Unverified50% confidenceFactExact time
VCR-Bench's current evaluation is based only on a subset of Kinetics-400, and the datasets and scenarios covered remain limited.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/8/2026
First Seen
Valid until: 1/6/2027
Sources
Related Entities
Related Claims
UnverifiedExisting model evaluation frameworks (such as HumanEval, MBPP, and various Arena benchmarks) primarily measure capability ceilings and rarely incorporate end-to-end task completion rate as an evaluation dimension.59% similarUnverifiedIn vertical domains such as industrial defect detection and pathology slide analysis, the zero-shot detection accuracy of large models such as GPT-4V and Gemini lags behind specialized models by 20–40 percentage points on the mAP metric.58% similarUnverified即使只有5-10%的训练样本存在错位，也会导致VLM模型在下游任务上出现显著的性能退化58% similarUnverifiedModels with smaller file sizes and lower parameter counts generally have more limited performance.57% similarUnverifiedCompared to MoE which adds new expert parameters, LRCC reuses existing low-rank factors, resulting in smaller additional parameter overhead.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/986902API
curl https://kongchang.com/api/v1/knowledge/claims/986902MCP
get_claim(id=986902)