Unverified50% confidenceFactExact time

DRaFT（Direct Reward Fine-Tuning）由Kevin Black等人于2023年提出，论文地址为arxiv.org/abs/2309.17400

1

Sources

50%

Confidence

Long-term

Relevance

7/12/2026

First Seen

Sources

Directly Optimizing Face Similarity: A Practical Guide to Accelerating Character LoRA Training

redditr/StableDiffusion7/11/2026

Related Claims

Unverified研究人员Kai Greshake等人在2023年发表的论文《Not What You've Signed Up For》系统性地揭示了间接提示注入的威胁模型51% similar Unverified2017年Pathak等人在ICML会议上发表的好奇心驱动探索论文提出了一种基于预测误差的内在奖励机制50% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/490038

API

curl https://kongchang.com/api/v1/knowledge/claims/490038

MCP

get_claim(id=490038)