If coding is solved, what now?: Measuring the sloppiness of code
Summary
The article argues that while LLMs can generate syntactically correct code, measuring the sloppiness of AI-generated code is still challenging. It introduces metrics such as Verbosity and Erosion from SlopCodeBench, discusses evaluation approaches (human vs. AI judges), and compares agent-generated code to human code, highlighting the persistence of slop and the limitations of current evaluation methods.