The limitations of ChatGPT OpenAI

We’ve all seen it on social media: the revolutionary model that can and will change everything. Will OpenAI’s ChatGPT Beat Google? Will OpenAI lose to the people who made it? In this article, we go beyond the hype and explore the limitations of ChatGPT and how it will change the world through predictions for the future.

To discover what the world will look like in a few years now that ChatGPT is here, let’s go back two years to OpenAI’s previous revolutionary model: GPT3. When that model came out, it became possible for the first time to generate coherent texts of more than 1 or 2 sentences. The model could also carry out assignments: write a short speech on topic X. The model was not limited to scientific articles. After a year, anyone could access OpenAI’s servers to use the model for a few cents at a time.

The news articles at the time also covered the party tricks: the recipe generator and creating an HTML page with a red button, but these options were not reliable or extensive enough to be useful in practice. In addition, GPT3 often “hallucinates” things that have nothing to do with the command. For example, GPT3 sometimes wants to mention in the scenario of a black dress that it is also on sale without any input.

The true revolution

GPT3’s true revolution lay elsewhere. The model could reliably perform simple tasks without the need for any training data: things like writing a paragraph on a specified topic, or answering questions such as: what color is the product in the product text: “Ford Fiesta 1.2D Bordeaux”? The options are: white, black, green, silver and red.

The next step was GPT3 fine-tuning, where the model can be trained with a very small number of samples (10-100), making many tasks more reliable and accurate. This approach made it possible to use a much smaller model. Over the years, new models were also added. Some models, such as GPT-J, were fully released and could run on an affordable computer at no extra cost.

limitations of chatgpt


This brings us to the current situation. Many people had predicted that GPT3 would make it possible to create intelligent chatbots that could replace helpdesk workers, but that didn’t happen. Yet the revolution in AI is already everywhere, but it is virtually invisible. For many people, Google Search is a tool that mechanically returns a list of pages based on a query, but in reality it is powered by the same kind of language models used for GPT3. You can also no longer assume that a text on a web page has been written by a human being. GPT3 and related models can do that too, as long as a human fixes the bugs.

Anyone who has used ChatGPT knows that it is more than just hype. It’s a huge step forward, perhaps more than the move from GPT2 to GPT3. Where GPT3 hallucinates, ChatGPT does exactly what is asked, sometimes to the point of being boring. ChatGPT knows the difference between reality and fiction and says when it doesn’t know instead of making something up. For me, the most impressive thing is that ChatGPT correctly answers questions where logical reasoning is required. This shows that ChatGPT actually has a very strong understanding of language and can derive the correct information. This makes it possible to develop an infinite number of new applications that have a much better understanding of texts and are much stronger in generating output. Just look at the following example.


“write a website text for a consultancy company that focuses on data-driven value creation”


“Welcome to our consultancy firm, specialized in data-driven value creation. We believe that data is the key to success in any business, which is why we help our clients improve their business processes through data analysis and interpretation.

Our team consists of experienced professionals with expertise in various industries and technologies. We partner with our clients to achieve their goals and create value through data analytics and use.

Whether you’re looking for ways to optimize your business processes, improve your decisions or increase your customer satisfaction, we’re here to help. Contact us today to see how we can help you achieve your goals through data-driven value creation.”

However, the power of the model also has its own limits: The answers are not always correct to use directly. In addition, the model has difficulty staying creative with longer texts: it often tells a story that stays too close to the assignment. The more precisely the model answers, the less creativity emerges.

The future 

Looking to the future, we see that the revolution will go much further. Here are some predictions:

The real power of these models lies in performing relatively simple – but increasingly complex – tasks normally performed by humans. For example, think of an automatic answering system on a website. Or a student who has his/her math homework done by ChatGPT. We will be seeing a lot of these types of applications in practice in the near future.

Help desk agents won’t disappear, but they will get an AI-based autocomplete feature that can answer virtually any query. The employees are then still responsible for checking the correctness.

The personal assistants Siri and Google Assistant will get a substantial update where they will become much more intelligent, they will also understand you if your voice stutters and the answers will sound much more human.

The AI will become part of existing applications. It’s not just autocomplete in Word. You can also think of smart summaries in a browser, or automatically taking minutes and summarizing a Teams meeting.

An attempt has been made to apply GPT3 in games so that the adventure takes place in an infinite world. In the next five years, the technology could be advanced enough to be applied in mainstream games.

In the coming years it will be possible to achieve results comparable to ChatGPT with an affordable computer. As a result, the price of generating high-quality texts is falling to such an extent that copywriters are going to have a hard time.

The true revolution, of course, will be in things that only one person has been able to predict. People like that often had names like Elon Musk and Steve Jobs in the past. We don’t know where he or she is or what the idea is, but it’s likely that the next self-made billionaire is already working on AI, and has an idea that no one thinks will work. However, the technology behind ChatGPT is going to make this possible… It’s going to be interesting times!

Be first to comment