Unverified50% confidenceFactExact time
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
1
Sources
50%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
GPT-5.6 Sol对决Fable 5：半价能否媲美旗舰？
bilibiliYoutube力工7/6/2026
Related Claims
Unverified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74% 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.74% similarUnverifiedIn actual Agent Tuning projects, the total cost of data preparation, hyperparameter tuning, and multiple experimental iterations is typically 5–10 times that of a single training run72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/345363API
curl https://kongchang.com/api/v1/knowledge/claims/345363MCP
get_claim(id=345363)