| Mr MXF

0001-01-01

~2 m

CoPilot 🔗	ChatGPT 🔗	Claude 🔗
ℹ️AI Model the nature of the AI in use model: architecture, hyperparameters generated dynamically e.g. AutoML	ℹ️AI Model the nature of the AI in use model: ❌ Unlikely - usually an engineer docs: ✅ Moderate - drafts by AI summaries: ✅ Moderate - AI evaluation summaries	ℹ️AI Model the nature of the AI in use model: ❌ Unlikely - usually an engineer bias docs: identify & document - demographic bias, group disparities, edge cases, failure modes docs: plain language explanations, capabilities, risk assessment
ℹ️AI Data the data used to make the AI function lineage: transformation logs provenance: real, synthetic or transformed data? annotation: tags, bounding boxes, entity labels sentiment scores: for human review or direct use	ℹ️AI Datasets the data used to make the AI function descriptive: ✅ High - AI generated descriptions / tags. provenance: ❌ Moderate - should be human + AI help quality / labeling: ✅ High - may be machine-generated ethical/privacy: ✅ Moderate - AI drafts + expert review	ℹ️AI Training Metadata the data used to make the AI function descriptive: AI insights - content distribution analysis quality / labeling: assessments & duplicate & cleaning ethical/privacy: Privacy risk assessments (PII) optimization: histories & reasoning, trade-offs, configs
ℹ️AI Inference Metadata generated during AI usage Confidence scores Predicted labels Explanation traces - Often real-time during inference	ℹ️AI Generated Content generated during AI usage watermarks / tags: ❌ Low - algorithmic, not AI “trained” usage: - logs and descriptions - ✅ Moderate - AI auto-summary / tag logs of generated outputs	ℹ️AI Inference Metadata generated during AI usage Explainability - AI self-explain - attention visualizations, feature importance, decisions, uncertainty quantification trend: AI systems becoming more self-documenting and self-evaluating
ℹ️AI Feature Metadata a copilot category feature - importance scores & statistical summaries synthetic labels generated or via expandability tools (e.g., SHAP, LIME)	ℹ️AI Governance & Compliance a chatGPT category transparency: ✅ Moderate–High \| AI draft of fairness / audit results Bias metrics: ✅ High - metrics explanation often AI-drafted	ℹ️AI Content Metadata a Claude category QA - AI doing QA on AI-generated content - accuracy, hallucination detection safety & toxicity, content quality, guideline conformance tagging & classification - AI generated tags: categories / topics, sentimentl, language