Personal and Work Access

Let's get to practice. The exercises later in the book are done in a chat with a model, so first you need access to one of the services.

Service, model, and chat

Three words come up constantly from here on, and they are worth keeping straight.

  • The service is what you open. ChatGPT, Claude, Gemini, and the like.
  • The model composes the answer inside the service. One service has several, and you pick from a list in the window.
  • The chat holds one conversation. You write in the window, and the answers arrive in the same place.

They differ in use as well. You choose a service once, switch models to fit the task, and start a new chat when the conversation has drifted.

The major services

There are four major services right now. Next to each is the company behind it: that is where you find the terms, the policies, and the data practices.

  • ChatGPT — OpenAI
  • Claude — Anthropic
  • Gemini — Google
  • Grok — SpaceXAI, formerly xAI, Elon Musk's company

Each has a free plan, and it covers everything we will do in this book. The limits differ and change several times a year, so there are no numbers here: check the website.

A personal account

Register a personal account with your personal email. You need it for the exercises: trying things and making mistakes is better done on your own account.

Your company may have corporate access of its own, and you can still use your own account alongside it.

On screen a corporate window looks almost like a personal one, but there is a contract behind it that governs what happens to your data. The legal terms are different.

On top of the contract the company adds its own: extra agreements and a security policy. They decide which models are turned on for you, which work systems are connected to the chat window, and what may be written in it.

Some settings check the text as you type and stop, say, a credit card number from going out. How that check works is covered in the chapter "Request Screening Inside the Company".

Conversations in a corporate account are available to the company. Corporate versions of the major services have an export that returns not only the time and the user name but the questions and the answers themselves. Who gets it is the company's decision: the security team, legal, your manager.

Other model makers

The four companies at the start of the chapter are not the whole picture. Models are trained and tuned by other companies and by individual developers. On Hugging Face, the largest open platform for models, users had posted more than 2 million models by the end of 2025.

The barrier to building and tuning models is lower than it looks from the outside. A developer takes a ready open model and fine-tunes it for one task. A single family of open models on that same platform has more than 113,000 modified versions.

Midsize companies build their own models for their own work.

A robot vacuum maker trains a model to tell what is in front of the machine: a chair leg, a cat, or a sock on the floor. Such a model does not talk and does not make anything up. It solves one narrow task, and solves it better than any general one.

Large governments run specialized projects of their own. For what that means for your data, see the chapter "Industrial Espionage".

You do not need accounts anywhere else: any one of the four services above is enough for the exercises in this book.

Knowing about the rest is useful for another reason. AI is not only a chat window. Models like these run in logistics, medicine, manufacturing, and banking, where they are called recognition, forecasting, or quality control systems.