“In a world where machines are developing more and more human-like capabilities, ChatGPT is a brilliant example of how artificial intelligence challenges our imagination and redefines the way we communicate. With a touch of science fiction and a pinch of vision for the future, ChatGPT has the ability to chat in a human way while showing amazing versatility. But what is really behind this fascinating AI model? Let's dive in and uncover the secrets of ChatGPT, its limits, and the exciting opportunities it offers in our digital age.”
This is the answer to ChatGPT's request to write an introduction to a contribution on the same topic and to formulate it in a friendly and funny way.
Since ChatGPT already provides a common thread for this article, it is a good idea to ask the topics that ChatGPT raises in an interview:
1. How does ChatGPT work?
“ChatGPT works on the basis of machine learning and neural networks. This means that, thanks to its extensive training data, it is able to answer questions and communicate in a human-like way.
In detail, ChatGPT works as follows: First, questions or instructions are sent to the model, which it processes and analyses. It uses its training data and neural network architecture to extract the relevant information and generate an answer. The response is then sent back to the user, creating a dialogue.
Thanks to its machine learning and neural networks, ChatGPT is able to answer a wide range of topics and questions and communicate naturally. However, it may contain errors as it is only a model and not a human brain.”
It is precisely these errors that are always pointed out at the beginning of the chat.
2. What are ChatGPT's strengths?
“ChatGPT is extremely versatile and useful in many areas of application. It not only masters natural conversations, but also offers flexibility in text generation. From creative writing projects to answering questions, it's a powerful tool. Accessibility via text interfaces and quick response time make it convenient and efficient. In addition, it can be improved through instructions and feedback and is well suited for customer service by allowing automated answers to frequently asked questions.”
3. What are ChatGPT's weaknesses?
“Although ChatGPT has impressive capabilities, there are a few weaknesses to be aware of. It can occasionally provide inaccurate or nonsensitive answers as it relies on textual patterns. The wording of the input is critical, as small changes can result in different answers. The model may also be too detailed and wordy when shorter answers would be appropriate. Despite efforts to filter inappropriate content, it may still generate inappropriate or offensive responses on occasion. ChatGPT also has difficulty dealing with ambiguous questions and doesn't always ask for clarifications. It has a limited knowledge base and only knows information up to its last training cut in October 2023. In addition, it can reflect bias and stereotypes in its answers based on the data on which it was trained. Despite these weaknesses, ChatGPT has the potential to improve many applications and enable human-like interactions in various areas.”
While ChatGPT admits these points quite openly and honestly, there is one significant weakness of the chatbot that it does not freely admit to itself and others. To do this, we temporarily interrupt our interview with ChatGPT and do some research on our own.
ChatGPT and hallucinations
The inaccurate or meaningless answers mentioned briefly at the beginning are primarily based on the fact that ChatGPT tends to so-called “hallucinations.” This does not mean — like hallucination in humans — a false perception or that the AI comes up with exciting, creative mythical creatures, fairy tales or fantasies.
Rather, it is about false information: ChatGPT is able to invent things without specifying that the information is incorrect, false, or incomplete.
An example of such a hallucination is a case from the USA in which a man sued an airline after he was hit by a serving cart on his flight. The man's lawyer, in contact with the airline's legal department, had presented some precedents, which — as it became known later — ChatGPT had provided him with. After a detailed examination of the files, it turned out that the cases were fictitious by artificial intelligence and could therefore not be used as a supporting tool in the process.
The case shows that ChatGPT's answers should be treated with caution and should always be checked for accuracy. You can find more chatbot hallucinations of this kind in our article”The 6 biggest chatbot fails and tips on how to avoid them”.
With a new test called SimpleQA But OpenAI now wants to fight hallucinations and false spending. The test was developed to verify the accuracy and reliability of AI models. The tool can ask questions to various AI models and then assesses the following two things:
- First, the accuracy of their answers.
- On the other hand, how certain the models themselves are as to whether the answer is correct.
What is particularly noticeable: Even the largest language models, such as GPT-4o or Claude-3.5-Sonnet, are far from the 50% mark. Through this test, which OpenAI has made available to the general public free of charge, language models can therefore become more reliable and trustworthy. Of course, there are also limits to SimpleQA, for example the fact that an AI model (such as GPT) should evaluate others, which calls into question the neutrality of the evaluation. Still, this test offers great benefits in AI development, as researchers and developers around the world can test and improve their models.
ChatGPT is quite vague and general
In response to the claim that ChatGPT is rather vague and general, the AI responded as follows:
“Yes, that's true. As a large language model of OpenAI, I am trained to answer general questions and inquiries and to speak on a variety of topics. This means that my answers are often general and vague as I don't have any specific information or details about a specific topic.
