cs249r_book/labs/plans/vol1/lab_01_ml_intro.md

📐 Mission Plan: 01_ml_intro (Deep Analysis)

1. Chapter Context

• Chapter Title: Introduction to ML Systems.
• Core Invariants: The AI Triad (D·A·M), The Bitter Lesson (Scale > Logic), and The Verification Gap.
• The Struggle: Moving from "Software 1.0" (Explicit Rules) to "Software 2.0" (Learned from Data).
• Target Duration: 45 Minutes.

---

2. Narrative Arc

You have claimed your track. Now, you must calibrate your "Architect's Intuition." You will witness the three physical laws that make AI Engineering a distinct discipline: the massive scaling gap, the take-off of learning systems, and the impossibility of brute-force testing.

---

3. The 3-Part Mission (The KATs)

Part 1: The Magnitude Gap (The AI Triad - 15 Mins)

• Objective: Quantify the 9-order-of-magnitude span of the ML landscape.
• The "Lock" (Prediction): "By what factor (ratio) does an H100 (Cloud) exceed an ESP32 (TinyML) in peak compute performance?"
• The Workbench:
  • Action: A slider that sweeps from TinyML -> Mobile -> Edge -> Cloud.
  • Observation: A Comparison Radar Chart and Dynamic Ratio Gauge.
  • The 5-Move Rule: Students must compare their specific chosen track against all 3 other archetypes.
• Reflect: "You say the gap is 10^9. Why does this physical asymmetry prevent a 'one-size-fits-all' software stack for AI? Justify using the D·A·M axes."

Part 2: Proving the Bitter Lesson (Historical Audit - 15 Mins)

• Objective: Contrast the "Feature Engineering" era with the "Deep Learning" era using real historical data.
• The "Lock" (Prediction): "If we increase human tuning effort by 10x, will it beat a 10x increase in machine scale over a 5-year period?"
• The Workbench:
  • Action: A "Historical Scrubber" slider (1980 -> 2024).
  • Observation: A Historical Accuracy Plot featuring actual model benchmarks (AlexNet, ResNet, GPT-3, GPT-4).
  • The 15-Iteration Rule: Students must "Step through Time" to see where the curves for "Rules" and "Learning" diverge.
• Reflect: "Reconcile this result with Richard Sutton's 'Bitter Lesson.' Why is human expertise a 'depreciating asset' in the current AI regime?"

Part 3: The Verification Gap Audit (Untestable Space - 15 Mins)

• Objective: Quantify the mathematical impossibility of exhaustive testing for your mission.
• The "Lock" (Prediction): "Can a test suite with 1 billion samples achieve even 1% coverage of your model's input space?"
• The Workbench:
  • Action: A Verification Calculator. Input resolution (e.g., 224x224) and test rate (samples/sec).
  • Observation: A Time-to-Test Counter. (Output: "Years to Test 1%: 10^300,000").
• Reflect: (Stakeholder Quality Lead): "Our CI/CD pipeline passed 100%. Why should we still invest in 'Continuous Monitoring' in production? Prove the necessity using the Verification Gap math."

---

4. Visual Layout Specification

• Primary: LandscapeRadar (Log-scale comparison).
• Secondary: ScalingLawPlot (Actual historical data points).
• Math Peeking: Toggle for the Degradation Equation math.