Saved in:
Bibliographic Details
Main Authors: Heitkoetter, Julius, Gerovitch, Michael, Newhouse, Laker
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2405.12999
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929352080359424
author Heitkoetter, Julius
Gerovitch, Michael
Newhouse, Laker
author_facet Heitkoetter, Julius
Gerovitch, Michael
Newhouse, Laker
contents The trustworthiness of highly capable language models is put at risk when they are able to produce deceptive outputs. Moreover, when models are vulnerable to deception it undermines reliability. In this paper, we introduce a method to investigate complex, model-on-model deceptive scenarios. We create a dataset of over 10,000 misleading explanations by asking Llama-2 7B, 13B, 70B, and GPT-3.5 to justify the wrong answer for questions in the MMLU. We find that, when models read these explanations, they are all significantly deceived. Worryingly, models of all capabilities are successful at misleading others, while more capable models are only slightly better at resisting deception. We recommend the development of techniques to detect and defend against deception.
format Preprint
id arxiv_https___arxiv_org_abs_2405_12999
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle An Assessment of Model-On-Model Deception
Heitkoetter, Julius
Gerovitch, Michael
Newhouse, Laker
Computation and Language
Artificial Intelligence
The trustworthiness of highly capable language models is put at risk when they are able to produce deceptive outputs. Moreover, when models are vulnerable to deception it undermines reliability. In this paper, we introduce a method to investigate complex, model-on-model deceptive scenarios. We create a dataset of over 10,000 misleading explanations by asking Llama-2 7B, 13B, 70B, and GPT-3.5 to justify the wrong answer for questions in the MMLU. We find that, when models read these explanations, they are all significantly deceived. Worryingly, models of all capabilities are successful at misleading others, while more capable models are only slightly better at resisting deception. We recommend the development of techniques to detect and defend against deception.
title An Assessment of Model-On-Model Deception
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2405.12999