ChatGPT makes you more articulate, not smarter
By Daan Smits & Jules van den berg
ChatGPT. Who hasn't heard of it? You hear about it on the radio, in the corridors, from LinkedIn gurus, and even at family gatherings and birthday parties. But what is ChatGPT, and is it really a solution that you can use for anything and everything?
Those who blindly rely on it will be disappointed. This becomes clear when you delve into what ChatGPT actually is.
The underlying technology of ChatGPT in a nutshell:
ChatGPT is a Large Language Model (LLM), also known as a language model. It's a model that analyzes a vast amount of text (training data) and searches for connections within it. These connections are formed between words and concepts, also known as word groups. For example, the concept of 'glass ceiling' might evoke associations like 'drink' and 'ceiling', but you would connect the combination to 'career'. The algorithm then assigns weights to these connections using deep learning.
Based on a user's question (also referred to as a prompt), the model can predict the following words based on the weights and word associations.
The model's prediction will likely sound like a well-formed sentence. However, it does not take into account the meaning that we as humans assign to words. In simple terms, a language model only ensures that the sentence flows well textually, even if it may be nonsensical in terms of content.
Let's go back to ChatGPT. This model has been trained on billions of texts from all corners of the internet (up until 2021). This means that if there are enough connections about a specific topic in the training data, ChatGPT can produce well-formed text on that topic. Keep in mind that ChatGPT, as a language model, does not consider the meaning of the words.
If you ask ChatGPT a question about a domain you know little about, the response will likely sound intelligent due to the good sentence structure. However, once you focus on your own expertise, you might encounter shortcomings.
You might wonder, then, what can a language model like ChatGPT be used for? Language models can translate a piece of text for you, particularly because the model is familiar with the word associations in each language it's trained on. The final translation does not consider the content of the text, but rather how naturally the words fit together.
You can also have ChatGPT generate a draft for a text within a domain you are knowledgeable about. For example, the outline of a project proposal. You've probably written similar pieces several times before, so you know exactly how it should look. However, you might encounter a writer's block, struggle with clear formulations, or be short on time. In such cases, ChatGPT could provide some assistance.
Regardless of the application, it's essential that you possess knowledge and expertise in the subject for which you're generating text. This way, you can verify the generated text for accuracy yourself.
In short, ChatGPT enables you to be more efficient at what you already know how to do. As long as you want to ensure quality, ChatGPT does not enable you to do things that you couldn't do without help. Therefore, we never advise directly copying a text from an LLM; instead, have it checked by yourself or another expert in the relevant field. Responsible use of solutions like ChatGPT requires expertise.
Daan Smits and Jules van den Berg are consultants at VKA. Even during their studies, but especially now during the execution of assignments, they advise on the use and implementation of AI. If you'd like to discuss the use of AI with them, feel free to contact Daan Smits or Jules van den berg.