It’s difficult to say my exact approach, but I’m really trying to approach it from all angles. I’ve used strings of variables to signify processes and use those to crunch ideas, but I’ve also tried to trick it into adding an alternate layer to its output.
I think the issue it most has trouble with is doing something really complicated with multiple prompts, where it almost forgets the initial goal after you try to align it more closely to what I wanted originally.
After trying it hundreds of times in dozens of different ways, there is one way it could do this properly and that is to make it Confirm exactly what I want from it before it does the thing, almost doing a little example and asking me to confirm “is this what you were asking for” before it does the actual process. So when I ask it to do something complicated before it just goes off, it would be great if there was a dummy prompt requiring me to sign off on the idea before it does the thing. Especially when the dialog is verbose.
I think this would solve more problems than just creative, writing or lyrics or anything, these confirmations would actually remove most of the mystery of how it loses sight in each session so quickly.
I imagine a visual flow chart or something like that when things go ten or 20 prompts in would really help users find out if the info it’s getting is even actually being crunched properly which often it is not. Language still is the barrier here because figurative speech can be messy which is why figures should be used to communicate it’s understanding with the user.
Some decent outputs;
From the mines to the stock floor,
Wealth amassed, but we crave more.
Forests burn, the oceans choke,
All for the sake of another smoke.
In a world of shadows, beneath the crimson sky,
Together we’ll rise, rise, rise,
Battling storms, as stars take flight,
We cling to love, embracing the night.
In sorrow’s maze, we reach for a hand,
Hope’s beacon shines, in shifting sand,
The chains that bind, like storms, they break,
In love we create, our hearts awake.
Together we’ll rise
The thing about some of these lyrics I’ve been able to get from it is that you actually do have to sing it to understand what it’s doing. It’s being told to analyze the execution of the lyrics, and to make sure that the imagery is unfolding in time with a listener.
Sometimes it works really well and you can see your own imagination form the imagery, but it does have to be sung because it’s broken down through syllables from the perspective of a listener. The big problem is to get it to understand the dynamic of what a song is, as it is being experienced real time by a listener with the added dimension of the lyrics unfolding in real time. It gets so caught up in the literal context that it forgets that without the following word, the image of the syllable itself is generating imagery so the distance between the words themselves actually will create the wrong impression unless it understands that before it does any generation. Getting it to understand the idea of creating imagery through words, and having the next lyric or phrase make imagery that adds to this, without being repetitive in its unfolding, being captivating in any way, Still a work in progress