My experience with AI tools

• Started with GitHub Copilot at mid-2024
  • currently Chat-GPT 4.1 Copilot
• Chat-GPT 5.2 for writer's block and casual Q&A
• Notebook-LM for text analysis and summarization
• Claude for coding assistance
• Radom AI tools offered as part of Gmail, etc

GREAT tool
with LIMITATIONS

• Very good at regenerating text, code, etc
• Should not be trusted in factual questions or reasoning
  • Hallucinations
  • Logical errors
  • Bugs in generated code
• What about ethics, data privacy, etc?
  • We have to remember and take good care

Feedback loop is essential

Treat AI as an RA who makes mistakes

• Manual checks

  • Verify outputs
  • Provide corrections
  • Update prompts

• Automatic feedback

  • Unit tests for code generation
  • Multiple agents for cross-verification (?)

Case study today

1. GitHub Copilot for autocompletion and code generation in VS Code

• main feature: very smart autocompletion
• configure your work environment for always-on experience

2. Notebook-LM for analyzing ANU policies and procedures

• main feature: references to source documents

Thank you!