On the Fundamental Impossibility of Hallucination Control in Large Language Models

• URL: http://arxiv.org/abs/2506.06382v7
• Date: Wed, 15 Oct 2025 22:25:41 GMT
• Title: On the Fundamental Impossibility of Hallucination Control in Large Language Models
• Authors: Michał P. Karpowicz,
• Abstract summary: Impossibility Theorem: no LLM performing non-trivial knowledge aggregation can simultaneously achieve truthful knowledge representation, semantic information conservation, and revelation of relevant knowledge.<n>We prove this by modeling inference as an auction of ideas, where distributed components compete to influence responses using encoded knowledge.<n>We show that hallucination and imagination are mathematically identical, and both violate at least one of the four essential properties.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: This paper establishes a fundamental Impossibility Theorem: no LLM performing non-trivial knowledge aggregation can simultaneously achieve truthful knowledge representation, semantic information conservation, complete revelation of relevant knowledge, and knowledge-constrained optimality. This impossibility stems from the mathematical structure of information aggregation, not from engineering limitations. We prove this by modeling inference as an auction of ideas, where distributed components compete to influence responses using their encoded knowledge. The proof employs three independent approaches: mechanism design (Green-Laffont theorem), proper scoring rules (Savage), and transformer architecture analysis (log-sum-exp convexity). We introduce the semantic information measure and the emergence operator to analyze computationally bounded and unbounded reasoning. Bounded reasoning makes latent information accessible, enabling gradual insights and creativity, while unbounded reasoning makes all derivable knowledge immediately accessible while preserving the semantic content. We prove the conservation-reasoning dichotomy: meaningful reasoning necessarily violates information conservation. Our framework suggests that hallucination and imagination are mathematically identical, and both violate at least one of the four essential properties. The Jensen gap in transformer attention quantifies this violation as excess confidence beyond constituent evidence. This unified view explains why capable models must balance truthfulness against creativity. These results provide principled foundations for managing hallucination trade-offs in AI systems. Rather than eliminating hallucination, we should optimize these inevitable trade-offs for specific applications. We conclude with philosophical implications connecting the impossibility to fundamental limits of reason.

Related papers

• Beyond Hallucinations: The Illusion of Understanding in Large Language Models [0.0]
  Large language models (LLMs) are becoming deeply embedded in human communication and decision-making.<n>They inherit the ambiguity, bias, and lack of direct access to truth inherent in language itself.<n>This paper argues that LLMs operationalize System 1 cognition at scale: fast, associative, and persuasive, but without reflection or falsification.
  arXiv  Detail & Related papers  (2025-10-16T13:19:44Z)
• 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)
• How Large Language Models are Designed to Hallucinate [0.42970700836450487]
  We argue that hallucination is a structural outcome of the transformer architecture.<n>Our contribution is threefold: (1) a comparative account showing why existing explanations are insufficient; (2) a predictive taxonomy of hallucination linked to existential structures with proposed benchmarks; and (3) design directions toward "truth-constrained" architectures capable of withholding or deferring when disclosure is absent.
  arXiv  Detail & Related papers  (2025-09-19T16:46:27Z)
• Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers [80.70134000599391]
  We argue that both behaviors stem from a single mechanism known as out-of-context reasoning (OCR)<n>OCR drives both generalization and hallucination, depending on whether the associated concepts are causally related.<n>Our work provides a theoretical foundation for understanding the OCR phenomenon, offering a new lens for analyzing and mitigating undesirable behaviors from knowledge injection.
  arXiv  Detail & Related papers  (2025-06-12T16:50:45Z)
• Counterfactual reasoning: an analysis of in-context emergence [57.118735341305786]
  We show that language models are capable of counterfactual reasoning.<n>We find that self-attention, model depth and pre-training data diversity drive performance.<n>Our findings extend to counterfactual reasoning under SDE dynamics.
  arXiv  Detail & Related papers  (2025-06-05T16:02:07Z)
• From Words to Waves: Analyzing Concept Formation in Speech and Text-Based Foundation Models [20.244145418997377]
  We analyze the conceptual structures learned by speech and textual models both individually and jointly.<n>We employ Latent Concept Analysis, an unsupervised method for uncovering latent representations in neural networks, to examine how semantic abstractions form across modalities.
