Unverified50% confidenceBenchmarkExact time
Among the 7 models that passed the test, adding aligned images led to a 15.9% label change, adding misleading images led to a 14.7% change, while the control group that only modified the prompt had just 8.8%.
1
Sources
50%
Confidence
Long-term
Relevance
10/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAmong all 13 models, only 37% of samples with label changes shifted toward the misleading image's meaning, while 63% went in the opposite direction, both below the 50% random baseline, indicating that images merely randomly disrupted model judgments rather than guiding them.79% similarUnverifiedThe 7 models that passed the surrogate test had a change rate of 15.9%, while the 6 models that did not pass had a rate as high as 25.8%, indicating that better-performing models are less susceptible to interference but still more so than the effect of prompt modification.72% similarUnverifiedOverall accuracy remained almost unchanged: agreement with humans was 54.4% without images, 53.7% with aligned images, and 54.7% with misleading images.65% similarUnverifiedIn the configuration using ten policy snapshots, using only 8 prompts achieved an average accuracy of 24.09%, approaching the 24.51% achieved with 14,080 prompts.64% similarUnverified靶向数据工程中，针对性数据占总微调数据的 5%-15% 通常能取得良好效果，过高比例可能导致模型在其他维度退化64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/977921API
curl https://kongchang.com/api/v1/knowledge/claims/977921MCP
get_claim(id=977921)