AI 101 for Marines
You already know the skill that makes AI useful. It is the brief you give a Marine.
Free. About 20 minutes. No account needed.
If you use AI every day, start at chapter 3.
- 6 chapters
- About 20 minutes
- No account needed
- 1What you are talking to2 min
- 2Brief it like a Marine4 min
- 3Catch the confident wrong answer4 min
- 4Get one real thing done today4 min
- 5When a chat is not enough3 min
- 6Self-check2 min
Your answers save in this browser. The optional email signup sends your address separately.
This browser is not saving your progress. New answers may be lost when you leave or reload.
What you are talking to
A chatbot generates text from learned patterns and the context it can see. It can sound confident when it is wrong.
Think of it as a sharp Marine who checked in yesterday. Give it your current facts and check what it uses. Memory and enabled tools can add context, and a confident answer can still be wrong. Anthropic, the company that makes Claude, gives its own users the same advice: treat the model like a brilliant new employee who lacks your context.
Which one to use: Claude, ChatGPT, and Gemini each have a free plan, and every exercise in this course works in any of them. Muster Out runs in your own Claude account.
Free plansClaude pricingChatGPT pricingGemini plans
Words you will see
- Chatbot: an app you type to, such as Claude, ChatGPT, or Gemini, that writes back.
- Model: the program inside the chatbot that writes the replies.
- Brief: what you type to it. Most people call it a prompt.
- Context: everything the model can see right now: your brief, what you pasted, and anything memory or search added.
- Memory: a setting that carries facts from past chats into new ones.
- Web search: a setting that lets it read current pages and cite them.
- Hallucination: a confident answer that is not true.
- Tool: a program that carries out a specific job.
Let it draft and explain. Before you act on a number, a date, or a rule it gives you, check it.
Brief it like a Marine
Give it what you would give a Marine who checked in yesterday: task, purpose, constraints, end state.
Task: Build a run plan. Purpose: Improve my PFT. Constraints: Four mornings, 45 minutes. End state: A weekly table.
When the reply is close
Send one specific fix: "Make week 3 easier." "Half as long." "Use only the numbers I gave you." The chat keeps everything above it, so you do not need to start over.
A clear brief gives it a better starting point. Check the reply before you use it.
Catch the confident wrong answer
It can be wrong in the same calm voice it uses when it is right. Learn four checks and use the one that fits the answer.
Why it happens
OpenAI's researchers explain it with a multiple-choice test. A blank answer always scores zero, and a guess sometimes scores, so a model that guesses looks better on the test. The way models are trained and scored rewards the guess. When a model does not know, it can still give you a fluent, specific, wrong answer.
Four checks
- 1Do the math. Dates, money, paces, counts. Ask it to show its inputs and calculation. Verify the result with a calculator or calendar.
- 2Compare it to what you gave it. Check added claims against your facts or a source, especially numbers it could not derive.
- 3Ask where it came from, then open it. For a rule, a fact, or a quote, ask for the source and read the line yourself. A source it names can be wrong too.
- 4Ask again in a new chat. If two fresh chats give two different answers to a fact question, neither one is settled, and the source decides.
Head it off in the brief
Two lines in your brief cut down on this before it starts: "If you are not sure about a fact, say so." and "Quote the line you used." Anthropic's guide lists both. It also says they reduce made-up answers without eliminating them, so the checks still apply.
Trust an answer after it passes the check that fits it: the math, your own facts, or the source.
Get one real thing done today
Pick a task you actually have this week. The page builds the brief from your answers. Paste it into any chatbot, then check what comes back.
Fill in what you know. Leave a box empty and the brief shows [fill this in], so you can see what the AI would have to guess.
Common sense applies: leave other people's private details, and anything you would not want made public, out of a public chatbot.
After you run it
These boxes record your own report after using a chatbot. The page cannot check what you did there.
Pick a task and build your brief before you leave.
When a chat is not enough
A new chat may use saved context or search. Muster Out adds a maintained board, source text beside each task, and date rules you can inspect.
Compare the written example
| Row | A new chat | A tool like Muster Out |
|---|---|---|
| Your facts | Supply your current facts and check what saved context contributes. | Saved once on your board. When you press Ask, the page sends your board to Claude with your question. |
| The source | Training, text you provide, and sources retrieved by enabled tools; check each citation against the source. | Each cited task quotes its paragraph and links to the official text. |
| Dates | It may calculate with a tool. Check the inputs and the rule. | Counted from your date by the rule on each task, and labeled by kind. |
| New MARADMINs | Search or scheduling may help where available; check what it covers. | A weekly check, offered at setup where scheduling is supported, files possible matches for your path. Read the official message before acting. |
| Still true | It can be wrong, so the checks apply. | It can be wrong too, and IPAC, your TAP counselor, your VSO, and your command decide what applies. |
Setup and other tools
Setup is one message in a Claude chat, on a phone or a computer, and takes about 5 minutes. You need Claude Pro ($20 a month) or higher, because the Free plan cannot store board data. The board lives in your own Claude account. Muster Out is an unofficial pilot. Muster Out covers resignation of a regular commission, separation at EAS, and retirement with qualifying active service. Enlisted coverage includes FMCR transfer at 20 to 29 years and direct retirement at 30 or more, subject to IPAC and MMSR-2 eligibility review. Medical separation, involuntary separation, and Guard or Reserve retirement with nonregular pay get an office referral. IPAC, your TAP counselor, your VSO, and your command decide what applies to you.
Next from Warfare Command: a FITREP reviewer, to review a draft fitness report before it goes up. It is not out yet. The finish page has the email signup.
Use the pattern yourself
You do not need Muster Out to use the idea. When you run the same brief twice, save it. When you run it every week, put it on a schedule. AI 102 to 104 cover exactly that.
Muster Out keeps your facts on a board and applies the date rule printed on each task.
Self-check
Get six of these eight right to pass. Your answers stay on this page unless you choose to send them at the end.
Put it to work
Run your brief today, then pick what comes next.
AI 102, 103, and 104
Save your goals and answer preferences, then check that new chats use them.
Your own standing instructions and a folder of saved briefs.
Not out yet.Learn to read PDFs, form photos, and spreadsheets, then verify each answer against the file.
A routine for reading a long document and checking every point it gives you.
Not out yet.Turn a brief you run every week into a small tool that runs on its own and tells you only what changed.
A weekly check you can test and inspect, where your chosen tool supports scheduling.
Not out yet.Get Warfare Command updates
Sign up for New lessons and Warfare Command Monthly. Confirm your choices in the email we send. You can change them in Email preferences.
This signup starts New lessons and Warfare Command Monthly after confirmation. Privacy
Leaving active duty in the next two years
Muster Out builds your separation or retirement plan from your date, with the source paragraph next to each task. One message in a Claude chat sets it up in about 5 minutes. You need Claude Pro or higher.