01Adamidis et al., 2026
"It's not the Language Model, it's the Tool: Deterministic Mediation for Scientific Workflows." — arXiv:2605.13245, May 2026.
Repeating the same analysis through a language model does not give the same result. The paper proposes a pattern in which the model picks and runs tools that work the same way every time — each one holding one researcher's exact procedure for one instrument — instead of writing analysis code, and tests it by running the same photoluminescence analysis on four platforms, four times each with the same request. The tool gave the same result in all runs, and the pattern is in daily use on two instruments serving users over about six months, on the local infrastructure where the data and the instrument are.