AI
Dating the Model: Hidden Dates in System Prompts Affect LLM Evaluation
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 28/09/2026, 20:00:00 UTC
Briefing created: 01/10/2026, 10:58:31 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 01/10/2026, 07:18:24 UTC.
Briefings
Recent studies reveal that LLM evaluation results fluctuate daily due to hidden date injection in system prompts, causing performance shifts of up to 14% on math reasoning and 6% on MCQA. This effect surpasses technical variables like batch size and is independent of prompting strategies such as chain-of-thought. The study analyzed nine LLMs across six datasets between September 2025 and September 2026, highlighting a critical need for standardized evaluation protocols to ensure reproducible research.
Key points
- LLM performance varies by up to 14% on math reasoning tasks solely due to the current date injected into system prompts.
- The date effect is the dominant source of non-determinism, exceeding variations caused by batch size and numerical precision.
Context and limitations
Analysis covers nine LLMs and six datasets from September 2025 to September 2026; findings apply to all tested models and datasets.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 01/10/2026, 10:58:31 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeier2yigvx6lyrpq4diklwi3ninj5svdnx44lkfosgzqbg6ofa6dbm