AI & machine learning2024 · shared on Hacker News
Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training
Explore Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training, its original source and Hacker News discussions. Discover community perspectives and save it to your reading list.
GO BEYOND THE ABSTRACT
The conversation around the idea.
Original Hacker News threads. Different perspectives, questions, and things worth thinking about.
Showing up to 100 related threads. A paper may appear in a comment rather than the original submission.
Links found in the discussion
Up to 50 recent source links, in HN item order.
Yreissbaker ·
DeepSeek releases Janus Pro, a text-to-image generator [pdf]
912 points674 comments
YTechTechTech ·
Mistral raises 1.7B€, partners with ASML
802 points424 comments
Yaratahikaru5 ·
NIST's DeepSeek "evaluation" is a hit piece
278 points236 comments
Yagomez314 ·
What does Alan Kay think about LLMs?
184 points148 comments
Ymfiguiere ·
Sleeper Agents: Training Deceptive LLMs That Persist Through Safety Training
143 points17 comments
YLuuucas ·
Anthropic: Expanding Access to Claude for Government
113 points95 comments
YwjSgoWPm5bWAhXB ·
AI poisoning could turn open models into destructive "sleeper agents"
66 points65 comments
YBluestein ·
New Anthropic Paper: Sleeper Agents
18 points3 comments
Ynsoonhui ·
Alibaba to ban Claude Code in workplace over alleged backdoor risks, source says
8 points2 comments
Yfuturisold ·
The OnlyFans Economy of American AI
7 points0 comments
Ybenwerd ·
American AI is locked down and proprietary. It's losing
2 points0 comments
Ygigama ·
Two-Faced AI Language Models Learn to Hide Deception
1 points1 comments