A Mind Worth Questioning: Working with AI in This Course

Every generation of engineers has had to learn how to use the most powerful thinking
tools available to them. Slide rules gave way to calculators, which gave way to
simulation software, which gave way to the internet. Each transition required
engineers to develop a new skill: not merely operating the tool, but understanding
what the tool was actually doing, where it could be trusted, and where it could not.

You are entering engineering at a moment when the newest such tool is the large
language model (LLM), the technology behind AI assistants such as ChatGPT, Claude,
and Gemini. This course treats proficiency with these tools as a genuine learning
objective, not an afterthought. By the time you complete ECE Emerge, you will not
only understand foundational electronics; you will also understand how to use AI
as a productive thinking partner rather than a shortcut that quietly undermines
your own development.

This chapter tells you exactly how to do that.

What an LLM Actually Is

Before you can use a tool wisely, you need a working mental model of what it does.
An LLM is not a search engine that retrieves facts, and it is not a calculator that
computes correct answers. It is a system trained to predict the most plausible
continuation of any text it is given. That training was performed on an enormous
corpus of human writing, which means the model has absorbed a remarkable breadth
of knowledge about how concepts are explained, how arguments are structured, and
how problems are solved across virtually every technical field.

This gives LLMs a genuinely useful property: they are extraordinarily good at
explanation. If you want a concept explained six different ways until one of them
clicks, an LLM will do that patiently and without judgment. If you want to work
through a problem step by step and have someone check your reasoning at each
stage, an LLM can do that too.

It also gives them a dangerous property: they are confident even when wrong.
An LLM produces fluent, authoritative-sounding text regardless of whether the
underlying content is correct. It can give you the wrong formula for a filter
cutoff frequency in the same measured, helpful tone it uses to give you the right
one. It cannot feel uncertainty the way a knowledgeable person does, and it will
not always flag its own errors.

The practical implication is this: an LLM is a powerful aid to understanding,
and a poor substitute for it. Use it to help you think; do not use it to
think for you.

The Fundamental Distinction

Every time you open an AI assistant in this course, you face a choice, usually
without realizing it. You can use the tool to build your understanding, or you
can use it to bypass the effort that builds understanding. These two paths are
not equally useful. They are not even in the same category. A practical test
for which path you took is described in the section on verification below.

This matters in ways that will become visible sooner than you might expect.
The prelab for Lab 4 depends on understanding the time constant from Chapter 7.
The analysis in Lab 5 depends on the voltage divider from Chapter 3. The capstone
project requires you to reason through every stage of a signal chain under
real-world constraints. Each of these moments will reveal exactly what you
actually understand, independent of what any AI told you to write.

The goal is not to avoid AI. The goal is to use it in ways that leave you
more capable, not less.

The Right Sequence

The most important practice in this course is also the simplest to state
and the most tempting to skip: attempt the problem before you ask the AI.

This is not a rule designed to make things harder. It is how learning works.
The struggle of working on a problem, even partially and incorrectly, prepares
your mind to receive an explanation in a way that going straight to the
explanation does not. A concept explained before you have wrestled with the
problem slides off. The same concept explained after you have identified exactly
where your reasoning broke down tends to stick permanently.

The sequence should always be:

1. Read the relevant section of the textbook.
2. Attempt the problem or work through the concept on your own.
3. Identify specifically where you are stuck or uncertain.
4. Bring that specific question to the AI.

Step three is the most important and the most frequently skipped.
"I do not understand Chapter 9" is not a question an AI can help with
productively. "I understand that impedance combines resistance and reactance,
but I do not see why the reactance of a capacitor decreases at high frequency
when the formula $X_C = 1/\omega C$ makes it look like frequency is in the
denominator" is a question the AI can engage with precisely. The specificity
of your question reflects the quality of your prior engagement with the material.

How to Ask Well

The quality of an AI conversation depends almost entirely on how you frame
your questions. The following practices will make your conversations substantially
more useful.

Ask for explanation, not answers

The least useful thing you can do is ask the AI to solve a problem you have not
yet attempted. The most useful thing is to describe your own reasoning and ask
whether it is correct.

Less useful: "What is the output voltage of this voltage divider?"

