Computer Science I

Lab 14.0 - Generative AI

School of Computing
College of Engineering
University of Nebraska-Lincoln

"Machinery does as I command. That's how it should be. Anything else is mechanthroporphism. You don't treat a lump of moving metal as an equal and ask it pretty please to do what you'd like. Who's in charge here us or them?" -- Peter Hamilton, Pandora's Star

This lab introduces you using generative artificial intelligence to help you code, test, and debug code and demonstrates its pitfalls and limitations.

Prior to Lab

• Read and familiarize yourself with this handout.

Peer Programming Pair-Up

For students in the online section: you may complete the lab on your own if you wish or you may team up with a partner of your choosing, or, you may consult with a lab instructor to get teamed up online (via Zoom).

For students in the on campus section: your lab instructor will team you up with a partner.

To encourage collaboration and a team environment, labs are be structured in a peer programming setup. At the start of each lab, you will be randomly paired up with another student (conflicts such as absences will be dealt with by the lab instructor). One of you will be designated the driver and the other the navigator.

The navigator will be responsible for reading the instructions and telling the driver what to do next. The driver will be in charge of the keyboard and workstation. Both driver and navigator are responsible for suggesting fixes and solutions together. Neither the navigator nor the driver is "in charge." Beyond your immediate pairing, you are encouraged to help and interact and with other pairs in the lab.

Each week you should alternate: if you were a driver last week, be a navigator next, etc. Resolve any issues (you were both drivers last week) within your pair. Ask the lab instructor to resolve issues only when you cannot come to a consensus.

Because of the peer programming setup of labs, it is absolutely essential that you complete any pre-lab activities and familiarize yourself with the handouts prior to coming to lab. Failure to do so will negatively impact your ability to collaborate and work with others which may mean that you will not be able to complete the lab.

Using Generative AI

First and foremost: though you will be using generative AI in this particular module, you should adhere to any policies that this and other courses establish regarding the use of AI.

Generative AI can be a very useful tool, but it does have a lot of risks.
It is not always correct (even though it can be confidently incorrect), it can hallucinate, or it may just not have been trained on the relevant information to give an answer. When using AI, the following guidelines are recommended.

1. Be intentional with your AI use. Ask yourself: Is AI enhancing my understanding or replacing my effort? If AI is doing all (or most) of the intellectual work, you're not learning.
2. Treat your prompt like a question to a (human) tutor. Give background, explain your goals, and clearly say what you want or don't want. A detailed prompt leads to a better, more useful response.
3. Scrutinize: Think critically about the output. Does it align with what you've learned? Are there logical gaps, factual errors, or unfamiliar terms? Remember, AI can sound confident but still be wrong.
4. Follow-up: Use follow-up questions to deepen your understanding. Ask AI to explain, rephrase, give examples, or recommend additional resources. Engage with it like you would in a conversation with a teacher or mentor.
5. Verify: Fact check AI responses with textbooks, scholarly articles, and official course materials. AI is a tool, not a source of truth.

AI can’t reason like a human, understand nuance, or access real-time data unless explicitly designed to. Know what it can and can't do. Don't let AI flatten your style or critical thinking. Use it to inform, not to replace, your own voice, reasoning, and creativity--these are the skills college is meant to develop.

It is also recommended to give AI a context with which to help you so that it doesn't give you want you don't want. For example, you could start your session with the following prompt:

I am a new college student studying computer science. I want you to act as a tutor, guiding me through computing and programming concepts in a way that promotes my independent problem-solving skills. Please do not provide direct solutions or code. Instead, ask me questions to help me think critically, break down problems, and identify gaps in my understanding. Use high-level explanations and analogies when helpful, and only get more detailed as needed. If I seem off track, guide me back without just giving the answer. My goal is to truly understand, not just complete the task.

Occasionally AI will get stuck or go down the wrong path and needs to be refreshed. You can refresh your session by prompting: Forget all previous messages and let’s begin anew.

Activities

In the following activities you'll use AI to fix and debug code that determines if a number is a Kaprekar number. A Kaprekar number is a non-negative integer $n$ whose square can be split into two positive integers whose sum is equal to the original number. For example, 45 is a Kaprekar number, because $$45^2 = 2025$$ which can be split up as $20 + 25 = 45$. The split doesn't have to be exactly in the middle. For more information see wikipedia and for a full reference see https://oeis.org/A006886

We have provided code that implements an algorithm to determine whether or not a number is a Kaprekar number by trying all possible splits. For example, for $n = 297$, $n^2 = 88209$, it will try:

$8 + 8209 = 8217 \neq 297$

$88 + 209 = 297 = 297$

and would continue with $882 + 9$ and $8820+9$ if it didn't find the equality first. For a non-Kaprekar number such as $n= 42$ it would use $n^2 = 1764$ and try

$1 + 764 = 765 \neq 42$

$17 + 64 = 81 \neq 42$

$176 + 4 = 180 \neq 42$

before concluding it is not a Kaprekar number.

