Elon Musk’s xAI has lost its bid for a preliminary injunction that would have temporarily blocked California from enforcing a law that requires AI firms to publicly share information about their training data.
xAI had tried to argue that California’s Assembly Bill 2013 (AB 2013) forced AI firms to disclose carefully guarded trade secrets.
The law requires AI developers whose models are accessible in the state to clearly explain which dataset sources were used to train models, when the data was collected, if the collection is ongoing, and whether the datasets include any data protected by copyrights, trademarks, or patents. Disclosures would also clarify whether companies licensed or purchased training data and whether the training data included any personal information. It would also help consumers assess how much synthetic data was used to train the model, which could serve as a measure of quality.
Don’t forget his child porn image generator, what sort of training data did he use to get that result?
If you ask an AI image generator for a bed shaped like a pineapple, it’ll give you one without having a single pineapple-shaped bed in the training data. It has beds and pineapples and it can mash the two together.
If you’ve got naked adults in the training data and you’ve got children in the training data, it’s going to be able to generate child porn.