BWhat changed
Built a temporally grounded question benchmark and compared6B models pretrained on chronological versus shuffled versions of CommonCrawl data at a fixed token budget.
Scientific scope
Built a temporally grounded question benchmark and compared6B models pretrained on chronological versus shuffled versions of CommonCrawl data at a fixed token budget. Chronological pretraining improved temporal freshness in the studied setting, not guaranteed continual learning, arbitrary knowledge updating or universal superiority. Released versus planned checkpoints must be checked separately at adoption.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Chronological pretraining improved temporal freshness in the studied setting, not guaranteed continual learning, arbitrary knowledge updating or universal superiority. Released versus planned checkpoints must be checked separately at adoption.
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.
- Understanding Data Temporality Impact on Large Language Models Pre-training
- Kyutai scientific publications index
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