AI can do a lot but it can also get a lot wrong.
METR, which runs the benchmark measuring how well models can complete long-duration tasks, found that Claude Mythos Preview ...