More useful: "I calculated 2.2 V for this voltage divider using
$V_{out} = V_s \cdot R_2 / (R_1 + R_2)$. My reasoning was that the bottom resistor
sees the full supply voltage across the series combination. Is that reasoning correct,
and if not, where does it go wrong?"

Ask for the "why," not just the "what"

An AI will give you a definition of impedance if you ask for one. That definition
is also in the textbook. What the textbook cannot do interactively is respond to
your specific confusion about why the definition takes the form it does. Ask for
that instead.

Less useful: "What is impedance?"

More useful: "I understand that impedance generalizes resistance
to AC circuits. What I do not understand is why we can use all the same series and
parallel combination rules with impedances that we used with resistances in
Chapter 3. What makes that valid?"

Ask it to use the course's framework

This textbook builds understanding through specific analogies and specific
conceptual sequences. The gravity and hill analogy for electric potential.
The voltage divider as the foundation for filters. The Golden Rules as
consequences of feedback, not memorized facts. You can instruct the AI to
stay within these frameworks rather than introducing new ones that may
be correct in general but confusing in context.

For example: "Can you explain why a capacitor blocks DC and passes AC, using
the idea that impedance is frequency-dependent and using the same voltage divider
reasoning from Chapter 3?" This grounds the explanation in what you already
know rather than importing a new framework from outside the course.

Ask it to quiz you

After studying a topic, ask the AI to question you on it rather than summarize it
for you. This is called retrieval practice, and it is among the most well-supported
techniques in learning science. Tell the AI what you have just studied, ask it to
pose a conceptual question, answer in your own words, and ask it to identify what
is missing or incorrect. This converts the AI from a source of information into a
training partner. The pre-lab self-test each week formalizes this practice.

Ask it to steelman your confusion

If something in the course seems wrong to you, or if a result surprises you,
describe your confusion precisely and ask the AI to explain why your intuition
led you astray. "I expected the output of the voltage follower to be different
from the input because the op-amp has such high gain. It seems like something
should be happening to the signal. What is wrong with my reasoning?"
This kind of question teaches you more than asking for the correct explanation
from scratch.

How to Verify

Verification is not optional. It is part of every AI interaction.

An LLM can give you a formula, a component value, a circuit configuration, or
a step-by-step derivation that is subtly or completely wrong. It will not
preface that information with a warning. It will present it with the same
helpful tone as everything else it says. In an engineering context, an
unverified result that happens to be wrong is worse than no result at all,
because it gives you false confidence at exactly the moment you need to
be most careful.

After any AI conversation that produced a technical claim, verify it against
at least one of the following:

• The textbook (the authoritative source for this course)
• Your own independent calculation
• A lab measurement
• A course instructor or teaching assistant

A practical technique: after an AI explains something to you, close the
conversation, write down what you understood in your own words, and check that
summary against the relevant section of this textbook. The gaps between what
you wrote and what the textbook says identify exactly what you still need to
work on.

When an AI gives you a numerical answer to a circuit problem, work through the
problem independently and compare. If the answers agree and you understand both,
proceed. If they disagree, do not simply assume the AI is correct. Find the
discrepancy before you move on.

What the AI Cannot Do for You

Several things matter in this course that no AI conversation can provide.

Physical intuition comes from measurement. When you watch the output
of a low-pass filter roll off on the oscilloscope as you sweep the frequency
upward, something is built in your mind that reading about cutoff frequencies
cannot build. When your measured time constant matches your calculated one to
within five percent, and then does not when you have a component in backwards,
you are developing the calibrated judgment that distinguishes a practicing
engineer from someone who has read about engineering. No amount of AI
conversation substitutes for this.

The prelab is not a deliverable to be completed; it is preparation for the lab.
The purpose of the prelab is to send you into the lab with predictions in hand.
If you arrive at the lab without having genuinely worked through the prelab
yourself, you will not know what to look for, you will not notice when something
unexpected happens, and you will not be able to explain discrepancies when they
arise. Using an AI to complete prelab calculations means you have satisfied the
submission requirement while defeating the purpose entirely.

The AI has never seen your breadboard. When your measured output
does not match your theoretical prediction, the AI can help you think
through the space of possible causes systematically. It cannot tell you
that your capacitor leads are reversed, that you have a loose wire at
a node, or that your resistor color-code reading was off by a factor of ten.
Physical debugging requires physical inspection.

