~/talk

Natural Language-Guided Programming

links:none on file
links:none on file
links:none on file
links:none on file
slides:none on file

Proposes enriching code with plain-English intent to drive completion and implements a prototype for Python data science that fine-tunes GPT-2 on ~300,000 Jupyter notebooks with three comment strategies (natural, docstring-inserted, none). Extracts 200 hand-validated test cases and gets human ratings showing the natural and docstring models produce about two thirds useful predictions while the no-comments model fails, notes Codex/Copilot outperform but at far greater scale and cost, and releases models, benchmarks and annotations.

AI-written summary · about

  1. 0:00Intro
  2. 0:39Motivation
  3. 2:04Natural Language-guided Programming (NLGP)
  4. 3:20Language models
  5. 5:11Prototyping an NLGP assistant for Python
  6. 8:37How good are the models?
  7. 9:21Creating a benchmark
  8. 10:36Evaluating the models
  9. 12:32Recent related work
  10. 13:58Conclusion
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.