- AI Dispatch
- Posts
- The Prompting Technique from MIT to 10X your ChatGPT output
The Prompting Technique from MIT to 10X your ChatGPT output
🚀 Level-up your prompt output with this secret technique
🕒 Estimated Reading Time: ~4 Minutes
Hey Hey AI Soldiers,
AI Dispatch guy here 👋
Ever tried to use ChatGPT to do anything involving big docs? Yup, it isn’t a pretty sight.
In the case of summary, it misses most details and mostly just says this is an article about this, this topic covers this. No actual info.
The folks at MIT, Columbia University, and Salesforce AI came up with a genius prompting technique that uses the power of .. you guessed it … refining it again and again. To get outputs that are not possible to generate with Vanilla prompts.
The topic of our discussion today is the technique in question ‘The Chain of Density’.
🔑 Chain of Density Prompt
CoD (Chain of Density Prompt) works better with GPT4 (Try with Claude 3, Llama 3, or GPT3.5 too).
The main trick is to make a response better without adding extra words to it and avoid making it bloated.
It’s really good for making summary of a large text body but can be used in other ways.
Enough Talk. Let’s see it in action.
This is the result I got with normal prompting vs when I used the Chain of Density Prompt in summarizing an article on Blue Ocean strategy for SMEs and startups.
Vanilla prompt ✨
Chain of Density-4th and 5th version ▶️
Pretty cool, right?
Here is the prompt 👇️
Article: {{article}
You will generate increasingly concise entity-dense summaries of the above article. Repeat the following 2 steps 5 times.
Step 1: Identify 1-3 informative entities (delimited) from the article which are missing from the previously generated summary.
Step 2: Write a new denser summary of identical length which covers every entity and detail from the previous summary plus the missing entities.A missing entity is:
- Relevant: to the main stories.
- Specific: descriptive yet concise (5 words or fewer).
- Novel: not in the previous summary.
- Faithful: present in the article.- Anywhere: located in the article.Guidelines:
- The first summary should be long (5-8 sentences, ~80 words), yet highly non-specific, containing little information beyond the entities marked as missing. Use overly verbose language and fillers (e.g., "this article discusses") to reach ~80 words.
- Make every word count. Rewrite the previous summary to improve flow and make space for additional entities.
- Make space with fusion, compression, and removal of uninformative phrases like "the article discusses".
- The summaries should become highly dense and concise, yet self-contained, e.g., easily understood without the article.
- Missing entities can appear anywhere in the new summary.
- Never drop entities from the previous summary. If space cannot be made, add fewer new entities.Remember: Use the exact same number of words for each summary.Answer in JSON. The JSON should be a list (length 5) of dictionaries whose keys are "missing_entities" and "denser_summary".So, what’s happening above? 🧐
-First, you write a long, wordy summary that doesn't really say much about the article
-Then you pick out some important phrases/details missing from that summary and rewrite it to be more concise while adding in those missing bits
-Repeat that process for a total of 5 times (resulting in 5th different versions) and you get a summary covering all the important entities and events.
A demonstration directly from the study, summaries getting better and denser each time 👇
That being said the last output isn’t always the best one, you have to choose the one that hits the sweet spot.
What else can CoD be used for? 🤔
Now other than summary, it can be used for a whole lot of things.
I used ChatGPT to create a similar prompt but for analyzing legal docs.
Have a look 👇️
Input 📨
Article Analysis: {{legal doc}}
You will generate increasingly concise, entity-dense analyses of the selected legal document above. Follow these 2 steps 5 times.
Step 1: Identify 1-3 significant entities (delimited) from the legal document that are missing from the previously generated analysis.
Step 2: Write a new analysis of the same length as the previous one, incorporating every entity and detail from the previous analysis plus the missing entities.A missing entity is:
- Relevant: Pertinent to the core themes and narratives.
- Specific: Descriptive yet concise (5 words or fewer).
- Novel: Not included in the previous analysis.
- Faithful: Present in the legal document.
- Anywhere: Found anywhere in the legal document.Guidelines:
- The first analysis should be lengthy (around 200 words), yet relatively non-specific, focusing minimally beyond the entities marked as missing. Use verbose language and fillers (e.g., "this document explores") to achieve the word count.
- Prioritize efficiency in word use. Redraft the previous analysis to enhance flow and incorporate additional entities.
- Create space by merging ideas, compressing details, and eliminating non-informative phrases.
- The analyses should evolve into highly dense and concise, yet complete and understandable without needing the literature.
- Missing entities can be integrated anywhere in the new analysis.
- Ensure all entities from earlier analyses are retained. If space is limited, introduce fewer new entities.I analyzed a rent agreement example between a restaurant and a corporation with it. ⬇️
Output: 📤️
Other use cases I can think of writing - no fluff essays, studying long PDFs and anything that involves a big chunk of text.
Tools of The Week 🆕
WHAT HAVE YOU LEARNED TODAY? 📚️
What Chain of Density (Cod) prompt is.
The full prompt and the process in the background.
Other possible use cases.
Okay, that’s for today. Let me know in the reply if you plan to try this prompt and for which use case!
AI Dispatch Guy – over and out. 🫡
Your feedback will help me write better content moving forward.