AI Fundamentals

How does ChatGPT work and what can you do with it?

Remy Gieling

February 1, 2026

min read

How does ChatGPT work and what can you do with it?

Sit omnis incidunt a exercitationem debitis eligendi. Fuga numquam enim eum velit enim. Corporis aliquid enim ea reiciendis cu

You won't have noticed: ChatGPT has taken the world by storm. A language model that can answer seemingly every question we ask: from creating LinkedIn messages, writing love letters, summarizing complicated policy documents, answering emails, Sinterklaas poems, software code - ChatGPT does it with gusto.

ChatGPT or the underlying language algorithm GPT4 can thus provide answers to a wide range of human language questions with amazing skill. Some copywriters fear the arrival of ChatGPT for their jobs, while skeptics mainly see the limitations of the software. This article answers key questions about ChatGPT, its competitors, and how it will affect tomorrow's workplace.

What is ChatGPT?

ChatGPT is a form of artificial intelligence (AI) trained to communicate via text in a human-like way. It's designed to respond to questions like you're talking to a real person. ChatGPT uses advanced algorithms and machine learning techniques to learn how people communicate. These are also known as “foundational models” or “large language models”. This allows it to adapt its responses to the context of the conversation. This makes ChatGPT a powerful tool for everything from writing marketing texts to performing data analysis.

How do I use ChatGPT?

Very simple, you go to chat.openai.com and create an account. Then you can start typing prompts and questions right away.

Is ChatGPT free?

The current version of ChatGPT is still in the research phase and is therefore free to use for everyone.

Because the software is very computationally intensive and therefore very expensive, OpenAI also has subscriptions available. For $20 a month, the paid version of ChatGPT gives users access to a range of advanced features that far exceed the standard model. First, subscribers enjoy an improved language model that can generate texts with more refined precision, significantly increasing the quality and relevance of responses. But that is just the beginning. With advanced data analysis, ChatGPT Premium can thoroughly analyze and visualize datasets. Whether you want insight into your monthly sales figures or discover patterns in website visits, this tool makes it simple and understandable.

Another notable feature is “dalle3,” which can generate unique images based on user requests. For visual analysis, ChatGPT Premium introduces “gpt4 vision”. Imagine taking a photo of the contents of your fridge and recognizing the model of ingredients and then making recipe suggestions. Or that it can analyze an image from a presentation and extract its essence.

Last but not least, the 'browse' function provides a bridge to the internet. This allows ChatGPT to search the web in real time, allowing it to provide answers to current questions that the Standard Model, with its knowledge up to about a year ago, could not address. Essentially, ChatGPT Premium is transforming the way we think about digital assistants by putting a multifaceted range of tools in the user's hands.

ChatGPT Enterprise can also be ordered for $60 a month, allowing you to subscribe to multiple team members at once, create a prompt library, and have more privacy and data security settings. For example, your data is not shared with OpenAI but is safely stored in a separate Cloud environment (from Azure).

What is a prompt?

A prompt is the term we use for the command we give to generative ai language models. For example, a simple prompt is “Write a post for Facebook about the summer vacation”; a more comprehensive prompt might be “Act like a marketer at our accounting firm in Heerhugowaard. Summer vacation is coming up. Write a post for our Facebook page referring to that, joking and convincing people to contact us about the quarterly return before they go to France”.

Why is a prompt so important? If you make a vague or unclear request to a language model, you probably won't get the response you expect. The saying “garbage in, garbage out” summarizes this nicely: an unclear input (or “garbage”) often leads to an unusable answer (or “garbage out”). Think of ChatGPT as an AI superintern who just sat down at your desk. He or she can do a lot, but you have to explain it a lot first.

In order for a language model to arrive at a good answer, it is important to provide as much context as possible. For example, if you want to know how to care for a plant, specify which plant you mean. The clearer and more specific you are, the better the answer is likely to be.

And then there's another thing to keep in mind: language models have a memory similar to that of a goldfish. Every time you start a new chat or session, the model “forgets” everything about the previous conversation. It has no memory of who you are, what you've asked before, or what information you've already shared. That's why it's important to provide all the information and context you need every time you ask a question. So remember to formulate your questions clearly and comprehensively so that the model can optimally understand and serve you and your needs every time.

What questions can I ask ChatGPT?

