Michal Bravansky

Berkeley, California

Michal Bravansky

About

I work on AI R&D at METR, with a current focus on our uplift studies. I did my undergraduate studies in Computer Science at UCL, where I was fortunate to have my thesis supervised by Prof. Tim Rocktäschel through BOLD Lab. In a past life, I co-founded Verifee, supported by Google.org, Microsoft’s Democracy Forward program, and O2 Czech Republic.

Updates

  1. I joined METR to work on AI R&D uplift.

  2. Our paper on LLM forecasting is at the AI Forecasting workshop at ICML. See you in Seoul!

  3. Rethinking AI Cultural Alignment is at BiAlign at ICLR. See you in Singapore!

Earlier updates

  1. We released Dataset Featurization, along with the code.

  2. Our paper on cultural adaptability is at COLM. See you in Philadelphia!

Papers

2026

  1. Elicitation Format Drives Divergent LLM Geopolitical Forecasts

    Hariharan, Ghetiu, Weiler-Ofek, Pe, Kantasit, Michal Bravansky, Tang

    Forecast @ ICML

2025

  1. Dataset Featurization: Uncovering Natural Language Features through Unsupervised Data Reconstruction

    Michal Bravansky, Vaclav Kubon, Suhas Hariharan, Robert Kirk

    Preprint

  2. Rethinking AI Cultural Alignment

    Michal Bravansky, Filip Trhlik, Fazl Barez

    BiAlign @ ICLR

2024

  1. Evaluating Cultural Adaptability of a Large Language Model via Simulation of Synthetic Personas

    Louis Kwok, Michal Bravansky, Lewis D. Griffin

    COLM

  2. When XGBoost Outperforms GPT-4 on Text Classification: A Case Study

    Matyas Bohacek, Michal Bravansky

    TrustNLP @ NAACL

2023

  1. Czech-ing the News: Article Trustworthiness Dataset for Czech

    Matyas Bohacek, Michal Bravansky, Filip Trhlik, Vaclav Moravec

    WASSA @ ACL