The LLM Belief Atlas: Visualizing How Language-Model Stances Depend on Elicitation

Abstract

What a language model appears to believe about a proposition is not a scalar but a signal that depends on how it is elicited and read: forcing a choice can fabricate a value, option-logprob scoring carries order bias, and sycophancy makes a stance a function of framing. We present the LLM Belief Atlas, a scalable in-browser WebGL instrument organized around a framework of belief spaces. The framework treats a model’s stance as a latent construct observed only through a four-dimensional measurement space, model, item, elicitation condition, and readout channel, and names seven subspaces, each bound to a facet of psychometric construct validity. The instrument co-registers three switchable belief readouts (Likert agreement, free-text reflection, and option logprobs) of the same elicitation instances and morphs between them at the million-point scale, while exposing uncertainty. On 76,048 elicited beliefs from four LLMs, framing conditions explain 34.2% of stance variance, against 46.8% for belief itself (and on par with it on the two natively-read models), but a within-model reliability measure does not predict cross-model reproducibility beyond noise.

Keywords: Large language models, belief elicitation, measurement, construct validity, dimensionality reduction, scalable scatterplot visualization, uncertainty visualization

Cite this paper

Fan Huang (2026). The LLM Belief Atlas: Visualizing How Language-Model Stances Depend on Elicitation (AIE-WP-2026-001, version 1.0). AI Explain. https://papers.ai-explain.org/papers/2026-001/

@techreport{AIE-WP-2026-001,
  title       = {{The LLM Belief Atlas: Visualizing How Language-Model Stances Depend on Elicitation}},
  author      = {Fan Huang},
  institution = {AI Explain},
  number      = {AIE-WP-2026-001},
  year        = {2026},
  month       = oct,
  note        = {Version 1.0, 26 October 2026},
  url         = {https://papers.ai-explain.org/papers/2026-001/}
}

Versions

1. Version 1.0Latest26 October 2026

   Initial release.

   PDF, 986 KBGitHub release

   SHA-256 fingerprint862804f8064ce29a880bb1c6c2ffdbf6d126dad76ba4171fa9567f8d15543115

# macOS / Linux
shasum -a 256 paper.pdf

# Windows PowerShell
Get-FileHash paper.pdf -Algorithm SHA256

Change history on GitHub