Unverified50% confidenceFactExact time
Constitutional AI、DPO等新方法正在尝试解决RLHF的部分问题，但针对未成年人场景的专门对齐方法仍在探索阶段
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/8/2026
First Seen
Valid until: 11/6/2026
Sources
Related Claims
UnverifiedAlignment research has multiple technical approaches, including RLHF-based behavioral alignment, interpretability research, Constitutional AI, and Scalable Oversight.65% similarVerifiedConstitutional AI and adjustable safety levels are research directions for next-generation alignment technologies.60% similarUnverified第一代AI视频工具（2022-2023年）以Runway Gen-1/Gen-2、Pika Labs为代表，解决从无到有的创作门槛问题59% similarUnverified人机协同机制（Human-in-the-Loop，HITL）最早在强化学习（RLHF）中广泛应用，现已成为企业AI部署框架的标准安全节点，可防止系统误判执行不可逆操作59% similarUnverified2023年美国一起涉及AI伴侣应用Character.AI的未成年人心理健康事件引发全球关注59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/707265API
curl https://kongchang.com/api/v1/knowledge/claims/707265MCP
get_claim(id=707265)