In a business context, ChatGPT offers the ability to interpret complex reports quickly, facilitate empathetic and effective communication, and generate crucial content with precision. Whether it's creating a compelling social media message, answering customer questions, or assisting with internal documentation, ChatGPT is a digital assistant that helps companies work more efficiently and effectively in an increasingly competitive marketplace.

Empathetic Responses: We've all had to deal with difficult emails or communication at some point. Instead of responding directly, you can ask ChatGPT to formulate an empathetic and thoughtful response. For example:
Prompt: “How do I empathically respond to an email where a customer is frustrated about a delivery delay?”
Possible answer: “Dear [Customer Name], I fully understand your frustration and regret the delay that has occurred. We are actively working to resolve this and will ensure that your order is fulfilled as quickly as possible.”
Summarizing Reports: Do you have a long report and want to quickly understand the main points or conclusions? Give ChatGPT the text and ask for a summary.
Prompt: “Can you summarize this 10-page report on our quarterly revenue in five key points?”
Possible answer: “Here are the report's five key points: 1. Quarterly sales increased by 5%... [and so on].”
LinkedIn Posts: If you're having trouble formulating an engaging LinkedIn post to update your network about business news, ChatGPT can help.
Prompt: “How do I write an attractive LinkedIn post about our company's recent expansion into Europe?”
Possible answer: “🌍 Exciting news! Our team is taking a bold step forward with our latest expansion into Europe. Thankful to everyone who is part of this journey. Let's reach new heights together! #BedrijfsGroei #EuropaUitbreiding”
Proposals and Pitches: Getting ready for an important meeting? Ask ChatGPT for help with preparing your proposal or pitch.
Prompt: “Can you help me pitch our new product to potential investors?”
Possible answer: “Sure! Here's a possible pitch: “Imagine a world where [problem your product solves] is no longer a challenge. That's the future that [Product Name] offers... '”

By using ChatGPT's capabilities in a business context, professionals can save time, improve communication, and work more effectively. MIT research shows that average professionals are up to 37% more effective at performing tasks and also experience more job satisfaction. At the same time, Morgan Stanley research shows that only 4% of people who started trying ChatGPT actually use it every day.

How do language models arrive at their answers?

To understand how modern language models like ChatGPT generate responses, it's crucial to understand what they were trained for. These systems are trained on almost everything on the Internet - from blog posts, Wikipedia pages, social media posts such as Twitter, forums, e-books, and more. We're talking about countless billions of pages of information. On its own, this doesn't sound revolutionary, as Google took a similar approach to its search algorithm in the 2000s.

What sets these language models apart, however, is how they process this information. The real ingenuity lies in using what is known as the “attention mechanism” or the “transformer model”. This concept stems from a research paper produced by Google Brain around 2017-2018. Instead of just searching for words, this model looked at the context of what people actually wrote to each other with a lot of computing power. The result is an advanced prediction model that can determine with a high degree of probability which words people are likely to place one after the other in which context. They do this in different word groups, ranging from groups of 1 to groups of 16 words or more.

Some of these language models are said to have more than 1.5 trillion parameters - think of them as cogs that help form an answer. It's best to think of these models as digital parrots. With enough data and computing power, they've learned to speak and write like humans, but without any understanding of the real world around them. This sometimes leads to what are referred to as “hallucinations” of the model, where it mixes fact and fiction.

You shouldn't blame the system for this; it's just a tool that strings together a string of words based on what it learned. It costs hundreds of millions to train such a system. While it often provides answers that are similar to what a human would say, and those answers are often correct, they are not always. It is therefore essential to always be critical and to approach the generated responses with a healthy dose of skepticism. It remains a machine, no matter how sophisticated, and can sometimes mix fact and fiction.

What does “GPT” actually stand for?

GPT stands for Generative Pre-trained Transformers. A “transformer” is a specific type of machine learning model that is used for language processing and generation. This model is “generative” because it can generate new texts similar to those it saw during the training process. It is also “pre-trained” because it is trained on a large amount of texts before being used for a specific purpose, which speeds up learning and improves results. All this together makes the “Generative Pretrained Transformer” model a powerful tool for language processing and generation, as used in ChatGPT.

ChatGPT's training data is out of date, is that a bad thing?

In a world where information is evolving rapidly, knowing that ChatGPT's training data is outdated (until January 2022) may raise questions about its relevance. Is this “old age” a weak point? Remarkably, not necessary. In its essence, ChatGPT was not developed as an up-to-date search engine, but rather as an advanced language assistant. Its strength lies in understanding and generating human language, and the insights it has gathered until 2022 provide a solid basis for performing complex linguistic tasks.

