Jev glossary
Evaluation & TrainingReinforcement learning with verifiable rewards
RLVR
Training against automatically verifiable rewards, which produced strong reasoning models that are slower and more expensive.
- Category
- Evaluation & Training
- Also known as
- Reinforcement learning with verifiable rewards
- Reading time
- 1 min
- Related terms
- 3
- Directory entries
- 1
- Docs
- docs.typesafe.ai
- Added
- 2026-09-24
Definition
RLVR created models that excel at tasks with checkable answers such as mathematics. The primer lists it as the second established adaptation path, with the trade-off that reasoning is slow and costly.
TypeSafe positions RLCD against it for a different job: fast, structured judgments where the reward is calibration rather than a verifiable final answer.
Tagstraining
Related terms
Definitions that connect to this one.
RLCDTypeSafe's post-training method: reinforcement learning that optimizes for calibrated decisions and probabilities instead of generated text.Evaluation & TrainingRLHFThe human-preference training method that turned pretrained models into chatbots, and the method TypeSafe deliberately did not use.Evaluation & TrainingSystem Two taskWork that needs extended, multi-step reasoning; the docs treat it as a signal to decompose rather than a job for Jev.Core Concepts
From the Jev directory
Directory entries that go deeper on this term.
Why TypeSafe trains decision models with RLCD instead of RLHF: calibrated probabilities where 0.2 outcomes happen about 20% of the time, and the case for machine-to-machine automation.
Jev Guides & ArticlesDocs#official#docs#evaluation
Official
From the Jev AI model community
Posts from builders shipping with Jev right now.