LLM Unlearning Without an Expert Curated Dataset

TL;DR

This paper presents a scalable method for generating synthetic forget sets using language models to enable effective post-hoc unlearning of specific knowledge in large language models, without requiring expert-curated datasets.

Contribution

The authors introduce an automated, multi-step prompting pipeline to generate high-quality synthetic datasets for unlearning, outperforming baseline synthetic data and matching expert-curated datasets.

Findings

Synthetic datasets outperform baseline alternatives

Multi-step generation improves data diversity

Synthetic data matches expert-curated quality

Abstract

Modern large language models often encode sensitive, harmful, or copyrighted knowledge, raising the need for post-hoc unlearning-the ability to remove specific domains of knowledge from a model without full retraining. A major bottleneck in current unlearning pipelines is constructing effective forget sets-datasets that approximate the target domain and guide the model to forget it. In this work, we introduce a scalable, automated approach to generate high-quality forget sets using language models themselves. Our method synthesizes textbook-style data through a structured prompting pipeline, requiring only a domain name as input. Through experiments on unlearning biosecurity, cybersecurity, and Harry Potter novels, we show that our synthetic datasets consistently outperform the baseline synthetic alternatives and are comparable to the expert-curated ones. Additionally, ablation studies reveal that the multi-step generation pipeline significantly boosts data diversity, which in turn improves unlearning utility. Overall, our findings suggest that synthetic datasets offer a promising path toward practical, scalable unlearning for a wide range of emerging domains without the need for manual intervention. We release our code and dataset at https://github.com/xyzhu123/Synthetic_Textbook.