OpenAI's 2019 language model, 1.5 billion parameters — the one OpenAI initially refused to fully release citing misuse concerns, which is now remembered as both a historic overcorrection and the moment the "scaling laws" thesis became culturally legible. GPT-2 is the ancestor of every model shaping Karthik's work on SupportWire today; it's worth remembering how fast the ground has moved in under a decade.