Looking for help with optimizing prompts to save more time

I try to optimize prompts to work quicker, but i still have a few issues:

  • How can i create a prompt which writes articles which are as unique as possible?
  • How can i create a prompt which generates titles which started only with a capital letter and therefore do not capitalize almost every word? So only the first word starts with a capital letter.
  • How can i create a prompt which will format titles and subtitles directly in H2 or H3? So i can copy-paste the text without having to change format/styling of titles in Wordpress. Now sometimes the generated text is only in bold, but the ‘H2’ or ‘H3’ is added as text at the beginning or the end of a title.

Does anyone have any tips?

Welcome to the community!

So, there’s a lot to unpack here. I’ll try and answer each question accordingly:

  1. What does “unique” mean to you? If you want to generate articles that sound authentic and aren’t written like how an AI would write them, I would hire a writer, or edit them yourself. These systems are great for reducing time to write articles, but there are plenty of examples showcasing why completely copy-pasting AI generated content into articles is a bad idea. Keep in mind too, unique may not always mean coherent or accurate either. Do you have a mechanism in place for checking accuracy?

  2. This one is going to be tough, because you’re going against the grain of what is the grammatically correct form to write headlines and titles, as well as what most professional writers ask for. Perhaps providing an example here in the beginning of your conversation will help, if you haven’t tried simply asking it to make only the first letter of the first word capitalized for each given title.

  3. This is another problem that could be solved wither with an example provided to the model or few-shot prompting /refining. Also, you can request it to format text specifically for something like a wordpress article using h2/h3 headers. I ask it to format my text for github readme’s without any problems. If it provides raw textual output, it’s a simple “Great! Now could you format your response properly for wordpress?” and voila.

Being as clear and precise about what you want as possible will help you easily achieve this. I wouldn’t recommend trying to make this all a single shot prompt, but depending on your prompt itself, it might be doable so long as you explicitly tell it what you want and provide an example for how you want it. Examples go far with these creatures.

Just, please, as a former Science and Technology editor, please don’t copy/paste its responses to directly turn into articles without editing them and verifying it for accuracy. This is why the WGA went on strike. It doesn’t produce quality articles like you think it does, and trying to send out unedited AI-generated content en masse throughout the internet is exacerbating a lot of people’s concerns about information on the internet right now. Please be careful and wise about this.

2 Likes

If you’re able to work with the API, you can adjust the temperature settings higher so that your answers are more ‘creative’, according to the hover-instructions there. You could also specify the unique elements you’d like such as a taboo couple, a sci-fi setting, or other imaginative features you can think of. Just give it a little nudge and refine it later.

The above person is correct to say that formatting is really hard to change because it’s one of those ‘core’ training features too deeply embedded to alter at a whim. You can try being explicit and try generating all of them and asking the model to collect and format after. So for example, in your first prompt you generate the titles, and it’s likely to give you some additional detail. Instead of copy pasting, try to ask in another prompt for the model to “gather” just the titles. Then you can try to fix the formatting. But, it might just be easier to copy that list and run the ‘lowercase’ formatting and capitalize just the first word like you want. You can even try those instructions to the model, but again, going against its training will be tricky.

If you can specify your formatting in the first prompt or in the presets, it has a much better chance of being successful. I think this should be available now since there have been a lot of developments since Nov 2023. Is it working better for you now?

1 Like

That is far away from truth, even it is popular way write scoop headlines in the States.

Something like this may work:

When writing headlines, always start the first word with a capital letter. All other words are written with a lowercase letter, except for proper nouns and other words that normally start with a capital letter according to grammar rules.

An example:

## Header with right use of capital and lowercase for AI