Sam Bowyer
I'm a probabilistic machine learning PhD student at the University of Bristol's Compass CDT. My research interests include probabilistic programming, deep learning, and using (mostly -- but not exclusively -- Bayesian) stats to improve LLM eval procedures. I've just finished an internship working on pretraining evals @ Cohere and will be rejoining full-time in September!
- links
- blog posts
- Bayesian LLM Finetuning (Compass Student Blog)
- Compass at AIUK 2025
- Special Report: Compass Away Day 2024
- presentations
- projects
- mrmr_eval -- a minimal tutorial on using mRMR for efficient LLM benchmarking
- bayes_evals -- a lightweight library for Bayesian analysis of LLM evals
- alan -- a massively parallel probabilistic programming language
- papers
- Efficient Benchmarking Is Just Feature Selection and Multiple Regression (preprint)
- Learning Generation Orders for Masked Discrete Diffusion Models via Variational Inference (ICLR 2026 DeLTa Workshop)
- Position: Don't Use the CLT in LLM Evals With Fewer Than a Few Hundred Datapoints (ICML 2025 Spotlight)
- Massively Parallel Expectation Maximization For Approximate Posteriors (AABI 2025)
- Using Using Autodiff to Estimate Posterior Moments, Marginals and Samples (UAI 2024)
- tennis