Seyed Hamed Hassani

Title: Seyed Hamed Hassani: Inventor Advancing LLM Security.

Introduction

Seyed Hamed Hassani is an inventor based in Philadelphia, Pennsylvania, and is affiliated with the University of Pennsylvania. His work includes an invention focused on defending large language models against jailbreaking attacks.

Latest Patents

Hassani is associated with 1 patent: “Methods, Systems, and Computer Readable Media for Defending Large Language Models (LLMs) Against Jailbreaking Attacks.” The method receives an input prompt, creates multiple perturbed versions, and submits them to an LLM to generate responses. It then estimates the group effect of the perturbed prompts and selects a response that agrees with that estimated effect.

Career Highlights

Hassani’s patent addresses a security challenge in generative artificial intelligence: attempts to manipulate language models into producing unintended outputs. The approach uses multiple prompt variations and compares their responses to help identify an output consistent with the group effect.

Collaborations

Hassani’s listed coworker is Alexander Beck Robey. His University of Pennsylvania affiliation connects his work with an academic research environment.

Conclusion

Seyed Hamed Hassani’s patent presents a prompt-variation and response-aggregation approach to strengthening large language models against jailbreaking attacks.