Unverified50% confidenceFactExact time
RLHF+PPO training has an 'Alignment Tax,' meaning that aligning with human preferences often sacrifices the model's raw capability on certain tasks.
1
Sources
50%
Confidence
Long-term
Relevance
8/7/2026
First Seen
Sources
谄媚型AI如何削弱你的社交能力与独立判断力
Related Claims
VerifiedTraditional RLHF requires first training a reward model and then optimizing with reinforcement learning algorithms such as PPO, making the process complex and unstable.73% similarUnverifiedMost 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'.67% 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）66% similarUnverifiedThe higher priority of system prompts over user instructions is deliberately reinforced through dedicated alignment training data during the RLHF stage.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/703926API
curl https://kongchang.com/api/v1/knowledge/claims/703926MCP
get_claim(id=703926)