Dispatch
Training on copyrighted books is still legally unsettled, and most published authors are already in the corpus
TechCrunch's August 23 explainer lays out why the intuitive answer is wrong: most published authors contributed to the training of the tools now competing with them, without knowledge or consent, and the law has not resolved whether that is infringement. The piece walks the competing fair-use readings rather than declaring a winner. For builders shipping on frontier models, the practical takeaway is that upstream corpus liability remains unpriced and unresolved as of this week.
Source
↳ Follow the thread