What is ChatGPT and what is it used for?
ChatGPT has opened up a world of infinite possibilities in a wide range of sectors and industries. It benefits different areas of the quaternary sector, such as education, research, the media, the IT industry and ICT, etc., and completely different sectors, such as finance, transport and retail. This is supported by a vast number of publications and scientific studies on the potential use of ChatGPT. These publications cover topics as diverse as its impact on improving reference search services and metadata cataloging in library and documentation services, and others that attempt to prove if, as with other AI-based content generators or bots, it displays a political bias. Consequently, companies and societies have formed high expectations around OpenAI’s ChatGPT.
Therefore, in order to shed a little more light on the matter, we will now delve into what ChatGPT is and the different GPT models that exist, where it extracts information from and some of its practical applications in different sectors to leverage a tool with a potential that has yet to be exploited.
What is ChatGPT according to ChatGPT?
Just as journalists observe, investigate, analyze, and meet celebrities or those implicated in news items, to gather and contrast information, in this article we have asked the protagonist of our research, ChatGPT, what it is and what it is for.
To do this, we registered on the website set up by the company OpenAI, founded by some of Silicon Valley’s leading investors, including Elon Musk, Peter Thiel, Greg Brockman, Sam Altman, Ilya Sutskever, Reid Hoffman and Jessica Livingston.
The process is very simple and consists of very few steps. The user must provide a verification email address and a password. After confirming that they own the account, they are then asked to provide a telephone number to receive a security code, which must be entered at the end of the registration process. Once all these steps have been completed, you can access ChatGPT and write all the questions and queries you wish to ask in the “send a message” section.
Specifically, this is the answer we received when we asked the question relevant to this section: what is ChatGPT?
“ChatGPT is a large language model developed by OpenAI based on the GPT (Generative Pre-trained Transformer) architecture. It is designed to generate human-like text in response to prompts or questions from users. ChatGPT has been trained on a massive amount of text data and can generate high-quality responses on a wide range of topics. It can be used for a variety of applications such as chatbots, question answering systems, language translation, and text summarization, among others”.
We also asked ChatGPT about its origins, and received the following response:
“ChatGPT is a large language model developed by OpenAI, an artificial intelligence research laboratory consisting of engineers and researchers who work to advance and promote artificial intelligence in a safe and beneficial manner.
The origins of the GPT (Generative Pretrained Transformer) architecture on which ChatGPT is based can be traced back to 2017, when a team of researchers at OpenAI first introduced the model in a research paper. Since then, the GPT architecture has undergone several iterations and improvements, culminating in the release of GPT-3 in 2020, upon which ChatGPT is built.
ChatGPT is specifically designed to engage in natural language conversations with humans, using its ability to generate human-like text to answer questions, provide information, and carry out tasks. The model is constantly being updated and refined, allowing it to improve its performance and accuracy over time”.
However, despite the answers provided, we decided to investigate ChatGPT further. To do so, we resorted to different bibliographic resources such as research articles and specialized books in order to contrast the information provided. We found the answers provided by this atypical interviewee, an AI program, which extracts all the information from large datasets, in text format on the public web (web pages, news articles, books and other online sources),to be accurate.
A bit of ChatGPT history
The OpenAI chatbot is the result of research and advances in the field of Natural Language Processing (NLP) and the AI discipline known as Machine Learning.
Some publications point to research in the field of natural language processing as a starting point, specifically in the area of language models based on the Transformer architecture. In 2017, Google published the scientific article entitled Attention Is All You Need, which presented this type of architecture as a more efficient alternative to models based on recurrent neural networks for natural language processing. From here, OpenAI started working on the GPT language model series.
