so, it's very import for private. we develop a DB-GPT project which can chat with local GPT. It's 100% private and security.

Have you tried churning out random numbers that start with 35 to see if the hit rate is similar? Also, how many times did you test it?

First try.

Ah, so it might have been a fluke. I took "4 out of 5" to be a ratio, but now I realize you mean just for this list.

You really should should re-ask it a few times to see. And then if it's not a fluke, try with a prng to see if the results are similar.

It may very well not be a ChatGPT thing, but an easily guessable thing just by stringing random numbers together.

ChatGPT generates strings of numbers that are the same length as IMEI codes, but are mostly invalid.

This isn't news.

The last one in the image is valid.

What's your point? Everything that randomly generates numbers of appropriate length could hit a real one eventually. It's not a "leak," especially because it's not even for an iPhone, as requested.

This isn't news.

25% hit rate on a 15-digit sequence in the first try is leak and news.

It's not news unless it is regular. Try it a bunch of times and see how the "4 out of 5" stands up. It very well might. Or you might also not ever find another one. You won't know until you try it.

Put it this way: you can find the same sequence of numbers in the decimal digits of pi. That's not news - that's the definition of random in action.

It would be news if it was predictable and repeatable, but until someone tries it enough times to rule out it having been a total fluke, we'll never know.