What it does
jeff uses independent weights, not TypeSafe's model. Choice and Score questions share an encoder pass by default; Noul questions run separately, and normalized scores are not calibrated probabilities.
Benchmarks & research
jeff
Use a self-hosted GLiFormer model behind a Jev-shaped API, with published comparisons showing where its answers differ from hosted Jev.
Bookmark: jeff
Keep this in your collection.
Leave a noteWhat would you try with this?
: jeff
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
To sort messages on your own computer, ask a developer to set up jeff. Start with its example about a billing message. It runs a different AI model from Jev, even though you can ask questions in the same format. Do not expect identical answers.
Try asking several questions together, then ask them separately. Questions that pick an option or rate something can affect each other. Yes-or-no questions run separately by default. Separating every question takes more work. The displayed scores are not proven chances that an answer is right.