ChatGPT: step by step, how to use and exploit the artificial intelligence of the moment

Although more than a hundred million people have already tried it Chat GPT since it has been enabled worldwide, the capability of this revolutionary tool still raises many questions. Is he really smart? What is needed? What to ask and how?

Basically it is a chat system based on the AI ​​language model, which is powered by algorithms and text prediction to create original content. The new version (GPT-4) processes up to 25,000 words, about eight times more compared to the previous.

To interview this digital oracle, you will first need to enter the site, enter an email, verify your account, enter a telephone number and enter the PIN you receive via text message. Everything is ready.

“Although chatbots are programmed with specific answers, people prefer to ask them general questions. This is how the Generative Neural Networks were born, such as ChatGPT or Google’s Bard, which reach greater breadth in their expressions”, highlights Marcella RiccilloPhD in computer science, expert in AI and robotics.

Without limits of insistence, you can demand all the clarifications you need, without paying a penny. If the argument he uses is somewhat vague, it will be necessary rephrase the sentence until you get the idea. All chats are saved in a history.

“Generative chatbots –adds Riccillo- compose their own sentences using words as puzzle pieces. But they don’t understand, hear or know if what they return is correct. To create the texts, they combine information from the data that has been used for their formation.

ChatGPT: why use it

ChatGPT’s ability to understand the context and the intention behind each query makes it a very useful tool for developing chatbots, improving the accuracy of search systems, and generating reviews in the form of articles or replies.

Unlike the Google search engine, which memorize the pages on a server, ChatGPT stores all information. Replies of him are expressed in a natural way, with detailed information, which they do very difficult to distinguish that the text was generated by an artificial intelligence.

“Currently these chatbots are being incorporated into the internet browsers. The idea is that a concise response is generated instead of delivering a list of sites in columns of importance,” warns Riccillo.

One of the most seductive aspects of ChatGPT is its ability to work within certain imposed limits. For example, you may be asked to define a concept in 10 statements.

Another option is copy entire articles or notes to summarize inside, for example, 100 words or three main ideas. These concrete numbers mean that the AI ​​sometimes works better and doesn’t stretch too much.

In turn, he is able to do this review the already written code and not just detect bugs, but it helps to understand the error through some concrete examples. Just paste the text and tell it to debug your code.

Unlike online translators, GPT Chat splits text into smaller units, such as words or sentences, and applies its neural network to generate the translation based on context and meaning.

“When translating text, it is convenient to check the results produced by generative chatbots, as they are known to be wrong. Creativity, common sense and experience continue to be attributes of man, which the machine does not have and may never have”, sums up Riccillo.

ChatGPT: how to ask the right questions

One of the keys to communication is in the armed with request. The term refers to the “input” or message that initiates the request. This input can be an order, a question, a statement, or even a conversation topic.

“ChatGPT is a language model that can be trained with specific data for each work area. An example is Provide the prompt with an article and text related to the topic to be discussed. In this way, the model will be able to generate more precise and relevant answers,” says Mauro Avendaño, director of Digital Technology Argentina.

Another of the advantages of ChatGPT is that it can be adapted to the level of knowledge of the interlocutor. The more complex the requests, the better the results that can be obtained. The key is to provide clear and concise instructions.

“To receive more precise conclusions, it is advisable to be very specific. For example, instead of asking What’s happening in the world? it is more accurate to say What happened to the last United Nations Security Council? This prevents the model from scattering into generality,” suggests Avendaño.

Furthermore, it is possible to generate dynamics between the chatbot and the user to simulate certain situations or scenarios in which the AI ​​can, for example, assume the role of a god, a public figure or a 17th century astronomer: “What Did Galileo Galilei think about the religion of his time?

“In the more technical and detailed aspect, the prompt has some adjustable parameters to improve the answers according to what we expect from them,” explains Avendaño and lists:

• The “max_length” parameter controls the maximum length of the response. If you need a short answer, set it low.
• The “temperature” parameter controls the randomness of the response. A low value makes the response more predictable, while a high value makes the response more surprising and creative.
• The “top_p” parameter controls the cumulative probability of words generated in the response. It can be used to generate more accurate and consistent responses, by setting “top_p” to 0.1.
• “Repetition_penalty” controls the penalty for repeating words or phrases in the generated response.
• “Num_beams” controls the number of responses that are generated at the same time.
• “Min_length” controls the minimum length of the generated response.