Verified65% confidenceFactExact time
Modern 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.
3
Sources
65%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
GPT-5.5 Codex Reasoning-Token Clustering Explained: Causes of Performance Regression and How to Respond
hackernewshackernews7/4/2026
Related Claims
UnverifiedRLHF的人类介入发生在训练阶段并固化为模型权重，属于隐式编码，而表征工程干预在推理时操控内部表征，具有即时性和可逆性83% similarUnverifiedIn LLM post-training, the model directly spews intermediate reasoning steps that should have been hidden to the user, or during RL reward computation, the reward model that should only evaluate the final answer sees the reasoning process, thereby distorting the optimization direction.79% similarVerifiedEach RL training step requires running inference (generating rollouts) and backpropagation (updating weights) simultaneously, making the compute requirements far higher than ordinary fine-tuning.78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/112827API
curl https://kongchang.com/api/v1/knowledge/claims/112827MCP
get_claim(id=112827)