I think your understanding of what “AI” is fundamentally flawed if you are asking this question.

The easiest and safest intro to understanding how models work is first to understand that they are not making decisions, they are sorting through data. It’s a very fast parallelization of sorting. There is no version of any model that has a concept of a novel idea.

Next steps from there are understanding what it can provide you, and that is only data it has been trained on and programmed to make relations between. If you get deep down into the lingo, it will referred to as an “alignment” or “relation”.

Example: “A” is a letter. “B” is a letter. They are both letters in the English alphabet. They are not letters of the Cyrillic alphabet. English speaking people will use A and B to write or speak. English is used by these countries…etc

That’s all it can give you.

If you go and ask it how to build a Linux distribution, it will tell you all the relations it knows about those keywords, but can’t go and generate new and novel “thoughts” about it. In more practical situations, it can only give you information about a single point of reference at a time, and a Linux distribution is thousands.