AI Control: Improving Safety Despite Intentional Subversion

AI Safety Fundamentals: Alignment - Un podcast de BlueDot Impact

Essayez BookBeat gratuitement pendant 60 jours

Profitez d'un univers entier d'audiolivres et d'e-books, allant des nouveautés aux classiques.

Sponsored

We’ve released a paper, AI Control: Improving Safety Despite Intentional Subversion. This paper explores techniques that prevent AI catastrophes even if AI instances are colluding to subvert the safety techniques. In this post:We summarize the paper;We compare our methodology to the methodology of other safety papers.Source:https://www.alignmentforum.org/posts/d9FJHawgkiMSPjagR/ai-control-improving-safety-despite-intentional-subversionNarrated for AI Safety Fundamentals by Perrin WalkerA podc...

Visit the podcast's native language site