Stop reasoning! When multimodal LLMs with chain-of-thought reasoning meets adversarial images
Recently, Multimodal LLMs (MLLMs) have shown a great ability to understand images. However, like traditional vision models, they are still vulnerable to adversarial images. Meanwhile, Chain-of-Thought (CoT) reasoning has been widely explored on MLLMs, which not only improves model’s performance, but...
Main Authors: | Wang, Z, Han, Z, Chen, S, Xue, F, Ding, Z, Xiao, X, Tresp, V, Torr, P, Gu, J |
---|---|
Format: | Conference item |
Sprog: | English |
Udgivet: |
IEEE
2024
|
Lignende værker
-
When LLMs meet cybersecurity: a systematic literature review
af: Jie Zhang, et al.
Udgivet: (2025-02-01) -
Adversarial Prompt Transformation for Systematic
Jailbreaks of LLMs
af: Awoufack, Kevin E.
Udgivet: (2024) -
SelfCheck: using LLMs to zero-shot check their own step-by-step reasoning
af: Miao, N, et al.
Udgivet: (2024) -
Philosophy, Adversarial Argumentation, and Embattled Reason
af: Phyllis Rooney
Udgivet: (2010-08-01) -
A comparison of chain-of-thought reasoning strategies across datasets and models
af: Konstantin Hebenstreit, et al.
Udgivet: (2024-04-01)