Suspected AI-generated text in FineWeb-Edu retrieval results
A user says they retrieved “AI-generated dreck” from FineWeb-Edu, which they describe as supposedly the most educational 8% of FineWeb.
TLDR
After downloading FineWeb-Edu, a user says they retrieved text they identified as AI-generated, mocking it with “Good old N=NP.” They contrast that result with the dataset’s intended selectivity: FineWeb-Edu is supposed to be the most educational 8% of FineWeb, which they say is itself supposed to be the best 2% of CommonCrawl.
Combined views
10.4K
2 Sources, first seen ago
162 likes9 comments23 saves5 reposts
