FollowAgents AI AGENTS DIRECTORY
Home / AI Agent Glossary / RLCD (Reinforcement Learning for Calibrated Decisions)
Ecosystem & Emerging Terms

RLCD (Reinforcement Learning for Calibrated Decisions)

Also called: Reinforcement Learning for Calibrated Decisions

RLCD is the name TypeSafe AI gives to the reinforcement-learning method it used to train Jev so that the probabilities the model outputs are honest about how often it is right.

Most reinforcement learning for language models rewards a good outcome: a response a person prefers (RLHF), or an answer a program can check (rlvr). Neither directly teaches a model to say how sure it is. RLCD, as TypeSafe describes it, optimizes for "epistemically honest probabilities" on decision tasks, so that when Jev reports 80% it should be right about 80% of the time (see calibrated-confidence).

That matters for Jev because its answers arrive as probabilities that software acts on. A threshold like "act automatically above 0.9" is only meaningful if 0.9 really means about nine times in ten.

RLCD is TypeSafe's own name. In the launch material we reviewed, the detailed reward design and training data are not described, so anything beyond the stated goal is not public knowledge.

How it differs

RLCD vs. rlvr: RLVR rewards a correct final answer that a program can verify. RLCD, per TypeSafe, rewards the quality of the stated probability, so a confident wrong answer is penalized more than a hesitant one. They optimize different things and could in principle be combined.

Common misconceptions

Often assumed: RLCD guarantees Jev's probabilities are accurate.
Actually: It is a training goal, not a guarantee. Calibration can degrade on inputs unlike the training data, so check it on your own labelled examples.

FAQ

What does RLCD stand for?
Reinforcement Learning for Calibrated Decisions, TypeSafe AI's name for the training method behind Jev.
How is RLCD different from RLHF?
RLHF rewards responses people prefer; RLCD, as TypeSafe describes it, rewards probabilities that match how often the model is actually right on decision tasks.
Is RLCD an industry standard?
No. It is TypeSafe's own term, and its training details had not been published in the launch material we reviewed.

Part of the Jev topic guide — read the full explainer →

Last checked: 2026-09-21

Related terms

Jev (TypeSafe AI's System One Model)

Jev is a model from TypeSafe AI, announced in September 2026, that takes program state and typed questions and returns typed decisions with calibrated probabilities instead of generated text.

Calibrated Confidence

A model's confidence is calibrated when the probabilities it states match reality: among all the answers it gives at 80% confidence, about 80% should be correct.

RLVR (Reinforcement Learning with Verifiable Rewards)

RLVR is reinforcement learning where the reward comes from an automatic check of the outcome, such as whether code passes its tests or a math answer matches the known solution.

System One Model

A System One model is a class of AI model, named after fast intuitive thinking, that answers structured questions in a single fast pass and returns typed values with calibrated probabilities instead of generated text.