The same questions arise repeatedly: What is this tool? Is it Google? Is it alive? Why did it provide false information? Can it be used at work? Am I already behind?
At a certain point, it is more efficient to answer these questions once in writing. That work became a book: A Beginner's Guide to ChatGPT, in English and in French. It covers the origins of ChatGPT, how to write an effective prompt, what ChatGPT Plus and the new plugins add, how the tool is being used in education, business and marketing, and the more difficult questions of privacy, copyright and employment.
Writing a book about a tool that changes every few weeks is an unusual undertaking, and we chose to proceed regardless.
Why write a ChatGPT guide for beginners?
Because beginners are largely underserved. There is no shortage of content about ChatGPT: threads promising "100 prompts that will change your life," and videos of people earning money through methods closely resembling earlier get-rich-quick schemes in a new form.
What is lacking is material for people who have never used it: shop owners, teachers and employees who have been told to "learn AI" without knowing where to start. Most available content moves directly to techniques and assumes the reader already knows what a prompt is.
Language is another factor. At present, most quality material is in English. A French-speaking reader in Quebec receives, at best, translated tips, often in European French and usually several months late. With a French edition, readers here do not need to wait for someone else's translation.
The news offers beginners little clarity either. Italy's privacy regulator blocked ChatGPT at the end of March. This week, OpenAI added a setting to disable chat history, so that those conversations are not used to train its models. For newcomers, such headlines add confusion: is the tool dangerous, and is it even permitted? The fundamentals must come first for any of this to make sense.
Why should beginners learn the vocabulary before the techniques?
Because vocabulary is where they first encounter difficulty. The natural approach when writing a how-to guide is to start with tips. For this audience, that order is reversed.
A reader who does not know what a "prompt" is cannot act on advice to "write better prompts." A reader who has not been told that the model has a limited memory within a conversation cannot understand why it forgot something said twenty messages earlier. A reader unfamiliar with the term "hallucination" may assume the machine is deliberately lying, or that they made an error.
A beginner's essential vocabulary fits on a single index card:
- Model: the system producing the text, and the fact that there is more than one.
- Prompt: what the user types, including all the context provided.
- Token: the units of text the model counts, which is why long documents reach a limit.
- Context window: how much of the conversation the model can retain at once.
- Hallucination: an invented answer stated with confidence.
Once these five terms are understood, the remaining material is far easier to absorb. We begin with them for every new learner.
Why are examples more effective than theory?
The principle "give it context" can be explained in a paragraph, and readers will nod politely. Placing a weak prompt beside a strong one, with both answers shown below, produces genuine understanding.
This is why we prioritize real examples over theory. Most readers do not need to understand the internal workings of a language model. They need to see what was typed, what was returned and the version that works.
Working in two languages adds complexity. An example that works in English does not always work in French. Humour does not always translate, and a business example set in California has little relevance for a reader in Rimouski. Examples for French-speaking readers must be written for them, not simply transferred.
How can a book about AI remain useful when the tool changes every month?
The central challenge of a book about ChatGPT is that the tool keeps changing during the writing process. GPT-4 was released in March, midway through the writing. Plugins were announced about a week later, and most users still cannot access them. Features appear, are renamed and move behind the Plus paywall.
Every screenshot becomes outdated within weeks, and any sentence beginning "Currently, ChatGPT can..." has a short lifespan.
Our conclusions, which apply to anyone writing about a rapidly evolving tool, are as follows.
Publish promptly. Waiting for the technology to stabilize means never publishing. The people who need guidance need it this spring, not next year.
Write for the reader, not for the version. The sections expected to remain relevant concern practices: how to ask clearly, how to verify an answer and why private information should not be pasted. The sections expected to age concern interface buttons.
Then update. Ebooks and print-on-demand make this possible in a way that was not feasible twenty years ago. A guide to a moving target must be treated more like a website than a book: a resource that is revisited and corrected.
Which content remains useful the longest?
Content about practices rather than interfaces. The pages most likely to remain useful in a few years are those about how to think with these tools, with very little about where to click.
This principle extends well beyond books. A help page, an FAQ or a training document for a business is more effective when the "why" is kept separate from the "which button," since the button will move. For businesses that need this type of clear writing in both languages for their own customers, we recommend professional bilingual copywriting rather than a machine translation of the English version.
Beginners are far more numerous than the technology sector generally assumes. They do not need another list of prompts. They need clear explanations of the vocabulary, one good example, and confirmation that scepticism is appropriate.