RESEARCH NOTE · PAPER YEAR 2023

Finding Biases in Code Generation

By Spyros Mouselinos

What influences a language model’s solution to a coding problem? This first-author paper with Mateusz Malinowski and Henryk Michalewski examines the influence of function names, problem specifications, and examples.

Published in Findings of ACL 2023, the work develops an approach to identifying biases in code generation. The research sits within my broader interest in distinguishing robust reasoning from learned shortcuts.

Read “A Simple, Yet Effective Approach to Finding Biases in Code Generation”