Mathematician Proposes "Timing Attacks" to Detect AI Contact Lens Cheating
AlexKontorovich · x · 2026-11-20
Mathematician Alex Kontorovich on AI exam cheating: once everyone wears AI contact lenses, even oral exams may only be verifiable via a crypto-style "timing attack"—watching how fast someone replies, since reading an AI answer takes extra time. Caveat: some of the strongest mathematicians would fail this test, as they are statistical outliers in how long they silently think before answering.
More from AGI Musings
- Emad Mostaque: 'human in the loop' breaks when AI outruns review — VraserX · 2026-11-20
- Google's Denny Zhou: "The best models make the best data" since 2026 — denny_zhou · 2026-11-20
- Investor Says Anthropic's Moat Is an 'Intelligence Moat' That's Hard to Catch — shaunmmaguire · 2026-11-20
- Paper Revisits How Longshoremen Unions Faced the Containerization Shock — Afinetheorem · 2026-11-20
- Median human earns ~$3k/yr in Asia — so whose values does Anthropic actually share? — NathanpmYoung · 2026-11-20
- Israeli bank exec: AI lets fewer devs ship more products, but no layoffs wave — ziv_ravid · 2026-11-20