A prompt doesnt give good results very short output

API
1

when i use the prompt

write 200 exactly 6 word sentences about what the person can be thinking in their brain, the subject is something cool you can do about yourself

i get only 6 lines of text in a text file

Thinking about brain capacity.
Exploring own capabilities.
Seeking knowledge to understand.
Many possibilities to ponder.
Discovering new ideas everyday.
Unlimited potential within me.

this is my python code

import os
from tqdm import tqdm
import openai

Set up your OpenAI API key

openai.api_key = ‘’

def generate_article(prompt, article_length):
response = openai.Completion.create(
engine=‘text-davinci-003’,
prompt=prompt,
max_tokens=article_length,
temperature=0.7,
n=1,
stop=None
)

generated_text = response.choices[0].text.strip()
return generated_text

def main():
# Path to the text file containing article prompts
text_file = ‘bci.txt’

# Length of the generated articles
article_length = 3000

with open(text_file, 'r') as file:
    prompts = file.readlines()

# Create a folder named 'generated_articles' if it doesn't exist
if not os.path.exists('generated_articles'):
    os.makedirs('generated_articles')

# Iterate over the prompts and generate articles
for i, prompt in tqdm(enumerate(prompts, start=1), total=len(prompts), desc="Generating articles"):
    filename = f"generated_articles/{i}.txt"  # Generate the filename based on the prompt's order
    
    # Skip generating the article if the output file already exists
    if os.path.exists(filename):
        continue
    
    article = generate_article(prompt, article_length)
    with open(filename, 'w') as file:
        file.write(article)

if name == ‘main’:
main()

2

The GPT models, in this case Davinchi-003 do not perform well with large numeric value tasks like this, counting words and even sentences is difficult for them, a more effective method would be ask the AI to create lets say 4 examples and then call it 25 times.

1 Like