Aligned vs. Monitorable: An X Debate Over Whether CoT Monitoring Is Necessary
sandersted · x · 2026-09-05
A debate on model alignment versus monitorability. sandersted argues that forced to pick one, he'd rather have an aligned model than a merely monitorable one — but stresses that monitoring is how you gain confidence in alignment, and frames the core crux as how much general monitoring depends on chain-of-thought monitoring. He draws an analogy to humans: we judge trustworthiness by observing behavior over time, but that only works for Ted-level power; a God-level power would demand stronger guarantees.
More from AGI Musings
- Consciousness Experts Clash in Nature Over Whether AI Could Ever Be Conscious — anilkseth · 2026-10-21
- Polymarket puts just 7% odds on OpenAI pausing AI training this year — Polymarket · 2026-10-21
- AI agents are overwhelming Gmail's spam filter, researcher warns — LuizaJarovsky · 2026-10-21
- Daniel Lemire: AI-generated papers are exposing academia's 'Profzi scheme' — lemire · 2026-10-21
- Universities block thesis downloads while each dissertation averages barely a handful of lifetime reads — generativist · 2026-10-21
- Berkeley wins Schmidt Sciences HAVI grant to decode whale and parrot communication with AI — begusgasper · 2026-10-21