Artificial Languages are Dead. Long Live Artificial Languages!
summary
Claims that LLMs upend programming by automating translation and replacing many hand-designed languages, demonstrated with concrete experiments: a working tokenizer with ChatGPT produced runnable Python after iterative prompts and edits; GPT-4 commit messages generated for many diffs are often more informative than original dev messages; teaching a pre-trained model to emit verified Dafny code uses a small corpus (≈50 examples) plus semantic retrieval of the five nearest examples, multi-step prompting and a verifier in Visual Studio, producing ~100 correct verified algorithms.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:04Artificial languages history
- 4:57AI models automate programming
- 8:58LLMs beyond chatbots
- 10:04Live demo: LLM programming
- 19:36LLM limitations and prompts
- 21:22AI-generated commit messages
- 29:42Prompt engineering rise
- 31:28LLMs and niche languages
- 39:16Teaching LLMs DSLs
- 43:17Multi-step prompting
- 46:16Future of language design
- 49:31Effective communication and language models
- 51:17Q&A
related talks
34mProgramming Languages for AI Programing Agents
01h 28mFuture of Software Programming with AI Bots at Hand
01h 02mAI Effects on Research and Education: A Programming and Software Language Perspective
32mHow generative AI is Changing the way people interact with computers
26mLanguage Model Agents Enable Semi-Formal Programming
45mProgrammer’s Apprentice Season 2: Future Directions in AI-assisted Coding
47mHow AI Will Bring Computing to Everyone
57mMachine Translation of Human Languages in the Age of LLMs: Is This the End of the Language Barrier?