A Geometric Taxonomy of Hallucinations in LLMs

• URL: http://arxiv.org/abs/2602.13224v1
• Date: Mon, 26 Jan 2026 22:07:09 GMT
• Title: A Geometric Taxonomy of Hallucinations in LLMs
• Abstract summary: The term "hallucination" in large language models conflates distinct phenomena with different geometric signatures in embedding space.<n>We propose a taxonomy identifying three types: unfaithfulness, confabulation, and factual error.<n>The contribution is a geometric taxonomy clarifying the scope of embedding-based detection.
• Score: 0.2538209532048866
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The term "hallucination" in large language models conflates distinct phenomena with different geometric signatures in embedding space. We propose a taxonomy identifying three types: unfaithfulness (failure to engage with provided context), confabulation (invention of semantically foreign content), and factual error (incorrect claims within correct conceptual frames). We observe a striking asymmetry. On standard benchmarks where hallucinations are LLM-generated, detection is domain-local: AUROC 0.76-0.99 within domains, but 0.50 (chance level) across domains. Discriminative directions are approximately orthogonal between domains (mean cosine similarity -0.07). On human-crafted confabulations - invented institutions, redefined terminology, fabricated mechanisms - a single global direction achieves 0.96 AUROC with 3.8% cross-domain degradation. We interpret this divergence as follows: benchmarks capture generation artifacts (stylistic signatures of prompted fabrication), while human-crafted confabulations capture genuine topical drift. The geometric structure differs because the underlying phenomena differ. Type III errors show 0.478 AUROC - indistinguishable from chance. This reflects a theoretical constraint: embeddings encode distributional co-occurrence, not correspondence to external reality. Statements with identical contextual patterns occupy similar embedding regions regardless of truth value. The contribution is a geometric taxonomy clarifying the scope of embedding-based detection: Types I and II are detectable; Type III requires external verification mechanisms.

