alt Hacker NewsI think the moat(s) is anything the models still are nowhere near performing on their own, OR, getting them to do that is seriously expensive.
I don't know about Astra, and have used Fable 5.1 only slightly (And burned 85% of my 5 hour quota in 1 prompt).
In my experience, LLMs are completely devoid of "thinking outside the box" capabilities. And I'm not talking about 6 thinking hats type of stuff, but slight of to the right field lateral thinking. Something a 6 year old does for fun when playing with their mates.
If this weren't so, we wouldn't have been through so many "<buzzword> engineering" in such a short amount of time.
I think this is probably true, but a lot of creative things are small-c creative: you brainstorm, you come up with a bunch of vanilla stuff, you connect some of those ordinary ideas.
The brainstorming thing, AI is really good at. "Gimme a bunch of 80s pop culture" is going to get you a longer list than you could think of yourself, and then you can just supply the judgement about what you want.