On the other hand, others take the research paper entitled Improving Language Understanding by Generative Pre-Training published in 2018 by the OpenAI research team as a starting point. The aim of the study was to address the problem of learning text representation, i.e. how to make a machine understand and generate natural language. To address this question, the researchers proposed a pre-trained generative language model called GPT (Generative Pre-trained Transformer). This model uses the neural network architecture called Transformer to predict the next word in a text based on the previous context. The model was trained on a large corpus of unsupervised text, meaning that it was not given any specific information about the meaning or context of the words in the text. Thus, it learned to generate coherent text from statistical patterns and word sequences. The researchers analyzed GPT’s performance and rated it as outstanding on a variety of natural language tasks, such as reading comprehension, text generation and machine translation. In addition, they found that the pre-trained model could be easily adjusted to specific natural language tasks, such as text classification or answering questions.
Over the years, the OpenAI research team has made improvements to the GPT model, leading to the development of more sophisticated versions. The initial GPT model had 117 million parameters and was built from a large dataset, in text format, from the Internet.
Similarly, it is worth noting that each new GPT model builds on the previous one, improving the model’s ability to process natural language and produce a coherent and useful output. This is explained in the next section.
The evolution of GPT models
Here is a summary of the evolution of the GPT models developed by OpenAI, from the very first to the most recent model:
- GPT-1: This was the first model of the GPT series launched in 2018. It was trained on a corpus (a set of written or spoken texts that have been collected and organized for linguistic analysis and use in natural language research) of approximately 40 GB of text and had around 117 million parameters. GPT-1 achieved great success in natural language tasks, including text generation, machine translation, responding to questions and text summarization. Specifically, text summarization is a very important aspect that refers to the process of summarizing an entire text into a shorter and more concise version, while preserving its meaning and most important information. This can be done manually or by using automated techniques, such as machine learning algorithms, abbreviated ML (Machine Learning) and natural language processing. The aim of text summarization is to help readers gain a quick and effective understanding of text content, especially in the case of long or complex documents.
- GPT-2: This second version was released in 2019 and had 1.5 billion parameters, approximately 10 times larger than GPT-1. In addition, GPT-2 achieved more remarkable results than its predecessor, generating coherent and convincing texts that were difficult to differentiate from those written by a human being.
- GPT-3: Launched in 2020, GPT-3 was, until recently, considered the largest and most advanced language model in the GPT series, with 175 billion parameters. GPT-3 has made a quantum leap in natural language processing, being able to perform very complex tasks, including creative writing, human conversation, language translation, code generation and answering questions.
- GPT-4: On March 14, 2022, OpenAI released GPT-3.5, as a “first test run” of the system. One year later, on March 14, 2023, and after fixing some bugs and implementing a number of improvements to the theoretical underpinnings, the company released the new version of one of the most advanced autoregressive language models available: GPT-4. This new variant produces safer and more useful answers than its predecessor versions, as well as being able to solve difficult problems with greater accuracy, as it possesses broader general knowledge and advanced reasoning capabilities. According to research conducted by OpenAI, “GPT-4 is a large multimodal model (accepting image and text inputs, emitting text outputs) that, while less capable than humans in many real-world scenarios, exhibits human-level performance on various professional and academic benchmarks.” This conclusion was drawn by the researchers after subjecting GPT-4 and GPT-3.5 to a series of available tests such as AP (Advanced Placement) exams in subjects such as history, biology, mathematics, and calculus. The tests confirmed that “GPT-4 is more reliable, creative, and able to handle much more nuanced instructions than GPT-3.5”.
Where does ChatGPT extract information from?
ChatGPT extracts information from large datasets of unstructured text, including news articles, books, websites, online conversations and other types of text. These datasets are selected and prepared, with the aim of providing relevant and high-quality information. During training, the ChatGPT language model learns to identify patterns and relationships in the data and to generate consistent and relevant responses based on the context of the conversation. However, the model can sometimes provide inaccurate or incomplete answers, as it is highly dependent on the quality and relevance of the training data.
Furthermore, the model does not have access to personal user information and does not store or use information from conversations for commercial or advertising purposes.
Applications or uses of ChatGPT
As explained in the previous section, the language model behind ChatGPT has been trained from large amounts of unlabeled text data, allowing it to generate coherent and relevant responses to different questions and comments from users. Its various applications by functional areas and sectors include:
- Marketing and Customer Service: ChatGPT can be used to resolve frequently asked questions and common problems and provide additional information about the products and services marketed by a company.
