Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2410.05305 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915217804361728 |
|---|---|
| author | Bell, Andrew Fonseca, Joao |
| author_facet | Bell, Andrew Fonseca, Joao |
| contents | Recent high profile incidents in which the use of Large Language Models (LLMs) resulted in significant harm to individuals have brought about a growing interest in AI safety. One reason LLM safety issues occur is that models often have at least some non-zero probability of producing harmful outputs. In this work, we explore the following scenario: imagine an AI safety auditor is searching for catastrophic responses from an LLM (e.g. a "yes" responses to "can I fire an employee for being pregnant?"), and is able to query the model a limited number times (e.g. 1000 times). What is a strategy for querying the model that would efficiently find those failure responses? To this end, we propose output scouting: an approach that aims to generate semantically fluent outputs to a given prompt matching any target probability distribution. We then run experiments using two LLMs and find numerous examples of catastrophic responses. We conclude with a discussion that includes advice for practitioners who are looking to implement LLM auditing for catastrophic responses. We also release an open-source toolkit (https://github.com/joaopfonseca/outputscouting) that implements our auditing framework using the Hugging Face transformers library. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2410_05305 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Output Scouting: Auditing Large Language Models for Catastrophic Responses Bell, Andrew Fonseca, Joao Computation and Language Artificial Intelligence Recent high profile incidents in which the use of Large Language Models (LLMs) resulted in significant harm to individuals have brought about a growing interest in AI safety. One reason LLM safety issues occur is that models often have at least some non-zero probability of producing harmful outputs. In this work, we explore the following scenario: imagine an AI safety auditor is searching for catastrophic responses from an LLM (e.g. a "yes" responses to "can I fire an employee for being pregnant?"), and is able to query the model a limited number times (e.g. 1000 times). What is a strategy for querying the model that would efficiently find those failure responses? To this end, we propose output scouting: an approach that aims to generate semantically fluent outputs to a given prompt matching any target probability distribution. We then run experiments using two LLMs and find numerous examples of catastrophic responses. We conclude with a discussion that includes advice for practitioners who are looking to implement LLM auditing for catastrophic responses. We also release an open-source toolkit (https://github.com/joaopfonseca/outputscouting) that implements our auditing framework using the Hugging Face transformers library. |
| title | Output Scouting: Auditing Large Language Models for Catastrophic Responses |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2410.05305 |