Chapter 0.2Chapter 0.2. Your First Hour with Claude and ChatGPT

The point

A chat box looks like the whole product. You type, and it answers. But around that box are a few controls, and each one does a job for you. Pick a stronger model for a harder question. Attach a file, and it answers from your file instead of guessing. Let it run code, which it writes for you, and it calculates instead of guessing. Ask for a document, and you get a page you can edit and keep. Put your files and instructions in a project, and every chat there starts with them. Before your first chat, you also make two quick choices: whether your chats may help train later models, and whether the app remembers you.

After this chapter you can open a free account with Claude and ChatGPT, and make those two choices in a minute. You can pick the control a job needs: the same chat or a new one, a stronger model, a file, code, a document or a project. And you can check that an answer came from your file, by finding the line it quotes.

Why it matters

On Monday you ask a chat whether a $7,800 invoice from Midwest Packaging needs the controller's approval. Brightline Wholesale Supply is the book's example company. The answer comes back at once: no, only invoices over $10,000 need it. That was a guess. Brightline's accounts payable (AP) policy sets the limit at $5,000, and the chat never saw it. If you had trusted the answer, the invoice would have been paid without the controller's approval.

So you attach the policy and ask again. This time the answer quotes line 4.1: yes, it needs approval. Then you ask the chat to total this week's six invoices by running code, and it shows each amount it added. You ask for the answer as a document, and you find it again on Tuesday. The chat box was the same all morning. Only the controls changed. So which control does each job need, and how do you know an answer came from your file?