It is important to note that as a language model, I have no knowledge of current events or specific details and my answers are therefore always general. If you need more specific or detailed information, I recommend reaching out to a trusted source.”
This also shows that ChatGPT is not suitable for every application and should generally not be used as the only source for research.
Interim conclusion: Impressively talkative, but should be treated with caution
Even though it can happen that the chatbot gives false information and hallucinates, this happens relatively rarely. In most cases, ChatGPT provides correctly formulated answers. If you are aware of the fact that the AI information is initially used for an initial overview and that not everything should be taken at face value, the obvious advantages, such as 24/7 availability and the ability to automatically conduct a complex conversation thanks to neural networks, cannot be denied.
This is how you find yourself positively surprised and amazed at the answers that ChatGPT plays out within a few seconds. The fact that the sentences are quite vague and general is in the nature of things — ChatGPT is not intended to provide users with specific information or expertise.
At this point, we'll finish the question-answer game with ChatGPT and turn to other important questions, but we've formulated and researched the answers to them ourselves:
4. Does ChatGPT replace all other chatbots?
In view of the fact that ChatGPT cannot provide users with specific information and that the knowledge base is currently ending in April 2023, the chatbot is not suitable for specific use cases such as customer communication by companies. In particular, companies that primarily use chatbots in customer service or marketing & sales depend on providing their (potential) customers with specific and tailored answers that are tailored both to the company and to the needs of customers. This is not possible with ChatGPT, so that in the worst case scenario, customers receive incorrect, outdated information or get products suggested by competitors when asked which products are recommended.
ChatGPT can answer many questions, even humorously and eloquently — but the innovative chatbot does not find an answer to the question of what the current delivery status of a package is.
It is therefore highly questionable whether ChatGPT is a substitute for chatbot providers or even consultants, copywriters, developers — as is currently speculated. However, it is a fact that all of these areas can benefit from the tool, whether for inspiration or assistance. Also use 92% of Fortune 500 companies OpenAI, which shows the growth and acceptance of the tool in various industries.
5. Is ChatGPT free?
The use of ChatGPT is available in both a free and a paid version.
Comparing the two ChatGPT plans
In Germany, the free version currently requires a one-time registration with an e-mail address. After successful registration or login, the chat can start directly in the input window. Whether in English, German or another language: Simply start with any question and get into conversation with the innovative chatbot.
OpenAI also introduced the option to use ChatGPT without an account and login at the beginning of April 2023. However, free access is not yet available for additional products such as the DALL-E 3 image generator and is only gradually being rolled out to all users. In Germany, too, it is now possible to use it without logging in.
The paid version of ChatGPT (Plus, Team and Enterprise Plan) has been available in the USA since February 02, 2023 and in Germany also since March 2023. The decisive difference to the toll-free version lies in the additional features that the paid plans offer.
In the section “12. How can I use ChatGPT?” This article explains all the details about the differences between the two tariffs.
6. Is everything ChatGPT says true?
A conversation with ChatGPT is sometimes amusing, interesting or even insightful. Nevertheless, AI's answers should be treated with caution and should always be checked — especially when it comes to information and facts. This is shown not least by the hallucinations already explained at the beginning, i.e. fictitious or false information that can occur when using ChatGPT.
ChatGPT itself makes it no secret that errors and incorrect content can occur. This is confirmed both as a warning on the homepage and in conversation with the chatbot:
“I am a computer program and base my answers on the information available to me. I do my best to provide accurate and accurate information, but it is possible that there may be errors or incompleteness. Please always verify that the information I provide is correct before using it.”
7. Does ChatGPT speak German?
As mentioned above, ChatGPT can communicate in German. Simply type in your question in German, ChatGPT will answer in the same language. An alternative is to explicitly ask the chatbot to only answer in German. But be careful: While ChatGPT plays English sentences almost always without errors, grammatical errors occasionally creep into the German formulations (see paragraph about this). However, since the recommendation is to proofread and check all ChatGPT content before using it anyway, the minor errors should be corrected quickly.
8. Is a ChatGPT ban to be expected?
After the initial euphoria about ChatGPT, a number of voices gradually came forward, especially in legal terms, who critically questioned the use of the tool. For example, in spring 2023, Italy's supervisory authorities published a press release that pointed out ChatGPT's data protection violations. They blocked the chatbot based in particular on the following three allegations:
- OpenAI has no legal basis for collecting personal data. This also applies to the processing of data to train the AI.
