Traces of Memorisation in Large Language Models for Code

Abstract

This talk is about my paper which was recently accepted at ICSE’24. In the paper, we explore the issue of memorisation in LLMs4Code. I will briefly discuss the paper itself, the experiments and the results. I will also talk about how the paper came to be, the initial idea and most importantly, the experiments and attempts that either failed miserably or never made it into the final paper at all.

Pre-print avaliable here