How likely is an idea, formed of two sub-ideas, to be interesting?

That is, novel/surprising and potentially useful?

Within a domain, many of the ideas have been tried, so the likelihood an idea is novel is low.

If you particle collide two ideas from different domains, most of the time it's novel... but also not potentially useful.

Only a small subset of ideas from very different domains combine to create something useful.

It takes patience to try all the combinations.

But you know what has patience? Computers!

But computers aren't able to do synthesis.

But now they can, with LLMs!