Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2407.00075 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866917939780780032 |
|---|---|
| author | Xue, Anton Khare, Avishree Alur, Rajeev Goel, Surbhi Wong, Eric |
| author_facet | Xue, Anton Khare, Avishree Alur, Rajeev Goel, Surbhi Wong, Eric |
| contents | We study how to subvert large language models (LLMs) from following prompt-specified rules. We first formalize rule-following as inference in propositional Horn logic, a mathematical system in which rules have the form "if $P$ and $Q$, then $R$" for some propositions $P$, $Q$, and $R$. Next, we prove that although small transformers can faithfully follow such rules, maliciously crafted prompts can still mislead both theoretical constructions and models learned from data. Furthermore, we demonstrate that popular attack algorithms on LLMs find adversarial prompts and induce attention patterns that align with our theory. Our novel logic-based framework provides a foundation for studying LLMs in rule-based settings, enabling a formal analysis of tasks like logical reasoning and jailbreak attacks. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2407_00075 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference Xue, Anton Khare, Avishree Alur, Rajeev Goel, Surbhi Wong, Eric Artificial Intelligence Computation and Language Cryptography and Security Machine Learning We study how to subvert large language models (LLMs) from following prompt-specified rules. We first formalize rule-following as inference in propositional Horn logic, a mathematical system in which rules have the form "if $P$ and $Q$, then $R$" for some propositions $P$, $Q$, and $R$. Next, we prove that although small transformers can faithfully follow such rules, maliciously crafted prompts can still mislead both theoretical constructions and models learned from data. Furthermore, we demonstrate that popular attack algorithms on LLMs find adversarial prompts and induce attention patterns that align with our theory. Our novel logic-based framework provides a foundation for studying LLMs in rule-based settings, enabling a formal analysis of tasks like logical reasoning and jailbreak attacks. |
| title | Logicbreaks: A Framework for Understanding Subversion of Rule-based Inference |
| topic | Artificial Intelligence Computation and Language Cryptography and Security Machine Learning |
| url | https://arxiv.org/abs/2407.00075 |