Developer Tracks Slop Artifacts in Claude Opus Code Comments
A Stanford researcher plotted em dash and negation artifacts in AI-generated code across model versions.
TLDR
Zhizhuo Zhou posted observations from using Claude for coding tasks. He reported noticing more slop comments and created plots of the two most frequent artifacts across successive Opus versions. Zhou added that AI writing has grown more circuitous and harder to understand overall. He stated plans to develop methods for quantifying AI slop in follow-up work. The post includes no data files or further measurements beyond the initial plots described in the text.
Combined views
4.2K
1 Source, first seen 32d ago