Publications and talks

Papers and thesis

1. Antoni Czolgowski, Abel Iyasele. Cultural Misalignment in Large Language Models: Detection, Measurement, and Mitigation Through Targeted Fine-Tuning. arXiv preprint arXiv:2609.04485 [cs.CL], September 2026. Extended version of a paper published in the Proceedings of OSSConf 2026, Zilina, Slovakia.

   Abstract | PDF | DOI | Code and data

   Three open-weight models compared with World Values Survey Wave 7, with distributional measurement, bootstrap confidence intervals and evaluation of the cross-cultural effects of LoRA fine-tuning. Improving representation for targeted groups sometimes made it worse for others.

2. Antoni Czolgowski (2025). Does Artificial Intelligence Discriminate? An Analysis of Bias in Google's Gemma 2-2b Model. B.Sc. thesis, SGH Warsaw School of Economics.

   A comparison of Gemma 2-2b responses with World Values Survey responses across 10,692 demographic subgroups. I found that Gemma assigned women higher life satisfaction even though the pooled survey data showed no statistically significant gender difference.

Talks

• Cultural Bias and Alignment in Open-Source LLMs, with a live PacketLLM demonstration. Invited talk, OSSConf 2026, Žilina, Slovakia, 1–3 July 2026.

Press and mentions

• CU Boulder MS-DS Student Story, May 2026.
• CU Boulder MS-DS program feature, 2026.