IF CODING IS SOLVED, WHAT NOW?: MEASURING THE SLOPPINESS OF CODE

Summary from the TLDR newsletter

Large language models are able to write almost perfectly correct code. It is pretty straightforward to get a model to generate code and then let that code be checked by hidden tests. Checking the sloppiness of this code often requires human intuition and time, and is an extremely difficult task in general. This post illustrates why this is by going through possible ways of measuring slop.