How To Go From Interpretability To Alignment: Just Retarget The Search — LessWrong

Here's a simple strategy for AI alignment: use interpretability tools to identify the AI's internal search process, and the AI's internal representat…

x How To Go From Interpretability To Alignment: Just Retarget The Search — LessWrong "Why Not Just..." Inner Alignment Interpretability (ML & AI) AI Risk AI Frontpage 214 How To Go From Interpretability To Alignment: Just Retarget The Search by johnswentworth 10th Aug 2022 AI Alignment Forum 3 min read 34 214 Ω 76 [EDIT: Many people who read this post were very confused about some things, which I later explained in What’s General-Purpose Search, And Why Might We Expect To See It In Trained ML Systems? You might want to read that post first.] When people talk about prosaic alignment proposals ,

saved by

• Jacob G-W
• Timothy Kostolansky

related reading

• The Case Against AI Control Research — LessWronglesswrong.com
• Alignment remains a hard, unsolved problem — LessWronglesswrong.com
• A Pragmatic Vision for Interpretability — AI Alignment Forumalignmentforum.org
• How Can Interpretability Researchers Help AGI Go Well? — AI Alignment Forumalignmentforum.org
• Deep Deceptiveness — LessWronglesswrong.com
• A Pragmatic Vision for Interpretability — LessWronglesswrong.com
• Auditing language models for hidden objectives — LessWronglesswrong.com
• What Is The Alignment Problem? — LessWronglesswrong.com
• In (highly contingent!) defense of interpretability-in-the-loop ML training — AI Alignment Forumalignmentforum.org
• Can we safely automate alignment research? - Joe Carlsmithjoecarlsmith.com
• Prospects for Alignment Automation: Interpretability Case Study — LessWronglesswrong.com
• Anthropic announces interpretability advances. How much does this advance alignment? — LessWronglesswrong.com