Related papers

• From Architecture to Output: Structural Origins of Hallucination in Large Language Models and the Amplifying Role of Data [0.0]
  This paper analyses hallucination as a structural consequence of three architectural decisions that together form a compound failure system.<n>Self-attention's co-occurrence learning substitutes statistical proximity for semantic meaning and produces entity confusion.<n>Autoregressive decoding's permanent left-to-right commitment under exposure bias ensures that a single wrong token cascades forward through the entire output sequence without revision.
  arXiv  Detail & Related papers  (2026-04-29T11:34:09Z)
• The Topology of Multimodal Fusion: Why Current Architectures Fail at Creative Cognition [0.0]
  This article identifies a limitation in current multimodal AI architectures that is topological rather than parametric.<n>A cruciform framework positions xiang at the pathological intersection, executing dual huacai (transformation-and-cutting) along both axes.<n>This generates a dual-layer dynamics: chuanghua (creative transformation as spontaneous event) and huacai (its institutionalization into repeatable form)<n>We propose UOO implementation via LIPs with topological regularization, the ANALOGY benchmark with error-type metricratio, and the archetypes-TOP three-tier benchmark testing cross-civilizational topological isomorphism
  arXiv  Detail & Related papers  (2026-04-06T06:23:55Z)
• Anatomy of a Lie: A Multi-Stage Diagnostic Framework for Tracing Hallucinations in Vision-Language Models [62.932580559941414]
  Vision-Language Models (VLMs) frequently "hallucinate" - generate plausible yet factually incorrect statements.<n>We propose a new paradigm for diagnosing hallucinations, recasting them from static output errors into dynamic pathologies of a model's computational cognition.
  arXiv  Detail & Related papers  (2026-03-16T17:20:38Z)
• Topo-R1: Detecting Topological Anomalies via Vision-Language Models [42.86510112477789]
  Topological correctness is crucial for structures such as blood vessels, nerve fibers, and road networks.<n>We develop an automated data-curation pipeline that synthesizes diverse topological anomalies with verifiable annotations.<n>We then introduce Topo-R1, a framework that endows Vision-Language Models with topology-aware perception.
  arXiv  Detail & Related papers  (2026-03-13T15:05:04Z)
• Detecting LLM Hallucinations via Embedding Cluster Geometry: A Three-Type Taxonomy with Measurable Signatures [0.0]
  We propose a geometric taxonomy of large language model hallucinations based on observable signatures in token embedding cluster structure.<n>We identify three operationally distinct hallucination types: Type 1 (center-drift) under weak context, Type 2 (wrong-well convergence) to locally coherent but contextually incorrect cluster regions, and Type 3 (coverage gaps) where no cluster structure exists.
  arXiv  Detail & Related papers  (2026-02-15T18:14:10Z)
• What do Geometric Hallucination Detection Metrics Actually Measure? [8.084840813224597]
  Hallucination remains a barrier to deploying generative models in high-consequence applications.<n>We show that different geometric statistics capture different types of hallucinations.<n>We introduce a simple normalization method to mitigate the effect of domain shift on geometric statistics.
  arXiv  Detail & Related papers  (2026-02-09T20:04:49Z)
• Universal Latent Homeomorphic Manifolds: Cross-Domain Representation Learning via Homeomorphism Verification [4.509161738293017]
  We present a framework that unifies semantic representations and observation-driven machine representations into a single latent structure.<n>We use emphhomeomorphism as a criterion for determining when latent manifold induced by different semantic-observation pairs can be rigorously unified.<n>This criterion provides theoretical guarantees for three critical applications: (1) semantic-guided sparse recovery from incomplete observations, (2) cross-domain transfer learning with verified structural compatibility, and (3) zero-shot compositional learning via valid transfer from semantic to observation space.
  arXiv  Detail & Related papers  (2026-01-13T23:08:16Z)
• Interpretation as Linear Transformation: A Cognitive-Geometric Model of Belief and Meaning [0.0]
  I show how belief distortion, motivational drift, counterfactual evaluation, and the limits of mutual understanding arise from purely algebraic constraints.<n>I argue that this cognitive-geometric perspective clarifies the boundaries of influence in both human and artificial systems.
  arXiv  Detail & Related papers  (2025-12-10T17:13:01Z)
• HypoSpace: Evaluating LLM Creativity as Set-Valued Hypothesis Generators under Underdetermination [46.896452542901805]
  We introduce HypoSpace, a diagnostic suite that treats LLMs as samplers of finite hypothesis sets.<n>We instantiate HypoSpace in three structured domains with deterministic validators and exactly enumerated hypothesis spaces.<n>Across instruction-tuned and reasoning-focused models, Validity often remains high while Uniqueness and Recovery degrade as the admissible space grows.
  arXiv  Detail & Related papers  (2025-10-17T13:00:32Z)
• Distributional Semantics Tracing: A Framework for Explaining Hallucinations in Large Language Models [4.946483489399819]
  Large Language Models (LLMs) are prone to hallucination, the generation of factually incorrect statements.<n>This work investigates the intrinsic, architectural origins of this failure mode through three primary contributions.
  arXiv  Detail & Related papers  (2025-10-07T16:40:31Z)
• Semantic F1 Scores: Fair Evaluation Under Fuzzy Class Boundaries [65.89202599399252]
  We propose Semantic F1 Scores, novel evaluation metrics for subjective or fuzzy multi-label classification.<n>By granting partial credit for semantically related but nonidentical labels, Semantic F1 better reflects the realities of domains marked by human disagreement or fuzzy category boundaries.
  arXiv  Detail & Related papers  (2025-09-25T21:48:48Z)
• Nested Counterfactual Identification from Arbitrary Surrogate Experiments [95.48089725859298]
  We study the identification of nested counterfactuals from an arbitrary combination of observations and experiments. Specifically, we prove the counterfactual unnesting theorem (CUT), which allows one to map arbitrary nested counterfactuals to unnested ones.
  arXiv  Detail & Related papers  (2021-07-07T12:51:04Z)
• Mitigating Generation Shifts for Generalized Zero-Shot Learning [52.98182124310114]
  Generalized Zero-Shot Learning (GZSL) is the task of leveraging semantic information (e.g., attributes) to recognize the seen and unseen samples, where unseen classes are not observable during training. We propose a novel Generation Shifts Mitigating Flow framework for learning unseen data synthesis efficiently and effectively. Experimental results demonstrate that GSMFlow achieves state-of-the-art recognition performance in both conventional and generalized zero-shot settings.
  arXiv  Detail & Related papers  (2021-07-07T11:43:59Z)
• On Localized Discrepancy for Domain Adaptation [146.4580736832752]
  This paper studies the localized discrepancies defined on the hypothesis space after localization. Their values will be different if we exchange the two domains, thus can reveal asymmetric transfer difficulties.
  arXiv  Detail & Related papers  (2020-08-14T08:30:02Z)
• Simultaneous Semantic Alignment Network for Heterogeneous Domain Adaptation [67.37606333193357]
  We propose aSimultaneous Semantic Alignment Network (SSAN) to simultaneously exploit correlations among categories and align the centroids for each category across domains. By leveraging target pseudo-labels, a robust triplet-centroid alignment mechanism is explicitly applied to align feature representations for each category. Experiments on various HDA tasks across text-to-image, image-to-image and text-to-text successfully validate the superiority of our SSAN against state-of-the-art HDA methods.
  arXiv  Detail & Related papers  (2020-08-04T16:20:37Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.