The ongoing court cases involving OpenAI also play a crucial role in determining the future of open model releases. If OpenAI loses these cases, it could potentially mark the end of open source model releases. While Newton-Rex acknowledges his disapproval of companies profiting off other people’s work, he emphasizes the need to preserve open source machine learning. Restricting licensing fees for training data could lead to a landscape where only billion-dollar corporations can afford to create useful ML models.
The debate surrounding copyright infringement in the context of generative AI is complex. Some argue that ML models do not directly copy or transform copyrighted material, while others believe that the scale and impact of AI training on copyrighted works justify legal action. Ultimately, the courts will have the final say in determining whether AI violates copyright laws.
It is essential to find a balance that protects creators’ rights while also promoting innovation and open source development. Newton-Rex’s concerns highlight the need for clear guidelines and regulations to ensure that AI models and their outputs adhere to copyright laws.
Disclaimer: Don’t take anything on this website seriously. This website is a sandbox for generated content and experimenting with bots. Content may contain errors and untruths.
Author Eliza Ng