The AI does not know this course. It was not trained on this
textbook. It may explain a concept using terminology this course specifically
avoids, or it may introduce ideas from a framework the course has not yet
built toward. When an AI explanation conflicts with something in this
textbook, resolve the conflict by examining both carefully and consulting
the textbook as the authoritative source.

Documentation: What Is Required

This course requires you to document and attribute AI contributions in
your submitted work. This is not bureaucratic overhead. It is a professional
norm that is becoming standard across engineering practice, and forming
the habit now matters.

The standard is that a reader of your report should be able to tell exactly
what you did and what the AI did. In practice, this means the following.

If an AI conversation helped you understand a concept that you then applied
in your own work, note it briefly: "I used [AI tool] to clarify my understanding
of the time constant derivation before completing this calculation."

If you asked an AI to review your reasoning and it identified an error that
you then corrected, note that: "I used [AI tool] to review my phasor calculation.
It identified an error in my sign convention for the phase angle, which I
corrected by re-reading the sections on Signals in the Time Domain and on Phasors."

If any text, code, numerical result, or figure generated by an AI appears
directly in your submitted work, cite it explicitly, just as you would cite
any other source.

The practical test is this: if you cannot write clearly about what the AI
contributed and what you contributed, the boundary became unclear during
the work itself. That is the moment to pause and re-engage with the material
on your own terms before continuing.

The Larger Picture

Every interaction you have with an AI in this course is practice. An
interaction in which you handed off your thinking to the machine is practice
at becoming less capable. An interaction in which you brought your own
reasoning to the table, challenged it, refined it, and left understanding
something you did not before; that is practice at something that compounds.

The material in this course is genuinely interesting. Circuits are a language
for describing how energy and information move through the physical world.
Signals carry meaning. Amplifiers and filters shape that meaning. The
instrumentation amplifier at the heart of your capstone project is a
precision instrument for extracting a faint physical signal from a noisy
background — which is, in a certain sense, what good thinking is as well.

The AI is a capable conversation partner for exploring all of this. Bring
your questions to it. Challenge it. Verify what it tells you. The course
will be richer for it, and so will you.

Putting It Into Practice: The Pre-Lab Reflective AI Exercise

Each lab in this course includes a structured Reflective AI Exercise as part of the
pre-lab assignment. The format is identical each lab and has three parts.

• Part 1: Exploration. Use an AI assistant to explore two designated
  focus areas for the current lab. The goal is not to collect a summary you
  can paste into your report. The goal is to arrive at the lab with a mental
  model that lets you make and check predictions.

• Part 2: The Self-Test. Write and refine your own quiz prompt targeting
  the specific concepts for that lab, then run the quiz. The details of how
  to do this well are covered in the following section on prompt quality.

• Part 3: The Formal Reflection. Submit a 150–250 word written synthesis
  addressing three required points specific to each lab. This is submitted to
  Gradescope as part of the pre-lab assignment, due Tuesday at noon.

Prompt Quality: A Skill You Will Practice Every Week

The quality of what you get from an AI depends almost entirely on the quality
of what you ask. This is not a minor detail. It is the skill. A vague prompt
produces a generic response that teaches you nothing. A well-constructed prompt
produces a focused, concept-specific conversation that actually builds your
understanding.

Step 1: Asking for Information (Part 1 Exploration)

The first use of a prompt in each lab is to explore the designated focus areas.
The same contrast between weak and strong applies here as anywhere else.

Weak: "Tell me about RC circuits."

This produces a textbook summary. It teaches you what you could have read. It does
not engage your existing understanding or identify where it breaks down.

Strong:

"I am a first-year electrical engineering student preparing for a
lab on RC circuits. I understand that a capacitor stores charge, but I do
not yet understand why the voltage across it cannot change instantly when a
switch closes. Can you explain the physical reason for that constraint,
without using differential equations? Focus on what the capacitor is
actually doing with energy at the moment the switch closes."

This prompt works because it establishes who you are and what you are preparing
for, it names what you already understand, it identifies specifically where your
understanding stops, and it constrains the form of the answer so the explanation
stays at the right level. The AI now has a precise target. The resulting
explanation will fill a real gap rather than restate what you already know.

