Elon Musk wants a pause in development, Bill Gates doesn’t think it’s necessary. Never before has a technology attracted so much attention in such a short time as the Chat-GPT language model. In Italy, access to the tool was cut off completely and Canada is investigating OpenAI.
The world is upside down: everyone is talking about and with the new chatbot.
There is growing concern about a power shift. As observed with the rise of Facebook and Google, dominance in the technology sector can quickly lead to astonishing power over society.
The impact of Chat GPT is set to be even more transformative than social media has been.
OpenAI explicitly prohibits the misuse of Chat GPT for essential services such as lending, employment and housing, as well as for spam, gambling, fake medicines and political influence. These prohibitions demonstrate the enormous potential of Chat GPT for both positive and negative applications. This groundbreaking language model has the potential to revolutionise the world of artificial intelligence. Everyone understands that.
What makes GPT-3 so unique and powerful compared to other language models?
As part of the Natural Language Processing (NLP) category of Large Language Models of Deep Learning, it understands and generates language in a way that was not possible before.
The developers of Open.Ai have trained the system by feeding it millions of texts, the training data.
The training data, called the corpus, included a wide variety of texts, including books, letters, articles, Wikipedia entries, as well as literary works such as the complete collection of the Gutenberg Project, which includes authors such as Heinrich Heine. In addition, the Oxford Poetry Corpus, which contains a collection of English-language poetry from the Middle Ages to the present day, was also used. The AI learned that poems often consist of short lines and technical texts often contain rare terms.
The settings, called parameters, are key to the accuracy and adjustability of the system. GPT-3 had 175 billion parameters and GPT-4 will probably have even more. The neurons learn on their own and adjust their settings automatically.
However, human assistance was needed in the beginning to correct wrong answers and test new parameters. So-called “supervised learning” leads to more precise and better quality texts. A corpus is a machine-readable collection of training texts. In order to recognise and distinguish the linguistic patterns of the texts, the system needs such a large amount of data. GPT systems constantly adapt their computations to new circumstances and redefine their connections to other neurons in order to increase their performance.
Chat GPT has mastered our language, can we therefore trust it?
A puzzling and potentially worrying peculiarity of Deep Learning systems is their opaque mode of operation. The term “black box” aptly illustrates this phenomenon.
Although ChatGPT can surprise with original prose, distrust is the order of the day.
This is what a teacher at George Washington University recently had to experience. The renowned law professor is facing serious accusations. ChatGPT, commissioned to conduct research, made allegations against the professor and provided false sources purporting to prove that he had harassed students during an Alaskan trip.
But the accusation lacks any basis. In addition, the professor has never made a trip to Alaska. The reputation of the professor and also of the university have been tarnished. This incident again shows how important it is to carefully check the sources that Chat GPT puts out. It remains to be seen whether the professor will take legal action. And above all: against whom? Against Open.AI or against the Chat GPT user who did not check the information thoroughly enough?
This example clearly shows the challenges that jurisdiction will face in the future.
It remains to be seen how international legislation will react to these challenges.
The professor’s example shows that the potential of this language model is enormous – for better or for worse.
Although Chat GPT impresses because of its convincing results, it is important not to switch off the mind when using it and to control everything carefully.
The system should be understood solely as an assistance and not as the sole source of information or decisions.
Not only society is facing new challenges, but also entire professions fear for their jobs. Chat GPT has the potential to revolutionise numerous industries and increase efficiency in various areas – including customer service, data analysis and automated text generation.
The technology can help relieve human workers by taking care of repetitive tasks.
If Chat GPT were to lead to increased unemployment, the question of a basic living income would become even more pressing than it is now.
The big language models make clear what should have been clear long ago: We need to find a consensus on how our democratic principles will reconcile and shape with this technology.