My Assessment of Compute Verification in Plan A (+ open questions)
ai
A technical assessment on LessWrong examines methods to prevent secret AI training. Researchers propose several approaches: restricting bandwidth between processor clusters, periodically erasing hardware memory with cryptographic verification, and recording network traffic to verify what code is actually running. Each method involves tradeoffs in feasibility and performance cost, with several open questions about whether these techniques can work at scale without impractical overhead.
Source: https://www.lesswrong.com/posts/2eznrbNo6S7k5M9mu/my-asse...
Listen to this story
Hear this and more stories in a personalized audio briefing.
Open The Chonkerton