Important if true, but waiting for replication:
"Researchers have found that the latest large language models (LLMs) rarely, if ever, attribute information to sources that don’t exist…The analysis…has not been peer reviewed."
https://www.nature.com/articles/d41586-026-03753-1