Learning guide · experiment · Suggested 15–25 minutes
MicroGPT for AI leaders
About this resource · How to use it · Learning engineering connections
What is this resource?
A guided activity for MicroGPT for AI leaders. Explain why model behavior and a useful score depend on the training setup. Start with LLM101 if next-token prediction is unfamiliar. You do not need to master the mathematics to compare outcomes.
How should I use it?
- Read Before you begin, then use Open the project to try the listed steps alongside this guide.
- Record an observation, explanation, transfer question and next test in the reflection panel. Saving is optional; Print one-page guide produces a paper copy.
How does this connect to LENS and learning engineering?
This activity connects a capability question to observable work and a possible next decision. When does a better score stop being evidence of a better outcome? A tiny browser model is a teaching example; it does not establish the behavior of production models. Use the reflection to distinguish what you observed from what still needs evidence.
Explain why model behavior and a useful score depend on the training setup.
Before you begin
Start with LLM101 if next-token prediction is unfamiliar. You do not need to master the mathematics to compare outcomes.
Try this
- Train the tiny model and inspect its generated output.
- Compare checkpoints and describe what changed.
- Explore the engagement-label example; ask what the score rewards and what it misses.
Keep this guide open while you explore.
Reflect
What changed in your explanation after trying this?
Observation → Explanation → Transfer → Next test
Pause & make sense
What are you taking away?
What changed in your explanation after trying this?
Help me make sense of this
- Separate what happened from why. Your observation is the result; your explanation is a claim that may need testing.
- Check another explanation. What else could account for the result? What comparison would distinguish them?
- Try without the support. Could you explain or perform the task in a new situation? Identify what evidence you would need.
- Choose one next move. Return to the concept, compare another example, or test one changed assumption.
Choose your next move
Return to the process, or examine how evidence could change a design.
Choose another question on the map