Subema on Nostr: While there is certainly still a gap when you have a perfect image of the thing you ...

While there is certainly still a gap when you have a perfect image of the thing you want to create in your mind, it's like with any other tool - you can't expect to hit the moon despite the hype around it.

There are best practices and pitfalls to avoid:

* prefer rolling back the chat and updating the prompt rather than explaining the issue
* don't put all the content in one session
* iteratively create a sort of handbook of your style, get back to it when there's a miss and update it
* try few-shot prompting
* for audio use models that are actually intended for that, not just generic multimodals that get you like 90% there
* some models can better match your tone (for copywriting I think chatgpt takes the crown, for nerd explanation claude hits the spot for me)

Not saying it will get you 100% of what you expect, just wanted to show that there is some non-transferable experience you have to feel out by yourself, and judging its capabilities on limited interaction can be misleading.