- OpenAI does not provide full information about the handling of personal data and what it is used for.
- The age verification mechanism is deficient, which is why children can easily access ChatGPT and may therefore receive content that is inappropriate for their age.
OpenAI responded to the demands of Italian regulators with appropriate adjustments, which allowed the tool to be reintroduced in Italy. For example, there is now an age limit for using AI and a form that allows users to object to the processing of personal data.
How golem.de Already reported at the beginning of the year, but there have already been bans or planning of bans in some institutions in other countries. For example, the International Conference on Machine Learning (ICML) has decided that authors may not use AI tools such as ChatGPT when publishing scientific papers. Although the IMCL is open-minded about ChatGPT and describes it as an “exciting” development, it is also pointed out that issues such as copyright or level of creation have not been clarified. However, the ban only applies to fully produced texts. According to ICML, texts that are edited or refined by ChatGPT are allowed. In 2024, the ICML wants to re-evaluate the ban on AI-based texts.
The New York Department of Education has also blocked access to ChatGPT on its networks to prevent students from using the tool. In the USA, there was also a joint lawsuit by several people against OpenAI, Microsoft, investors and cooperation partners. The allegation is that personal data, including from adolescents and children, was used in the development of GPT-3.5 into GPT-4 without their permission.
In Germany, too, there are already discussions about ChatGPT. For example, the commissioners for data protection and freedom of information in Hesse, Thuringia, Baden-Württemberg, Rhineland-Palatinate and Schleswig-Holstein have already initiated an investigation procedure. In addition, the data protection authorities of all federal states prepared a list of over 40 questions in April 2023 and sent them to OpenAI to answer. So far, there has been no response from the company in this regard. There is currently no sign of ChatGPT being blocked or banned in Germany. However, it is likely that adjustments and further protective measures, such as those already implemented in Italy, will need to be made.
A step in this direction is AI Act of the European Union which came into force from August 2024. To ensure that AI systems comply with European values and remain safe for society, the AI Act sets criteria for AI applications and clear transparency requirements and strengthens supervision of AI systems, such as ChatGTP. These rules are to be implemented within the next two years. It should therefore be noted that ChatGPT does not yet fully respect many of these requirements and adjustments are therefore necessary, particularly in the area of data protection. It remains to be seen how ChatGPT will evolve in relation to these new requirements.
9. Are there alternatives to ChatGPT?
Everyone is talking about ChatGPT — and not only that, but the number of users and requests to ChatGPT continues to skyrocket. This ensures that the servers are currently often overloaded and that not every user receives an answer. So if you want to talk or have planned ChatGPT for a work step or a specific task, you can also use alternatives.
Because how do you say so beautifully? “The competition never sleeps.” For example, Google Gemini with 6.9 million additional visits in September 2024 felt a significant increase in usage. But it is not just the big players that should be kept in mind: even smaller providers, such as Mistral Le Chat, a French AI start-up, are also gaining in importance. With new AI models such as Ministral 3B and 8B, which are designed for data protection, efficiency and use on end devices, the start-up is showing that innovations can also take place away from the big names.
ChatGPT is therefore by no means the only provider on the market that provides a well-functioning, free chatbot. There are several providers that are based on the same language model as ChatGPT (GPT-3.5 or GPT-4), such as:
- YouChat
- ChatFlash by Neuroflash
- Perplexity AI
- Jasper Chat
- Microsoft Copilot (formerly Bing Chat)
- ChatSonic or
- GitHub co-pilot.
However, there are also chatbot providers such as Claude, DeepMind Sparrow or Google Gemini (formerly Google Bard), which are based on other language models. The various providers differ not only in terms of language models, but also in terms of prices, information retrieval (with or without Internet sources), the languages that can be used and the ability to code. Our article explains the detailed differences between chatbot operators”20 alternatives to ChatGPT“On. If you are particularly interested in the chatbot Google Bard (now Google Gemini), you will also find interesting information in our article”What is Google Gemini (formerly Bard)? Everything you need to know”.
10. What ChatGPT applications are there?
ChatGPT's capabilities are impressive and more diverse than often thought. While it can be observed on LinkedIn and Twitter that many users have e-mail, posting or blog texts generated by OpenAI AI, there are numerous other possible applications:
- Use for research purposes, e.g. to identify approaches to specific problems.
- Preparation of analyses, summaries and protocols
- Create tables based on unstructured data sets
- Database queries: ChatGPT can help you to formulate the right database queries in everyday life to evaluate data.
