antagonlabs/

CritiqueBank-11M

CritiqueBank-11M The largest open dataset for training adversarial AI critics

CritiqueBank-11M is a large-scale dataset of 11.7 million adversarial critique examples designed to train AI systems that detect reasoning flaws in autonomous AI outputs. Each example pairs a rationale (an AI-generated trading signal) with a critique that identifies potential flaws, biases, or blind spots.This dataset powers MiniCrit, an… See the full description on the dataset page: https://huggingface.co/datasets/antagonlabs/CritiqueBank-11M.