Intelligently choosing when and where to log by considering possible paths through basic blocks: https://blog.acolyer.org/2017/11/03/log20-fully-automated-optimal-placement-of-log-printing-statements-under-specified-overhead-threshold/
(Improves both runtime performance and debuggability!)
miniblog.
Related Posts
Design principles for autocomplete: https://jeremymikkola.com/posts/2019_03_19_rules_for_autocomplete.html
(I've read persuasive defences of sorting autocomplete more intelligently than alphabetically. Nonetheless, the articles lists a ton of great heuristics.)
I'm intrigued by the idea of deep learning systems 'hallucinating' high resolution images from low res. It makes sense: if it's a photo of known type (e.g. people, birds) you have a lot of additional data to interpolate intelligently.
Going beyond QuickCheck: using an SMT solver (based on Rosette) to intelligently generate test inputs that are diverse and hit interesting code paths!
https://youtu.be/Br16rvT_C00