Slopfest #93 · 2019 · Vienna
Too Dangerous to Release
OpenAI said GPT-2 was too dangerous to release, then released it. We did the same with the afterparty.
Museum status: In storage: Basement B, next to the Christmas decorations.
- Venue
- Semperdepot, the event hall
- Attendance
- 57,100
- Fun fact
- 1.5 billion parameters, 1 paragraph released
What happened
The committee wrote a festival program so persuasive that it was judged too dangerous to publish, and released only the first paragraph. Rumor filled the gap. By October, guests were expecting a herd of English-speaking unicorns from the Andes, and the committee, not wishing to disappoint, hired two ponies and a man from Graz.
“All models are wrong, but some are useful.”
Elsewhere in 2019
OpenAI withholds its GPT-2 language model over misuse fears, releasing it in stages; a demo tells of unicorns in the Andes.
Tip: use ← and → to flip through the archive.



