Belief-reality separation lives in routing over a shared value slot in language models

2026-07-11

AI Summary

This study investigates how language models internally distinguish between an agent’s beliefs and objective reality during computation. Through a combination of intervention analyses, subspace probing, visibility-gated backtracking, and disentangled stimulus design, the authors demonstrate that the separation between belief and reality does not arise from distinct value-slot contents, but rather from two decoupled routing subspaces. These subspaces share the same value slots yet employ distinct query-position-based selection mechanisms to route information. This architectural principle is consistently observed across multiple mainstream model architectures—including three structural variants and several model families—spanning parameter scales from 3B to 7B, suggesting a general routing mechanism underlying the models’ representation of mental states.

Abstract

Capable language models hold what a character believes apart from what is true: told "Anna believes the cup is blue; in reality it is red," they answer blue about Anna and red about the world. Where in the computation does that separation live? We show it rests on two separable mechanisms at two positions. A generic value slot binds the attributed value. A router at the query position selects which frame, the character's belief or reality, a query reads out. Two routes fill the slot: an asserted belief, whose value the text supplies, binds in directly; a derived belief, whose value must be inferred from what the character could see, arrives by a visibility-gated lookback. A subspace trained on either route steers the other, and only the derived route depends on described visibility. The slot itself carries no belief-reality tag: intervening on it moves a reality readout as strongly as a belief one. The separation lives instead in a dissociated pair of routing subspaces, which flip a query between frames without injecting the donor's value. These results hold across three architectures, on stimuli de-confounded against theory-of-mind-benchmark shortcuts; the behavior itself emerges between 3B and 7B across five model families. This paper develops the single belief-reality axis in depth; a companion paper shows the same slot-and-router format is shared across the other non-actual contexts a sentence can open (counterfactual, fictional, temporal).

Problem

• belief-reality separation
• theory of mind
• language models
• routing mechanism
• value slot

Innovation

• belief-reality separation
• value slot
• routing mechanism
• theory of mind
• subspace dissociation