I’ve read several topics here dealing with “long form content” but I’m unable to produce what I want.
I need to write a text of around 500/1000 words about a certain topic.
And this text must be divided into several paragraphs (i.e. several questions or sub topics).
So I’ve tried different things:
asking directly to write an article about topic “xxxxxxxxx” divided into X paragraphs
asking to generate an outline for an article about topic “xxxxxxxxxx” and then generate a paragraph for each item of this outline (which obviously is not good because there is no connection between the paragraphs so there is often repetitions among the paragraphs)
asking to extract keywords from the topic and write paragraphs about topics linked to these keywords
Well I’ve tried many things and I can’t find the right method so if you have any pointer to what I could do to get closer that would be awesome!
Let the completion modes complete what you type in whatever size sections you like. But the more you let it go alone, the more it will predict completions outside your scope. Here is an example of speaking to it cooperatively in first person:
This is a list of ideas, news and topics to be discussed on [subject]:
-**generate (this is pressing the submit or generate button and you will see some great and helpful words on the above completion topic.
The following is a blog outline using these ideas including an introduction and conclusion
//now you can let it do it all at just 500 or 1000 words or for Maximum quality possible for you, if you’re using davinci 2 or another very advanced one and/or, you can just start typing each section or paragraph yourself then let it finish (this seems that it will give you the closest thing to what you’re looking for or the desired quality that is possible for you. No matter if you choose the human-guided quality one that relays your plain English in sufficient pieces or the lazy button pushing, id just make sure a human reads through it first before publishing. That seems to be the best practice and very helpful.
There was or is a youtube account called skolo or something that was making small programs for blog processes with OPENAI and showing you how to do it and the components and all that so if you want a more automated approach, maybe you can look there
Feel free to ask any questions about strategy or utilization for generating words or images. I don’t work here, I’m just a guy but I tried all the big image generators and most of the word ones, and a lot of other stuff.
(edited a typo, a clarification, a word change and an offer to field other questions.)
Hi Dave,
I had watched some of your videos, including this one, and I must admit I couldn’t replicate your results but I’ll watch it again, and try again, that’s what it’s all about anyway
One thing I forgot to mention is that I’m using OpenAI to write in French and sometimes it seems like I’m having difficulties to have the AI understand what I want
Hi Derrick,
Thanks for the input. Since I need to write in a fully automated manner, started with a topic or most of the time an article title (and then have it reviewed by an editor) I cannot use the “write the beginning and have the AI complete it” method.
But I will explore your suggestions and maybe reach out to you again if I get stuck
Sure, no problem. I was not exaggerating when I said the component-by-component human guidance would give the Maximum results possible for you. But it is not usually necessary, and seldom is to get the maximum results. With the principles of the first post and an understanding and slight adjustment of a few sliders on the controls to the right, you could use an example like this
This is 500-1000 grammatically correct french words outlined, generated, and translated to a french language blog post regarding the subject of [ The best types. of cheese ]. The clever, relatable and well-formatted 500-1000 coherent french language article are as follows:
That’s it; that should do it. Note that this is instead of a please write and is used here and it’s much more consistent when you’re thinking with the machine instead of trying to let it guess what you want it to say in the completion provided by it.
Obviously (to some) replace the cheese stuff in the brackets whether you leave the brackets there to guide you or not, it would give you something kinda good enough in a way usually then it can be improved in the settings on the right before you generate the results you’re going to use. and you could always set up a no-code automation to take whatever good enough close thing you get that only needs to be fed one last digestible piece of further instruction.
And that is how you become a sorcerer and why if Openai does not give us a 14-day warning before cutting all the poor people off from Codex, this should be the biggest and fastest transfer of wealth to the richest 1% of the population in human history.
An important note: Good luck and good, smart idea reading and watching this Dave Shapiro guy even if we can’t replicate all his coding, adjustments and implementations ourselves doesn’t mean we can’t learn a lot from his experiments and posts on github, reddit, twitter, youtube, here etc. For my money, he’s the premier expert in the world on this general topic though it is a topic in its infancy for everyone. I saw him getting 2 replies on Reddit talking about finishing jobs before his credits run out, back in the first part of this year and now he’s actually entering a phase where he’s getting some recognition and I’m glad about it even if he can’t offer as much personal free help to us peasants, like he usually does. I’d still use that as a general idea of how this all works, with words, at the highest known level at the moment. I’ve never spoken to him in my life but I’ve learned more than from anybody putting my eyes on his stuff regarding building even simple cognitive architectures to coincide with these large nlp systems, models and datasets.
This is getting long so I’ll end it here but the short version is the part between the lines should work or some variation of that and first person is still the key if you do it a (usually) lower quality way.
I intended to bold only the word “maximum” but don’t want to make edits on that post too. So I’ll just note and edit this one prior to your response if one is forthcoming. There is always more to say or a way to do better I suppose.
Some of the grammatical errors were inserted by my careless use of incorrect Grammarly but it seems comprehensible. I hope something like this solution works for you until you acquire an interface where you can just enter one french word and get your elite french article.
Here is 2 minutes of effort making it instead of 15 seconds of it but they both work and should give you an idea of what you need to do
This is 500-1000 grammatically correct french words outlined, generated, and translated to a French language blog post article, regarding the subject of [ The best types of cheese ]. The clever, relatable, persuasive and well-formatted 500-1000 word coherent French language article with an introduction paragraph at the start, main body paragraphs and then a conclusion paragraph at the end, is as follows:
Hello Dave, I wanted to watch your video again but it looks like you made it private. Any chance it could be made available again? (the title of the video is “Write an entire blog post with GPT-3 fully automatically”)
Thank you
G.
