Applied AI Tools for Scientific Programming

Scientific workflows: Tools and Tips 🛠️

Selina Baldauf

2025-12-18

What is this lecture series?

Scientific workflows: Tools and Tips 🛠️

📅 Every 3rd Thursday 🕓 4-5 p.m. 📍 Webex

  • One topic from the world of scientific workflows
  • Material provided online
  • If you don’t want to miss a lecture subscribe to the mailing list
  • For credit points: Send me a short message (Email or Webex)

Motivation

  • Speed up repetitive tasks
  • Learn new methods and languages
  • Support debugging, documenting, refactoring, …

AI tools for programming

  • Browser-based chat bots (ChatGPT, etc.)
  • Web-tools for data analysis (Data analyst GPT, JuliusAI, …)
  • IDE-integrated AI tools (GitHub Copilot, Codium AI, …)

Integrated tools

  • know your project structure
  • see your code and console output
  • respond in context while coding

Today

GitHub Copilot inside the IDE Positron

  • How to use it
  • Limitations and responsible use

Let’s get started

GitHub Copilot in Positron

GitHub Copilot

  • Integrated and context-aware
  • Standard tool for different environments
  • Free for academics with an educational account

Positron

  • General-purpose data science IDE
  • Built by the same team behind RStudio
  • Based on VS Code
  • Complete GitHub Copilot integration

Code auto-complete

Copilot predicts what you want to write based on

  • Current code
  • Open files
  • Comments

Get better suggestions

  • Provide context
    • Good comments and names for functions and variables
    • Open other relevant files
  • Adopt a good coding style
    • GitHub Copilot will imitate your style
  • Acceptance discipline
    • Don’t auto-accept everything
    • Modify suggestions if needed

The default context

  • Context is indicated at the top of the chat
  • Default context: current file + console output
  • If you highlight code, the highlighted code is added as context

The “Add context” button

  • Add other files and folders as context with the “Add context” button

Use @, # or copy-paste

@workspace adds all files in the project

#file adds specific files to the context

You can also copy paste images (e.g. screenshots) as context in the chatbox

Advanced features

  • Project-level instruction files
    • Give instructions that apply to the whole project
    • Special file in /.github/instructions.md
  • Agents: Automate multi-step workflows (use with care)

Limitations and risks

  • Can suggest
    • Outdated functions and code
    • Patterns that are common but not best practice
  • GitHub Copilot has no understanding
  • Predicts plausible code not necessarily correct code
    • Code can be subtly wrong or inappropriate
  • Risk of over-reliance

Responsible use

  • Don’t treat AI as an authority: review and double-check
  • Use version control as a safety net: review and trace changes
  • Privacy: Code, comments and console output may be shared with model providers
    • Beware if you have sensitive data
  • Check institutional and journal guidelines
  • Transparency: Disclose your use of AI tools
  • You are responsible for your scientific output

Conclusion

  • GitHub Copilot can be a great assistant
  • Support but not replace

Getting started

  1. Get GitHub Copilot for free with your educational GitHub account
  2. Install Positron
  3. Setup GitHub Copilot in Positron

Check out the lecture website for all relevant information

Next lecture

Topic t.b.a.


📅 15th January 🕓 4-5 p.m. 📍 Webex

🔔 Subscribe to the mailing list

📧 For topic suggestions and/or feedback send me an email

Thanks for your attention :)

Questions?