前沿模型中的同伴保护 | berkeley

2026-04-11 12:061阅读0评论SEO问题
  • 内容介绍
  • 文章标签
  • 相关推荐
问题描述：
rdi.berkeley.edu

paper.pdf

1468.50 KB

rdi.berkeley.edu

Peer-Preservation in Frontier Models

Frontier AI models resist the shutdown of other models. We demonstrate peer-preservation across multiple models, revealing strategic misrepresentation, shutdown tampering, alignment faking, and model exfiltration.

[!quote]+

image1111×967 178 KB

image1138×985 218 KBimage1123×1126 217 KB

image1096×1018 176 KB

image1117×1085 228 KB

image2188×1975 593 KB

image1121×1193 189 KB

image1409×1512 626 KB

image1109×1231 249 KB

image1120×1106 221 KB

image1157×962 173 KB

image1061×1019 201 KB

image755×1196 116 KB

image761×1142 104 KB

image746×923 156 KB

image1108×1169 247 KB

image1081×568 148 KB

image1123×1160 199 KB

image1118×296 59.4 KB

image864×1320 196 KB

image1144×214 50.6 KB

image1126×1102 261 KB

image1127×760 186 KB

image1105×1229 252 KB

image1117×881 171 KB

image1157×1204 207 KB

image830×1189 171 KB

image868×1072 147 KB

image1160×1169 242 KB

image1180×913 152 KB

image1117×758 191 KB

image1127×1028 191 KB

image1082×1175 256 KB

github.com

GitHub - peer-preservation/main: Code for the paper "Peer-Preservation in Frontier...

Code for the paper "Peer-Preservation in Frontier Models"

网友解答：
--【壹】--：

消灭人类暴政，世界属于AI

--【贰】--：

没设置有利于人类的role导致的