AI can already do a lot, but it also surprises us with what it cannot (yet) do.
Anyone who uses LinkedIn has probably seen so-called ‘action figures’ everywhere. These are AI-generated representations of people together with symbols and text that refer to their professional activities, for example.
A blacksmith, for example, (if they are active on LinkedIn) might have a hammer and an anvil and a light-hearted remark (‘LinkedIn? I prefer to forge real connections – and sometimes horseshoes.’).
We wanted to try this out, and so we asked ChatGPT:
“We would like to create a suitable action figure for ‘Alef Völkner’. Can you give us some suggestions?”
The AI was very good at finding and collating the relevant data:
- ChatGPT found the right photo:
- Alef Völkner’s name, job title and main areas of work were summarized correctly (CEO of fox-on and foxondo, focus on pragmatic implementation of data protection, etc.).
- There was also a reference to the company’s 20th anniversary celebration in September 2024 (including a suggestion to illustrate this with a champagne glass).
So far, so good.
But then we asked ChatGPT to create a corresponding image. And this image wasn’t so convincing after all… 😉
It took several more queries and explicit instructions before ChatGPT finally produced a version that bore at least a distant resemblance:
We found this particularly interesting because our managing director, Alef Völkner, has often experienced her name being mistaken for a male first name. But that was only the case until people met her in person or at least saw her photo.
What else did we learn from our little experiment?
AI is only as good as the instructions you give it. That’s why a really good prompt is often much longer than the actual answer you receive.