gpt2-15b-params-webtext-40gb-feb-2019

Summary

This records the starting point for GPT-2: a model with 1.5 billion trainable numbers, taught purely by reading through 8 million web pages without any human-labeled examples, in early 2019. It matters because it sets the baseline scale from which later models grew, and it shows that the system treats this as a fixed historical fact rather than something derived from other claims.