ChatGPT: understand OpenAI and artificial intelligence – 02/22/2023 – Tech

Artificial intelligence has made the news in recent weeks thanks to ChatGPT. As much as the robot knows how to communicate very well —and that was precisely what drew attention—, there are still many doubts about this tool from the company OpenAI.

Understand some of the main points involving ChatGPT and other similar systems, specialized in language, which should appear more and more in the coming years.

What is ChatGPT?

A system based on artificial intelligence that allows you to talk to a robot in a text chat format: the person sends a message and it responds. The tool became famous due to the quality of its writing, which looks like it was made by a human. She is supplied with data from billions of texts extracted from the internet and books and uses this information to formulate the content she writes.

What is ChatGPT used for? What are your main functions?

One of the simplest uses is to answer questions, but it’s important to keep in mind that the tool doesn’t always provide true information. His specialty is making well-crafted texts without necessarily presenting precise content.

Through these questions, it is possible to ask ChatGPT to perform tasks such as translations (by saying something like “can you translate this into Portuguese?” and include a text in the sequence), summarize information, help to improve already written content and create schedule.

How much does it cost?

The service can be used for free or with a $20 monthly plan. In the paid plan, the system is faster and more stable, in addition to promising the early release of updates. For businesses, it’s billed as you go.

How to use?

It is necessary to access the service’s website and create a registration. After that, when entering the service with your email and password, the user is already taken to the chat screen.

Does ChatGPT have an owner? Who is OpenAI?

ChatGPT belongs to OpenAI, who created the tool. OpenAi was created in 2015 as a non-profit research center, but started to act in search of profits from 2019, with some restrictions. It takes money from venture capitalists specializing in the technology market, such as Sequoia Capital and Bedrock Capital.

What are the roles of Elon Musk, Microsoft and Google in ChatGPT?

Billionaire Elon Musk was one of the founders of OpenAI, responsible for ChatGPT. He stepped aside in 2018 citing a conflict of interest with his car maker, Tesla, which would invest more in artificial intelligence.

Microsoft is an OpenAI investor. He started to put money in the company after Musk left, adding US$ 3 billion (R$ 15.6 billion) between 2019 and 2021 and, in early 2023, announced another US$ 10 billion (R$ 52 billion). In addition, artificial intelligence systems such as ChatGPT will be applied to the technology giant’s products, such as the Bing search engine.

In the development of artificial intelligence, Google and OpenAI are competitors. The technological base used in ChatGPT, however, is part of a technique developed by professionals from the search company, published in the form of a scientific article in 2017.

How was ChatGPT made?

ChatGPT is built on top of another OpenAI product, GPT-3 (which stands for “pre-trained generative transformer”). The expression refers to a language model, that is, a rather complex mathematical account that calculates the probability in strings of words.

By calculating these probabilities of associations, the model is able to generate texts that mimic those produced by humans. It’s good to point out that he doesn’t understand what he’s talking about, he just puts the words together.

How does ChatGPT learn?

The area of ​​artificial intelligence has gained prominence in recent years thanks to techniques that allow computers to notice patterns in large data sets. An example is to supply a system with thousands of photos of cats and let it detect what is common there, in a process called training. Afterwards, confronted with an image, he can say whether it is a feline or not.

In the past, a human being would need to translate into programming all the parameters that indicate it was an animal, which is practically impossible.

In the case of GPT-3, the foundation of ChatGPT, there are billions of texts extracted from books and the internet. The system, or model, is capable of noticing patterns of word associations in writing to generate new texts.

What are the limits of what ChatGPT knows?

ChatGPT was trained with data available on the internet up to 2021, so it does not have information about events that occurred after that date.

It is able to maintain context only during conversations. That is, if the user says “my name is Raphael” in a message and then asks what his own name is, ChatGPT will be able to respond. In a new iteration, the same question would be left without the reply.

What is an LLM (Large Language Model)? Is ChatGPT the only one out there?

They are algorithms such as GPT-3, the base of ChatGPT, created from the analysis of a large amount of text (over billions) and capable of manipulating words for tasks such as generating sentences, translating and summarizing content. Competitors to OpenAI in this area, such as Google and Meta (which owns Facebook), have their own models.

What is the difference between ChatGPT, GPT-3 and GPT-4?

