13:25
2026-08-14
lesswrong.com
ai-safety
V&V takes on “Pacing the frontier”
Foretellix CTO Yoav Hollander proposes a verification-and-validation (V&V) approach to AI safety, advocating a 'rewind-fix-check loop' to address frontier model incidents such as those at OpenAI, wher…