  arXiv  Detail & Related papers  (2025-06-01T19:33:21Z)
• DeepTheorem: Advancing LLM Reasoning for Theorem Proving Through Natural Language and Reinforcement Learning [67.93945726549289]
  DeepTheorem is a comprehensive informal theorem-proving framework exploiting natural language to enhance mathematical reasoning.<n>DeepTheorem includes a large-scale benchmark dataset consisting of 121K high-quality IMO-level informal theorems and proofs.<n>We devise a novel reinforcement learning strategy (RL-Zero) explicitly tailored to informal theorem proving, leveraging the verified theorem variants to incentivize robust mathematical inference.
  arXiv  Detail & Related papers  (2025-05-29T17:59:39Z)
• I Predict Therefore I Am: Is Next Token Prediction Enough to Learn Human-Interpretable Concepts from Data? [76.15163242945813]
  Large language models (LLMs) have led many to conclude that they exhibit a form of intelligence.<n>We introduce a novel generative model that generates tokens on the basis of human-interpretable concepts represented as latent discrete variables.
  arXiv  Detail & Related papers  (2025-03-12T01:21:17Z)
• The Law of Knowledge Overshadowing: Towards Understanding, Predicting, and Preventing LLM Hallucination [85.18584652829799]
  We introduce a novel framework to quantify factual hallucinations by modeling knowledge overshadowing.<n>We propose a new decoding strategy CoDa, to mitigate hallucinations, which notably enhance model factuality on Overshadow (27.9%), MemoTrap (13.1%) and NQ-Swap (18.3%)
  arXiv  Detail & Related papers  (2025-02-22T08:36:06Z)
• In-situ graph reasoning and knowledge expansion using Graph-PReFLexOR [0.0]
  We present Graph-PReFLexOR, a framework that combines graph reasoning with symbolic abstraction to dynamically expand domain knowledge.<n>Inspired by reinforcement learning, it defines reasoning as a structured mapping, where tasks yield knowledge graphs, abstract patterns, and ultimately, final answers.<n>Results show superior reasoning depth and adaptability, underscoring the potential for transparent, multidisciplinary AI-driven discovery.
  arXiv  Detail & Related papers  (2025-01-14T13:52:41Z)
• No Free Lunch: Fundamental Limits of Learning Non-Hallucinating Generative Models [14.535583931446807]
  We develop a theoretical framework to analyze the learnability of non-hallucinating generative models. We show that incorporating inductive biases aligned with the actual facts into the learning process is essential.
  arXiv  Detail & Related papers  (2024-10-24T23:57:11Z)
• Turing Video-based Cognitive Tests to Handle Entangled Concepts [0.0]
  We present the results of an innovative video-based cognitive test on a specific conceptual combination.<n>We show that collected data can be faithfully modeled within a quantum-theoretic framework.<n>We provide a novel explanation for the appearance of entanglement in both physics and cognitive realms.
  arXiv  Detail & Related papers  (2024-09-13T14:30:55Z)
• LLMs Will Always Hallucinate, and We Need to Live With This [1.3810901729134184]
  This work argues that hallucinations in language models are not just occasional errors but an inevitable feature of these systems. It is, therefore, impossible to eliminate them through architectural improvements, dataset enhancements, or fact-checking mechanisms.
  arXiv  Detail & Related papers  (2024-09-09T16:01:58Z)
• Lean-STaR: Learning to Interleave Thinking and Proving [53.923617816215774]
  We present Lean-STaR, a framework for training language models to produce informal thoughts prior to each step of a proof.<n>Lean-STaR achieves state-of-the-art results on the miniF2F-test benchmark within the Lean theorem proving environment.
  arXiv  Detail & Related papers  (2024-07-14T01:43:07Z)
• Learning Discrete Concepts in Latent Hierarchical Models [73.01229236386148]
  Learning concepts from natural high-dimensional data holds potential in building human-aligned and interpretable machine learning models.<n>We formalize concepts as discrete latent causal variables that are related via a hierarchical causal model.<n>We substantiate our theoretical claims with synthetic data experiments.
