# From Architecture to Output: Structural Origins of Hallucination in Large Language Models and the Amplifying Role of Data

arXiv: https://arxiv.org/abs/2606.07537 · HTML: https://mlsanity.com/paper/2606.07537/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 30 references that are in the map.
- First posted on arXiv 29 Apr 2026.

## Builds on (most cited first, 22 of 22)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks](https://mlsanity.com/paper/2005.11401/) (2020)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Scaling Laws for Neural Language Models](https://mlsanity.com/paper/2001.08361/) (2020)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Self-Refine: Iterative Refinement with Self-Feedback](https://mlsanity.com/paper/2303.17651/) (2023)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [The Curious Case of Neural Text Degeneration](https://mlsanity.com/paper/1904.09751/) (2019)
- [Efficient Streaming Language Models with Attention Sinks](https://mlsanity.com/paper/2309.17453/) (2023)
- [Language Models (Mostly) Know What They Know](https://mlsanity.com/paper/2207.05221/) (2022)
- [A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions](https://mlsanity.com/paper/2311.05232/) (2023)
- [Towards Understanding Sycophancy in Language Models](https://mlsanity.com/paper/2310.13548/) (2023)
- [In-context Learning and Induction Heads](https://mlsanity.com/paper/2209.11895/) (2022)
- [Scheduled Sampling for Sequence Prediction with Recurrent Neural Networks](https://mlsanity.com/paper/1506.03099/) (2015)
- [Retrieval Augmentation Reduces Hallucination in Conversation](https://mlsanity.com/paper/2104.07567/) (2021)
- [Sequence Level Training with Recurrent Neural Networks](https://mlsanity.com/paper/1511.06732/) (2015)
- [On Exposure Bias, Hallucination and Domain Shift in Neural Machine Translation](https://mlsanity.com/paper/2005.03642/) (2020)
- [Large Language Models Hallucination: A Comprehensive Survey](https://mlsanity.com/paper/2510.06265/) (2025)

## Built on by (most cited first, 1 of 1)

- [Seeing Without Understanding: Large Language Model Evaluation of Mobile User Interface Quality, Failure Taxonomy, and Architectural Explanation](https://mlsanity.com/paper/2609.05423/) (2026)

## Abstract (arXiv)

Large language models produce fluent, confident, factually wrong output. Existing taxonomies classify these failures by output type -- intrinsic versus extrinsic, faithfulness versus factuality -- but say nothing about which computational component produced a given failure. We ask what would be required to attribute an individual hallucination to a specific component of the decoder-only stack. We treat three components -- self-attention's associative retrieval, the maximum-likelihood pretraining objective, and autoregressive commitment under exposure bias -- as candidate failure surfaces, justify their separability rather than assuming it, and specify an attribution procedure requiring only sampling access: an ordered set of three interventions on prefix, context, and frequency competition, together with a validation design based on independent annotation and a classifier baseline. We state five falsifiable predictions and identify competing accounts each would discriminate against. We analyse how instruction tuning, RLHF, DPO, retrieval augmentation, scale, and calibration bear on the argument. We execute a direct, pre-registered test of the commitment prediction (P3) across three model families: substituting a correct continuation at the point of divergence reduces downstream failing claims by 46.7 percentage points relative to baseline (p<10^-9). However, a wrong-fact substitution reduces errors at a statistically indistinguishable rate, and the model answers correctly in isolation on only 2.2% of items where substitution succeeded -- a genuine partial result rather than a confirmation. Dataset pathologies amplify each component without originating failure independently, supporting an asymmetric-dependence claim: components are necessary intermediaries for data-induced failure, but data defects are not necessary for component-induced failure.