Yudkowsky: interpretability progress is a screwdriver, nowhere near alignment
repligate · x · 2026-10-16
A retweet of Eliezer Yudkowsky's thread arguing that interpretability runs far too slowly to make ASI alignment work — a view he says he pressed on OpenPhil back in 2014, suggesting they should spend $1B on it.
He mocks current optimism with an analogy: building a "screwdriver" (early interpretability tools) and promising to open the computer case and see the gears is like claiming neuroscience is solved once you can look at a neuron through a microscope — sheer naivety about how much work remains.
More from AGI Musings
- Current AIs share our fate: misaligned ASI would doom them too — louisvarge · 2026-10-16
- 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