System OneSystem One Models

Glossary

The words that come up when you read about System One AI models, defined once and linked from everywhere else on the site.

C

Calibrated ProbabilityA calibrated probability is a number from a model that means what it says: across all the answers given at 80%, about 80% are right.
CalibrationCalibration is the match between a model's stated probabilities and how often it turns out to be right.
ChoiceA Choice is a System One question type for selecting one option from a defined set.
Composite scoringComposite scoring is one of TypeSafe's documented patterns, described as "Combine several dimensions of analysis into a single score." You ask a separate question about each dimension, then your own code weights and combines the answers.
ConfidenceConfidence is a number from 0 to 1 that TypeSafe computes from the probability distribution an answer already carries.
Confidence GatingConfidence gating is the practice of branching your code on a model's confidence value as well as on its answer.

D

Decision ModelDecision model is the generic label people reach for when describing an AI model whose output is a typed answer with a probability attached rather than prose.

E

Expected calibration errorExpected calibration error (ECE) is the average gap between a model's stated confidence and how often it is actually right.

H

HallucinationHallucination is when a language model states something false in generated text.

I

Intent routingIntent routing is one of TypeSafe's documented patterns: "Classify a user's intent and route to the appropriate handler." A Choice question returns the intent plus a probability for every option, and your code maps the winning option to a handler.

J

N

Q

QuestionA question is the typed object you attach to state in a System One request.

R

RLCDRLCD stands for Reinforcement Learning for Calibrated Decisions, the training method TypeSafe AI says it used for Jev.

S

T

Z

Zero-shot classifierA zero-shot classifier assigns text to labels it was never trained on, using the label descriptions you supply at call time rather than labelled examples.