This is an older post from September 2024 that I never published (until now). Might be of interest to someone. I tested the same question today June 2:nd but in Swedish with the exact same result.

Earlier this week I was testing the possibility to generate images in ChatGPT and I found that really amazing. There is no doubt that AI technology is here to stay for obvious reasons. There are so many tasks where AI can make a really huge difference. I have, however for a long time anticipated some sort of setback for AI. I think the running point is what data the model is trained with and how it is used and I think there are thousands of times when the technology is used wrong. The reason for this is the wide and very fast adoption without much thought. Since this is kind of a new way to use computers we need do learn how to master it.

Computers have always been fantastic at doing repetitive task very accurate, and still are. Given a correct algorithm and valid input the output will always be correct and the same over and over again.

However with AI this might no longer be true, at least not from the users point of view. Especially when we ask computer things in the real world the algorithms start guessing. Another problem with this is that, for instance ChatGPT answers in a way that makes it sound like facts with references and all even though it says at the bottom of the page that it needs fact checking. And just that step is the thing I’m afraid will be missed.

An example I tried relates to swimming. I’m a swimming referee and have knowledge about the regulations regarding competitive swimming and I know it is quite well documented and published on the web, so it felt natural to ask a question about that.

So I asked ChatGPT this question:

How far can you swim under water in a competitive breast stroke race?

Immediately I got this answer:

In a competitive breaststroke race, swimmers are limited to how far they can swim underwater. According to the rules set by the international governing body, FINA, a swimmer can only swim underwater for 15 meters after the start and after each turn. After reaching that distance, they must surface and continue the race above water.

This rule is in place to ensure fairness and consistency across the sport. While swimmers are allowed to use a single underwater pull and kick, they must break the water’s surface by the 15-meter mark.

That really sounds very convincing. It also references the FINA (as World Aquatics was called earlier) rules, note that FINA and 15 meters are emphasized and it also adds a section about why the rule is in place which makes it sound even more convincing. The thing is that it is completely wrong, there is no 15 meter limitation in breast stoke and to my knowledge there never was.

The obvious risk here is that someone takes this as truth and uses it in a real world case. And it also might be regarding something much more important than a rule about swimming competitions.

If you read the regulations that ChatGPT refers to you can easily see that this is wrong. Since I know this is wrong I asked ChatGPT about it:

Is there really a 15 meter limit in breast stoke?

And the answer surprised me somewhat:

You’re right to question that! I made an error in my initial response.

Unlike other strokes like freestyle and backstroke, there is no 15-meter underwater limit in breaststroke. Instead, the rules for breaststroke are different:

• Swimmers are allowed one full arm pull and one kick underwater after the start and after each turn.
• They must surface before the second arm pull.
• The swimmer’s head must break the surface during each stroke cycle.

So, while there isn’t a specific distance limit like in other strokes, swimmers cannot stay underwater for an extended period in breaststroke, as they must follow the rule about surfacing during the stroke cycle.

And that is the correct answer, there is no 15 meter limitation.

But … now ChatGPT added another error. In the also very convincing points in the middle it adds ”They must surface before the second arm pull” with emphasizing and all.

That was information I did not really ask about and it is completely wrong that the swimmers must surface before the second arm pull. They are allowed to initiate the second arm pull when submerged but the head must reach the surface before the widest part of the second arm pull.

So now I got an answer that is correct (at least to my first question), but that is only because I knew it before, if I did not know the answer I probably would stop at the first answer and believed that ”FINA rules” says there is a 15 meter limit in breast stoke.

This is of course known to experts using these tools (I’m not one of them) but I’m a little concerned about what will happen when common people in positions all over the world gets these tools in their hands and starts trusting them.

I’m not too much worried about AI taking over all our jobs and in the end the world, I’m more worried about the consequences of miss use of this fine technology in the near future. It is obvious that you cannot use ChatGPT in this way, but I’m afraid that some users might think they can. Instead you still have to ask an expert or go to the source just like before AI.

You can se the chat here.

This text was written by me. Without AI. All mistakes are mine. I did run this text through AI and asked for improvements. The result was probably more grammatically correct and much more ”bombastic” and powerful, but it just wasn’t me so you’ll have to stick with this.