[Feature Request] RMU Unlearning

[ruidazeng]

Tasks

• Benchmark
• Unlearning method
• Evaluation
• Dataset
• None of the above

Feature request

RMU is 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.

The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning

Paper: https://arxiv.org/abs/2403.03218
Site: https://www.wmdp.ai/
Representation Engineering: https://www.ai-transparency.org/

Motivation

RMU apparently performs really well on the WMDP dataset

[ruidazeng]

Times Article about RMU: https://time.com/6878893/ai-artificial-intelligence-dangerous-knowledge/
GitHub for WMDP/RMU: https://github.com/centerforaisafety/wmdp

[ruidazeng]

GitHub repo for representation engineering: https://github.com/andyzoujm/representation-engineering

Paper for Representation Engineering: https://arxiv.org/abs/2310.01405

Center for AI Safety Blog about Representation Engineering:
https://www.safe.ai/blog/representation-engineering-a-new-way-of-understanding-models

Center for AI Safety Video about RMU: https://www.youtube.com/watch?v=2U5NNiGC9yk