About
A companion to the practical guide on using reasoning LLMs for science.
Prompting collects vetted prompts for common scholarly tasks — manuscript summaries, literature reviews, scientific editing, interactive journal clubs, mock qualifying exams, citation fact-checking, and professional correspondence — alongside a builder for composing your own. It accompanies the guide, whose guiding idea is to treat every AI query as an experiment, and every output as a result to be checked.
The RTIEC framework
A general structure for an effective prompt. Task, Information, and Constraints form the minimum effective prompt; Role and Examples are optional but can improve output on complex tasks.
- Roleoptional
- What persona the model should adopt (e.g. “an expert biochemist”).
- Task
- What you want the model to do (e.g. “summarize a manuscript”).
- Information
- Information needed for the task (e.g. the pasted manuscript).
- Examplesoptional
- Demonstrations of what the output should look like.
- Constraints
- Limits on how the output is presented (e.g. “< 200 words”).
Privacy
The library and the deterministic builder run entirely in your browser. Only the meta-prompt generatorsends text to a third-party model (via the Vercel AI Gateway). Don't enter confidential, proprietary, or unpublished information into it.
Source & license
The prompts are drawn from Large Language Models for Science and Research: A Practical Guide and the companion essay Treat every AI query as an experiment, by James M. Dewar (Department of Biochemistry, Vanderbilt University School of Medicine). They are licensed CC BY-NC 4.0 — adapt them for non-commercial academic use with attribution.
- Guide (Zenodo): 10.5281/zenodo.19177102
- Writing on AI for science: reasoningwithai.substack.com
- Contact: james.dewar@vanderbilt.edu