Hacking Productivity with LLMs
summary
Eighteen months of experiments with GitHub Copilot and ChatGPT on R projects produce mixed, concrete results: generating an entire HTML widget package from scratch fails, refactoring input types yields messy but functional code overloaded with tests, and adding Shiny reactivity to an HTML widget succeeds after light review. Using an LLM as scaffolded tutor speeds learning, locating resources and debugging, while complex PyArrow Python/C++ work needs human mentorship (Alenka) for conventions and final decisions, highlighting the stretch zone sweet spot.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:46When to use LLMs
- 2:34Hard coding task
- 4:20Easy coding task
- 5:45Right LLM difficulty
- 7:31LLMs as tutor
- 13:27Human better choice
- 16:57LLMs and human balance
- 17:42Key takeaways
- 18:03Q&A
related talks
01h 15m10x Development: LLMs For the working Programmer
40mUnlocking Productivity with GitHub Copilot Best Practices and Tips
45mSupercharge your coding: Integrating ChatGPT and Copilot in long-term projects
31mGithub Co-Pilot: Addictive or Effective?
01h 02mPANEL - Unleashing the Future: The Impact of Gen AI on Development Productivity
57mUnlocking Success: Navigating the Code of Life with GitHub CoPilot
01h 12mAI-Assisted Programming Today and Tomorrow
10mRethinking collaboration: LLMs, teams, and cognitive load