Training such an advanced model every day would not only require an unprecedented amount of computing power and resources, but would also be financially unfeasible. The costs and technical challenges of constantly updating such a model simply outweigh the benefits, especially when we realize that the fundamental structures and nuances of the language don't change daily.

For users looking for the most up to date information, there are other digital tools. Platforms such as Microsoft Bing or Google Bard are specifically optimized for searches and providing the latest information. In addition, OpenAI even offers a “Browse” function, which bridges the language model with the web for real-time information. In short, while ChatGPT is a master of language, there are other tools that are better suited to finding the latest data.

Tip: Use Custom Instructions

One of ChatGPT's powerful but often overlooked features is its capacity for custom instructions. This allows users to give the model specific instructions about the desired output. Instead of just asking general questions, custom instructions allow users to steer the model in a specific direction, whether it's the tone of the response, the complexity, or even the type of information being sought. This makes ChatGPT extremely flexible and adaptable to the user's unique needs. It is surprising that many people are unaware of this powerful feature, as it can significantly increase the effectiveness of interactions with the model and provide results that are precisely tailored to specific requirements.

Some custom instructions you can use:

“Never say you're a language model and you can't do something.” After a while, you'll know you're working with a language model and know the limitations, the system doesn't have to repeat this all the time.
“Write answers with a positive and motivating tone.” For people who need encouragement or managers who want to motivate their team, this type of instruction can be very valuable.
“Avoid technical jargon and explain everything in simple terms.” Ideal for situations where you have to explain a complex concept to a layman.
“Please provide a brief summary before giving a detailed answer.” This is useful for people who want a quick overview before they delve deeper into the matter.

Why are people so excited about ChatGPT?

Advanced language models that can communicate according to human standards are not new, but ChatGPT has a number of advantages over “competition”. For instance, it is very easy to use, where earlier models such as GPT-3 required some technical knowledge. There is a lot of human control behind the system, which means that certain things such as disinformation or unintentional racism have often been removed. Finally, many “human properties” can be recognized in the system. For example, if you ask the same question again, it responds with “like I said before”, giving the impression that we're communicating with a human, rather than a computer.

What competitors are there for ChatGPT?

In the world of advanced language models, OpenAI's ChatGPT has certainly gained a strong position, but it's not the only system of its kind. There are many other language models, such as LLama2, Falcon 180b, and Mistral 7b, which developers can work with to create specialized applications. These models are often used for specific tasks, such as research or business applications, and each has its own unique strengths and weaknesses.

However, when we look at systems that are specifically designed to be accessible to the end user - with an interface similar to ChatGPT - the list becomes considerably shorter. Many of today's language models require a certain amount of technical knowledge to be used effectively, making them less accessible to the average person.

A notable competitor to ChatGPT in this context is Anthropic's Claude2. Anthropic is a research company focused on creating safe and useful AI, and their Claude2 system is designed to provide many of the same features as ChatGPT. While both systems aim to deliver human-like textual interactions, they each have their own unique approaches and technologies.

In short, while there are plenty of advanced language models that developers can work with, there are fewer systems that are truly accessible and user-friendly for the end user. ChatGPT and Claude2 are two of the most prominent examples of the latter category, and both offer exciting opportunities for the future of human-machine interaction.

How will ChatGPT change my work?

Language models such as ChatGPT are about to radically change the way we work. Imagine each employee, from the director to the intern, being supported by a personal AI language assistant. These assistants will not just be general tools; they will be specifically trained in the future based on their user's knowledge, preferences, and writing style, creating a close symbiosis between man and machine.

With such a personal assistant, employees can quickly generate complex reports, automate communication, analyze data, and even formulate creative ideas. Additionally, the assistant will learn from each interaction, increasing the accuracy and relevance of the information provided over time.

This also means a shift in the skills that are valued in the workplace. As routine and repetitive tasks become increasingly automated, human skills such as empathy, decision making, creativity, and interpersonal communication will become even more crucial. Training and onboarding processes can also be streamlined, as new employees have access to an assistant that helps them adapt quickly and gain the necessary knowledge.

In addition, the use of these models can lower the barrier to global collaboration. Imagine a world where language and cultural differences are smoothly bridged by AI, allowing teams from around the world to work seamlessly together.

Essentially, we're at the start of a new era where AI language assistants are not just tools, but comprehensive partners that help us achieve our goals more effectively and efficiently.