Unverified50% confidenceFactExact time
Baseline bias refers to the default tilt a model naturally exhibits toward a certain value dimension when there is no steering at all, originating from the distribution of pretraining corpora, the selection tendencies of instruction fine-tuning data, and the group consensus of human annotators during preference alignment.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedBenchmark Selection Bias refers to the tendency of vendors, when choosing evaluation items themselves, to preferentially include task types for which their own models have been specifically optimized, thereby creating a favorable tilt in the overall score.77% similarUnverifiedA common limitation shared by mainstream bias benchmarks is that they measure a model's explicit choices in specific controlled contexts, rather than implicit biases that emerge in open-ended generation, long-text reasoning, or complex role-playing scenarios.75% similarUnverifiedJudge models exhibit systematic biases: favoring output styles similar to their own training distribution, giving higher scores to longer responses (verbosity bias), and being sensitive to the position of answers in the prompt (position bias).71% similarUnverifiedThis identification bias is statistically known as Selection Bias, where the sample is determined by observable characteristics rather than being randomly drawn from the overall population.70% similarUnverifiedConfidence-based routing is also referred to in academia and industry as 'selective prediction' or 'classification with a reject option,' with the core being granting the model the right not to answer.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038548API
curl https://kongchang.com/api/v1/knowledge/claims/1038548MCP
get_claim(id=1038548)