Module 2 · Proof: build useful evidenceWed, Sept 960 min

Model judgment for executives

Build enough model fluency to choose a sensible approach, question a confident claim, and set a useful quality bar.

Faculty: Prof. Ani Adhikari · Dr. Vinitra Swamy

Learning objectives

Supervised vs unsupervised learningGenAI & LLMsTokens, context, transformers (high level)Agentic systems

🧠 ML vs GenAI vs agents🧠 LLM limits🧠 Model ≠ agent

1. Read
Read the session guide above. Note the ideas that change how you see your organization's current AI question.
2. Test
Take the module quiz in Learn & Memorize to check understanding.
3. Drill
Run the spaced-repetition flashcards to lock in the terms from this module.

Question 1 of 6Model judgment for leaders

Supervised vs unsupervised learning
Supervised = model learns from labeled examples (X→Y). Unsupervised = model finds patterns in unlabeled data (clustering, segmentation).
LLM
Large Language Model: a deep network trained on vast text to predict next tokens; the engine behind generative AI. Limits: hallucination, context window, no true reasoning.
Agentic AI
Systems that use a model plus tools, memory, and loops to pursue a goal with limited supervision: vs a single model call. Example: Anthropic's agent patterns.
Data silos
Data trapped in separate systems/teams. The #1 blocker to AI: if data isn't accessible and quality, no model can save the use case.
RAG
Retrieval-Augmented Generation: feeding an LLM relevant retrieved context at query time to ground answers in your own data and reduce hallucination.
Build vs buy vs partner
Build = custom IP/control; Buy = vendor SaaS (faster, but vendor lock-in & data risk); Partner = co-build with consultants. Decision depends on data reality and core competency.

Build for Module 2: Use Case Map + Technical Brief + Prototype Demo