Hugging Face and multiple institutions release XYBench benchmark
Understand the story
Researchers from Hugging Face, Carnegie Mellon, NVIDIA, and Johns Hopkins announced on the official blog in November 2026 the release of the XYBench benchmark, which evaluates models' ability to identify implicit false assumptions in user questions. The benchmark contains 8,115 real questions based on implicit false assumptions. According to the Hugging Face blog (first-hand source), the XYBench paper won the EMNLP 2026 Best Resource Paper Award. The event is still in the release stage, with no further developments reported.
AI overview of reporting · Updated 1 h前
Reporting timeline
Explore different aspects of the story through its reporting.
- Hugging Face BlogXYBench: Helping with the problem behind the question
Researchers from Hugging Face, Carnegie Mellon, NVIDIA, and Johns Hopkins released XYBench, containing 8,115 real-world questions based on hidden false assumptions; it won the EMNLP 2026 Best Resource Paper award.
Story heat trend
Not enough continuous observations to show a trend yet.