ChatGPT: Let the Robot do it!

My fellow Integration Developer Daniel Graversen has recently demonstrated what is possible with the new ChatGPT, basically an AI based QnA bot.

Seeing and trying this out by myself really astonished me and made me a bit worried at the same time! Just a few weeks ago I would not have expected AI to work so well with answering technical or other complex questions in at least the next 5-10 years. So I’m quite shocked about what is now already possible! And it makes me think how quickly this will change our work environment in the near and farther future. In particular I’m wondering how it will affect us Integration developers and the tools we’re using?

I believe a lot of what we do today will be assisted by AI tools that make things much quicker and more efficient. When starting to work on a new topic or trying to remember how we did something the last time we needed it (sounds familiar? ;-) ), we will probably spend less time Googling around for hints and potential solutions. Instead, we might type a question in ChatGTP and the bot will produce an apparently often pretty accurate and helpful answer within seconds. We will always be required to use our own discernment, experience and knowledge to judge the quality and plausibility of the answer. Furthermore, we will in most times have to do further research or tweak the answer (for example when it includes programming code) to come to a final conclusion or solution, but I think such bots will become good sources for suggestions, pointers and new ideas.?

A good example of a case where the bot can already produce almost ready-to-use results is the following: In addition to what Daniel documented in his article, I’d like to add one more example of how incredible the cognitive abilities of this system are and how it can speed up our daily Integration work. I wanted to see if ChatGTP can also come up with the XSLT code that defines the transformation logic between a sample of the source and destination XML documents. And of course it can! However, in some cases it needs at least two tries. It seems that a click on the “Regenerate Response” button below an answer makes ChatGTP come up with a different approach or maybe a second guess. So as mentioned above, we still need to understand what we’re doing, but can be greatly supported by this amazing tool.?

My question was the one in the screenshot below.

Note: the source of the XSLT files is the blog post "Use cases of node functions in message mapping" by Eaksiri Sontisirikul about how to implement such a mapping in the graphical message mapping editor of SAP Cloud Integration - which is something ChatGPT will most certainly never be able to do (but could probably easily be adapted, integrated into the product and trained to do so!).

Question:

As you can see, the aim is to flatten the XML.?

The answer was this:?

This looked already quite promising at first. Impressive is also the plain text explanation below the code snippet! However, the XSLT doesn’t really work. The result of the transformation comes out as follows:?

We expected four Employee elements, not two half empty ones. I could now use the result as a basis for my own modifications on the XSLT. But I first wanted to give ChatGTP another chance and clicked the Regenerate button, not really expecting a better result... But I was wrong! It's a pretty different solution and actually does exactly what I needed! See below:?

Again also the explanation of the script is astounding.

The correct result of this XSLT is as follows:?

As somebody who doesn't use XSLT every day (and probably even if), it would have definitely taken me longer to tweak the first result than just click a button and get the correct one!

I can't tell how exactly this is possible. But some hints of how the AI was trained are given here: https://help.openai.com/en/articles/6783457-chatgpt-faq

Let me know what you think about the demonstrated abilities of ChatGPT and how it will influence our work environment.