Unverified50% confidenceSolutionExact time
Common approaches to compressing thinking length include constructing short-thinking–correct-answer paired data for SFT, or designing length-penalty reward functions for reinforcement learning training.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
Unverified业界探索的思考力度校准方向包括基于强化学习训练模型的元认知能力、通过监督微调学习推理深度分布、以及设计动态停止机制73% similarUnverified团队认为在长序列上训练是解决拼接片段接缝问题的方案，且依赖稀疏注意力工作降低训练成本，属于未来的模型迭代72% similarUnverifiedTraining thinking-length preferences into the model weights is more stable than using system prompts at deployment time to request concise answers, as the latter is an external constraint that is unstable while the former becomes the model's default behavior.69% similarUnverifiedLearning technology should adopt a 'big picture first, details later' approach—establish an overall understanding before filling in details, and get a complete project running before diving deep into foundational steps.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961535API
curl https://kongchang.com/api/v1/knowledge/claims/961535MCP
get_claim(id=961535)