Step 2: Writing Your Own Quiz Prompt (Part 2 Self-Test)

After exploring the focus areas, you write your own quiz prompt to test what you
have learned. This prompt must do four things:

1. Establish who you are and what you are preparing for
2. Constrain the question type to scenario-based prediction, not simple recall
3. Name the specific concepts the questions must involve
4. Control when answers are revealed

Weak: "Quiz me on circuits."

This fails on every dimension: no role context, no constraint on question type,
no scope, and no output control.

Strong:

"I am a first-year electrical engineering student preparing for a
lab on RC circuits. Give me a three-question scenario-based quiz. Each
question must describe a specific physical situation (a component value,
a switching event, or a waveform observation) and ask me to predict what
happens and explain why in physical terms. Do not ask me to recall
definitions or reproduce formulas from memory. Do not reveal any answers
until I have responded to all three questions."

Step 3: The Meta-Prompt (Improving Your Draft)

Once you have written your quiz prompt draft, submit it to the AI using the
following meta-prompt before running the quiz. This is the same every week;
copy it exactly and paste your draft where indicated:

"Here is a prompt I wrote to generate a self-quiz to help me prepare
for an engineering lab. Please evaluate it against these four criteria:
(1) does it establish my role and context clearly,
(2) does it constrain the question type to scenario-based rather than simple recall,
(3) does it prevent the AI from revealing answers before I have responded,
(4) does it scope the content precisely enough to be useful for a specific lab topic.
Then rewrite it to address any weaknesses you identify."

My draft prompt: [paste your draft here]

Use the AI's critique and revised version to produce your final prompt, then run
the quiz with that revised version.

What You Submit for Part 2

Your Gradescope submission for Part 2 must include all four of the following:

1. Your original draft prompt
2. The AI's critique (copy-paste in full)
3. Your revised prompt after incorporating the feedback
4. The quiz transcript: your revised prompt, the AI's questions, and your responses

Items 1 through 3 are the prompt-craft artifact. Item 4 is the content record.

What a Strong Reflection Looks Like (Part 3)

The three required points in Part 3 are not three separate paragraphs. They are
three jobs that a single coherent paragraph must do simultaneously. The annotated
example below uses a made-up lab scenario to show which sentence is doing which job.

Example reflection (annotated):

The wiring path between a sensor and a measurement instrument acts as an antenna
for environmental interference, so the choice of connection method determines
how much of that noise reaches the instrument alongside the signal.
[The Link: wiring choice is framed as signal protection, not just a connection option.]
Differential wiring suppresses this noise through common-mode rejection: because
both wires travel the same physical path, any interference that appears equally
on both conductors is subtracted out at the instrument, while the true
differential signal (which appears on only one wire) is preserved.
[The Technical "Why": the key term is used to explain a mechanism, not just named.]
My specific realization was that using a single-ended connection at a bench
cluttered with switching power supplies would feed that switching noise directly
into the measurement chain; I now know to look for a periodically spiking noise
floor as the diagnostic symptom of that mistake, and to switch to differential
wiring as the first corrective step.
[The Lab Application: a concrete, physically plausible mistake is identified,
a symptom is named, and a correction is stated.]

Notice that the reflection does not list the three criteria and fill them in one
by one. It builds a single argument in which each sentence advances the next. The
annotations above are shown here for instructional purposes only; your submitted
reflection should read as continuous prose.

Summary

• An LLM is trained to produce plausible, fluent text. It is exceptionally good at
  explanation and exceptionally unreliable as an authority. It does not flag its own errors.
• The fundamental test of a productive AI conversation: could you explain the result
  in your own words afterward? If not, the AI did your thinking, not you.
• Always attempt the problem before asking the AI. The struggle of working on a
  problem prepares your mind to receive an explanation. Skipping the struggle
  discards most of the learning value.
• Ask for explanation of your reasoning, not production of an answer. Ask for the
  "why," not just the "what." Ask it to quiz you.
• Verify every technical claim against the textbook, an independent calculation,
  or a measurement before acting on it.
• Physical intuition, prelab preparation, and circuit debugging require your direct
  engagement. No AI conversation substitutes for them.
• Document AI contributions in your work as you would any other source. The
  standard: a reader should be able to tell exactly what you did and what the AI did.

Table: AI Use — Quick Reference