Unverified70% confidenceFactExact time
A common approach for generating synthetic data is to have a model rewrite, expand, or role-shift based on real user conversations, producing training samples that are fictitious in form but retain the semantic information of the original.
2
Sources
70%
Confidence
Long-term
Relevance
9/17/2026
First Seen
Sources
Related Entities
Related Claims
Unverified合成数据通常由一个强大的教师模型生成问答对、推理链或代码样本，再用于训练学生模型79% similarVerifiedSynthetic data refers to data artificially generated through algorithms rather than collected directly from the real world; it simulates the statistical characteristics of real data distributions but does not correspond to real individuals or events.78% similarPartially VerifiedSynthetic data is artificially created through computer programs, simulation engines, or generative models rather than directly collected from the real world76% similarUnverified典型的合成数据技术包括 Self-Instruct、Evol-Instruct 以及基于种子数据的扩写和改写76% similarUnverified大模型本质上是token生成器，训练目标是生成看起来自然的文本，而非精确的数据结构73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/928642API
curl https://kongchang.com/api/v1/knowledge/claims/928642MCP
get_claim(id=928642)