I put GitHub Copilot's AI to the test - its mixed success at coding baffled me In some challenges, the GPT-4-based model triumphed. In others, it failed. How do you know when to count on it?