Reddit user builds 103B-token Usenet corpus (1980-2013)

The corpus spans pre-web human-only discussions from 1980 to 2013, offering zero AI contamination. It gained 30K views on r/MachineLearning and is intended for fine-tuning local models.