ChatGPT – konkrete Anwendungen für Unternehmen

Chat GPT – concrete applications for companies

In the first part of this series, we looked at whether ChatGPT could be used for industrial applications. And if so, what needs to be considered. Now it’s time for the application. We want to know how well the tool works in practice. We looked at three specific application areas.

Text creation – amazing results, but with pitfalls

We all know how extraordinary ChatGPT’s text generation capabilities are. After all, this was the original purpose of the AI tool. With a short briefing, it produces solid texts with enviable speed. In this way, ChatGPT can save users valuable time and help them overcome writer’s block. In particular, people who are unsure of their wording or find it difficult to start texts will benefit from the AI’s services. Those who usually procrastinate and put off writing texts can use Chat GPT to make a playful start, relieve pressure and outsmart themselves.

On the other hand, the tool should not be overused, and in some cases, extreme caution is required. The first problem with widespread use is uniformity. The standard text generated by ChatGPT has a matter-of-fact but very impersonal tone. This is now instantly recognisable to experienced readers and can bore the audience in the long run. It also runs the risk of getting lost in a sea of generic content and failing to stand out from the competition.

Of course, not every text produced in a company needs to be of the highest literary quality. It is therefore important to consider carefully which texts shape the perception of a company and deserve to be written by a human hand.

Even product copy, which ChatGPT can produce very well, needs a human reviewer to check the accuracy of the statements made. It is easy for ChatGPT to make inaccurate, untrue or fictitious statements – which sound deceptively true due to the sophisticated wording. It is therefore necessary to double check the facts.

Another prerequisite for good text quality is a very good and detailed briefing, containing all the necessary technical data. You can also improve the result by telling ChatGPT in advance what kind of text you expect.

Quality of text

The more specific you can be about the changes you want to make, the better and more unique the result will be. Examples of appropriate commands are:

  • Use fewer hyperbolic adjectives
  • Write in a looser style
  • Shorten the text by XXX words
  • Focus more on the topic YYY

Be careful with secret or sensitive information!

Under no circumstances should classified information be entered into ChatGPT. The tool collects all the information you enter during your instructions. It may use this information for other users when they reply. This means that critical information can quickly become public. The processing of internal company content requires its own closed tool based on the ChatGPT interface.

Pros

  • Creating text is quick and easy
  • ChatGPT makes it easy to start writing
  • Often provides very creative ideas for headlines
  • It is possible to improve the result significantly through precise prompting

Disadvantages

  • Generated texts often sound bland and unexciting
  • Press releases and some other texts often have a very American flavour – they almost sound like a loveless translation, not an original
  • Texts need to be carefully checked for accuracy of statements, it still happens that parts of the text are freely invented

Programming – great time saver, but not too complex

ChatGPT is also very popular for creating code. The tool has programming capabilities in the most common languages such as C++, Java and Python. It is particularly useful for simple or repetitive tasks such as file input/output, data manipulation or database queries.

We did a little test here: ChatGPT was asked to write a program to log data from a thermostat. It completed the task with flying colours. Of course, integration of libraries or customisation depending on the naming of devices, etc. still needs to be done independently.

ChatGPT should write a programme that logs data from a thermostat

Things get more difficult when more complex problems need to be solved, or when the code needs to be understood in its overall context in order to build on it in a meaningful way. Here the tool is less reliable or provides results that could have been solved more elegantly. In this sense, ChatGPT is a valuable, time-saving tool. However, it still requires a person who knows what they are doing and who may have specialised domain knowledge. After all, a specialist is best placed to judge whether a piece of code is correct, useful, efficient or suitable for a particular case.

Nevertheless, the coding capabilities of ChatGPT should not be underestimated. Beginners can quickly make great progress with the tool. It can also be used to learn programming. ChatGPT will give you tips on the code you create or answer questions about individual commands or functions.

The liability pitfall

One thing must not be forgotten: ChatGPT assumes no liability for generated code. It is therefore crucial that users understand, comprehend and check the functionality and logic behind the generated code. Therefore, we must not get too comfortable with the tool and trust it blindly. In the long run, it is also not a good idea to have code generated by unskilled people. They often lack an understanding of the basic principles of a programming language, making it difficult for them to spot potential problems or security holes.

Debugging

Another big area where ChatGPT can save a lot of time and hassle is debugging. Aside from writing code, finding and fixing bugs is one of the most important tasks for programmers. Especially when working with regular expressions, errors can easily creep in. Finding and fixing these errors can take several hours and is not the most fulfilling task for a programmer. That’s why a tool that finds bugs, suggests solutions and provides ready-made replacement code is a great help.

In our test, ChatGPT immediately detected the built-in comma error in this code snippet:

Test: ChatGPT was able to immediately recognise the built-in comma error in this code snippet

However, it also found other ‘bugs’ whose ‘fixing’ did not help.

