AI in Research @ CREST — 2026-2027 cycle

AI models and their tools — Claude Code among them — are moving fast, and many of us are still figuring out how to use them productively in research. Practices are far from settled: shifting from chat interfaces to agents that act on our machines; managing context, permissions, and subscription limits; citing AI assistance in papers; choosing the right model for mathematics, coding, or sensitive data; and deciding what to automate versus what still needs careful human judgment.

This 2026-2027 cycle accompanies the Claude Code licences offered to CREST members. It provides short 1h30 trainings, held in person and online, and recorded, to get started with the tool, show concrete uses and their limits, and exchange on how they change our work.

Organizers: Nicolas Brosse, Émilien Schultz. Questions and ideas: Discussions.

Upcoming trainings

Trainings take place in person and online, and are recorded. Dates are indicative and to be confirmed.

Getting started with Claude Code

November (TBC) · Émilien Schultz

From installation to a first real project. We set up Claude Code Desktop, go through the basics of working with an agent — giving instructions, letting it read and edit files, granting permissions — and point to the resources worth bookmarking. Hands-on: together, we build and publish a personal web page on GitHub Pages.

For everyone, no prior experience needed. Bring a laptop with your Claude Code licence activated and a GitHub account.

How Claude Code works: models, context and inference

November (TBC) · Nicolas Brosse

What actually happens when you send a request? We open the box: what a language model is, the main model families and how to choose between them, what the context window is and why it fills up, and how inference works and what it costs. Understanding these mechanisms explains much of the tool’s behaviour — its strengths, its mistakes, and its usage limits.

For everyone; no technical background required.

Integrating AI in a data science workflow

December (TBC) · Émilien Schultz

A complete analysis of survey data, run with Claude Code from start to finish. We cover how to structure a project so the agent works well (README.md, CLAUDE.md), how to stay in control with version control and permissions, how to manage context over long sessions (/clear, claude --resume), and how to track your usage.

For researchers who analyse data in R, Python or Stata. Attending “Getting started” first is recommended.

Doing mathematics with AI

December (TBC) · Nicolas Brosse

Can AI help with mathematical research? We look at what current models can and cannot do: exploring a problem, running numerical experiments, drafting proofs and checking them. Live examples from statistics, and a discussion of how to verify what the model produces.

For researchers working on theory or methodology.

Latest