The language of AI data

Reinforcement Learning from Human Feedback (RLHF)

RLHF is a training method where people rank or rate a model's outputs, and those human preferences are used to train a reward model that then fine-tunes the AI to behave the way people prefer.

AI Red Teaming

AI red teaming is the practice of deliberately probing an AI model with adversarial inputs to find harmful, unsafe or incorrect behaviour before the model is deployed.

Data Annotation

Data annotation is the process of adding labels, categories or structured judgments to raw data (text, images, audio or video) so that it can be used to train or evaluate a machine learning model.

AI Model Evaluation

AI model evaluation is the process of measuring how well a model performs against a defined standard, using test data and criteria that reflect the real task the model is meant to do.

Ground Truth

Ground truth is the set of correct answers that a machine learning model is trained on and measured against, established by trusted human judgment or verified records rather than by the model itself.

Data Provenance

Data provenance is a verifiable record of where data came from and how it was produced, including who created each piece of it, when, and under what process.

Synthetic Data

Synthetic data is training or evaluation data generated by a model or an algorithm rather than collected from the real world or produced by human judgment.

Inter-Annotator Agreement

Inter-annotator agreement is a measure of how consistently two or more people assign the same labels to the same data, used to judge whether a labelling task is reliable.

EU AI Act (data and testing requirements)

The EU AI Act is the European Union's regulation for artificial intelligence, which sets obligations by risk level, including requirements around data quality, documentation and testing for higher-risk AI systems.