SciSpace AcademyExplore SciSpaceExplore
Tutorial

How to Write AI Prompts for Academic Research

Omkar Jadhav, Product Manager, SciSpaceEnglishStart here

Watch the tutorial37 min

TL;DR

Learn to prompt an AI research agent with the CORE framework for structured literature reviews, and how file formats and context affect results.

Who it's for

This session is for researchers, PhD students and academics who use SciSpace Agent and want more specific, less generic output.

What you'll learn

  • The CORE prompt framework: Context, Output, Role and Exclusions, with examples such as asking the agent to act as a peer reviewer or to limit databases and dates.
  • A live comparison of a lazy prompt and a CORE prompt for a literature review, ending in Deep Search results and a DOCX literature review.
  • How one line in a prompt makes SciSpace Agent plan before it runs a task, followed by a run that includes a literature search, an image and an ROI report.
  • Which file formats work best as agent inputs, with a test showing that a DOCX upload used about 3.5 times the credits of the same content in Markdown.
  • How to name and describe attachments so the agent reads only the relevant files.
  • What a context window is, why to trim PDFs before uploading, and when to start a new thread.

Steps shown in the video

  1. How to give an AI agent context and specify the output format

  2. How to name and describe uploaded files for an AI agent

Chapters

  1. Why does an AI research agent give generic answers?
  2. What is the CORE prompt framework for AI agents?
  3. How to give an AI agent context and specify the output format
  4. Role and exclusions in AI prompts: peer reviewer, databases, dates
  5. Lazy prompt for an AI literature review: what goes wrong
  6. Writing a CORE prompt for a structured literature review
  7. Why tell an AI agent what to exclude, not what to include?
  8. CORE prompt results: Deep Search and a DOCX literature review
  9. Why AI agents should plan before running a research task
  10. The one line that makes SciSpace Agent plan first
  11. Watching the agent run its plan: literature search, image, ROI report
  12. Which file formats work best as AI agent inputs?
  13. DOCX vs Markdown upload: 3.5x credit difference
  14. How to name and describe uploaded files for an AI agent
  15. Describing attachments so the agent reads only relevant files
  16. What is an AI agent's context window?
  17. When should you start a new chat with an AI agent?
  18. Trim PDFs before uploading for better AI output
  19. Start a new thread for every new task
  20. Recap: five tips for better results from an AI research agent

Questions this video answers

What is the CORE prompt framework for AI agents?

The CORE prompt framework: Context, Output, Role and Exclusions, with examples such as asking the agent to act as a peer reviewer or to limit databases and dates.

Watch this part of the tutorial
Why tell an AI agent what to exclude, not what to include?

Key takeaways: tell the agent what to exclude, not only what to include.

Watch this part of the tutorial
Which file formats work best as AI agent inputs?

Which file formats work best as agent inputs, with a test showing that a DOCX upload used about 3.5 times the credits of the same content in Markdown.

Watch this part of the tutorial
What is an AI agent's context window?

The CORE prompt framework: Context, Output, Role and Exclusions, with examples such as asking the agent to act as a peer reviewer or to limit databases and dates.

Watch this part of the tutorial
When should you start a new chat with an AI agent?

What a context window is, why to trim PDFs before uploading, and when to start a new thread.

Watch this part of the tutorial

Do this in SciSpace

Do this in SciSpace