Training & post-training
Data labelling
Assigning trusted categories, annotations or expected answers to examples.
Definition
Data labelling converts raw information into supervised examples using human experts, crowd workers, rules or other models.
Why it matters
Label quality, reviewer expertise and agreement are central to reliable financial-services models and evals.
Related concepts
- Supervised fine-tuning
Fine-tuning on curated examples of desired inputs and outputs.
- Golden set
A curated set of representative inputs and trusted expected outcomes.
- Inter-judge agreement
How consistently two or more graders assess the same outputs.