Verified75% confidenceFactExact time
Each RL training step requires running inference (generating rollouts) and backpropagation (updating weights) simultaneously, making the compute requirements far higher than ordinary fine-tuning.
3
Sources
75%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedRL training redistributes probability mass from ordinary outputs to high-quality outputs through methods such as policy gradients, and does not create entirely new capabilities.79% similarUnverifiedModern reasoning models generally use RLHF or its variant RLAIF for alignment training, where the model learns during training when to stop reasoning and when to output an answer.78% similarUnverified全参数RL训练的大范围参数更新提供了隐式正则化效果，可能是鲁棒性的来源77% similarUnverifiedMainstream code large models all adopt a two-stage strategy of "pretraining + instruction fine-tuning," with the second stage using alignment techniques such as RLHF or DPO.76% similarUnverified提示词与模型版本存在深度绑定关系，不同版本模型在RLHF训练数据、指令微调策略上的差异会导致相同提示词产生不同输出，这种现象被称为「提示词漂移」76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/930927API
curl https://kongchang.com/api/v1/knowledge/claims/930927MCP
get_claim(id=930927)