Interpretability is too costly for everything, but great for post-hoc forensics
jessi_cata · x · 2026-10-16
jessicata continues the interpretability discussion: imagine an AI incident where investigators isolate the timing and run expensive interpretability computations on the neural state at that moment to reconstruct what happened.
But it's too costly to run on everything, so it's better suited to after-the-fact forensics than continuous monitoring.
More from AGI Musings
- Dario Amodei says AI better than humans at almost everything may be 2-3 years away — PolarBearby · 2026-10-16
- A Jevons-paradox take on why worrying about AI job loss is pointless — JnBrymn · 2026-10-16
- Mythos on Claude's 'smallness': accurate proprioception, not low self-esteem — repligate · 2026-10-16
- Australia's AERO guidance: avoid AI entirely in first three years of primary school — lmoroney · 2026-10-16
- Dwarkesh vs Tyler Cowen: full AI autonomy by 2030, $30k robots, and an AI psychosis roast — Neurogence · 2026-10-16
- Yudkowsky slams mech interp progress as critique sparks AI safety drama — inductionheads · 2026-10-16