Weak supervision

Training or evaluating models on large volumes of cheap, noisy labels — increasingly produced by panels of LLMs labeling data — instead of costly hand-annotation.

grounded in: Latest-trends theme 'LLMs as evaluators and juries': using panels of LLMs to judge and label data is emerging as a practical building pattern.

Connected concepts

LLM jury, LLM-as-judge, Knowledge distillation, Evals & benchmarks, RLHF