Unverified50% confidenceFactExact time
If non-descriptive prompt changes are used instead, the matched-prompt penalty effect does not appear.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe probability of reviewers proactively raising their scores after a rebuttal is generally not high.65% similarUnverified负面指令（如'不要添加未被要求的内容'）往往比模糊的正面指令（如'请简洁回答'）更有效64% similarUnverifiedModels labeled 'obliterated' are uncensored versions with refusal mechanisms removed; in testing they more easily miss certain requirements in prompts, and uncensored does not mean better.63% similarUnverifiedUncensoring primarily affects a model's tendency to refuse, rather than its fundamental reasoning and generation capabilities, so capability loss is not significant for alignment-insensitive tasks such as coding.63% similarUnverifiedNegative Prompting, by explicitly telling the model what not to do through exclusionary constraints, is often more precise and controllable than positive style descriptions.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1046214API
curl https://kongchang.com/api/v1/knowledge/claims/1046214MCP
get_claim(id=1046214)