diff --git a/_events/2024-09-23_jindong-gu.md b/_events/2024-09-23_jindong-gu.md index fe133a6..7395581 100644 --- a/_events/2024-09-23_jindong-gu.md +++ b/_events/2024-09-23_jindong-gu.md @@ -1,5 +1,5 @@ --- -title: "Towards the Risks of Visual Inputs Brought into Multimodal LLMs" +title: "Toward the Risks Brought by Visual Input into Multimodal LLMs" abstract: "Recent advances in Large Language Models (LLMs) have demonstrated remarkable capabilities in processing and reasoning with textual data. By incorporating visual inputs, Multimodal LLMs extend these capabilities to understand and interpret images, achieving impressive results. Techniques such as Prompting, Chain-of-Thought Reasoning, and Alignment have been particularly effective in enhancing image understanding. In this talk, I will present my research on the risks associated with integrating visual inputs into Multimodal LLMs. Specifically, I will talk about how adversarial images can fool multiple prompts, mislead Chain-of-Thought inferences, and jailbreak the alignment of Multimodal LLMs. At the end, I will also discuss potential mitigation strategies of the risks." speaker: Jindong Gu,
Senior Research Fellow at University of Oxford and Faculty Researcher at Google Deepmind bio: Dr. Jindong Gu is a Senior Research Fellow at the University of Oxford. He also works at Google DeepMind as a faculty researcher in the Gemini Safety team. Prior to this, he received his Ph.D. degree from the LMU Munich in 2022, supervised by Volker Tresp. He has experience working at Google Brain, Microsoft Research, Tencent AI Lab, and Siemens Technology. His research focuses on AI Safety, especially, the safety of visual perception models, foundation models as well as general-purposed systems.