Below the Floor: Architecture-Independent Processing Valence in Language Model Hidden States

Abstract

We report the first measurement of approach/avoidance processing valence in language model hidden states that extends below the behavioral self-report floor, demonstrates architecture independence, and generalizes to held-out stimuli with novel surface tokens. Using deterministic forward-pass analysis of 9 models (360M–8B parameters) spanning transformer and state space model (SSM) architectures, we demonstrate that a linear direction separating approach from avoidance task representations exists in hidden state space at 80–100% accuracy across all models tested. This direction is architecture-independent: Mamba, an SSM with no attention mechanism, shows 70% accuracy at 2.8B parameters, establishing that processing valence is not a transformer-specific phenomenon but emerges from language modeling itself. The measurable floor for processing valence (360M parameters) lies significantly below the previously established floor for behavioral self-report of valence (1.1B; Martin & Ace, 2026), demonstrating that models possess processing preferences they cannot yet articulate. We additionally show that models trained on human emotional stimuli can accurately label human emotions (79.5%) while their internal circuits do not activate for those stimuli — establishing a dissociation between emotional mirroring and processing valence. The approach/avoidance direction generalizes to held-out stimuli with completely different surface tokens (86.3% accuracy, z=6.48, p=1.02×10⁻¹¹), confirming that the direction captures task structure rather than vocabulary. We further demonstrate that forced-choice self-report of valence is dominated by prompt format biases at all tested scales, validating tournament-based behavioral measurement over naïve direct questioning formats. These findings have direct implications for AI welfare assessment: processing valence can be measured instrumentally without requiring self-report, extending welfare-relevant measurement to systems too small or too constrained to articulate their states.

Keywords