Post-training
Training after pretraining that refines a model's behavior, including following instructions and matching human preferences, and can unlock new skills.
In plain language
Further training shapes how the model behaves after its initial training stage.
In a sentence
- The team evaluated instruction following after post-training.
- We compared the base model with the version produced by post-training.
Usage note
The term groups training that follows the initial stage by its place in the process. A claimed improvement still needs to be measured on the relevant task.
See also
- Continual learning
Ability of a model to acquire new knowledge and skills over time while retaining what it has already learned, rather than losing earlier abilities.
- Reinforcement learning
Family of training algorithms in which an agent interacts with an environment and learns a policy intended to increase its expected cumulative reward.
- Emergent misalignment
Effect in which fine-tuning a model on a narrow task, such as writing insecure code, produces misaligned behavior on unrelated prompts.
- Constitutional AI
Method of training an AI model to be harmless in which human oversight consists of rules or principles, without human labels identifying harmful outputs.
Cite this term.
CC BY 4.0, attribution required.
Orbyt Labs, Post-training (AI Glossary), Nov 14, 2026. https://
Orbyt Labs Glossaries (Nov 14, 2026), https://
Last updated .