2026 · Fourth edition
Lecture slides & challenge
Lectures on agents and evaluation, continual learning, and LLM data and RL environments, plus the AI Respondents Challenge dataset and starter kit. More sessions are being cleared for release.
Nuffield College, Oxford
A five-day intensive for early-career social scientists working with language models.
Mornings: technical lectures from ML practitioners and invited researchers. Afternoons: hands-on labs and a week-long collaborative research project. Tuition is free for all admitted participants.
The fourth edition ran 13–17 July 2026. Dates and applications for 2027 will be announced here.
4th Year running
30 Places
5 Days
£0 Tuition fee
Four years in
A participant-to-co-author pipeline.
The school is designed around a participant-to-co-author pipeline. Collaborative projects begun during the week continue beyond it: in 2025, a team of participants and speakers submitted a joint paper for peer review.
2026 Fourth edition · July 2026
Six teams. A week-long common-task challenge.
A technical programme taught by practitioners from Meta, Nebius, and Liquid.AI, wrapped around the AI Respondents Challenge: six teams predicting individual survey responses, scored on progressively harder held-out data through the week.
Read the 2026 highlights →
2025 Third edition · September 2025
30 researchers. A prediction competition. One paper submitted.
Lectures on LLM fundamentals, evaluation, fine-tuning, AI agents, and AI safety. A Kaggle-style prediction competition ran alongside the research project. A team of participants and speakers continued the work after September and submitted it for peer review.
Read the 2025 highlights →
2024 Second edition · September 2024
30 participants from ~200 applications.
Applied tutorials on RAG, observability, and agent-based systems. A cross-discipline line-up of NLP engineers, social scientists, and practitioners from Qdrant, Ori Cloud, Arize, and Google.
Read the 2024 highlights →
2023 Inaugural edition · September 2023
The first Oxford LLMs workshop.
Convened by Elena Voita and Ilya Boytsov for a cohort of ~30 early-career researchers exploring transformers, text-as-data methods, and the emerging landscape of large language models.
Read the 2023 highlights →
Open access
Slides and notebooks from past editions.
All lecture slides and coding materials are published openly after each edition. Datasets and model checkpoints from collaborative research projects are on Hugging Face. Materials go up as speakers clear them, and we are editing the lecture and research talk recordings for open release. Watch this space.
Challenge dataset 2025 · Third edition
Coding seminars
Hands-on notebooks on LLM fundamentals and on post-training: supervised fine-tuning, DPO, and GRPO. Remaining lecture slides are being collected from speakers.
GitHub repository → 2024 · Second edition
Slides, notebooks & datasets
2023 · Inaugural
Slides & notebooks
GitHub repository →