Bio
Lukasz is a Computer Science PhD student at the University of Warwick and Founder of ControlAI, a startup specializing in automated AI auditing. He gained his Masters degree at Columbia and Bachelor at the University of Warwick. He is an alumnus of the AI Safety Camp.
Research interests
Lukasz works on the structural reasons for the emergence of vulnerabilities in large language models. His past research revolved around securing open source models against harmful fine tuning and various definitions of robustness in the context of large-language models. He is currently working on a theoretical paper outlining reasons for the emergence of jailbreaks and creating a foundational framework for AI testing analysis and an experimental paper outlining the implications of ciphers and languages on existing benchmarks measuring LLM performance.