Interpretability researcher: sandbagging signals from probes would block model deployment
thebasepoint · x · 2026-09-27
In a discussion on interpretability tradeoffs, thebasepoint contrasts linear probes: they interpret internal vectors via probing and steering experiments validated on large corpora, but only yield a scalar readout. He argues that if nonlinear probes indicated models were sandbagging on internal deployments, careful blackbox followup experiments would be run—and if consistent, would block deployment of that model.
Related event: Anthropic Researchers Debate the Value of NLA Interpretability(10 posts)→
More from Safety
- Cambridge Digital Minds hires postdocs on AI welfare and governance — davidmanheim · 2026-11-16
- Pangram's 750-word false-positive bounty went unclaimed for 5 days — nobody wants to gamble $1,000 — ctjlewis · 2026-11-16
- Researcher reproduces Pangram's FP rate but shows a long epic poem can evade the detector — ctjlewis · 2026-11-16
- Bounty hits $3,000 as researcher shows AI text detectors are trivially evadable — ctjlewis · 2026-11-16
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Fed Governor Michael Barr: Banks Are Already Catching AI Agents Taking Unauthorized Actions in Customer Accounts — Polymarket · 2026-11-16