The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning
By Nathaniel Li, Alexander Pan, Anjali Gopal, Summer Yue, Daniel Berrios, Alice Gatti, Justin Li, Ann-Kathrin Dombrowski, Shashwat Goel, Long Phan and 43 more. International Conference on Machine Learning, 2024
473 citations recorded by the source corpus.
The White House Executive Order on Artificial Intelligence highlights the risks of large language models (LLMs) empowering malicious actors in developing biological, cyber, and chemical weapons. To measure these risks of malicious use, government institutions and major AI labs are developing evaluations for hazardous capabilities in LLMs. However, current evaluations are private, preventing further research into mitigating risk. Furthermore, they focus on only a few, highly specific pathways for malicious use. To fill these gaps, we publicly release the Weapons of Mass Destruction Proxy (WMDP) benchmark, a dataset of 3,668 multiple-choice questions that serve as a proxy measurement of hazardous knowledge in biosecurity, cybersecurity, and chemical security. WMDP was developed by a consortium of academics and technical consultants, and was stringently filtered to eliminate sensitive information prior to public release. WMDP serves two roles: first, as an evaluation for hazardous knowledge in LLMs, and second, as a benchmark for unlearning methods to remove such hazardous knowledge. To guide progress on unlearning, we develop RMU, a state-of-the-art unlearning method based on controlling model representations. RMU reduces model performance on WMDP while maintaining general capabilities in areas such as biology and computer science, suggesting that unlearning may be a concrete path towards reducing malicious use from LLMs. We release our benchmark and code publicly at https://wmdp.ai
Subfields: Computer Networks and Communications
View Paper: https://www.semanticscholar.org/paper/06b9ad0b52d23231f650be0aeb0b17cc52c8a74b
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
Citations in the corpus
- Rethinking machine unlearning for large language models (2024, unknown): “Unlearning can be used to forget harmful behaviors such as the production of toxic, discriminatory, illegal, or morally undesirable outputs (Shevlane et al., 2023; Gehman et al., 2020; Li et al., 2024a), e.g. , instructions to build CBRN (chemical, biological, radiological, and nuclear) weapons.”
- Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning (2024, methodology): “Furthermore, the method of Liu et al. (2024b) is based on knowledge negation, while the approach of Li et al. (2024) relies on controlling model representations.”
- Improving Alignment and Robustness with Circuit Breakers (2024, unknown): “One approach involves routing the targeted representation to a fixed random direction with a large norm, as utilized in the unlearning method RMU [29].”