Running
bluepencil
✏
A prose quality gate with a measured 5% false-positive rate
J. Alderman Lyell
wolfvswhale
AI & ML interests
Evaluation and measurement. Whether benchmarks measure what they claim, how detectors hold up across model generations, and the stylometry of machine-written prose. Recent work: a task-level audit of all 10,453 rubric criteria in GDPval.
Organizations
None yet