A new research paper offers nine ways to plan work between people and AI. It names who acts, who decides, and what proof counts. It does not test whether the plan works.
WIRED reports a proposed deal involving Character.AI and Disney during bankruptcy. The deal still needs approval. The report raises a question about user data and complaints; it does not settle who is liable.
People do not always use the labels in a database. Wagner and Mädche tested a map from everyday terms to data fields in a simulation. It beat a simpler matching method on the measures they report.
A conference used AI checks, author repairs, and reviewer decisions to examine research figures and tables. Follow-up closed 20 issues. Some results still had not been fully reproduced.
MIT Technology Review covers an AI island simulation. Copies that explored and kept notes later did better than fresh copies on selected tasks. The study is not yet peer-reviewed.
A research team mapped tactics from fact-checked claims in 51 languages. In a planned experiment, its guide helped people judge unfamiliar claims. Across all claims, the average gain was not clear.
