The Best Way to Summarize a Paragraph Using GPT-3

ChatGPT Update

ChatGPT was recently released, so I wanted to update this post with valuable new prompts for summarizing. One of the benefits of ChatGPT is that the prompts don’t have to be as complicated as in other GPT-3 models.

As you can see, this prompt is straightforward. The actual prompt text is only two lines and many fewer words than in the past prompt, which was written for the text-davinci-002 model.

Summarize the following:
---
{{COPY&PASTE CONTENT HERE}}
---
BEGIN:

This next prompt is similar but returns the summary in a list format.

Summarize the following in list format:
---
{{COPY&PASTE CONTENT HERE}}
---
BEGIN:
List:
-

Best GPT-3 Summarization prompt

In a recent Reddit thread, a user asked about the best ways to summarize a paragraph using GPT-3. Another user responded with the best prompt I’ve come as far as maximally compressing the information in a short text.

Personally, seeing other people’s prompts has been the best way for me to learn about prompt writing and that’s why I wanted to share this prompt.

We introduce Extreme TLDR generation, a new form of extreme summarization for paragraphs. TLDR generation involves high source compression, removes stop words and summarizes the paragraph whilst retaining meaning. The result is the shortest possible summary that retains all of the original meaning and context of the paragraph.

Example

Paragraph:

INSERT TEXT TO SUMMARIZE HERE

Extreme TLDR:

Compared to similar prompts, this prompt is quite effective at summarizing short texts. However, it should be noted that longer texts may be more difficult for this prompt to handle accurately. With some quick mental math, 100 tokens = 75 words, I’d estimate the maximum size this prompt could possibly work for would be 2000-3000 words, given that you’re using the Davinci model with the max_tokens set to 4096.