Chat GPT API Not making articles that are very long, any tips?

IslanderMark

Registered Member
Joined
Sep 2, 2020
Messages
88
Reaction score
21
Have a python script to generate content through the GPT API.

Each prompt basically gets me around 350-500 words worth of content. I want at least 750 per realistically.

I have the max tokens set to 4,000 which is basically the max i believe, temp to .3, have tried bumping it up but found it was not better as it began to go way off course.

I cant get a prompt for the AI to get more words out for the life of me right now.

Does anyone have any experience with the API and have had similar issues they were able to figure out?
 
be more specific in the prompt how the article should be structured and you can also tell it a range of words it should use.
it doesnt always follow it though.. sometimes its more, sometimes its less, but i works.
 
I want at least 750 per realistically.
Have you mentioned that in your prompt?
Have you also mentioned what topics to cover in your prompt.
You need to cover all the info in brief because that AI is just like a smart baby.
The better you explain your requirements to the baby the better response you get from it.

In simple words either you yourself create the best possible article structure/outline and feed it in your prompt to the AI
Or
Ask the AI to create the outline first and the expand it into an article.
 
be more specific in the prompt how the article should be structured and you can also tell it a range of words it should use.
it doesnt always follow it though.. sometimes its more, sometimes its less, but i works.
I give it more rules to follow in terms of headings and lists and whatnot.
I actually also just messed with more settings in the playground and it began to work a lot better. Just got one article over 1000 words.
 
Have you mentioned that in your prompt?
Have you also mentioned what topics to cover in your prompt.
You need to cover all the info in brief because that AI is just like a smart baby.
The better you explain your requirements to the baby the better response you get from it.

In simple words either you yourself create the best possible article structure/outline and feed it in your prompt to the AI
Or
Ask the AI to create the outline first and the expand it into an article.
Mentioned the word count which did nothing. I tell it the main focus and am now giving it more ideas to write by saying it needs a certain amount of headings and lists and stuff like that. Thanks.
 
Firstly, it is important to keep in mind that the length and quality of the content generated by the GPT API may vary based on several factors, including the prompt, the parameters used, and the complexity of the task. While increasing the max tokens may allow for longer responses, it may also result in lower quality content if the model is not able to maintain coherence and relevance.

One approach you could try is to experiment with different prompts and parameters to see if you can generate longer and more relevant content. For example, you could try breaking down your topic into smaller subtopics and generating prompts for each one, or you could try adjusting the temperature or other parameters to see if this improves the quality of the responses.

Another approach you could consider is to use a different AI language model or API that is better suited to generating longer responses. For example, some models may be better at maintaining coherence and relevance over longer texts, while others may specialize in generating summaries or outlines.

Ultimately, it may require some trial and error to find the best approach for your specific needs, but I hope this information is helpful in guiding your experimentation.

very obvious AI text lol
 
try and tell it to make an article 1000 words or longer. or have it write an article then say expand or talk more about topic a b c.
 
if it stops, i use to do this, but for coding tasks:

"Please continue write where you accidentally stopped, do NOT re-write any text from previous answer, just continue, exactly where you stopped, here let me remind you:

blab blalb ba bla bblala balla end"


just copy the last rows from his previous answer and he wil lcontinue
 
I give it more rules to follow in terms of headings and lists and whatnot.
I actually also just messed with more settings in the playground and it began to work a lot better. Just got one article over 1000 words.
Can i know what rules did you talking about? Example please so GPT can create more than 1000 words?
 
if it stops, i use to do this, but for coding tasks:

"Please continue write where you accidentally stopped, do NOT re-write any text from previous answer, just continue, exactly where you stopped, here let me remind you:

blab blalb ba bla bblala balla end"


just copy the last rows from his previous answer and he wil lcontinue
Always find it funny that people ask it "please". But really, no need to waste tokens on making prompts long winded.

Just try "continue". If the token context is already full and kicking out important details from earlier prompts and responses, there's nothing you can really do but try and break the response into smaller responses by prompting for it in pieces, then combining them.
 
Can i know what rules did you talking about? Example please so GPT can create more than 1000 words?
Something like "Write an article about heart disease around 1000 words long"

Keep in mind that with longer text it is more likely to repeat info but the temperature (randomness) will cause it to repeat it with different wording. With the API and API playground you can increase the penalty for reusing the same words or phrases, giving you more control than the basic ChatGPT interface.

Even so, how much unique info it can provide on a topic depends on how much it was trained on that topic. Some things it will be able to give more details on than others.
 
Your using free account credit it limited.

Upgrade to plus and fund the account.
 
if it stops, i use to do this, but for coding tasks:

"Please continue write where you accidentally stopped, do NOT re-write any text from previous answer, just continue, exactly where you stopped, here let me remind you:

blab blalb ba bla bblala balla end"


just copy the last rows from his previous answer and he wil lcontinue
This post is about the API which provides different output than it would when going through the normal GPT chat. But i tried to implement this in a way that works with the API and it didn't work, but this was before i figured out a pretty good workaround
 
this post is about the API not normal GPT chat. I pay for the API usage

Send them any project you are working on that is not yet possible to create with gpt3.5, i mean a tool that does things 3.5 woudlnt be able to do.

and you will be granted gpt4 access fast.

alternatively, if you don't have any project, try apply with some fake site hehe, its not like they are asking u heavy documentation.
 
Send them any project you are working on that is not yet possible to create with gpt3.5, i mean a tool that does things 3.5 woudlnt be able to do.

and you will be granted gpt4 access fast.

alternatively, if you don't have any project, try apply with some fake site hehe, its not like they are asking u heavy documentation.
I got on the waitlist liek a week ago... too late to request again with another story. hope they approve me within a few months
 
Back
Top