discover the best techniques for 'commanding' an AI

All about Artificial intelligence

For those who want to create an attractive CV for the market, there are some skills that present themselves as almost essential differentiators. Mastery of strategic software, such as Excel, or knowledge of languages, such as English and Spanish, are examples of skills desired and valued by companies. Within this context, new knowledge has gained relevance, albeit in a subtle way: the so-called prompt engineering.

Just as it is essential to formulate efficient commands in search tools, such as Google and Bing, to find the desired information, knowing how to interact effectively with artificial intelligence is essential to make the most of these new technologies. This is precisely where prompt engineering works, focusing on the strategic design of instructions, the so-called “prompts”, with a focus on maximizing the assertiveness with which models respond to user requests.

Mastering this skill can be crucial for Brazilians, especially with the population's growing interest in Generative AI. According to the Boston Consulting Group (BCG), 84% of the Brazilian population already knows about Generative AI applications, such as ChatGPT, surpassing the global average of 82%.

The country's taste for technology was also measured by a survey by the Talk Inc studio, which revealed that 63% of the local population has already used some AI service for professional or personal purposes. Of this audience, 70% say they started using the resource less than a year ago.

These numbers indicate that AI has already been incorporated into the daily lives of Brazilians. Therefore, mastering these tools has become essential, especially in the job market. If this skill was previously limited to the area of ​​technology, today it is crucial for practically any professional, as it allows you to reach new levels of productivity and innovation.

Different approaches

The most exciting thing is that mastering this new form of “language” does not require in-depth knowledge of programming or data science, but an understanding of approaches to effective commands.

Let's imagine AI as a genie in a digital lamp, ready to grant your wishes. Who has never seen a joke or meme that makes fun of the fact that the magical character delivers a request in the wrong way precisely because the request was not that precise? With AI, the same can happen. In this case, prompt engineering focuses on preventing this from happening, ensuring the ability to formulate assertive desires so that they are met without surprises.

To do this, it is essential to know some of the different techniques that integrate the concept and that contribute to understanding how the tools work and deal with commands. In this scenario, the main methods are:

Zero-shot prompting

Ideal for simple and direct tasks, this technique consists of formulating a clear and concise command, without providing prior examples to the AI. Using the genie analogy again, it's like asking “bring me a coffee”, without necessarily explaining what type of coffee you expect to have.

Few-shot prompting

When the task is more complex and requires a specific style, the few-shot method is ideal for making the solution more effective. This method consists of providing the AI ​​with some examples that illustrate the type of response desired. It's as if the user showed the genie a type of coffee and asked for something similar.

Chain-of-Thought prompting

For tasks that require logical reasoning and problem solving, chain-of-thought is the most suitable. The technique encourages the AI ​​to “think aloud”, describing the step-by-step process and reasoning behind constructing its response. It's like teaching the genie in the lamp the entire process from preparation to delivery of a type of coffee, and asking him to prepare another similar one.

Least-to-most prompting

Focusing on complex and extensive problems, the least-to-most prompting technique consists of dividing the task into smaller, progressive steps. In this process, AI itself is used to divide an initial request into several steps, ensuring a response that is closer to what is expected. It's as if the person asked the genie: “Make me a cappuccino”. Instead of responding to the request immediately, the genius would create a script with intermediate steps: “First, I must find the coffee beans. Then select the best ones. Then grind the beans…” until finally arriving at the delivery of the cappuccino.

Read more:

Self-consistency

In this approach, the user asks the AI ​​to generate multiple solutions to a problem and then choose the most consistent one. The goal is to ensure that the final answer is as accurate as possible. Using the genie analogy, it would be like asking: “Prepare coffee ten times following this recipe. Then, analyze how many stayed loyal to the recipe. These will probably be the correct results.”

Prompt: we need to learn!

By mastering prompt engineering techniques, you not only optimize your interactions with AI, but also become an active agent in creating innovative solutions. Think about how you can apply these skills in your everyday life and how they can transform your relationship with technology. After all, the future of AI is in your hands – and your words have the power to shape it.