2026
Across 597,744 scored responses from nine open models at every effort setting, choosing effort per question would save 43.8% of tokens within half an accuracy point of always thinking hard.
I study when more reasoning is worth it.
PhD Fellow at Integreat and UiT The Arctic University of Norway. I work on test-time compute and reasoning efficiency in large language models.
Before that an MSc by Research at Oxford on an Oxford-Man Institute scholarship, a BSc at WU Vienna, and six years in quantitative finance and machine learning engineering in Zurich.
PhD ends in 2027. Open to research visits in spring 2027 and to postdoc and research roles from autumn 2027. Get in touch →
Research
2026
An exact decomposition of accuracy per token into answer availability, correctness and cost, over 785,541 responses on twelve benchmarks, where relative cost changes with the workload and can reverse.
arXiv (earlier version) Workshop version Explorable and BibTeX
2026
Logic puzzles whose length, intrinsic difficulty and distractor density move independently, which show task length as the dominant constraint across 22 reasoning models.
Paper (OpenReview) Interactive site Code Dataset Explorable and BibTeX
2020
The paper version of my Oxford thesis, machine learning on financial statements to forecast stock movements, in the top 0.3% of SSRN papers by downloads.
Essays and experiments
- Counterpoint The Spearhead Problem Language models raise the floor. I worry about what they do to the ceiling.
- Counterpoint The "Trustworthy AI" Trap Trustworthiness is a skill of the user, not a feature of the software.
- Note A slime mould on the front page What the thing on the front page is, how it works and why I put it there.
- Live Physarum 260,000 agents on your GPU, food you can plant.
- Site cogniload.dk.fo Results for every evaluated model, the dataset and the code.
- Generator Make a CogniLoad puzzle The reference generator, in the browser.
- Scorer Score a model on CogniLoad Accuracy and error buckets from model outputs, nothing uploaded.
- Explorer Model capacity Fitted response curves of all 22 models.
- Calculator Efficiency decomposition The gap between two models, split exactly.
- Toy Where to spend thinking Three ways to spend a token budget across questions.
Background
- 2024 – PhD Fellow, Machine Learning · Integreat and UiT
- 2022 – 24 Machine learning and data engineering · Orderfox, CH Media and Smallpdf, Zurich
- 2020 – 21 Quant · Rothschild & Co, Zurich
- 2019 – 20 ML research scientist · avax / Deep Impact, Winterthur
- 2018 – 20 MSc by Research, Engineering Science · University of Oxford
- 2014 – 17 BSc Economics · WU Vienna
In progress
- Whether a frozen language model can use another model’s internal states.
- An open dataset of scored responses at every effort setting of nine models.
- Forecasting where extra reasoning helps, model by model.
- The thesis, one argument about single questions and single models.