Show HN: Reproducing three AI text detectors, and why filtering hurt my model

Category: ai-ml

Tags: ai-detection, reproducibility, text-classification, experiment

Score: 5.3/10 (Innovation: 5, Technical: 5, Documentation: 7, Utility: 4)

Reproduces three AI text detectors (perplexity, near-duplicate similarity, embedding density) and evaluates their impact on a downstream sentiment classifier. The experiment shows that filtering AI-generated reviews can hurt model performance. The repository provides scripts and data to rerun the analysis.

Target audience: machine learning researchers, NLP practitioners

Repository: https://github.com/abduldattijo/ai-slop-detection-reproduction · Python · 2 stars

View on Hacker News