self-generated

Definition

Data or reasoning traces produced by the model itself rather than collected from humans or the web. It underpins synthetic-data pipelines and self-improvement loops, where a model samples candidate solutions, filters them with a verifier, and trains on the survivors. The attraction is that it scales past the limits of human-labelled data; the standing risk is error amplification, since unfiltered self-generated data can entrench a model's own mistakes.

Explainers using this term

• Building a Dataset in Practice — Collect, Clean, BlendSelf-Instruct: Aligning Language Models with Self-Generated Instructions