This pattern can be observed in many debugging attempts with ChatGPT. For some issues, the AI identifies bugs much faster than a human programmer would. For other problems, it delivers incorrect or completely unusable results. The more the tool is used and trained through user feedback, the better it will get over time. It’s possible that ChatGPT will soon offer more advanced programming and debugging solutions.

Consider the sources

However, one thing must not be forgotten. All this code does not come from nowhere.

Real people have painstakingly compiled the data that ChatGPT uses to create code. They share this code with the developer community on platforms like GitHub or Stack Overflow. The code is available for free to other users, under the condition that those who benefit from it contribute back to the platform.

ChatGPT breaks this law of participation and reciprocity. In the jargon of the community, it is a ‘taker’, not a ‘maker’. As a result, more open source platforms are resisting the unrestricted extraction of their expertise without compensation by restricting access to their services. For instance, Stack Overflow is insisting that operators of large language models like ChatGPT pay for data usage. If the AI loses access to certain key data sources, the quality of its code generation will eventually decline.

Pros

  • Significantly reduce development time
  • Access to a wide range of web development frameworks and libraries that make development faster and easier
  • Easy introduction to programming or new programming languages
  • Learning opportunities through explanations and the ability to ask questions about the functions of individual commands
  • Debugging features save valuable time and nerves

Disadvantages

  • Limited understanding of context, especially with long and complex code
  • Not always aware of best practices or required security standards
  • Incorrect or inefficient results for complex queries
  • Risk of users trusting the tool too much and not developing their own deeper understanding of the code and underlying concepts

3D printing – an interesting gimmick, but still with weaknesses

Our final attempt is to create an STL file to create a 3D model for additive manufacturing. A major hurdle in using 3D printers is creating a script that allows the machine to understand the desired design.

For designers, learning a full programming language to turn an idea into a three-dimensional object is a long and complex challenge. ChatGPT can help bridge this gap. When a designer provides precise specifications for their idea, the AI can quickly generate an STL script.

We wanted to put this to the test and asked ChatGPT to programme an OpenSCAD script for a christmas star.

3D printing: using ChatGPT to programme an OpenSCAD script for a Christmas star

ChatGPT immediately went to work and created the code. The result is not bad for a first try. Three or four dots are clearly visible. However, something seems to have gone wrong and the star has probably fallen from the Christmas tree and broken on the floor.

3D printing: First attempt to programme an OpenSCAD script for a Christmas star using ChatGPT

So we ask ChatGPT to make the star symmetrical on the second try.

Second attempt to programme a christmas star for 3D printing using ChatGPT

Unfortunately, things go downhill from here. The tool seems to have little imagination as to what the result of the generated code will look like. It delivers several drafts of individual mini-jags arranged in a circle:

Second attempt

Attempt #4:

Attempt #4

Sometimes opening a new chat can help. Unfortunately this was not the case here:

another failed attempt

ChatGPT doesn’t seem to know what a star looks like.

Still no presentable result after further attempts

After several attempts to create a star made up of only one part, rather than several, here we are:

One last attempt
Still no star after 12 attempts

After 12 attempts, we gave up and manually intervened in the code. With a few tweaks, the circle was removed. The five ‘bars’ were long enough to meet at the end to form a point.

After 12 attempts, we gave up and manually intervened in the code

Not exactly the design we would have liked, but clearly much better than anything we could have achieved in the short time without the help of ChatGPT.

Conclusion: Even someone with very limited programming skills can create a 3D model with a little imagination and a lot of patience.

Pros

  • Quick creation of simple shapes and designs
  • Easy to use for amateurs and beginners
  • ChatGPT comments on the code, so you can learn from the designs you create
  • Possibility to ask about the function of individual commands to better understand the code and possibly make manual additions

Disadvantages

  • Difficulty in formulating more complex designs or changes in a way that the tool can understand and implement them
  • Design errors or changes are sometimes difficult to understand
  • Many iterations required to achieve the desired result
  • Apparently ChatGPT has a very limited horizon of what everyday things look like (e.g. stars!) or what the model of the generated code looks like

Conclusion: It’s up to us

We can use ChatGPT to find solutions that are actually beyond our capabilities. As a helping hand, source of inspiration and sparring partner, it provides us with valuable insights . It helps us to complete tasks better and faster. What the model cannot do, however, is read minds, solve very specific problems without supervision, and think critically. ChatGPT is a powerful tool – but it’s still a tool. It’s up to us how we use it and what we get out of it.

Images: reichelt elektronik, Adobe Stock


Deepen your knowledge with more magazine articles on the exciting topic of artificial intelligence, and keep your finger on the pulse of technological developments:

Part 1: ChatGPT – ready for industrial use?
Topic-Explorer – AI & Machine Learning

Leave a Reply

Your email address will not be published. Required fields are marked *