Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.27083 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914604924272640 |
|---|---|
| author | Ye, Xiaotian Wang, Xiaohan Zhang, Mengqi Wu, Shu |
| author_facet | Ye, Xiaotian Wang, Xiaohan Zhang, Mengqi Wu, Shu |
| contents | Counterfactual tuning (CFT) has emerged as a promising paradigm for Large Language Model (LLM) unlearning by training models to generate alternative fictitious knowledge in place of undesired content. However, in this work, we find that this paradigm still underperforms other paradigms in some aspects, and identify two previously overlooked pitfalls underlying this gap: (1) knowledge conflict, where mutual inconsistencies within counterfactual corpora induce conflicting gradients that disrupt parameter optimization, and (2) hallucination spillover, where fitting false targets instills a persistent fabrication bias, inflating hallucination rates on unrelated domains. To systematically diagnose these issues, we introduce RWKU+, an extended benchmark equipped with novel trade-off metrics and gradient-level diagnostic tools. Our work further discusses the limitations and overhead of the paradigm, aiming to provide insights and actionable guidance for more rigorous LLM unlearning research. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_27083 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | On the Hidden Costs of Counterfactual Knowledge Training in LLM Unlearning Ye, Xiaotian Wang, Xiaohan Zhang, Mengqi Wu, Shu Computation and Language Cryptography and Security Counterfactual tuning (CFT) has emerged as a promising paradigm for Large Language Model (LLM) unlearning by training models to generate alternative fictitious knowledge in place of undesired content. However, in this work, we find that this paradigm still underperforms other paradigms in some aspects, and identify two previously overlooked pitfalls underlying this gap: (1) knowledge conflict, where mutual inconsistencies within counterfactual corpora induce conflicting gradients that disrupt parameter optimization, and (2) hallucination spillover, where fitting false targets instills a persistent fabrication bias, inflating hallucination rates on unrelated domains. To systematically diagnose these issues, we introduce RWKU+, an extended benchmark equipped with novel trade-off metrics and gradient-level diagnostic tools. Our work further discusses the limitations and overhead of the paradigm, aiming to provide insights and actionable guidance for more rigorous LLM unlearning research. |
| title | On the Hidden Costs of Counterfactual Knowledge Training in LLM Unlearning |
| topic | Computation and Language Cryptography and Security |
| url | https://arxiv.org/abs/2605.27083 |