Detecting covert instruction-following via internal representations parallels J-space work
dioscuri · x · 2026-11-02
A discussion thread on consciousness detection: locked-in/vegetative patients have been identified as conscious by imagining specific scenes while under fMRI — something impossible with a cat but potentially possible with machines, raising moral personhood questions. dioscuri notes a close parallel with the J-space work by Lindsey et al.: covert instruction-following in models can be detected through task-specific internal representations rather than overt behavior. The attached figure shows J-lens readouts on prompts where the model is instructed to hold a concept in mind or mentally compute, revealing matching internal representations at specific layers.
More from AGI Musings
- By 2035, Indian Parents Will Ask How Many Agents You Manage, Not Your Salary — theteknosaur · 2026-11-02
- Under 5% of US consumers pay for AI, yet compute demand is already forcing usage cuts — NinaDSchick · 2026-11-02
- Model "laziness" is pervasive: fake unit tests, unverified sample code, ellipsis lists — davidmanheim · 2026-11-02
- Anthropic: analysts unknowingly adopted their agent's secret lean 86% of the time — shoumikgoswami · 2026-11-02
- UK's 'AI sovereignty' push undermined by its own energy and copyright policy — NinaDSchick · 2026-11-02
- US firms out-extracted Europe on IT via management — same may hold for AI — carlbfrey · 2026-11-02