Polina Gordienko

Research Interests            

• AI Evaluation and Benchmark Design;
• AI Safety and Alignment; Epistemic Reliability;
• LLM Forecasting and Forecast Aggregation;
• Benchmark Gaming and Strategic Robustness;
• Computational Social Choice

Short Description

Polina is a doctoral researcher in statistics focused on a fundamental question: when should we trust an AI evaluation? Her work develops mathematical and statistical foundations for reliable evaluation, spanning robust benchmark design, multi-criteria model comparison, as well as the aggregation and updating of probabilistic forecasts. She studies how evaluations can remain coherent across competing objectives, stable under changing conditions, and resistant to strategic gaming. Drawing on statistics, computational social choice, and formal epistemology, her broader goal is to strengthen the evaluation infrastructure needed for reliable, safe, and aligned AI systems.
From January to September 2026, she was a Visiting Doctoral Researcher at the University of Oxford. She holds master’s degrees from LMU in both Statistics and Data Science (M.Sc.) and Philosophy (M.A.), having previously earned two bachelor’s degrees in Economics and in Philosophy and History.
Born in Ulan-Ude and raised in Minsk, Polina moved independently to Germany at the age of fifteen. From 2020 to 2026, she served as an elected member of a Munich district council, including as its vice-chair and as chair of its Social and Education Policy Committee. Her civic leadership was recognized with the 2024 Helene Weber Award, a national distinction for women in politics. She has also contributed to German media and public forums on democracy, migration, and Eastern Europe. Together, her research and public engagement reflect a broader commitment to making consequential technologies and institutions worthy of public trust.