I had an example In the artikel Use of AI in production design, where I ask Copilot for a list of object and the tools answered with a list of over hundred of objects.

But when I asked the same question in a swedish manuscript I only got eleven objects, but there are far more in the text.

The first english document was 83 pages and 16 865 words, the second in swedish was 117 pages and contained 17163 words. A thrird try on the last document with 90 pages and 13007 words gave the same result.

So the question, is the difference due to the language or something else? Both are manuscripts copied and pasted into MS-Word. The latter one is more dialogue so that may be a reason.

The challenge is that it’s very hard to debug there kind of systems, ans we don’t have the source code. You can just look att the input and the output.

Reminds me of riding a horse, your signals as a rider and the reaction of the horse. This takes a lifetime to master., so I don’t expect AI to be better.

Disruptive architecture