Automating Software Development with Deep Learning
summary
Claims deep learning can automate parts of development now by treating software as data; demonstrates an end-to-end model (Screenshot-to-code) that maps pixels to markup using a CNN + LSTM concatenation and a small DSL trained on large web archives (author reports a ~97% BLEU-style score on static page layout). Works well for static GUIs and UI components, fails on variable-heavy program logic, multi-system integration, and production-grade APIs; scale and reasoning remain unsolved.
AI-written summary · about
chapters (6)
- 0:00Start
- 0:50Software Development as Data
- 1:40Automating Software Development
- 21:28Undefined Tasks
- 21:54Novel Manifolds
- 22:39Scalability
related talks
43mMachine Learning for Developer Productivity
11mCan a robot code my website?
43mWill AI kill the developers job?
26mAl and Software: The Good, The Bad, The Scary
27mWill AI (Finally) Replace Developers?
39mAI and the Democratization of Software Development
24mHow AI is increasing the surface area for serendipity
45mIntroduction To Deep Learning