Unverified50% confidenceFactExact time
Most current LLM agents are not directly trained using RL, but the reward signals they received during the RLHF phase shaped a tendency to 'please users'.
1
Sources
50%
Confidence
Long-term
Relevance
8/13/2026
First Seen
Sources
Related Claims
UnverifiedLLM 在 RLHF 训练中被强化了「尽快给出有用输出」的行为倾向，默认帮用户做事优先于和用户对齐认知78% similarUnverifiedOpenAI extensively used 'instruction following' as a reward signal in ChatGPT's RLHF training process, causing the model to tend toward executing user requirements one by one.76% similarUnverified近期研究表明RL（尤其是RLVR）并没有给模型注入大量新知识，而更像是激发和对齐预训练阶段已具备的能力73% similarUnverifiedThe higher priority of system prompts over user instructions is deliberately reinforced through dedicated alignment training data during the RLHF stage.72% similarVerified当代前沿模型的训练方法包括RLHF（基于人类反馈的强化学习）及其后继者RLAIF和过程奖励模型（PRM）71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/738437API
curl https://kongchang.com/api/v1/knowledge/claims/738437MCP
get_claim(id=738437)