Introducing v0.5 of the AI Safety Benchmark from MLCommons

作者：Bertie Vidgen, Adarsh Agrawal, Ahmed M. Ahmed, Victor Akinwande, Namir Al-nuaimi, Najla Alfaraj, Elie Alhajjar, L. Aroyo, Trupti Bavalatti, Borhane Blili-Hamelin, K. Bollacker, Rishi Bomassani, Marisa Ferrara Boston, Siméon Campos, Kal Chakra, Canyu Chen, C. Coleman, Zacharie Delpierre Coudert, Leon Derczynski, Debojyoti Dutta, Ian W. Eisenberg, J. Ezick, Heather Frase, Brian Fuller, Ramu Gandikota, Agasthya Gangavarapu, Ananya Gangavarapu, J. Gealy, Rajat Ghosh, James Goel, Usman Gohar, Sujata Goswami, Scott A. Hale, W. Hutiri, Joseph Marvin Imperial, Surgan Jandial, Nicholas C. Judd, Felix Juefei-Xu, F. Khomh, B. Kailkhura, Hannah Rose Kirk, Kevin Klyman, Chris Knotz, Michael Kuchnik, Shachi H. Kumar, Chris Lengerich, Bo Li, Zeyi Liao, Eileen Long, Victor Lu, Yifan Mai, P. Mammen, Kelvin N. Manyeki, Sean McGregor, Virendra Mehta, Shafee Mohammed, Emanuel Moss, L. Nachman, Dinesh Jinenhally Naganna, Amin Nikanjam, Besmira Nushi, Luis Oala, Iftach Orr, Alicia Parrish, Çigdem Patlak, William Pietri, Forough Poursabzi-Sangdeh, Eleonora Presani, Fabrizio Puletti, Paul Röttger, Saurav Sahay, T. Santos, Nino Scherrer, Alice Schoenauer Sebag, P. Schramowski, Abolfazl Shahbazi, Vin Sharma, Xudong Shen, V. Sistla, Leonard Tang, Davide Testuggine, Vithursan Thangarasa, E. A. Watkins, Rebecca Weiss, Christoper A. Welty, Tyler Wilbers, Adina Williams, Carole-Jean Wu, Poonam Yadav, Xianjun Yang, Yi Zeng, Wenhui Zhang, Fedor Zhdanov, Jiacheng Zhu, Percy Liang, Peter Mattson, J. Vanschoren · 发表于：arXiv.org · 年份：2024 · DOI：10.48550/arXiv.2404.12241 · 被引用次数：71 · 研究领域：Computer Science

This paper introduces v0.5 of the AI Safety Benchmark, which has been created by the MLCommons AI Safety Working Group. The AI Safety Benchmark has been designed to assess the safety risks of AI systems that use chat-tuned language models. We introduce a principled approach to specifying and constructing the benchmark, which for v0.5 covers only a single use case (an adult chatting to a general-purpose assistant in English), and a limited set of personas (i.e., typical users, malicious users, and vulnerable users). We created a new taxonomy of 13 hazard categories, of which 7 have tests in the v0.5 benchmark. We plan to release version 1.0 of the AI Safety Benchmark by the end of 2024. The v1.0 benchmark will provide meaningful insights into the safety of AI systems. However, the v0.5 benchmark should not be used to assess the safety of AI systems. We have sought to fully document the limitations, flaws, and challenges of v0.5. This release of v0.5 of the AI Safety Benchmark includes (1) a principled approach to specifying and constructing the benchmark, which comprises use cases, types of systems under test (SUTs), language and context, personas, tests, and test items; (2) a taxonomy of 13 hazard categories with definitions and subcategories; (3) tests for seven of the hazard categories, each comprising a unique set of test items, i.e., prompts. There are 43,090 test items in total, which we created with templates; (4) a grading system for AI systems against the benchmark; (...