Basic Precautions to Take When Interacting With a Generative AI

With the rise of generative artificial intelligence tools like ChatGPT or Midjourney, talking with an AI is no longer limited to experts. In just a few clicks, they produce text, images, or sound. With their immense potential, they now find their way into both our professional and personal lives.

But before integrating these assistants into our workflows, it is essential to understand how they function and to know the precautions to take. A generative AI has no emotions, no consciousness, no convictions, no opinions. When it produces an answer, it is simply reproducing learned patterns, without creating new meaning. Even if the answers appear smooth and coherent, an AI does not “think.” It does not truly understand the words it generates. This is important to keep in mind to avoid overestimating its abilities.

The AI revolution raises essential questions. How can we protect our data? Avoid being misled by errors or biases? What are the environmental impacts of these technologies? How can we use prompts effectively? Here is an overview of the basic precautions to take when interacting with a generative AI.

Protecting your privacy during interactions

When interacting with an AI, it can be tempting to provide personal details in order to receive more precise answers. However, this information may be stored, analyzed, or even reused to train future models.

It is therefore strongly discouraged to share usernames, passwords, banking information, as well as confidential documents, intimate stories, or highly specific details about your private life.

In practice, if you need to illustrate a personal or professional problem, choose fictional or anonymized examples. Referring to “Company X” or “Mr. Y” helps reduce risks. For example, instead of asking: “Write a message to my son Paul, who lives in Lyon and failed his baccalaureate,” prefer: “Give me an example of a letter for someone close to you who has failed an exam.”

Checking the reliability of responses

Generative AIs can produce very convincing answers that are sometimes entirely false. They may invent references, cite nonexistent authors, provide inaccurate figures, distort facts, and more. It is therefore essential to maintain a critical mindset and to keep a habit of verifying information.

To avoid being misled, ask the AI to provide its sources, cross-check the information with other reliable websites, and remain cautious when faced with statements that are too specific or, conversely, too vague.

AI should be considered a support tool, not an infallible authority. It is not designed to contradict you. If your initial assumption is false, it may still attempt to satisfy you—even if that means giving incorrect, incomplete, or misleading answers.

Understanding and limiting biases

Generative AIs rely on vast datasets collected from the Internet, which reflect existing inequalities and stereotypes. These models are therefore trained on sometimes imperfect data, which may lead them to produce sexist, racist, or discriminatory content.

To limit these biases:

• Formulate your questions in a neutral and inclusive way: ask open, judgment-free questions so the AI responds equitably.
• Ask the AI about its limitations: some are capable of acknowledging their areas of uncertainty.
• Never delegate your critical judgment: every user remains responsible for evaluating the relevance of the content.

Considering environmental impacts

AI is a powerful tool that should be used sparingly. We often forget that behind what appears to be a simple request lie massive, energy-intensive infrastructures. Training AI models requires considerable resources, especially water and electricity. Their daily use also has a significant environmental cost:

Datacenters associated with AI consumed nearly 460 terawatt-hours (TWh) of electricity in 2022, representing about 2% of global production. This share could double by 2026 due to the rise of generative AI. And producing a single high-definition AI image consumes as much energy as fully charging a smartphone (Source: Délégation régionale académique au numérique éducatif d’Île-de-France).

To reduce AI’s environmental footprint:

• Avoid repetitive or unnecessary requests: reserve AI usage for situations where it provides real added value.
• To quickly obtain factual information, instead of systematically using AI, use a search engine with a targeted query.

Using prompts effectively

The clearer and more precise a prompt (instruction given to the AI) is, the more relevant the answer will be. As a central element in the quality of expected responses, it can also be seen as a creative skill. Properly formulating requests, trying different approaches, comparing results… all of this allows for better use of AI while keeping control.

It is essential not to include personal data and not to formulate illegal or discriminatory requests.

Best practices for writing prompts:

• Define the format: summary, list, tutorial…
• Specify the tone: formal, educational, humorous…
• Set the length: number of words or points to cover.
• Ask for examples to enrich the answer.
• Reformulate if the result does not meet your expectations.

Generative artificial intelligence can be considered a support tool: it facilitates brainstorming, provides inspiration, and helps structure ideas. However, the content it produces is only a starting point, which must be revised to ensure the accuracy of the information and to reflect each person’s style and thinking.

Effective production therefore relies on active interaction, where the AI suggests and the user selects or adapts. By integrating these simple precautions, it becomes possible to take advantage of these tools while remaining aware of their limitations.

References:

• Délégation régionale académique au numérique éducatif d'Île-de-France

  What Is the Environmental Impact of a Generative AI?

• Info.gouv.fr

  Generative AI: How to Use It Properly?

• France Num

  How to Create Effective Prompts: Beginner’s Guide