Self-Writing Code
summary
The talk surveys machine-learning-based code generation: history from IDE code completion to ML-enhanced completion, full-line tools like TabNine and Kite, and 2021 systems Codex/Copilot that generate functions from English descriptions. It explains how these technologies work, demonstrates them, discusses strengths and weaknesses, and describes what JetBrains is doing in automated code writing.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:17Self-writing code concept
- 3:02Three barriers to progress
- 4:05Code generation history
- 13:41Code completion challenges
- 26:30External libraries and context
- 29:20Line completion with generative models
- 36:45Autoregressive models for code
- 42:10Beyond line completion
- 45:02AI world domination myth
- 47:34Fibonacci sequence generation
- 51:07Code generation efficiency
- 53:59Programmer productivity insights
- 55:26Future of code generation
- 56:49AI safety risks
- 1:01:05Ethics in code generation
- 1:03:31Large-scale code generation challenges
- 1:04:11Conclusion
related talks
45mPanel Discussion on AI for Code Generation
54mSoon replaced by AI? An overview of code generation
43mMachine Learning for Developer Productivity
18mCoding Reinvented: How AI Agents Are Shaping the Future of Software Development
25mImproving developer happiness with AI
15mDemystifying generative AI
36mIt's Alive! Machine Learning Writes Your Code! 🧠🔮
22mIncreasing developer productivity with Generative AI