Saved in:
Bibliographic Details
Main Authors: Liu, Hude, Hu, Jerry Yao-Chieh, Zhang, Jennifer Yuntong, Song, Zhao, Liu, Han
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2509.21473
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908559144386560
author Liu, Hude
Hu, Jerry Yao-Chieh
Zhang, Jennifer Yuntong
Song, Zhao
Liu, Han
author_facet Liu, Hude
Hu, Jerry Yao-Chieh
Zhang, Jennifer Yuntong
Song, Zhao
Liu, Han
contents We formalize hallucinations in generative models as failures to link an estimate to any plausible cause. Under this interpretation, we show that even loss-minimizing optimal estimators still hallucinate. We confirm this with a general high probability lower bound on hallucinate rate for generic data distributions. This reframes hallucination as structural misalignment between loss minimization and human-acceptable outputs, and hence estimation errors induced by miscalibration. Experiments on coin aggregation, open-ended QA, and text-to-image support our theory.
format Preprint
id arxiv_https___arxiv_org_abs_2509_21473
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Are Hallucinations Bad Estimations?
Liu, Hude
Hu, Jerry Yao-Chieh
Zhang, Jennifer Yuntong
Song, Zhao
Liu, Han
Machine Learning
Artificial Intelligence
Computation and Language
Computer Vision and Pattern Recognition
We formalize hallucinations in generative models as failures to link an estimate to any plausible cause. Under this interpretation, we show that even loss-minimizing optimal estimators still hallucinate. We confirm this with a general high probability lower bound on hallucinate rate for generic data distributions. This reframes hallucination as structural misalignment between loss minimization and human-acceptable outputs, and hence estimation errors induced by miscalibration. Experiments on coin aggregation, open-ended QA, and text-to-image support our theory.
title Are Hallucinations Bad Estimations?
topic Machine Learning
Artificial Intelligence
Computation and Language
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.21473