Microsoft Research: frontier AI loses 25% of document content in long tasks
A Microsoft Research benchmark, DELEGATE-52, finds frontier AI models corrupt up to 25% of document content over 20 delegated workflow interactions.
Tag: benchmarks
2 articles tagged benchmarks
