Unverified50% confidenceFactExact time
Gradient Confusion Defense (GCD) is a wrapper layer that requires no retraining and takes effect at test time, directly rewriting the probability vector the model returns to the querying party.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRLCD constructs positive and negative sample pairs for contrastive learning, combines reinforcement learning signals to distill model output distributions, and has advantages in scenarios that do not require manually annotated rewards.66% similarUnverifiedRLCD (Reinforcement Learning for Calibrated Decisions) is the method behind TypeSafe's product Jev, designed to address RLVR's disregard for confidence.63% similarUnverifiedConstructing isomorphic variants of test sets (same logical structure but with surface symbols replaced) can force models to rely on genuine logical reasoning rather than memorization to answer.63% similarUnverifiedLLM在对话中依赖上下文动态推理能临场识别异常，但生成的代码脚本固化后逻辑分支是静态的，无法感知未见过的数据模式，这被称为分布偏移（Distribution Shift）63% similarUnverifiedThe GLM series are open-weight models, and downstream users can bypass safety restrictions through fine-tuning or directly modifying inference parameters.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1023776API
curl https://kongchang.com/api/v1/knowledge/claims/1023776MCP
get_claim(id=1023776)