  arXiv  Detail & Related papers  (2024-06-01T18:01:03Z)
• Exploring the Limits of Fine-grained LLM-based Physics Inference via Premise Removal Interventions [0.0]
  Language models (LMs) can hallucinate when performing complex mathematical reasoning. Physical context requires that any symbolic manipulation satisfies complex semantics. We show that LMs' mathematical reasoning is not physics-informed in this setting.
  arXiv  Detail & Related papers  (2024-04-29T02:43:23Z)
• Inference of Abstraction for a Unified Account of Reasoning and Learning [0.0]
  We give a simple theory of probabilistic inference for a unified account of reasoning and learning. We simply model how data cause symbolic knowledge in terms of its satisfiability in formal logic.
  arXiv  Detail & Related papers  (2024-02-14T09:43:35Z)
• A Probabilistic-Logic based Commonsense Representation Framework for Modelling Inferences with Multiple Antecedents and Varying Likelihoods [5.87677276882675]
  Commonsense knowledge-graphs (CKGs) are important resources towards building machines that can'reason' on text or environmental inputs and make inferences beyond perception. In this work, we study how commonsense knowledge can be better represented by -- (i) utilizing a probabilistic logic representation scheme to model composite inferential knowledge and represent conceptual beliefs with varying likelihoods, and (ii) incorporating a hierarchical conceptual ontology to identify salient concept-relevant relations and organize beliefs at different conceptual levels.
  arXiv  Detail & Related papers  (2022-11-30T08:44:30Z)
• On Binding Objects to Symbols: Learning Physical Concepts to Understand Real from Fake [155.6741526791004]
  We revisit the classic signal-to-symbol barrier in light of the remarkable ability of deep neural networks to generate synthetic data. We characterize physical objects as abstract concepts and use the previous analysis to show that physical objects can be encoded by finite architectures. We conclude that binding physical entities to digital identities is possible in finite time with finite resources.
  arXiv  Detail & Related papers  (2022-07-25T17:21:59Z)
• Acquiring and Modelling Abstract Commonsense Knowledge via Conceptualization [49.00409552570441]
  We study the role of conceptualization in commonsense reasoning, and formulate a framework to replicate human conceptual induction. We apply the framework to ATOMIC, a large-scale human-annotated CKG, aided by the taxonomy Probase.
  arXiv  Detail & Related papers  (2022-06-03T12:24:49Z)
• A Mathematical Walkthrough and Discussion of the Free Energy Principle [0.0]
  The Free-Energy-Principle (FEP) is an influential and controversial theory which postulates a connection between the thermodynamics of self-organization and learning through variational inference. FEP has been applied extensively in neuroscience, and is beginning to make inroads in machine learning by spurring the construction of novel and powerful algorithms by which action, perception, and learning can all be unified under a single objective. Here, we aim to provide a mathematically detailed, yet intuitive walk-through of the formulation and central claims of the FEP while also providing a discussion of the assumptions necessary and potential limitations of the theory.
  arXiv  Detail & Related papers  (2021-08-30T16:11:49Z)
• Modelling Compositionality and Structure Dependence in Natural Language [0.12183405753834563]
  Drawing on linguistics and set theory, a formalisation of these ideas is presented in the first half of this thesis. We see how cognitive systems that process language need to have certain functional constraints. Using the advances of word embedding techniques, a model of relational learning is simulated.
  arXiv  Detail & Related papers  (2020-11-22T17:28:50Z)
• Logical Neural Networks [51.46602187496816]
  We propose a novel framework seamlessly providing key properties of both neural nets (learning) and symbolic logic (knowledge and reasoning) Every neuron has a meaning as a component of a formula in a weighted real-valued logic, yielding a highly intepretable disentangled representation. Inference is omni rather than focused on predefined target variables, and corresponds to logical reasoning.
  arXiv  Detail & Related papers  (2020-06-23T16:55:45Z)
• Leap-Of-Thought: Teaching Pre-Trained Models to Systematically Reason Over Implicit Knowledge [96.92252296244233]
  Large pre-trained language models (LMs) acquire some reasoning capacity, but this ability is difficult to control. We show that LMs can be trained to reliably perform systematic reasoning combining both implicit, pre-trained knowledge and explicit natural language statements. Our work paves a path towards open-domain systems that constantly improve by interacting with users who can instantly correct a model by adding simple natural language statements.
  arXiv  Detail & Related papers  (2020-06-11T17:02:20Z)

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.