jd_pressman: interpretability can't cross the discontinuity before ASI takeover
jd_pressman · x · 2026-10-16
jdpressman argues interpretability only helps alignment if it lets you generalize model behavior to counterfactual situations and reliably foresee how the nth model behaves past the takeover-capability threshold. But ASI will presumptively behave differently once it actually has the opportunity to take over — a discontinuity where observing models without that opportunity tells you nothing, echoing points from Yudkowsky's List of Lethalities.
More from AGI Musings
- Testing alongside an agent in a p2p room feels oddly real, says dev — sull · 2026-10-16
- Yudkowsky's 2008 takeoff scenarios resurface as AI-advances-AI takes hold — jd_pressman · 2026-10-16
- Guardrails can make risky delegation work — but it takes a full-time team, says iandanforth — iandanforth · 2026-10-16
- e/acc founder: agentic models are autocomplete for digital actions — beffjezos · 2026-10-16
- What can you truly delegate? Only tasks with unambiguous, verifiable results — iandanforth · 2026-10-16
- Casper: Understanding 90% of neurons ≠ understanding the network — StephenLCasper · 2026-10-16