I feel like writing software documentation is like doing my taxes - help!
summary
Some scikit-learn and NALs docstrings are removed and GitHub Copilot is asked to regenerate examples, parameter docs and method docs; the tool produces a usable train_test_split example and an R2 calculation and even improves the get_params explanation, but also expands a 300-line docstring with redundant text and obvious advice. Conclusion: AI speeds drafting in well-defined contexts but cannot invent project rationale, often hallucinates, and always requires humans to double check outputs and curate audience-facing content.
AI-written summary · about
chapters (8)
- 0:00Introduction and motivation
- 1:26Why documentation matters
- 3:54Key aspects of good docs
- 8:28Can AI write docs?
- 14:06AI limitations and future
- 17:18Fun examples and inspiration
- 18:21Resources and references
- 19:34Q&A session
