Information leakage in Artificial Intelligence systems, particularly in Large Language Models (LLMs), poses significant security and privacy concerns. Sensitive information such as the model's architecture details, system configurations, or proprietary data can unintentionally be exposed through model outputs or interactions. Detecting and mitigating such leaks are crucial for maintaining the integrity and confidentiality of LLM systems.
We seek excellent student assistants motivated to be part of ongoing research in these areas and contribute to cutting-edge research in LLM security. As a student assistant working on information leakage detection, you will contribute to developing methods and tools to identify and prevent unauthorized disclosure of sensitive information in LLM models. This work is essential for enhancing the security of LLM applications and ensuring compliance with privacy regulations.
Your tasks include:
Research existing methods for detecting information leakage in AI models, focusing on LLMs.
Collect datasets of LLM model outputs that may contain leaked information; analyze the data to understand patterns or features indicative of information leakage.
Implement prototype to detect information leakage in LLM outputs; test and refine the algorithms based on performance metrics.
If you are intrigued by this cutting-edge subject, please get in touch with us at info@trust.tu-darmstadt.de to obtain further information. To facilitate the process, please include a summary of your academic background and a copy of your transcripts.
Knowledge of machine learning concepts and familiarity with LLMs.
Proficiency in programming languages such as Python; experience with CUDA is a plus.
Strong analytical and problem-solving skills.
Ability to conduct thorough literature reviews and synthesize information.
Good communication skills for documentation and teamwork.
Ability to work independently and strong motivation
Stellenmerkmale
Dein Beschäftigungsumfang
Teilzeit (befristet)
Dein Gehalt
Nach Vereinbarung
Dein Arbeitsplatz:
vor Ort
Dein Büro:
Raum Darmstadt