Hello there! I'm Zihui Wu, a Ph.D. student from the 2022 cohort at Xidian University, diving deep into the fascinating world of AI and machine learning.

My academic journey began with a focus on adversarial examples and adversarial training. In May 2023, I brought to life a method called AdvFunMatch. This method was groundbreaking as it trained a ResNet-18 model to achieve SOTA robust accuracy. Sadly, despite my efforts and excitement, another team managed to publish similar work concurrently, which was a tough setback. However, it has only strengthened my resolve to push the boundaries in innovative research.

These days, I'm captivated by the security of large language models. This area is ripe for exploration and full of potential.

• July 2024: I unveiled JailbreakFunction, pioneering work that exposed jailbreak vulnerabilities in the function calling capabilities of large language models. This was a thrilling discovery that pushed boundaries!

• October 2024: I launched GlitchMiner, an innovative method designed to efficiently detect Glitch Tokens, ensuring models remain free from hallucination-inducing glitches.

Looking ahead, I am eager to continue my journey of discovery and hope to contribute more groundbreaking work to the field. I am optimistic that good fortune will accompany my endeavors in the world of academia, and I am particularly hopeful about building a network of collaborations that can transform solitude into synergy.

Thank you for stopping by! If my research intrigues you or if you have any insights to share, don't hesitate to reach out. You can contact me at [zihui@stu.xidian.edu.cn]. Let's connect and explore the future of AI together!