How a Unicorn Story Set the Rules for AI Launches

In February 2019 an AI lab showed off a model that wrote a news story about English-speaking unicorns, and said it would not release the full model. The headlines called it too dangerous. Nine months later all of it was public, and the danger never quite arrived. This is what actually happened, and why almost every AI launch since follows the playbook GPT-2 started. CHAPTERS 0:00 Cold open: the unicorn story 0:46 I. A machine that guesses the next word 2:19 II. One and a half billion 4:05 III. Valentine's Day 5:50 IV. Too dangerous? 8:32 V. Nine months 11:16 VI. The playbook 13:12 So was it dangerous?