We're sunsetting PodQuest on 2025-07-28. Thank you for your support!
Export Podcast Subscriptions
back
“Testing for AI Scheming” by Guive
39:05
Share
2025/1/7
LessWrong (30+ Karma)
AI Chapters
Transcribe
Chapters
Why Testing for AI Scheming is Hard?
The Experiment: Testing AI's Response to Deletion
Evidentiary Challenges in the Experiment
Objection 1: The AI Might Not Believe It Will Be Deleted?
Objection 2: The AI Might Allow Deletion to Help Future AIs?
Indexical Goals: What Are They?
Time Preference and Risk Tolerance in AI
Objection 3: Acausal Trade as a Reason for Deletion?
Objection 4: Imitating Characters vs. Genuine Scheming?
Predictions and Practical Challenges
Conclusion: What Does It All Mean?
Shownotes
Transcript
No transcript made for this episode yet, you may request it for free.
