Snorkel
Day 3 after signup
Joules-to-GDP: How much useful work does a model do for the energy it uses?
Discover how AI models maximize economic value per joule.
Why it works
- Lead with a concrete research finding to give readers a clear reason to keep reading.
- Explain the company’s contribution before introducing the related resource, so the call to action follows naturally from the news.
How it is built
- Brief greeting
- Study finding
- Research contribution
- Dataset invitation
- Professional sign-off
Email copy
Hi there, Researchers at Stanford’s Hazy Research lab just released a new paper featuring tasks built in partnership with Snorkel AI’s research team. Joules-to-GDP examines how efficiently AI turns energy into economically valuable work. Gemma 4 31B delivered 2.5x the net economic value per joule of GLM 5.2, the model with the highest output value, while using less than a quarter of its energy. Joules-to-GDP uses professional tasks called GDPWorks, modeled after GDPval, mirroring the kind of multi-step work analysts, engineers, and other specialists do every day. Snorkel AI partnered to develop and expand the GDPWorks task set through our Open Benchmarks Grants program, contributing 83 novel tasks that bring the combined dataset's coverage to 80 occupations. If your team is training models for this kind of professional work, our GDPval+ dataset includes expert-authored tasks in the same format. You can request samples here. Regards, Bryan Wood Head of Forward Deployed Research, Snorkel AI Unsubscribe Snorkel AI, 101 2nd Street, , San Francisco,CA,94105,United States, Unsubscribe Manage Preferences