ChatGPT is an application derived from GPT-3, the algorithm that allows the generation of text from the analysis of billions of websites and books. It’s as if this model were the chat engine: people interact through conversation and an adaptation of it that is generating the phrases behind the scenes. The GPT-4 is a new version of this system that is under the hood and, according to The New York Times, should be launched in the first quarter of 2023.

What is artificial intelligence?

It is an area of ​​computing that seeks to imitate human thinking and learning ability to do complex tasks, and may be able to make decisions. It appears in technologies such as answering robots, virtual assistants (such as Siri and Alexa) and electronic games.

In their most recent versions, these systems are based on algorithms created from computational analysis of large data sets. By detecting patterns in x-ray images, for example, it can help identify disease.

What is generative artificial intelligence?

It is a field of artificial intelligence focused on generating new content. They are systems capable of creating images, texts, audios and videos.

What is the difference between strict and general artificial intelligence?

The existing artificial intelligence mechanisms today are all extremely specialized, which is why they are called strict. A system capable of playing chess is not suitable for translating texts, for example.

The general one is the one often seen in fiction: a machine capable of understanding everything or learning any skill, like a human being. Some experts say it will never happen, others say it’s a matter of years.

What is NLP, or Natural Language Processing?

It is the area of ​​artificial intelligence that includes domains related to language. It contains mechanisms linked to tasks such as comprehension and voice generation, analysis of written content, text production and translation.

What is machine learning (machine learning)?

Area of ​​computer science, used in artificial intelligence, that gives computers the ability to learn without being explicitly programmed. These are the techniques used so that a computer can detect patterns in a data set without much human intervention, used in ChatGPT.

What are neural networks and deep learning?

Deep learning is a type of machine learning that uses specialized algorithms, separated into several layers, to understand complex structures in data. The way information is processed is inspired by the connections of human neurons, which is why it is called an artificial neural network.

The ChatGPT algorithm was created using deep learning techniques.

Does ChatGPT lie?

No, but it can pass incorrect information. Since it is an artificial intelligence system, he has no intention of saying anything wrong. Users, however, have already reported wrong answers to questions asked on the platform – its specialty is generating and processing texts, not serving as an oracle.

What is bias in artificial intelligence and how does it impact models like ChatGPT?

Bias is a phenomenon that occurs when the results produced by an artificial intelligence algorithm are impaired by problems in the technology’s training process. An example of this is when a hiring system chooses many more men than women, because it was based on a bank of resumes with a greater presence of men.

As ChatGPT was based on data extracted from the internet, it can run into problems when reproducing behaviors found online — such as providing false information or reproducing racist speeches.

The system is also trained to avoid answering some requests, but it can be very strict or manipulated to circumvent these constraints (as when giving instructions on how to make a cocktail molotov). OpenAI says it is investing in research to correct this type of bias.

In the process of trying to identify abusive and illegal content on the platform, the company was criticized for outsourcing work in Kenya paying less than US$ 2 (R$ 10.40) per hour to employees, according to Time magazine. They needed to analyze potentially violent and offensive content, such as pornography and sexual abuse.

Does ChatGPT speak Portuguese?

Yes, but it works better in English.

Does ChatGPT think or understand what we write?

No. The idea of ​​a robot that does something similar to human thinking, artificial general intelligence, only exists in fiction. ChatGPT also does not “understand” the meaning of what is said and what it responds to. What it does is detect word association patterns to form your texts.

Is ChatGPT a chatbot? What is that?

Yes. ChatGPT is a chatbot, that is, a robot made to talk to human beings. He’s not the only one of his kind. With different capabilities, systems like this have been around for years and are used for tasks such as customer service.

Can ChatGPT do famous things that other artificial intelligences have already done, like play chess?

No. The existing artificial intelligence systems today are focused on specific tasks. ChatGPT has language-related capabilities, so it would not be able to generate images, play chess or detect financial fraud, for example. All these functions can be done by robots, but a different one is needed for each one.

What is the impact of ChatGPT and artificial intelligence for fake news?

Generative artificial intelligences can be used to create an avalanche of fake content without much effort. In the same time that a person writes a text, ChatGPT can create several. Therefore, there are studies that try to create systems to automatically identify content generated by robots.