Choose the fix that matches the failure

Do not train a model to memorize tonight’s menu

Your assistant writes too much, misses a newly added dish, and forgets the user’s current budget. Those are three different failures. A single “better model” decision can hide the actual cause.

Use prompting for task instructions and examples, retrieval for relevant external information, and explicit state for current user choices. Consider fine-tuning when a repeated behavior remains deficient after a sound baseline and you have suitable examples.

Check your understanding

The assistant misses a menu item added today. What should you inspect first?

Separate changing facts from repeatable behavior

A daily price change belongs in a source of current data. A consistent output convention might be improved through instructions, examples, structured output, or eventually training. An incorrect total belongs in arithmetic code.

Fine-tuning changes learned behavior through examples; it does not create a live connection to a menu. Before investing, write down the failure frequency, the target behavior, the available data, and how you will measure improvement.

Check your understanding

Match the failure to the first intervention.

Make an intervention memo

The Adaptation Lab presents specific failures. Choose an intervention and inspect the rationale, including why attractive alternatives miss the cause. This is decision practice, not a model benchmark.

Your memo should name a baseline, one change, an evaluation set, and a stop condition. For example: try two concise examples, then compare held-out requests. Escalate only if the measured shortfall remains important.

Check your understanding

What evidence would justify investigating fine-tuning?

Build checkpoint

Open the Adaptation Lab and follow the practice above. Each lab explains its inputs, its actual logic, and its limits. Save your observations below; your check-ins and reflection remain in Product Academy.

For an AI-assisted code change, download the lab bundle from the lab page. Ask your coding assistant to explain the relevant function first, change one behavior, and add one test that could fail. Compare the result with your prediction.

Save your experiment and check-in

Record your prediction, what you tried, what happened, and the evidence for your conclusion. Name one thing you can now do independently and one uncertainty to revisit.

Key takeaways

• Diagnose the failing component before choosing a technique.
• Keep changing facts in retrievable sources.
• Require a measurable behavior gap before training.