What is computer science research?
On 22 October 2026, I spoke for about thirty minutes at the 3iL Ingénieurs Research Discovery Day, to second-year engineering students, over Teams and on several sites at once. The programme announced "Gemini, ChatGPT, Claude and disability: anatomy of an asymmetric guardrail", the title of the study I ran in August.
I opened with a lab-seminar talk, six-line title included, then interrupted myself: "Sir, we don't understand a thing!" We then started again from scratch, in four parts: what research is, what is particular about computer science, how to do some with next to nothing (the guardrail study, redone step by step), and what LLMs change about the job.
The thread of the talk is a grid of five conditions, built up as we went and kept at the bottom of the screen until the end:
- C0, understand: why it works, where it stops, what would make it fail;
- C1, name your observations;
- C2, write down in advance what refutes;
- C3, always a competing hypothesis;
- C4, test outside what you built.
The last one comes from computer science, where you build both the machine and the tests it passes. The grid is then applied to a viral screenshot, which meets none of the conditions, then to the study, and finally to AI tools themselves. These tools take over a good part of C1, and a little of C3 and C4; C0 and C2 still need someone to do them.
The slides are online, in French: C'est quoi, la recherche en informatique ?. Press Space to advance; to show everything at once, add ?statique to the address. The seven backup slides, kept for questions, come after the references.