- Use for brainstorming
- Preparation of invoices and calculations
- Create, explain, correct, and optimize programming codes
- Generate images, videos, and audio tracks
- Collection of pros and cons on a specific topic
- Filter keywords from a text
- Creating websites and apps
- Spelling and grammar check
- Obtaining advice, tips and advice
- Preparation of translations
- rewording texts
With new updates in November 2024, users can now also communicate with ChatGPT via voice chat. Instead of typing text, they can simply use their voice. This feature makes productive conversations easier during activities such as shopping, planning, writing, or brainstorming, as you can continue working on other tasks in parallel. For now, however, this feature will only be available to paid tariffs.
ChatGPT has also introduced real-time video, screen sharing, and image uploads in the mobile app — ideal for visual communication, presentations, or sharing complex content. The features are available for Team, Plus, and Pro users (excluding the EU, Switzerland, Iceland, Norway, and Liechtenstein) and are subject to daily usage limits.
Subject-specific applications, e.g. in online marketing:
- Market research, e.g. finding the 10 largest e-commerce companies including links and social media accounts
- Sorting and ranking keywords by search intent
- Generating typical questions about a topic
ChatGPT therefore has a wide range of applications not only in everyday private life, but also in a professional context. However, as already emphasized, it is important to always be aware of the limits of the tool. The chatbot can also make mistakes and sometimes reveal incorrect information.
Interesting fact: OpenAI is currently working on new advanced AI reasoning technologies such as o-1, nicknamed “Strawberry” or Swarm. Strawberry should make it possible to handle complex tasks such as evaluating market opportunities and strategic decisions. The goal is to move AI towards artificial general intelligence (AGI) develop further so that systems are able to think critically and teach themselves how to solve problems on their own. Swarm, on the other hand, is a platform that was developed to simplify complex computing tasks. The new developments could make ChatGPT an even more valuable tool for companies and individuals.
11. Where does ChatGPT get its knowledge from?
The chatbot develops the answers that ChatGPT users have to their questions using a neural network that works in a similar way to the neural system of the human brain. Within this artificial neural network, there are interconnected nodes called neurons. These take in information from other neurons or from outside, process it and pass it on to the next neuron, creating numerous chains of information.
The data used to train Chat-GPT comes largely from various Internet sources, such as
- books: novels, non-fiction and textbooks
- web pages: online forums, blogs, news sites
- Scientific work: Technical and research articles, scientific publications
- social media
This data serves as the basis for AI's learning process and ensures that it is increasingly able to conduct human-like conversations.
12. How can ChatGPT be used?
You can get free access to ChatGPT easily and quickly in this way:
- Create a free account with OpenAI: Just do that this link Follow and enter your email address and telephone number.
- After the registration has been confirmed — confirm the email in your inbox — chatting can begin. To do this, simply enter the question or concern in the chat field at ChatGPT.
OpenAI announced in April 2024 that registration for the free version of ChatGPT was gradually being abolished. So far, it has only been possible to use the core product without logging in.
Whether with or without registration: in both cases, since May 13, 2024, the free version of ChatGPT has been based on OpenAI's latest and most powerful AI model to date: GPT-4o. Here, the “o” stands for omni, which means something like “everything.” The model should be significantly more powerful and work faster than its predecessors. It is not based on several separate neural networks, but on a single neural network and is able to process content such as text, code, images (PNG/JGP...), audio (MP3) and also videos (MP4). All details about ChatGPT4O and the differences with GPT-4 can be found in our article GPT-4 is here — everything you need to know about it.
💡 Tip: Users with a free account can also use the latest version, GPT-4o - but access is limited and after reaching the limit, they automatically switch to the smaller standard model, GTP-4o-mini. So if you want to use the latest version, GPT-4o, even more intensively and process more complex issues, a paid ChatGPT account (Plus Plan, Team Plan or Enterprise Plan) must be set up. The prices for these plans are $20 per month upwards and include beneficial features and improvements:
- Priority access to new features and enhancements
- Higher usage limits for GPT-4o
- Access to GPT-4, GPT-4o and GPT-4o-mini
- Individual instructions (“custom instructions”) and creation of your own GPTs: Presetting certain instructions and preferences, which the chatbot automatically implements in conversations without being asked to do so.
- Using ChatGPT Plug-ins
- Using current data that goes beyond the October 2023 knowledge base
- Possible deactivation of the function that the data is used for training purposes
- AI-based image generation with the Dall-E 3 Image Generator
- Advanced data analysis (mathematical calculations and code generation)
- file uploads
13. ChatGPT and copyright — how do they go together?
The question of copyright in relation to artificial intelligence is not new, but is being discussed more sharply due to the success of ChatGPT. In the case of ChatGPT, it must be set directly twice:
- On the one hand, it is up for discussion who owns the content that ChatGPT generates. The AI does not belong to itself, but to OpenAI. Whether this in turn means that the generated content belongs to OpenAI is frequently discussed.
