Neat idea in LLVM: adding a pass to ensure that generated code actually respects the requested calling convention:
miniblog.
Related Posts
Generated Code Generates Overconfident Coders: https://www.deeplearning.ai/the-batch/issue-180/
A study of programmers found that using a LLM for completion produced buggier code but users were more confident in it.
I wonder if this generalises to other completion techniques?
A survey of parser generator usage in major language implementationss: https://notes.eatonphil.com/parser-generators-vs-handwritten-parsers-survey-2021.html
Most have a handwritten parser. GCC and Go moved from generated to handwritten. SQLite and Python <3.10 built their own parser generators!
Thoughtful article on the long term consequences of AI generated art, copyright, the art industry and historical innovations (e.g. the invention of the camera):
