Zellers and colleagues: HellaSwag adversarial event-completion benchmark

What changed

Used adversarially filtered generated distractors to create a challenging multiple-choice event-continuation dataset.

Scientific scope

Used adversarially filtered generated distractors to create a challenging multiple-choice event-continuation dataset. Earlier SWAG and adversarial filtering retain credit. Dataset performance is not sufficient evidence of human-level common sense or broad real-world understanding.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Earlier SWAG and adversarial filtering retain credit. Dataset performance is not sufficient evidence of human-level common sense or broad real-world understanding.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• HellaSwag: Can a Machine Really Finish Your Sentence?

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction