Fraud detection by distance is broken

[iocanel]

While working on #566 I realized that our distance based fraud detection is broken.
By broken I mean that it constantly calculates as 0 the distance between any pair of cities and thus fails to correctly detect frauds.

I think that the issue has to do with the capabilities of the model. ChatGPT 4 is capable to calculate distances, while 3.5 is not.

[geoand]

In that case, we should probably use the latest OpenAI model

[iocanel]

I verify, that going with gpt-4 solves the issue. I am working on a PR.

[iocanel]

Resolved as of #568