Models such as ChatGPT are causing considerable discussion, both in science and among companies. The idea of having a machine employee who can formulate texts, answer e-mails and program seems tempting. And indeed, ChatGPT, as a large language model, is often a real help here. However, to understand how these technologies work and why they can’t replace traditional search engines, we first need to look at the basic technologies.

What are Transformers?

Transformers are a deep learning model that was unveiled by researchers at Google in 2017. Designed specifically for natural language processing (NLP) tasks, it has revolutionized the way models process human language. The Transformer model is based on self-attention mechanisms that allow it to consider different aspects of a text at the same time, rather than one after the other. This allows them to grasp the context of each word in a sentence much better. GPT builds on this. It stands for “Generative Pre-trained Transformer”. GPT was developed by OpenAI and is known for generating amazingly accurate text. The word “pre-trained” already signals that GPT has been trained on an enormous amount of text before it is further optimized for specific tasks. During this training process, the model captures both the nuances of our languages and global knowledge, although this is not explicitly intended.

What is ChatGPT?

ChatGPT is a specific application of GPT that has been optimized for conversations or chats. It is trained to give human-like responses in a chat context. The main difference with traditional chatbots is that ChatGPT is not based on fixed rules or scripts, but on the analysis of billions of words and phrases from the web. Although ChatGPT is trained on an extensive amount of data, it has some limitations:

1. Training Data Limit: ChatGPT is only as good as the data it was trained on. It can’t know anything about events or developments that happened after the last training date.
2. No own understanding: The model can generate text and respond to queries, but it doesn’t really understand the content. It emulates human communication based on statistical patterns rather than genuine understanding.

Plugins extend ChatGPT

ChatGPT plugins are extensions that allow the ChatGPT model to provide additional features and services beyond just text output. Plugins can be thought of as “eyes and ears” for language models. They allow the model to access current, personal, or specific information that is not included in the training dataset. For example, a special plugin allows ChatGPT to browse the web. This expands the model’s access to information, allowing it to go beyond its training corpus and retrieve up-to-date information.

Why isn’t ChatGPT replacing Google?

ChatGPT and Google Search are two impressive technologies, but they serve different purposes, so they can’t be easily compared or swapped out 1:1. Google is primarily a search engine that specializes in scouring the sprawling internet to provide users with relevant results based on their search queries. It presents links to real websites that were (still today) created by humans. This means that you can come across both top-notch sources of information and less reliable content.

In contrast, ChatGPT is a system that generates responses based on the data it was originally trained on. It doesn’t provide direct links to sources, which is why the answers run the risk of being inaccurate or misleading. Not to be neglected is the fact that false, misleading, questionable content was also learned during the training.

What does wonk.ai Write do?

Wonk.ai Write combines research based on generic knowledge (search engines) and specific company knowledge. By having access to company knowledge, the editorial team can create specific content that is relevant to your company. Write also supports the structuring of topics and the identification of topics. As soon as all relevant sources have been collected and reviewed by the editorial team and, if necessary, adapted, Write generates a first draft, which editors can then use as a starting point for individual work. More about Write.

Dr. Frederik Bäumer

Dr. Frederik Simon Bäumer ist ein Absolvent der Universität Paderborn, Fakultät für Wirtschaftswissenschaften. Er promovierte im Bereich Wirtschaftsinformatik mit "summa cum laude". Seine Dissertation konzentrierte sich auf die "Indikatorbasierte Erkennung und Kompensation von ungenauen und unvollständig beschriebenen Softwareanforderungen". Zuvor schloss er sein Masterstudium in Management Information Systems mit dem Schwerpunkt auf semantischer Informationsverarbeitung mit Auszeichnung ab. Dr. Bäumer lehrt heute an der Hochschule Bielefeld.

mehr erfahren