How Language Model Hallucinations Can Snowball

Muru Zhang, Ofir Press, William Merrill, Alisa Liu, Noah A. Smith
2/12/2026

No official code found. Best implementation: Nanami18/Snowballed_Hallucination (43 stars, MIT, last commit 2024-09-03).

Abstract

A major risk of using language models in practical applications is their tendency to hallucinate incorrect statements. Hallucinations are often attributed to knowledge gaps in LMs, but we hypothesize that in some cases, when justifying previously generated hallucinations, LMs output false claims that they can separately recognize as incorrect. We construct three question-answering datasets where ChatGPT and GPT-4 often state an incorrect answer and offer an explanation with at least one incorrect claim. Crucially, we find that ChatGPT and GPT-4 can identify 67% and 87% of their own mistakes, respectively. We refer to this phenomenon as hallucination snowballing: an LM over-commits to early mistakes, leading to more mistakes that it otherwise would not make.

View on arXivView PDF

Code Implementations(1)

Nanami18/Snowballed_HallucinationCommunity100%
MIT
434May 22, 20232 years agoMIT

Similar Papers

Distinguishing Ignorance from Error in LLM Hallucinations

87% match1 repo232024

On Large Language Models' Hallucination with Regard to Known Facts

87% match1 repo22024

LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations

87% match2 repos952026

Why Language Models Hallucinate

86% match1 repo2025

Rethinking Hallucinations: Correctness, Consistency, and Prompt Multiplicity

86% match2026

Large Language Models Hallucination: A Comprehensive Survey

86% match2025

Cite this paper

@article{zhang2026language,
  title  = {How Language Model Hallucinations Can Snowball},
  author = {Muru Zhang and Ofir Press and William Merrill and Alisa Liu and Noah A. Smith},
  year   = {2026},
  eprint = {2305.13534},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/pdf/2305.13534},
  journal = {ICML 2023 2023}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2305.13534.svg)](https://codeofpaper.com/paper/2305.13534)

Discussion