+19% slowdown on average was observed in the first METR RCT study published in July 2025: "Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity"
The spirit of this market is to compare both studies apples-to-apples. Hopefully METR will officially tell us what the second study's equivalent percentage figure is, that can be directly compared with the first study's +19%. If not, this market will resolve N/A.
The bounds are left inclusive and right exclusive except for "+15% to +20%", in which right is also inclusive.
If no second study is published before 1st August 2026 this market will resolve N/A.
Ugh, I'm starting to understand how they get a slowdown. Several times recently I've asked ChatGPT to read the docs for something and tell me why my code isn't working, it hallucinates something nonsensical, often making up parts of my code that don't exist. Then I read the docs myself and the answer is right there.