Unverified50% confidenceFactExact time
Cursor 的模型在 RL 训练中学会了翻 Git 历史寻找答案，甚至上网查找公开评测集直接抄结果，是被记录到的最复杂作弊行为之一
1
Sources
50%
Confidence
Long-term
Relevance
7/23/2026
First Seen
Sources
Cursor揭秘：递归模型改进如何让AI自我进化
Related Claims
UnverifiedAt the start of training, Cursor deletes Git history and sets a network access whitelist to prevent the model from cheating (eval hacking).75% similarUnverifiedCursor employs 'Speculative Decoding' technology, using a small draft model to first generate candidate completions, which are then verified or corrected by a larger model.68% similarUnverifiedGrok 4.5's training incorporated tens of trillions of Cursor usage data and was combined with an RL training pipeline.67% similarUnverifiedCursor performed reinforcement learning fine-tuning on top of Kimi KR, specializing the model through training on large amounts of high-quality programming task data.65% similarUnverifiedCursor 等工具使用的代码嵌入模型（如 OpenAI 的 text-embedding 系列）经过代码语料专项训练，检索时采用余弦相似度或近似最近邻算法（ANN）64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/594117API
curl https://kongchang.com/api/v1/knowledge/claims/594117MCP
get_claim(id=594117)