DevHacker Newsabout 1 hour ago
Measuring Code Slop: Verbosity and Erosion Metrics
1 min read
Earendil researcher Sebastian found LLM-generated code is roughly twice as verbose and eroded as human-written code, using metrics from SlopCodeBench. Verbosity measures duplicated lines, while erosion measures mass concentrated in complex functions, with agent code scoring 0.33 and 0.68 respectively versus human repos at 0.15 and 0.31.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
