typed/recipes

Glossary

The terms that come up on every page, in plain language.

State
The text you send for the model to judge: an email, a ticket, a post, a JSON document.
Choice
A question with a fixed list of options. The model returns the most likely option and a probability for every option. Jev allows up to 255 options.
Score
A question with ordered levels you write, from 2 to 10 of them. The answer is probability-weighted, so it can fall between levels.
Noul
A yes/no question. The answer is one number from 0 to 1: the probability the statement is true.
Confidence
How sure the model is of its Choice or Score answer. Your code uses it to decide whether to act or escalate.
Calibration
Whether a model's probabilities mean what they say. A calibrated model that says 0.8 is right about 80% of the time.
ECE
Expected calibration error. Lower is better. On the published typed-decisions benchmark, Jev scores 0.144 and Laya 0.213.
Zero-shot
Answering a new kind of question with no task-specific training.
Fine-tuning
Training a model further on your own labeled examples. Possible with Laya's open weights; not possible with Jev.
Cascade
A fast model handles every request and passes only the uncertain ones to a slower, more capable model.
RLCD
Reinforcement Learning for Calibrated Decisions, TypeSafe's name for the training method behind Jev.
System One model
A model that returns typed decisions with probabilities instead of generated text. The name comes from Kahneman's System 1.

Continue learning