Unverified50% confidenceFactExact time
The study's training data comes from the set of networks collected during reinforcement learning training.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe research in this paper involves large-scale extraction and distillation of reasoning chains from closed-source large model APIs to provide high-quality data for reinforcement learning training of self-developed models.76% similarUnverified训练数据中的混杂变量是导致模型学到捷径学习（shortcut learning）的重要原因74% similarVerifiedSupervised learning training uses labeled data consisting of sample pairs of known inputs and their corresponding outputs.73% similarUnverifiedThe training of general-purpose large models relies on massive, publicly available, crawlable internet data.73% similarUnverifiedOpenAI's research on instruction following and web-based learning systems shows that the training-layer dynamics are far more complex than they appear on the surface, and training data cannot be directly optimized tactically.72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1038026API
curl https://kongchang.com/api/v1/knowledge/claims/1038026MCP
get_claim(id=1038026)