- Education: The chatbot can be used in online or language learning applications to provide immediate feedback to students by answering questions and providing examples.
- Research: Within this field, it benefits researchers by exploring specific topics and obtaining useful information from large data sets.
- HR: ChatGPT can be used in human resource management to answer frequently asked questions from employees, provide information on company policies and procedures, and online training.
- Health: In this case, the OpenAI application enables patients to ask questions about their symptoms and get recommendations for treatment or referrals to specialists.
ChatGPT today
ChatGPT’s success is resounding, or at least that is what the figures show. The chatbot has reached “10 million users in just 40 days”, according to Brett Winton, CEO of ARK Investment Management LLC (better known as ARKN Invest), in a publication on Twitter.
chatGPT at >10 million daily users in 40 days
Instagram took 355 days to get to 10 million registered users pic.twitter.com/Q7MWKNvyoN
— Brett Winton (@wintonARK) January 24, 2023
A statement of intent by the head of this global asset manager specializing in thematic investment in disruptive innovation, which is in addition to an article published on the ARKN Invest website entitled OpenAI Plugins Transform ChatGPT Into An App Platform. This publication analyses the product extension for ChatGPT, which allows the AI software to interact with external data and services through plugins developed by different companies such as Expedia, FiscalNote, KAYAK, Klarna, Milo, Shopify, Speak and Slack, etc. These add-ons extend its capabilities, turning it into a kind of application platform that allows you to search for information on a website in real time, book hotel rooms and flights, rent vehicles, etc.
However, there is still a long way to go before ChatGPT, like other natural language models, will see improvements in its ability to understand and produce natural language, as well as achieve greater fluency and efficiency in human-machine interaction. So, we will have to wait for further developments.
Moreover, in March 2023 Microsoft announced that ChatGPT will be incorporated into the most popular products of its Microsoft 365 suite. Named Copilot, the technology will be integrated into its popular workplace productivity tools, such as Word, Excel, PowerPoint, Outlook and Teams, which are used by thousands of users around the world. Copilot’s advantages include combining the power of Large Language Models (LLMs) with business data and Microsoft 365 applications; and the data security and privacy environment it provides, based on Microsoft’s AI principles and the Responsible AI standard. This means that Copilot’s large language models are not trained with specific content or individual user-supplied prompts or, in other words, that the data provided is not used to retrain the model, nor is it sensitive to possible data leaks.
Finally, in terms of process automation, the use of ChatGPT offers a plethora of possibilities. Regarding the e-mails managed daily by companies, it is possible to create compositions that, for example, identify the sender, classify the e-mail by type (e.g., complaint, gratitude, reservations, registrations, etc.), and show a list of the main points contained in the e-mail. This can be achieved without using any programming language, simply by giving a series of guidelines in natural language concerning the interesting aspects of the e-mails. This technology could be applied, for example, to automate the hotel booking process. The process starts the moment an email is received. The e-mail is identified by ChatGPT, which provides a series of data based on the instructions given, such as the type of e-mail, and then automatically sends the customer to the payment gateway to complete the booking. The e-mails can be classified, and the workflow can even be conditioned accordingly.
Finally, in terms of process automation, the use of ChatGPT offers a plethora of possibilities. Regarding the e-mails managed daily by companies, it is possible to create compositions that, for example, identify the sender, classify the e-mail by type (e.g., complaint, gratitude, reservations, registrations, etc.), and show a list of the main points contained in the e-mail. This can be achieved without using any programming language, simply by giving a series of guidelines in natural language concerning the interesting aspects of the e-mails. This technology could be applied, for example, to automate the hotel booking process. The process starts the moment an email is received. The e-mail is identified by ChatGPT, which provides a series of data based on the instructions given, such as the type of e-mail, and then automatically sends the customer to the payment gateway to complete the booking. The e-mails can be classified, and the workflow can even be conditioned accordingly.
