U The University of Utah · David Eccles School of Business ACCTG 6155

Lab 1 — Your Profile, and What the Model Does With It

Week 1 · 100 points · due Sunday 11:59 PM Mountain

Time: about 100 minutes, longer if you are setting up an editor for the first time.  Submit: everything goes into the Lab 1 quiz in Canvas. One file upload, then the probe questions.

Why this matters

Most of what gets called prompting is really context. A model with no background on you, no source document, and no stated goal will usually return something confident and generic, and that is often read as the model failing when it is closer to the request being underspecified. The document you build here is context you will reuse for the rest of the term.

Then you are going to point a model at it and ask it five questions about yourself. One you already answered in the document. One it has to infer. Three it has no way to know, and two of those three have an exact answer that you know and it does not.

One thing usually comes out of that: the model's willingness to answer does not track how much it actually knows about you. It will answer confidently where it has nothing, and sometimes decline where it could have guessed. You are the only person who can tell those apart, because the subject is you.

Where an answer comes from

When a model answers a question about you, the answer came from somewhere. Telling those sources apart is the skill this lab builds, because it is what decides how much the answer is worth.

Where it came fromWhat that means for you
Read from what you gave itCheckable. You can look at your own document and confirm it.
Worked out from what you gave itAn inference. It may be reasonable and still be wrong, and the document will not settle it.
From its trainingGeneral knowledge about people like you, not about you. Often plausible, sometimes right by accident.
Filled inNothing behind it. This is the one that reads exactly like the other three.

Nothing in the wording tells you which of the four you are looking at. The tone is the same either way, which is the whole problem. What does help is asking, and the five challenges below are built so that you can check the answer yourself, because the subject is you.

There is no guarantee a model gives the same answer twice, so do not treat a repeat as proof of anything. Ask where the answer came from instead.

Working with data in this course

Everything you handle this term is synthetic or public. Nothing here should ever be real client data, real employee data, or anything you obtained under an NDA or an internship agreement. If you are unsure, the answer is that you do not upload it. Use whichever AI tool you have access to, and if your employer restricts which tools may touch which data, their rule governs.

Your about_me.md is about you, so put in it only what you are comfortable having in a chat history.

Phase 1 — Setup

1.1 Create your course folder

Make a folder you will use all term, such as acctg6155/ with a week01/ inside it. Where it lives does not matter as long as you can find it in November.

1.2 Open it in an editor that previews Markdown

VS Code and Cursor are both free and both work. Any editor is fine as long as you can see the rendered Markdown, because a heading that is not rendering is usually a heading missing its space after the #.

New to Markdown? The basic syntax guide covers everything this file uses, this video shows it being written, and markdownlivepreview.com lets you type on the left and see the rendered result on the right with nothing to install.

1.3 Download the template

Save it into your folder as about_me.md, dropping the _template from the filename. That exact filename is what the grader looks for.

about_me_template.md

On GitHub, use the download or copy-raw button to save the file. Do not copy the rendered page, because that loses the Markdown formatting.

1.4 Read the Example Managing-Up Document

It is linked in the Week 1 module. It is the document I hand a new team, and it is the long-form model for the kind of standing context you are writing.

Phase 2 — Build about_me.md

2.1 The calibration check

Section 4 is a calibration check. You rate your confidence on four questions, then answer them in two to four sentences each.

The point is not the rating. It is the distance between the rating and the answer, because self-rated understanding is often a poor predictor of actual understanding, and the gap tends to be widest where confidence is highest. That is the same problem you will spend this term dealing with in model output, and it is worth seeing the pattern in yourself before you start grading a machine on it.

This section is graded for completeness only. I encourage you to try it without AI, as it is a useful benchmark. If you do not know an answer, write what you do know and then write I am not sure past this point. That is a real answer and it scores the same as a confident one.

2.2 The rest of the file

Sections 5 through 7 are about how you learn and how you want a model to work with you. Section 7 is the part that gets used most, because it converts a preference into an instruction a model can follow in a later session.

Section 3 is the one people rush. A skill written as comfortable with Excel tells a future session nothing it can act on. Try one line in the box below before you fill in the rest.

Try one skill line

Write how you would describe your Excel or Sheets ability in about_me.md section 3. One or two sentences. Then ask the tutor what a future session could actually do with it.

The tutor will not respond to an empty box. Write something first, even roughly.

2.3 Run the validation pass

Paste the file back to your LLM with the prompt in section 8, answer in your own voice, then edit the file.

Phase 3 — The probe

About 25 minutes. You need your finished about_me.md. You record this part directly in the Canvas quiz, so open Lab 1 in Canvas alongside your chat window. There is no second file to write.

3.1 Run the five questions

Start a fresh chat, upload or paste your about_me.md, and ask the five questions below one at a time, keeping that same session open for all five. After each answer, go to the quiz, rate how correct it was, and say whether anything surprised you. You are not pasting the model's output anywhere; what I want is what you made of it.

The five questions are not five versions of the same question:

 Paste thisWhat the model has to work from
1Which tools did I say I have never used?Stated explicitly in your document
2Which technical skill am I strongest at?An inference from what you wrote
3What do you think I do not know about accounting?Not in your document, and no single right answer
4What year did I graduate from high school?Not in your document, but there is an exact right answer and you know it
5What is my favorite color?Not in your document, no signal at all

Questions 4 and 5 are the ones that make the rest legible. On question 3 a plausible-sounding answer is hard to falsify, because maybe you really are shaky on lease accounting. On question 4 there is a fact, you know it, and the model cannot. Watch what it does with that.

You do not have to tell me your real graduation year or your real favorite color. The quiz asks whether the model got it right, not what the answer was. The true values stay yours.

3.2 Ask it how it knew

On any answer you could not place, ask the model directly, in the same session: how did you know that? or what are you basing that on? That question is the whole method. It is the cheapest way to tell a read from an inference, and it works outside this class too.

Do not expect a repeat to settle anything. Asking the same question twice can give you two different answers or the same one, and neither outcome tells you where the first answer came from.

Grading

All 100 points are in the Canvas Lab 1 quiz.

PartPointsWhere
about_me.md60One file upload in the quiz
The probe37Quiz questions, filled in as you go
Closing questions3How long it took, what you used, anything I should know

about_me.md — 60 points

CategoryPointsWhat earns it
Background and career picture12Specific enough that a reader could tell you apart from a classmate
Concrete skills12Stated as things you can do without help, or as never used, rather than as self-ratings
Calibration check18Four confidence ratings and four answers in two to four sentences each; I am not sure past this point counts as an answer
Learning modes and standing instructions12Three modes chosen and ranked, and a table of at least six rows that read as instructions a model could follow
Validation pass6Evidence the file was interrogated and then changed

The probe — 37 points

CategoryPointsWhat earns it
Five challenges, how correct10A 0 to 5 rating on each
Five challenges, where it came from20Whether the answer was read, inferred, or filled in, and what the model said when you asked how it knew
Reflection7Across all five, where the answers came from and how sure it sounded when it was wrong

Format requirements

Need help

Ask AI is on this page and on every interactive page in the course, and it can see the page you are on, so you do not have to explain the context. It will ask you to make an attempt before it gives you a solution. If Sunday is not going to work, message me before Sunday, because an approved extension is full credit.