What you will learn
  • Describe a specific knowledge gap to an assistant.
  • Request an explanation that connects mechanism and example.
  • Recognize where an analogy stops being accurate.
  • Use a worked example and a new question to test understanding.
  • Distinguish familiarity with an explanation from independent performance.

Before you begin

Read USE03 on clear instructions. Choose a concept from a class, hobby, or earlier lesson that you find partly confusing.

Locate the gap before requesting more detail

You read that a machine-learning model has weights. The word is familiar, but you do not know what a weight does. Asking for ‘more detail about AI’ may produce additional unfamiliar terms. A better request identifies the gap: you know inputs and outputs, but not how changing a number inside the model changes its output.

An explanation should begin from what you can already use. State your starting knowledge honestly. ‘I understand multiplication and graphs, but not derivatives’ gives a clearer boundary than an age label alone. Describe the exact sentence or step that lost you. The assistant can then build a short bridge instead of restarting an entire subject.

Avoid requesting only an easier tone. Shorter words can still hide a missing mechanism. A useful explanation answers what the thing is, what it does, how a small example works, and when that example stops applying. These questions keep simplicity connected to technical accuracy.

Move from a concrete case to the general idea

A worked example is a small problem with its solution made visible. Consider a rule that estimates a fictional delivery time: time equals two minutes per floor plus three minutes of preparation. For four floors, the estimate is eleven minutes. The number two controls how strongly the floor count affects the estimate; three adds a fixed starting amount.

This example can introduce a weight and a bias in a simple linear model. The weight multiplies an input; the bias shifts the output. It does not imply that every neural network has only one input or that all real delivery times follow a straight line. Once the concrete case is understood, the assistant can explain the broader pattern and its limits.

Ask for a second example that changes one condition. What happens if the per-floor value becomes three? The four-floor estimate becomes fifteen minutes. The learner can now connect a changed parameter to a changed prediction. Merely memorizing the definition ‘weights are model parameters’ would not demonstrate that connection.

A prompt that makes the explanation teachable

Weak prompt: ‘Explain neural-network weights like I am five.’ This may produce a friendly analogy without enough technical content, and the age request does not say which mathematics you know.

Improved prompt: ‘I know multiplication and can read a simple graph. I understand model inputs and outputs but not weights and biases. Use one numerical example with one input. Show how changing the weight changes the output, then define weight and bias accurately. Give one limitation of the example. Finish with a similar question for me, without its answer until I attempt it.’

The request controls the teaching sequence: a concrete example, a technical definition, a boundary, and a learner attempt. It does not ask for a full neural-network lecture. That narrower scope makes it easier to see whether the missing idea was actually explained.

Follow-up prompt: ‘I answered fifteen minutes when the weight became three. Now give a case where only the bias changes, and ask me to predict the result.’ Verification: Work the arithmetic yourself and compare the technical definitions with ML06. If the assistant gives an incorrect result, separate the arithmetic error from the concept rather than assuming the whole topic is beyond you.

Use analogies with an exit door

An analogy maps an unfamiliar idea onto a familiar one. Describing a model weight as a volume knob can suggest stronger or weaker influence. But real weights can be negative, interact with other values, and be transformed by later calculations. A volume knob is therefore a partial picture, not the mechanism itself.

After an analogy, ask: ‘Which parts correspond to the real system, and where does this comparison break?’ Then request the literal explanation. For weights, that means multiplication of an input value by a learned parameter. A good analogy lowers the initial barrier; the literal explanation prevents the picture from becoming a misconception.

Watch for person-like language. A model may be described as ‘deciding what matters,’ but that phrase can conceal numerical operations. It is acceptable shorthand only when the article or tutor also explains what the system actually computes. You should be able to replace the metaphor with a technical sentence.

Check whether you can use the idea

An explanation can feel clear while you are looking at it. To test understanding, close the answer and explain the concept in your own words. Then solve a changed example. If you cannot begin, identify whether the obstacle is a missing definition, an unclear step, or a prerequisite skill.

Ask the assistant to respond to your explanation with a precise correction: quote the mistaken claim, explain why it is mistaken, and pose a targeted follow-up. Avoid requests for a general rating such as ‘How smart is my answer?’ A useful correction describes the gap between your reasoning and the concept.

Do not turn every question into an endless conversation. When you can define the idea, work a new case, and describe one limitation, move to practice. If an explanation still conflicts with your textbook or instructor, check the source and discuss the conflict. Google’s machine-learning glossary offers reference definitions for terms such as weight and bias; it does not replace the worked practice you do yourself.

Important terms

Worked example
A problem whose solution steps are shown and explained.
Analogy
A comparison that transfers part of a familiar idea to an unfamiliar one.
Mechanism
The actual process that produces an effect.
Weight
A model parameter that scales a value, such as an input.
Bias
An added model parameter that shifts a calculation’s result.

Mini project: Explain, predict, and teach back

  1. Choose one confusing term and write what you already know and exactly what you cannot explain.
  2. Request a numerical or concrete example, a literal definition, and a limitation.
  3. Without looking at the response, explain the idea in three sentences and solve a changed case.
  4. Ask for feedback on your actual explanation. Finish when you can correct the error yourself and handle another changed case.

Common mistakes and debugging

  • Asking for more detail without locating the gap: identify the step or term that caused confusion.
  • Keeping only the analogy: translate it back into the real mechanism.
  • Reading the answer while testing yourself: close it before attempting a fresh problem.
  • Accepting praise as feedback: ask which claim is correct, which is wrong, and why.

Independent challenge

Use the delivery example with a negative input change: compare four floors with two floors while keeping the same weight and bias. Explain why the change in predicted time does not depend on the bias.

Check your understanding: 10 questions

  1. Why describe your existing knowledge?

  2. What makes an analogy incomplete?

  3. For time = 2 × floors + 3, what is the prediction for four floors?

  4. What changes when the weight becomes three and the bias stays three?

  5. What is stronger evidence of understanding than rereading a clear answer?

  6. In your own words, what does “Worked example” mean?

  7. In your own words, what does “Analogy” mean?

  8. In your own words, what does “Mechanism” mean?

  9. In your own words, what does “Weight” mean?

  10. In your own words, what does “Bias” mean?

Quiz answers

Reveal all 10 answers after your attempt
  1. It tells the assistant which ideas can support the explanation and which prerequisites still need teaching.
  2. It maps only selected features; other properties of the familiar object may not apply to the real system.
  3. Eleven minutes: 2 × 4 + 3 = 11.
  4. The four-floor prediction becomes fifteen minutes because each floor now contributes three minutes.
  5. Explaining the idea without the answer and correctly solving a new example.
  6. A problem whose solution steps are shown and explained.
  7. A comparison that transfers part of a familiar idea to an unfamiliar one.
  8. The actual process that produces an effect.
  9. A model parameter that scales a value, such as an input.
  10. An added model parameter that shifts a calculation’s result.

Summary

Describe the knowledge gap, request a concrete mechanism and bounded analogy, then test yourself on a changed example. Understanding means being able to use the idea.

Continue learning

Continue with USE05 to apply this careful questioning to research and source-based answers.

Choose a connected learning path

Sources and further reading

Prepared 2026-09-18. Draft; conceptual and arithmetic review complete. Example dialogue is illustrative.