Unverified50% confidenceFactExact time

PPO is currently the most widely used algorithm in LLM alignment training, such as RLHF.

1

Sources

50%

Confidence

Medium-term (~90 days)

Relevance

9/20/2026

First Seen

Valid until: 12/19/2026

Sources

Score Centering: A Numerical Solution for Stabilizing Off-Policy Reinforcement Learning

Related Entities

PPOconcept RLHFconcept

Related Claims

UnverifiedRLHF遵循先用监督微调（本质是行为克隆）建立基础能力，再用PPO等算法优化人类偏好目标的范式71% similar UnverifiedProximal Policy Optimization (PPO) is the primary policy gradient algorithm tool for robot reinforcement learning.70% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/937388

API

curl https://kongchang.com/api/v1/knowledge/claims/937388

MCP

get_claim(id=937388)