Classification is predicting which category an input belongs to: spam/not-spam, fraud/legitimate, intent of a support ticket, sentiment of a review. It is the m
Predicting which category an input belongs to, spam or not-spam, fraud or legitimate, the intent of a support ticket, the sentiment of a review. It is the most common business ML task and now spans classic trained models and LLMs prompted with label definitions.
It is an economic choice. Prompted LLMs win on flexibility and cold start; trained classifiers win on cost and speed at volume. Mature pipelines often start with prompts and graduate to small fine-tuned models as volume grows.