- On the other hand, ChatGPT generates its content from a variety of sources, which in turn have been created by other people. It cannot therefore be ruled out that the AI reproduces content that is actually protected by copyright.
There are many problems and uncertainties that arise as a result. What is certain, however, is that ChatGPT's content is not protected by copyright, as it is not the creation of a human being and the process is not considered creative. Nevertheless, the content published by ChatGPT should not be used for a scientific paper, such as a dissertation, for example. The reason for this is that ChatGPT can generate its content based on other creations, so this content could be considered plagiarism in a dissertation.
Due to the challenges posed by the use of artificial intelligence such as ChatGPT in terms of copyright, the topic is currently being discussed a lot at European level. The EU is currently planning to legally oblige AI providers to provide information on whether the data used to train AI is protected by copyright or not. How exactly these EU plans will be implemented and what they mean for the use of ChatGPT remains to be seen.
14. Why isn't ChatGPT working?
The ChatGPT does not work is usually due to the fact that the servers are overloaded because there are too many accesses. Especially when the day starts in the USA and people there become active on ChatGPT, congestion can quickly occur. Therefore, a tip to avoid these peaks is to use ChatGPT primarily in the morning and at noon (German time).
More tips if ChatGPT doesn't work:
- Wait and see
- Refresh the page
- Logging out and logging back in
If this is too lengthy and annoying for you, you can create a paid ChatGPT account. As already mentioned above, this account costs money, but ensures that you usually have access even during peak phases.
ChatGPT: Application examples and discussions on the web
The chatbot developed by OpenAI is currently a hotly debated topic in the media landscape and on various social media platforms. The posts show how ChatGPT is used in a variety of ways.
Stefan von Gagern gives his community in his LinkedIn post For example, specific tips on how to get maximum output from ChatGPT:
ChatGPT can also be useful in everyday life: Whether it's finding ideas for new recipes, planning a vacation or writing a poem - ChatGPT usually has a helpful answer ready. We also let ChatGPT be a bit creative and asked the AI for a short vegetarian recipe suggestion:
The user Paul Meyrat poststhat he has given AI a somewhat specific assignment, among other things: “Write a rap in the style of 50 Cent on the subject of emancipation.” In his opinion, the result: “amazing.”
We tested it ourselves and this is what came out:
Whether this fits the 50 cent style is negotiable. And whether it does justice to the issue of emancipation is also in the eye of the observer. However, it is still impressive how quickly and easily texts of this type can be generated.
Johan von Hülsen (managing partner at Wingmen) gives another interesting perspective in a newsletter issue in December:
Google has nothing against AI texts. From a Google perspective, it also makes no sense to penalize generated content. But there is a lot to be said for rewarding content that contains things that others don't have. And that is exactly what an AI cannot do, which can only reproduce and remix what others have done. Honestly, no copywriter can do that either, to whom you give a quick but ill-founded briefing and a few cents per word.
If you do this, then your texts lack the differentiating factor that sets you apart from your competition in the long term and gives you permission to rank because you help users with their questions and problems better than your competition.
This statement sums it up very well — even outside of an SEO perspective. In general, there is nothing wrong with AI-generated texts, whether because you want to use them for a speech, a blog post, a social post, etc. However, nuances, formulations or special insights are usually missing, which ultimately make the content unique and worth listening to or reading.
As mentioned above, it is therefore highly questionable whether ChatGPT is a replacement for chatbot providers at the current stage of development. For companies in particular, it is important to attach great importance to data protection and, above all, to never release unwanted misinformation to customers.
FAZ also uses an AI chatbot in both customer service and marketing, successfully answering a variety of recurring service requests automatically and correctly. Watch the free on-demand webinar now and learn how a company uses an AI chatbot to automate customer communication.
conclusion
Even when it comes to our own AI chatbot solution moinAI, we repeatedly emphasize that it is not a substitute for human employees, but ideally complements them, for example by automating repetitive and idle tasks.
ChatGPT is similar: A wonderful tool for getting inspiration, getting simple tasks done and really saving time. Or just for fun things, such as lyrics (more or less in the style of 50 cent) or to learn how to get a heart in the milk froth of a cappuccino — just as moinAI CEO Patrick Zimmermann recently had ChatGPT explain it to him.
Speaking of fun things: A popular topic should not be missing appropriate memes. With that in mind, we'll end the post with this image and the notice it contains.
.svg)
