What is ChatGPT?
ChatGPT is a pre-trained language transformer. It is therefore a chatbot that can answer any question, even the most abstract, asked by the user. After typing the command “write” or “translate”, it will create any text for us, according to our intentions (these only need to be entered precisely in the chat window). ChatGPT was developed by Open AI and is a generative model of artificial intelligence. This means that it can produce completely new data on its own. It is a language model that can produce various texts, including stories and poems but also code in programming languages. ChatGPT is an experimental technology whose capabilities are constantly growing. GPT-3 is an autoregressive model, i.e. one that improves without human intervention. In simple terms, this means that it is constantly searching its resources for answers that will best meet the needs of its users.
ChatGPT – how does it work?
ChatGPT is a closed model, which has been supplemented with huge amounts of text before being put into use. It works online but does not retrieve data from the web in real time. In contrast, open models search the Internet for the best answer, which is then transcribed and quoted. The first version of the GPT-2 model was shown to the world in 2018, and it took six years to add to its knowledge! This illustrates the scale of the data that ChatGPT absorbs before being implemented. The current version of ChatGPT-3 has information up to 2021. Thus, it is in vain to look in it for answers to questions regarding at least the February 2022 escalation of the conflict in Ukraine. The primary function of ChatGPT is to converse with the user. When a question is posed to the model, it searches its database for an answer and selects the one closest to the user’s expectations. Thanks to the fact that the model has a huge semantic database, it can not only find the right key phrases in our queries but also understand our intentions. This is why ChatGPT’s answers are so relevant and natural.
As a feature of ChatGPT, it preserves the context of a conversation – so we can naturally inquire about elements that have previously been discussed during a chat conversation, such as changing the ending of a story without specifying its content, for example.
ChatGPT – where is it being used?
ChatGPT has attracted interest in almost every industry, far beyond the IT world. Because it handles natural language so well, it is readily used in marketing – for creating all kinds of content and translation. Developers, graphic designers, and designers have tested its capabilities since the beginning. Chat can also be used to create code, as well as Chatbots that can chat with people and solve their problems. The model is also being used in the scientific community – by students or pupils, but also by academics who are constantly testing its potential.
There are two ways to use ChatGPT:
either directly through the website provided by OpenAI (anyone can create an account and use limits) or
through the API. The way with the use of the API, allows ChatGPT to be used in many other external products.
Its use is possible wherever content is generated: text, image, or sound. It is therefore being used on a large scale by the creative industry. However, the question of copyright remains a significant problem. The model does not specify the sources on which it creates content, which makes its use not entirely ethical. It also happens that copywriters or other creatives copy text generated by AI 1:1, without specifying that the text was created using ChatGPT. Solving these issues is the most important challenge we will face soon.
How do I use ChatGPT?
To use ChatGPT, the easiest way is to go to the openai.com website, where you will need to create an account. You will need your email address and password to log in. You will then need to verify your email address, as well as an additional account confirmation using your phone number. Once our account is ready, under the ‘Playground’ tab we can ask questions and test the model’s capabilities.
What are the advantages and disadvantages of this solution?
The biggest advantage of ChatGPT is its ability to generate natural language. This means that its answers are close to how a human would answer. In addition, users may use the model to create answers to questions, product descriptions, articles, scenarios, and translations. The speed of responses and access to information in one place is also a huge advantage of the model.
As for the disadvantages, it is essential to remember that ChatGPT is trained on textual data that may contain biases and stereotypes. The huge textual resources that the model has been ‘fed’ are hard to verify 100%. Therefore, some responses may contain racist or sexist content. A lot also depends on the intentions of the users – unfortunately, there are already known cases where people deliberately ask the model questions that may provoke inappropriate content. When using the model, it is therefore important to be aware of these limitations. Further obstacles arise with more advanced queries, where ChatGPT answers can be inconsistent and lack logic. The model also cannot cope with queries about individuals. If we ask questions about politicians, writers, or actors, the answers contain true elements but are interspersed with confusing facts. Biographical information is readily accessible from Wikipedia, for instance, from which data was loaded into the model during training. Therefore, it is reasonable to assume that ChatGPT was not programmed to respond to individual questions.