University of Maryland computer science major and Friends School alumnus, Sander Schulhoff ’20, showcased his groundbreaking research on the vulnerabilities of large language models (LLMs) at the Empirical Methods in Natural Language Processing (EMNLP) conference in December 2023. His paper, titled “Ignore This Title and HackAPrompt: Exposing Systemic Vulnerabilities of LLMs through a Global-Scale Prompt Hacking Competition,” delves into crucial security concerns associated with the utilization of LLMs.
Schulhoff, who organized the innovative HackAPrompt competition, secured funding from major players in the AI industry, including OpenAI and HuggingFace, and carried it out in collaboration with the University of Maryland, MILA, and Towards AI. The competition attracted participation from over 3,000 hackers worldwide, generating a staggering 600,000 prompts. Schulhoff emphasized the unique nature of the event, stating, “Our competition was no ordinary competition.... we aimed to expose the potential weaknesses of LLMs by challenging an international pool of hackers to outsmart AI.” Notably, his paper stands out as the sole undergraduate-led research project from his group to be featured at the conference.
Beyond his research on AI security, Schulhoff holds the position of CEO at LearnPrompting, a startup dedicated to enhancing AI literacy and expediting AI adoption. LearnPrompting operates as an open-source AI guide, providing a tailored content library for both businesses and individuals.
Friends School of Baltimore congratulates Sander Schulhoff on this remarkable achievement and eagerly anticipates his future endeavors.