Unverified50% confidenceFactExact time
蒸馏攻击是指通过大量调用目标模型API收集输入输出对，训练规模更小的学生模型来复制目标模型能力
1
Sources
50%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
Related Claims
Unverified蒸馏攻击通过大规模API调用收集输入-输出对，训练效率可能是常规预训练数据的数十倍74% similarUnverifiedAdversarial training improves a model's ability to recognize and resist attack prompts by incorporating known jailbreak samples into training data, but faces the ongoing challenge of novel variants.68% similarUnverified强化学习中策略网络通过最大化累积奖励来优化参数，具有信息瓶颈特性，倾向于只编码对任务目标有用的信息66% similarVerifiedTraining data poisoning can implant persistent backdoors at the model weight level, posing far greater risks than application-layer jailbreaks, but the implementation threshold is higher and typically requires control over the training pipeline.65% similarVerifiedThrough higher-quality training data, more optimized training strategies, and more refined post-training alignment, smaller-scale models can fully surpass larger-scale but lower-quality-trained models on practical tasks.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/400315API
curl https://kongchang.com/api/v1/knowledge/claims/400315MCP
get_claim(id=400315)