VLMs preserve linearly separable visual magnitudes and can compare them, yet collapse at symbolic mapping because visual and textual number spaces remain fractured and disjoint.
Ariba Khan, Stephen Casper, and Dylan Hadfield-Menell
3 Pith papers cite this work, alongside 2 external citations. Polarity classification is still indexing.
3
Pith papers citing it
2
external citations · external index
years
2026 3representative citing papers
Linear probes on residual-stream activations identify a shared preference vector in LLMs that tracks choices across prompts and causally steers decisions even for anti-correlated personas.
Direct research on AI consciousness is intractable, so the field should prioritize studying perceived AI consciousness and its societal consequences.
citing papers explorer
-
Dissociative Identity: Language Model Agents Lack Grounding for Reputation Mechanisms
VLMs preserve linearly separable visual magnitudes and can compare them, yet collapse at symbolic mapping because visual and textual number spaces remain fractured and disjoint.
-
Probing Persona-Dependent Preferences in Language Models
Linear probes on residual-stream activations identify a shared preference vector in LLMs that tracks choices across prompts and causally steers decisions even for anti-correlated personas.
-
AI and Consciousness: Shifting Focus Towards Tractable Questions
Direct research on AI consciousness is intractable, so the field should prioritize studying perceived AI consciousness and its societal consequences.