Cómo mejorar las respuestas de LLMs

Modern artificial intelligence can be a powerful tool for obtaining personalized answers to our questions, but we often do not take advantage of its full potential. Discover how to significantly improve the quality of answers from large language models (LLMs) using a simple but effective technique that will allow you to get truly useful information tailored to your specific needs.

How to get more effective answers from AI models?

One of the simplest and most effective techniques for improving the responses of large language models is to ask them to ask you questions step-by-step. This strategy is particularly useful because both you and the machine are unaware of certain aspects relevant to solving your query. The absence of shared experience and knowledge can significantly limit the value of the answers you get.

When you ask a simple query such as "I want to decide what to invest ten thousand dollars I have in a bank account", most LLMs will respond with a generic and lengthy text that probably does not fit your specific needs. This happens because the model lacks contextual information about your particular situation.

Why is it important to add context to your prompts?

Providing additional context is the first step in improving responses. For example, if you add information such as "I am from Mexico City, I am twenty-five years old and I am employed", you are already giving the model more specific data to work with. However, even with this additional context, the answer may still be too general.

The real trick is to include specific instructions that allow the model to get the information it needs to give you personalized advice. The key phrase is: "Ask me all the questions you need to know about me to give me better advice."

How to structure your prompt correctly for best results?

To maximize the effectiveness of this technique, you should include the following instructions in your prompt:

1. "Ask me all the questions you need to know about me to give me better advice."
2. "Ask me one question at a time" - This prevents the model from bombarding you with multiple questions at once.
3. "Don't limit yourself in the number of questions" - Some LLMs have token restrictions per response, so this instruction encourages them to be thorough
4. "Only answer my request when you have asked me all the questions you need" - This ensures that the model gathers all the necessary information before giving you a final answer

How do different AI models respond to this technique?

When applying this technique with different AI models, varied results are observed:

• ChatGPT-4o: Asked 19 specific questions about the investment situation before providing a detailed answer about investment accounts, risk levels and ETFs.
• Gemini 2 Flash: Asked 8 questions and provided an answer focused on prioritizing emergency funds, paying off debt and crowdfunding options.
• Claude: Also responds with specific questions tailored to the initial query.

The difference in the number and quality of questions demonstrates how each model processes the instruction differently, which can result in final answers with different levels of personalization.

In what other scenarios can you apply this technique?

This strategy is not limited to financial investment queries. You can apply it in a variety of contexts such as:

• Personalized diet optimization
• Exercise routines tailored to your needs
• Health consultations
• Travel planning
• Professional decisions
• Any situation where you need personalized advice

The key is to allow the model to discover what additional information it needs to provide you with the best possible answer, even information that you yourself would not know is relevant to ask.

How to choose the right AI model for your needs?

It is advisable to test this technique with different AI models to determine which one offers the best results according to the type of query:

• ChatGPT-4o
• Claude 3 Sonnet
• Gemini 2
• Flame 3
• Other available models

Each model has its strengths and weaknesses, so experimenting with several will help you identify which is most effective for your specific needs. Some models will ask more questions and provide more detailed answers, while others may be more concise but equally useful.

This simple but powerful technique will allow you to take full advantage of the potential of great language models, getting truly personalized and valuable answers. We invite you to try it out and share your experiences in the comments, with which model have you obtained the best results?