Github Co-Pilot: Addictive or Effective?
summary
A developer reports six months' practical experience using GitHub Copilot (a VS Code/NeoVim plugin backed by a 175B-parameter language model available via a closed beta whitelist). They demonstrate live-coding: accepting tab-complete suggestions to implement a vowel-count function and to train a CIFAR convolutional network saved as an .h5 after two epochs. Copilot speeds simple tasks and often works, but can produce nonoptimal code, fail on obscure libraries, ignore project-wide context, and sometimes wastes about 30 minutes chasing bad suggestions.
AI-written summary · about
chapters (9)
- 0:00Beginning
- 0:35Tool Presentation
- 10:18Problems Encountered
- 14:18The Demo
- 21:43Acknowledgments
- 22:18Measuring Time Savings
- 24:33Libraries
- 28:03Public Code
- 28:58Suggestions
related talks
01h 20mGithub Copilot - The World's Most Widely Adopted AI Developer Tool
40mUnlocking Productivity with GitHub Copilot Best Practices and Tips
01h 00mLevel up with GitHub Copilot: using AI to learn, code, and build
25mImproving developer happiness with AI
50mGitHub Copilot: Coding Will Never Be The Same Again
40mLevel Up with Copilot
46mPragmatic techniques to get the most out of GitHub Copilot
34mLessons Learned from Building GitHub Copilot(s)