How do I generate a proper summary with OpenAI's API when the output mirrors the input?

Below is some advice to help you get a proper summary:

1. Be Explicit in the Prompt

Ensure that your prompt clearly instructs the model to summarize the text. For example, instead of passing the entire Earth description with no explicit instruction, modify your prompt (wPrompt) to contain something like:

PYTHONCopied
wPrompt = "Please summarize the following text in a concise paragraph:\n\n" + long_text

2. Consider the Context Length

If your input text is too long (near or exceeding the model’s token limit), the model may simply echo parts of it. For very long texts, consider breaking the text into smaller sections and generating intermediate summaries. Then, recursively combine these summaries into a final one.

3. Adjust API Parameters if Needed

While parameters such as max_tokens and temperature are important, the key is to have a prompt that clearly specifies the summarization task. For summarization tasks, a lower temperature (like 0.3–0.5) usually works well, but focus first on the clarity of your instruction.

Example Code

Here’s how you might adjust your code:

PYTHONCopied
import openai

# Ensure you set up your API key
openai.api_key = "YOUR_API_KEY"

# Explicit instruction to the model for summarization
wPrompt = (
    "Please summarize the following text in a concise paragraph:\n\n"
    "'''"
    "Your long text (e.g. the Wikipedia description of Earth) goes here."
    "'''"
)

resp = openai.Completion.create(
    model="text-davinci-003",
    prompt=wPrompt,
    max_tokens=1024,
    temperature=0.5
)

print(resp.choices[0].text.strip())

Summary

• Explicit Task: Clearly instruct the model to "summarize" in the prompt.
• Input Size: For very long texts, consider summarizing piecewise and then combining the summaries.
• Parameters: With proper instruction, your current parameters should work.

This approach should help yield a proper summary rather than an echo of the input text.