The promise of innovation can hide invisible risks. Educating for AI and with AI, first and foremost, means educating students to solve problems and optimize the possibilities for interacting with this technology. Understand how the complexity of AI can compromise its educational use and how to turn that challenge into a learning opportunity.

First of all, what does “hallucination” mean in the context of AI?

In the context of artificial intelligence, “hallucination” is the term used when a language model generates false information, invents data, or provides answers with no factual basis. Unlike a simple typo or calculation error, a hallucination is characterized by appearing convincing—but being incorrect. This is especially concerning in educational, legal, or medical settings, where the accuracy of information is essential.

Users who subscribe to paid AI services have encountered an unexpected situation: the free version, which is theoretically more limited, seems to operate with greater stability and quality.

In search of more reliable answers, many users opt for paid subscription plans. However, practical experience has shown the opposite to be true: less coherent answers, a higher incidence of errors, and inconsistent results. Interestingly, free systems continue to offer more effective interactions.

When asked about this, the systems themselves explain that the paid version, by offering more features and experimental models, tends to have more bugs—the well-known phenomenon of choice overload. Simply put, more options don’t always mean better performance.

This paradox reveals something fundamental: generative AI still operates in a highly stochastic manner—that is, based on probabilistic processes that do not always converge on the best solution (thus creating a cycle of trial and error with no guarantees).

See also:

• Learning About Artificial Intelligence from the Perspective of Jiddu Krishnamurti and the Inquiry-Based Learning (IBL) Methodology
• Vibe Coding: The Future of Programming Is Here
• 6 Movies to Spark Conversations About Diversity and Inclusion in the Classroom

Learn how to avoid this mistake

Even though they promise improvements, many paid versions enable experimental features that compromise stability. The abundance of features and contextual adjustments increases the likelihood of glitches, with no real guarantees of quality—and the time lost due to bugs is unlikely to be made up for.

Furthermore, it is important to recognize that these shortcomings are part of the evolutionary process of technology. The purpose of this text is not to criticize AI itself, but to offer guidance so that educators know how to use it more efficiently, thoughtfully, and in a way that supports learning.

Factors to consider before signing an IA:

• Try out the free version thoroughly. See if it already meets your needs;
• Compare it with other platforms. The most popular one isn't always the best choice for you;
• Avoid paying for vague promises. Read reviews and ratings first;
• Be wary of the "most powerful" option. In AI, less can be more;
• Use it as a tool, not as an oracle. Always double-check the result.

What NOT to do when you get an error in prompts:

• Do not repeat requests without revising them;
• Don't use vague instructions like "improve this";
• Do not praise what is wrong;
• Don't appeal to people's emotions with AI;
• Do not keep broken contexts;
• Don't insist on continuing a conversation that has gone off track; it's easier to start a new conversation using the information gathered up to that point.

Tips for prompts to avoid hallucinations:

• Use clear, well-structured commands;
• Avoid vague requests such as "improve the text";
• Specify what should not be done;
• Break down complex tasks into steps;
• Provide fixed, unchangeable references;
• Request a technical review before generating the output;
• Establish clear response standards;
• Include the rule: "If you're not sure, speak up."

Using AI in the classroom goes far beyond simply asking students to come up with quick answers. It’s about opening technology’s “Pandora’s box” through computational thinking: observing the process, repeating attempts, identifying patterns, and, above all, approaching each step critically and practically.

This procedural work is what ensures that humans remain in control of technology. If this process is not understood and consciously maintained, AI can reverse the roles—using human behavior as fuel to reinforce its own stochastic error cycles. This not only compromises learning but can also generate “hallucinations” in the user (ironically speaking), undermining the experience of the very person who should be in control of the system: the user themselves.

About the author:

Share on social media: