Artificial Intelligence (AI) is a technological field that is rapidly advancing and finding applications in various aspects of human life. There are many types and focuses of artificial intelligence. Currently, one of the most talked-about is ChatGPT, which falls into the category of so-called generative AI (artificial intelligence). Below, you will find the definition of what it actually is and how it works. We will learn why ChatGPT sometimes lies and the specific technologies it is based on.

Generative AI (Generative Adversarial Networks – GAN)

Generative artificial intelligence, also known as GAN (Generative Adversarial Networks), is a technique that was first introduced by Ian Goodfellow and his colleagues in 2014. GANs are designed to generate new data that is similar to real data, based on training data. ¹

ChatGPT and other GANs work based on two main components:

1. Generator: This part of GAN tries to create new data that is as faithful as possible to the training data. The generator learns to create data based on the patterns it has learned from the training data.
2. Discriminator: The discriminator acts as a judge and tries to distinguish between real data and data generated by the generator. Its task is to determine whether the data is true or not.

Generative artificial intelligence has wide-ranging applications, including generating images, text, sound, and many others. It is a key element in the creation of deepfake videos, realistic visualizations, and many other applications.

Why Does ChatGPT Sometimes Lie (and Quite Convincingly)?

The generator generates data, and the discriminator evaluates data. ChatGPT works by having these two components compete with each other. The generator tries to create data based on the training model that convinces the discriminator that it is real, while the discriminator tries to detect fake data. This way, the generator keeps improving in creating realistic data.

Truth or lie? In the article ChatGPT (OpenAI) – Introduction, pricing, limitations, and use cases, I wrote that it is necessary to constantly monitor ChatGPT. This is because neither the Generator nor the Discriminator are perfect. If the generator generates a lie, the discriminator usually catches it, but not always. That’s when ChatGPT can convincingly lie.

The good news is that it is continuously evolving, and the frequency of “lying” is decreasing. I have seen discussions with a bot (somewhere on the internet) that could be convinced that 1+1 is not 2 if you argued with it. Now it seems that it won’t be convinced anymore

What Technologies is ChatGPT Based On?

ChatGPT is primarily based on the following technologies:

1. Machine Learning: Machine learning is a general term for methods that allow computers to “learn” from data and experience. This includes methods such as support vector machines, decision trees, and neural networks. ²
2. Deep Learning: Deep learning is a specific form of machine learning that focuses on using deep neural networks to solve complex problems, such as image recognition and natural language processing. ³
3. Recurrent Neural Network (RNN): ChatGPT uses a recurrent neural network (RNN) with Long Short-Term Memory (LSTM), which allows it to capture and retain context and dependencies between words in the text. ⁴
4. Attention Mechanism: ChatGPT uses attention mechanism, which allows the model to selectively focus on important parts of the text and give them more weight when generating responses. ⁵

These technologies together form the basis for ChatGPT and enable it to communicate effectively with people and generate text based on text inputs.

1. Wikipedia, Generative artificial intelligence [online]. [accessed 2024-01-28]. Available at: https://en.wikipedia.org/wiki/Generative_artificial_intelligence
2. MIT, management school, Machine learning explained [online]. [accessed 2024-01-28]. Available at: https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-explained
3. Geeks for geeks, Introduction to Deep Learning [online]. [accessed 2024-01-28]. Available at: https://www.geeksforgeeks.org/introduction-deep-learning/
4. Wikipedia, Recurrent neural network [online]. [accessed 2024-01-28]. Available at: https://en.wikipedia.org/wiki/Recurrent_neural_network
5. Wikipedia, Attention (machine learning) [online]. [accessed 2024-01-28]. Available at: https://en.wikipedia.org/wiki/Attention_(machine_learning)