Chain of Density Prompt with ChatGPT

Mastering the Chain of Density for Effective Summarisation

In this video, I introduce the concept of “chain of density.” This is a structured technique that helps transform a lengthy transcript—like a YouTube video, podcast, or even an academic paper—into a series of progressively denser summaries. Instead of a single summary, I refine the content multiple times, making it increasingly concise and impactful.

This method is not about reducing quality—it’s about focusing on the key entities and core concepts, stripping away unnecessary filler words, and honing in on the most valuable points. Whether it’s for blog posts, YouTube descriptions, or executive briefings, “chain of density” ensures that each iteration of your summary becomes richer in information while staying accessible for the reader. This process is particularly helpful for those of you trying to create attention-grabbing and informative content summaries that really resonate.

I also provide some practical examples using ChatGPT to generate summaries that balance readability with substance. My goal is to help you make your content engaging, easy to digest, and packed with value—perfect for SEO optimisation and keeping your audience hooked. If you’re interested in tools and techniques that can level up your summarisation skills and boost your content’s effectiveness, stay tuned!

Transcript of Chain of Density structured prompt lecture

[00:00:00.00]
Hello, my name is Laurel Papworth, and I thought what we’d do today is have a look at chain of density. It’s where you take a large document. In my case, it’s a transcript for a YouTube video. It’s about 40, 45 minutes long. And then make it denser. So you don’t just say, summarise this document. Chain of density means create a chain of denser and denser summaries. The core concept here is around the fact that a GPT, or the T on GPT is transformer. We have to take a word soup, and then we have to make sure that the AI goes into that word soup, that document, and extracts out all of the core concepts. I do it for my podcast. You can do it for an academic paper, a executive briefing, a news release, anything of that style. So let’s get started.

[00:01:03.29]
I’m going to type in here, please summarise this YouTube video lecture transcript. I’m going to specify in five sentences or less. No, we’ll just say five sentences. And then I put the transcript in, and then I hit return. So it’s created me a paragraph. In this lecture, Laurel Papworth explains, and there’s quite a lot of phaff in here, so this stuff is not helpful. What we want in chain of density is we want to get this denser and denser. We wanted to have more impact and have more key elements. Don’t worry about what the lectures about. It’s how to use about three different ways to use ChatGPT. But for the purposes of this exercise, I just need to do any text or any transcript. I now want to use chain of density and make this denser. And I’ve asked it sometimes to do this and it knows what it is, sometimes it doesn’t. It’s okay. It’s a fairly new style of prompt, and it’s only just been ingested and not in all the models. Although I’m in 4O Omni, let’s have a look. So we can use that or not, but I’ll say, please regenerate this summary with more entities or key points, but they’re called entities.

[00:02:50.19]
But stay within five sentences. So I don’t want a bigger summary. I want a denser one. Let’s go down. And it got rid of those first… How many sentences? 1, 2, 3, 4, 5, 6, 7, 8, 9. It got rid of those nine, 1, 2, 3, And got closer to three or four words. So okay, let’s make it denser again, but still five, five sentences, so include more key points. Stay readable. That’s a problem with density. It can become unreadable. Now, we write for non-technical people. Stay dense but simple. Five sentences only. Really, I just want to make it a bit more readable. And that now is something I’m going to add to my YouTube description field. I’m going to put it on my blog post as the summary on the blog post. Make sure that it’s a key a, I guess, summary. Yeah, of course, it’s a dense summary. So I hope you can see. I didn’t spend a lot of time reading through these because I know how they work, but you can give it a go yourself. There’s lots of filler words in here based on a brief description. We don’t need that.

[00:04:33.24]
We can just say brief description. I wanted to get away from the weasel words and the filler words and simply get into a very clear explanation. And then I allowed it a little bit more, I guess, readability for anything missing or unclear. That wouldn’t be in a super dense summary. The other thing I would say is if you actually ask it, and I would do this at the end so you’re not making up your other prompts, please use chain of density and make this denser. If you have any questions, Please let me know. I want to see if it knows what chain of density is. It’s just gone ahead and made it denser again. I actually like this one the best because it’s highlighting some of the key points. It seems to know exactly what the high points are of the presentations, which is good.

[00:05:35.11]
I hope you found that interesting. We went through chain of density and making sure that summaries get denser and denser and full of entities or the core concepts. And it’s a really useful tool, really useful structured prompt. My name is Laurel Papworth. If you’d like to see the resources and the links for the academic papers on this topic, please go to laurelpapworth.com. You’ll also find my courses and keynote presentations and things like that there. And otherwise, I will see you in the next lecture video. Thank you.

Resources for Chain of Density summaries

“From Sparse to Dense: GPT-4 Summarization with Chain of Density Prompting” by Griffin Adams et al. (2023)

• Abstract: This paper introduces the CoD prompting method, demonstrating how GPT-4 can generate increasingly dense summaries by iteratively adding missing salient entities. The study includes human preference evaluations and provides annotated CoD summaries for further research.
• Read the full paper

“GPT-4 Summarization with Chain of Density Prompting” presented at the NewSum 2023 Workshop

• Abstract: This work explores the application of CoD prompting in GPT-4 for summarization tasks, highlighting its effectiveness in producing concise yet informative summaries.
• Access the workshop paper

“Unlocking GPT-4 Summarization with Chain of Density Prompting” by KDNuggets

• Overview: This article provides a practical guide on implementing CoD prompting with GPT-4, including step-by-step instructions and insights into its benefits for summarization tasks.
• Read the article

“Chain of Density Prompting” by Zilvinas Medelis

• Overview: This resource offers an in-depth explanation of the CoD technique, discussing its methodology and applications in generating high-quality, dense summaries.
• Explore the resource

“Chain of Density Prompting Implementation” by Richard Wong

• Overview: This GitHub repository provides an implementation of the CoD summarization approach, allowing users to experiment with the technique in their own projects.
• Access the repository

Thank you to Happyscribe for AI transcription and my mate ChatGPT for helping with resources and summarising my off the cuff transcript/video!