Capabilities, limits, and the words we will use

List what Grok is generally good at, what it cannot reliably do, and define the basic terms used in the rest of the course.

Where supported, Grok can answer questions, draft and rewrite, brainstorm, analyze text, help with code, reason through multi-step problems, look at images, search the web, search X, work with uploaded files, and on some products generate images, video, or voice. That list is a menu, not a promise that your account includes every item.

What it cannot reliably do: know unpublished private facts, replace licensed professional advice, guarantee legal/medical/financial correctness, remember your life unless you (or a product memory feature) provide it, or remain current without search tools.

Core terms: Prompt (your instructions). Context (everything in the conversation and attachments). Context window (how much it can hold). Hallucination (a fluent falsehood). Token (a chunk of text). Tool (an extra capability such as search). Plan (what your subscription allows).

We will ask Grok to map its own limits — and then we will not take the map as gospel.

Create two lists for a new Grok user.

LIST 1 — Generally strong uses
Everyday questions, writing help, brainstorming, explaining ideas, coding assistance, structured analysis. For each, add a one-line "watch-out".

LIST 2 — Do not outsource
Medical diagnosis, legal conclusions, live prices without a source, private data about a person, anything you would be ashamed to be wrong about in public.

Then define these terms in one sentence each: prompt, context, hallucination, tool use, knowledge cutoff.

Constraint: If a capability depends on a plan or product surface, mark it "where supported". Do not invent button names.

You are collecting a working map, not a contract.

The watch-outs matter more than the strengths. If List 2 is shorter than three items, the model is being a salesman. Push back.

Critique your previous lists. Where were you too confident? Add a third list: “Questions I should ask my product’s current help docs, not you.”

This installs the habit of using Grok as a coach and the vendor docs as the spec.

• Memorizing a feature list from a YouTube video dated last year.

  Re-check grok.com and docs.x.ai when it matters.

• Using jargon (tokens, context window) without a picture.

  Context window = the size of the desk, not the size of the library.

• Assuming image, voice, or file upload is always there.

  Mark them “where supported” until you see them.

A one-line glossary in your notes beats a 40-page PDF you will not reopen.

Write your own two lists after the demo. Star any item you have not personally seen in your Grok surface.

Hint: Stars become experiments, not beliefs.

Pick one thing you wish Grok could do (for example: ‘see my calendar’). Ask whether it can, and what workaround exists if it cannot (usually: paste a snippet).

• Wide capability, conditional reliability.
• Professional advice and private facts stay human-owned.
• Learn five terms: prompt, context, hallucination, tool, cutoff.
• “Where supported” is part of literacy, not hedging.