Want to get more from AI? Treat every prompt like an experiment

TL;DR - A Nature comment piece by James Dewar arguing that researchers should treat each AI prompt as a designed experiment whose output is a result requiring verification, offered as ten practical tips. It matters because it reframes everyday LLM use as a methodological practice subject to the same scrutiny norms as lab work.

• Core thesis: an AI query is an experiment, not an oracle — every output is provisional evidence that must be checked before use.
• Implies scientific-method habits carried over to prompting: stating the question precisely, controlling/varying inputs, and validating outputs against independent sources.
• Framed as ten actionable tips for practitioners rather than empirical results; it is opinion/guidance, not a study, despite appearing in Nature.
• Content available here is thin (title plus a one-line abstract), so the specific ten tips and any supporting examples are not visible and are not reproduced above.