Feed Entities Podcasts Deep Dives Developers About

[KongchangAI]

Real-time AI knowledge base. Every claim is traceable to its source, with data freshness under 24 hours.

Powered by 240+ first-hand AI sources

Knowledge

Feed
Hot Entities
Podcasts

Developers

CLI
MCP Server
REST API
About

© 2026 KongchangAI. The freshest, most accurate AI knowledge

Verified65% confidenceTradeoffExact time

传统RLHF依赖人工标注员的偏好判断，成本高且存在标注质量瓶颈

3

Sources

65%

Confidence

Long-term

Relevance

8/5/2026

First Seen

Sources

QwenGrowthPlan千问成长计划：真实任务驱动AI模型迭代新范式

Related Claims

UnverifiedRLHF的价值判断来自人类标注者的主观偏好，而CAI更依赖规则驱动的自动化批判机制77% similar UnverifiedRLHF依赖大量人工标注构建奖励信号，成本更高但更能捕捉隐性的人类偏好77% similar VerifiedRLHF存在局限性：人类标注者有偏见和不一致性、奖励模型可能被过度优化（奖励黑客）、流程成本高昂难以扩展77% similar Unverified人类标注员在RLHF评分时倾向于给措辞流畅、结构清晰、长度适中的答案打高分，导致对齐税和输出多样性下降75% similar UnverifiedRLHF（基于人类反馈的强化学习）会放大规避风险效应，因为人类评分者倾向于给保守、无争议的输出打高分69% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/687607

API

curl https://kongchang.com/api/v1/knowledge/claims/687607

MCP

get_claim(id=687607)