07. Causal Inference for Generative AI
Generative AI
Causal Inference
Evaluation
Lecture Notes
Causal evaluation of prompts, model upgrades, RAG, hallucination reduction, judges, safety interventions, personalization, and GenAI product outcomes.
This course specializes causal evaluation for generative AI systems, where outputs are stochastic, measurement is difficult, and human behavior often changes in response to generated content.
Notebook links open rendered HTML pages generated from the source notebooks under notebooks/lectures/. Code is visible by default; rendering is configured not to execute live notebook code, so local LLM or GPU-heavy cells are not triggered during website builds.
Notebook Sequence
- 01. Generative AI as a Causal System
- 02. Estimands for Generative AI Changes
- 03. Prompt Changes as Interventions
- 04. Model Choice and Model Upgrade Experiments
- 05. Decoding Parameters, Temperature, Top-p, and Output Variability
- 06. RAG, Context, and Grounding Interventions
- 07. Hallucination Reduction as a Causal Estimand
- 08. LLM Judges, Human Labels, and Measurement Error
- 09. Preference Data and Causal Interpretation
- 10. Safety Interventions and Policy Evaluation
- 11. User Trust, Reliance, and Overreliance
- 12. Human Creativity, Productivity, and Quality Effects
- 13. Code Generation and Developer Productivity
- 14. Synthetic Data Generation and Causal Validity
- 15. Multimodal Generative AI Evaluation
- 16. Personalization, Memory, and Long-Context Effects
- 17. Marketplace, Content, and Creator Ecosystem Effects
- 18. Interference, Contamination, and Feedback Loops
- 19. Observational Evaluation of Generative AI Rollouts
- 20. Cost, Latency, and Quality Tradeoffs
- 21. Monitoring, Regression Testing, and Eval Drift
- 22. Executive Readouts for Generative AI Experiments
- 23. Governance, Auditability, and Content Risk
- 24. Capstone: Evaluating a Generative AI System
How To Read This Track
- Work through the notebooks in order if you want the full course arc.
- Treat each notebook as a lecture plus lab: read the discussion, inspect the code, and rerun locally when you want to experiment.
- For AI-heavy notebooks, expect some brittleness when live model calls are enabled; that instability is part of the course material rather than something hidden from the reader.
The .ipynb sources remain in the matching folder under notebooks/lectures/.