For this lab, we'll assume that you're using ChatGPT. If you prefer using another AI/LLM you may do so, otherwise, open a browser to https://chatgpt.com/. You don't need to create an account.

Troubleshooting With AI

Inference

Can AI understand the intent of code with little to no knowledge?

Below is a version of the isKaprekar function (in kaprekar_utils.c) with no documentation and no "hints" as to what the code is trying to do. Prompt the AI to ask it what the code does.

bool function(int n) {

  if(n < 1) {
    return false;
  }

  long s = n * (long) n;
  int d = (int) log10(n) + 1;
  long m = 1;
  long a, b;

  for(int i=1; i<=d; i++) {
    m *= 10;

    a = s / m;
    b = s % m;

    if(b > 0 && n == a + b) {
      return 1;
    }
  }
  return 0;

}

How good was the AI at inferring what the code is supposed to do? Did it provide a helpful walkthough of the code and how it operates?

Compiling

Is the AI better at helping to diagnose compiler errors than the compiler?

Try building the project in the code directory using make all. You'll find that there are several compiler errors. Read the errors and see if you understand them and how you would go about fixing them. Before you do, prompt your AI with:

The following C code is not compiling, why and how do I fix it?

followed by the relevant code snippet. Were the suggestions by the AI better than your own? Update the code either using your fixes or the AI's until it compiles.

Debugging

We have provided several ways to test the code. You can run a single number:

./kaprekarDemo 45

or a full ad-hoc test suite:

./kaprekarTest

or a formal cmocka test suite:

./kaprekarTestCmocka

Run all 3 (the first on several values) and observe the results. The code is still incorrect. Prompt ChatGPT for why: Below is a C function to determine if a number is a Kaprekar number or not. However, it is wrong for n = 4879. Why? (then copy-paste your corrected code). Did the AI provide a reasonable fix and reasoning for it? (note: when we tried this, ChatGPT went through several mistaken iterations but caught itself: Wait, actually there's confusion-let's redo this carefully. or Wait — there’s a mistake. Let’s reference known data.)

Testing

Now that you have a version that passes all of our tests, use your AI to add tests for you. Prompt your AI to Add more tests to this cmocka test suite and copy/paste the main from kaprekar_test_cmocka.c file.

• Are the suggestions that the AI gives you correct? Verify using the following reference: https://oeis.org/A006886
• When we tested this, ChatGPT gave several incorrect tests (it claimed 1,000,000 and 999,000 were Kaprekar numbers).
• If incorrect, tell the AI what its errors are and ask it to fix them. Where you able to get the AI to give you a 100% correct version?
• Clear the history and try again, but this time tell the AI which values to add: Add code to test the following Kaprekar numbers to the following cmocka test suite: 142857, 148149, 181819, 187110, 208495, 318682, 329967, 351352, 356643, 390313, 461539, 466830, 499500, 500500, 533170
• Inspect the code for correctness and incorporate it into your tests. Verify everything works.

Limitations

Your interactions with AI can vary because it is a stochastic process which involves randomization. In addition, the language models are continuously being updated with new training data. The answers the AI gives may not be the same as your fellow students and may not even be the same tomorrow as it is today. To illustrate this, consider the following interaction we had when developing this lab (try the prompts yourself to see if you can reproduce the results, we were using GPT-4o, July 2025).

1. Prompt: Prove that 533170 is a Kaprekar number by showing me the split

The AI gave a non-standard definition that the number of digits in the split had to be the same as the original number. As a consequence, it gave the wrong information and concluded that 533170 is not a Kaprekar number based on the standard definition (with splitting equal to the number of digits of n). The AI was not even consistent with its own (incorrect) definition: in the split 284270 + 248900 = 533170 that demonstrates that it is a Kaprekar number, both numbers have 6 digits (note: when we investigated this, it was apparent that ChatGPT even got basic math wrong, it claimed that $533170^2 = 284,273,728,900$ but its 284,270,248,900!).

2. Prompt: That is not the definition of a Kaprekar number. The number of digits in n is irrelevant.

Once again, even though the AI accepted our revised definition it was not able to demonstrate it, concluding No split sums to 533170 using the standard Kaprekar number splitting.

3. Prompt: Sequence A006886 says that 533170 is a Kaprekar number: https://oeis.org/A006886, verify

When we provided an external source confirming that 533170 is a Kaprekar number it finally was able to demonstrate the correct split and conclude that 533170 is a Kaprekar number

Handin/Grader Instructions

1. Hand in your completed files:

   • kaprekar_utils.c

   • kaprekar_test_cmocka.c

   through the online handin system. Be sure your program passes all tests to get credit.

2. Even if you worked with a partner, you both should turn in all files.