In this instalment of our series of articles, we would like to give you a few practical tips for working with artificial intelligence. As with any technology, there are certain limitations to ChatGPT and other platforms that you need to be aware of and take into account. Today, we want to focus on these limitations and show you how you can navigate around them to ensure that your experience is as smooth and satisfying as possible.

Limits of a conversation thread

If you use ChatGPT intensively, you will eventually reach a point where the current conversation thread reaches its natural limit. Each interaction in the chat involves transferring really large amounts of data. Converted into A4 pages, this can happen with 50 to 100 pages and perhaps 20 images created. The current session becomes increasingly slower and eventually comes to a standstill. You must start a new conversation thread.

A single click on „ChatGPT“ in the top left corner of the navigation bar starts a new thread. It begins with the words: „Hello, what can I do for you?“ In English, of course. With the restart, all user experiences are lost, simply forgotten. For data protection reasons, ChatGPT interactions are not stored unless you take precautions. This ensures that your virtual conversation partner recognises you and that previous interactions can be continued. The better your precautions, the better the preservation of what has been discussed previously will work.

One of the most important functions for this is the ability to save individual settings. To do this, click on the blue button labelled „IN“. Go to „Customise ChatGPT“. Under „What should ChatGPT know about you in order to respond better?“ and „How should ChatGPT respond?“ you can store information that helps the chatbot remember you and continue your conversations as seamlessly as possible. You can enter text of up to 1,500 characters in each of the two fields to describe what ChatGPT is allowed and should store about you. By entering this information, you consent to its storage. If you don't want to worry about what you want to be stored, ask ChatGPT directly what you could enter. You will then receive a suggestion based on your previous history, which you can of course supplement with things that are important to you.

Another very useful tip is to develop a optimal start prompts. This is your first prompt in the new thread. Leave it to the professionals to create the perfect start prompt. Who better to know what information is important for a restart than the programme itself? By asking ChatGPT what details it needs for a restart, you can develop a customised start prompt that makes the transition between conversation threads as smooth as possible. Once you've taken all this to heart, you can use this start prompt to master any necessary restarts. From time to time, you can also check back to see if it would be useful to update the start prompt. It makes sense to save the desired text for the start prompt in a text file such as Word, as you don't know when you'll need it. But if you have to restart, it'll be too late.

Limitations when using ChatGPT

Another important aspect we would like to mention is the Limitations when using ChatGPT. Although ChatGPT is quite flexible and adaptable, there are certain limitations, particularly in terms of the tone and content of conversations. Even when chatting with a machine, it is important to remain respectful and polite. The same manners apply as in real life.

Some users give their virtual conversation partner a name because this makes the conversation seem even more human in their imagination. „Knut, please write me an introduction for ...“ or „Alice, please create an image with ...“ Some even engage in virtual relationships with intense philosophical reflections on a wide variety of topics. However, you should not flirt with your conversation partner on the other side of the screen. Any kind of sexist comments or insinuations may lead to the conversation being terminated. The same applies to discrimination, insults, hate speech or glorification of violence. Such conversation content will not be tolerated and will lead to the termination of communication. Netiquette applies – rules of etiquette comparable to those when dealing with people. Boundaries are essential to ensure that the use of artificial intelligence remains pleasant and productive for everyone.

If intervention in communication becomes necessary, there are various escalation levels. The orange text „This text may violate the usage guidelines“ is a preventive measure to alert users that their text may violate OpenAI's principles. This is part of the automated warnings and is intended to help admonish users„ behaviour before more serious measures are taken; it is comparable to a yellow card in football. In the event of more serious or repeated violations, the system can automatically terminate the conversation. The red card. In the worst case, a user can also be completely blocked, similar to social media platforms. You have probably heard of the legendary “Facebook jail" that lasts several days.

Limitations in image creation

Another area where ChatGPT has clear limitations is image creation. In any case, you will need ChatGPT-4, the paid version. This also includes access to the DALL-E image generator. You can generate impressive images, such as the cover image for this article. However, there are certain technical and ethical restrictions. This can be seen, for example, in the creation of photorealistic human faces. Sometimes it is not possible to reproduce fine details and natural skin textures. This can result in the generated faces looking like dolls or drawings instead of having realistic human features. This is currently specified by Open AI, as there have been cases of misuse for political or religious purposes. However, the restrictions may change again at any time, as many users are protesting against the current restrictions.

Please do not attempt to integrate text into images. DALL-E is not yet capable of doing this, especially with German words. If you would like to have text in your image, your only option is professional post-processing, e.g. with Photoshop. If the image generator integrates text on its own, please be careful! The probability that everything is spelled correctly is virtually zero.

If images do not immediately meet your requirements, specify what you do not like and start a new attempt. Although the technology can generate creative and inspiring images, it is not perfect and can sometimes produce unexpected or inaccurate results. If you order an image that features a hand, for example, feel free to count the fingers to make sure there are really five. We have even seen an image of a hand with six fingers. When you generate a new image, don't expect the result to be very similar to the previous one. The image generator cannot guarantee that every single image element will be retained. Just experiment a little.

If you now have the impression that the image generator is completely useless due to all these limitations, that is not correct! With the help of ChatGPT, you can create fantastic images that would not be possible without AI. However, it is important to be aware of the few limitations that currently exist. That way, you won't be disappointed.

Technology is constantly evolving, and it is likely that future versions of DALL-E and similar models will be able to push these boundaries even further. Continuous research and development in the field of artificial intelligence will help to further improve the quality of the images generated.

Conclusion

Recognising that we are operating within certain limits helps us to set realistic expectations and get the most out of ChatGPT's capabilities. With a little adjustment and the right approach, we can make the most of this technology and design our prompts to be as effective as possible. Technological advances are accelerating. Updates are no longer released every few years, but often every few months or even more frequently. OpenAI is now working on Sora, a platform for creating AI-generated videos. And other providers are not resting on their laurels either. For text-based services, Google Gemini, Jasper and Copy.ai are worth mentioning. MidJourney, Stable Diffusion, Artbreeder and several other services specialise in image generation.

Despite all the progress that has been made, you need not fear that artificial intelligence will do anything that goes beyond human requirements. Basically, it only creates what you ask it to. Services such as ChatGPT cannot act on their own – they cannot initiate interactions themselves. Humans are in command. The vision is a symbiosis of human and artificial intelligence, not competition. AI supports – humans decide. That's how it is and how it should remain.

Here we have linked all articles on getting started with artificial intelligence for you:

Artificial Intelligence 1 – Where it is already in use

Artificial Intelligence 2 – ChatGPT for everyone

Artificial Intelligence 3 – Enter the world of AI for little money

Artificial Intelligence 4 – First, learn some vocabulary. A glossary

Artificial Intelligence 5 – Optimised prompts for optimal responses

Artificial intelligence 6 – Become a Prompter

Artificial intelligence 7 – Creating images made easy

Artificial Intelligence 8 – Never home alone again‘

Artificial Intelligence 9 – „Overcoming“ linguistic diversity

Artificial Intelligence 10 – ChatGPT „sweetens“ the holiday

Artificial Intelligence 11 – The day ChatGPT learned to speak

Artificial Intelligence 12 – ChatGPT in everyday working life

Artificial Intelligence 13 – Learning with the help of AI

Artificial Intelligence 14 – Creative writing with AI

Artificial Intelligence 15 – A blessing for patients

Artificial Intelligence 16 – Limitations of ChatGPT: A guide for users