(Part of my series of texts about prompt engineering - see also: https://laibyrinth.blogspot.com/2023/11/forget-prompt-engineering-there-are.html )
'Prompt engineering' is claimed to fix a lot of issues with ChatGPT, or even to avoid them in the first place.
And yes, it's true, working with ChatGPT can be quite complicated. The output might be not what you want - or even be faulty!
You can get stuck, arrive at a point where it's seemingly impossible that ChatGPT does the job and completes the task.
'Prompt engineering' promises to provide a workaround for this.
But you don't need prompt engineering - because you don't need a workaround!
You can solve most issues - by getting meta with ChatGPT.
If there is a problem - just talk with ChatGPT about it!
ChatBot AIs have the unique ability to reflect on (and to understand) the task they do, as well as the conversation you are having with them, and any clashes that may happen between what you envision and the output they provide.
If there is such clash, or an error happens, just tell ChatGPT about this and ask it to fix it, and it will try to do so.
Example:
As mentioned, I wrote multiple "producing music with ChatGPT" tutorials. The main idea was that ChatGPT outputs each note, chord, drum sequence, etc. - every element of the song / track - and I just copy the data into my DAW (this is a bit oversimplified - it's actually more complex. Read about it, if interested - https://laibyrinth.blogspot.com/p/how-to-create-music-with-chatgpt.html ).
This worked much better than expected. But it would not have needed to work well at all!
If it failed to do the job, if 'my' idea had not worked - I could just have told ChatGPT about it!
And in such a case, it would point out where we did get wrong, the errors we made, and potential problems in the future - and then provide a fix to it.
ChatGPT has the ability to self-enhance and to fix issues with a task it does - by itself.
So, you can use ChatGPT to work on something - or you can use ChatGPT to *work on working on* things. You can go meta.
If the performance is sub-par - always ask ChatGPT to fix it.
Second example:
Other music producers read my tutorials, and told me they were thankful for it, because they tried to get ChatGPT to generate music or elements of track themselves, and the AI did a great job; but they also mentioned that, in some cases, ChatGPT failed to provide the output they desired, and wondered what went wrong.
But they did not need to ask me about these issues at all!
They just could have told ChatGPT!
They could have typed something like: "Dear ChatGPT, I wanted to work on creating music with you, but it seems the output you created for [insert description of what you wanted here] is not what I envisioned at all! What can we do about this?"
And ChatGPT would have tried to provide a solution for it!
Another example: many people use ChatGPT to create texts; maybe promotional texts, or text for business, or even (short) stories, and so on.
As beginners, they might not like the output ChatGPT provides.
Maybe it's the "wrong" style, sentiment, mood, length - whatever.
In these cases - just tell ChatGPT about it.
You could do it the following way: "You wrote a great promotional text for my upcoming album release. But it sounds a bit 'technical', even overly 'professional' to me. I'm more of a low-brow DIY underground producer, ya know? So let's get down with that street jive for the promo text!"
And ChatGPT will do this.
So whenever you run into an error with ChatGPT - just point it out, and try to find a fix in a conversation with ChatGPT.
Be meta!
Many times, it will be as simple as that.
Other times, it will not go that easy, and issues do remain. Then be meta about that too - address this new (meta) issue, and try to fix it.
It must be said that there are rare cases where all of this might not lead you to the desired results. As I said, ChatGPT's intelligence and resources are vast, but not infinite. It cannot complete *every* possible task (but some impossible ones, though!).
In these cases, leave it at that, use your human brain for the specific task, or do things differently.
But in most cases, going meta will provide a solution.
No comments:
Post a Comment