Everyone’s using ChatGPT, but almost no one understands how it ACTUALLY works. Read this article to write better prompts than 99% of ChatGPT users:
This post is the result of me diving HOURS deep into machine learning. But I have Zero technical background. (yes, capital Z). I understood this, so you will too.
I’ll teach you:
- How ChatGPT works (kinda)
- How to point it’s attention for better prompts
Let’s dive in:
What’s even ChatGPT?
Don’t even start with: “It’S a nAtuRal LaNguAge pRoCesSing mOdEL”
No one knows what that means.
ChatGPT works with 3 things:
- Natural Language Processing
- Machine Learning
- Deep Learning
Wait. DON’T GO. I won’t get technical. Just know this…
Computers can’t understand words. So the 1st thing ChatGPT does is turning your input into numbers. Characters as numbers… Weird – but cool.
Next, it assigns every subword (3-4 characters) a value. Why? Because not every word is equally important.
Here’s the catch:
Multiple algorithms then determine how important which word in what context. You can pull 4 levers to point the algorithm’s attention:
- Be more specific
- Use more descriptive language
- Break up the prompt
- Use formatting and structure
How to use each ChatGPT Algorithm:
1. Give more specific instructions
This is what most of us do.
Higher specificity = higher importance
Pull this lever by specifying your request. But you kinda knew that already.
Next… Instead of saying;
- “Can you tell me about dogs?”, say “Can you tell me about the different breeds of dogs that are good with children?”
- “What is the best restaurant in town?” say; “What is the best Italian restaurant in town that serves gluten-free option?”
- “How can I improve my writing skills, say; “What are some exercises I can do to improve my ability to write persuasive essays?”
2. Use more descriptive language
The more context you give ChatGPT, the better it is at pointing it’s attention. Present information in this order:
- Start with a clear problem statement
- Give context
- Provide instructions & reqs and watch your results improve.
Also… Instead of saying;
- “I need help with a coding problem,” say; “I’m trying to debug a Python script that isn’t returning the expected output, and I’m not sure where to look for the error.”
- “I want to make pizza,” say; “I’m looking for a recipe for a thin-crust, Neapolitan-style pizza with fresh mozzarella and basil.”
- “I’m feeling sad,” say; “I’m feeling overwhelmed and anxious, and I’m having trouble finding motivation to do anything.”
3. Break up the prompt
Long prompts make it harder for ChatGPT to assess importance. For better output, slice up instructions into shorter sentences. It’s like making a cocktail. Add too much stuff and you’ll end up with cough syrup instead of a drink.
Just keep it simple. Instead of saying;
- “Can you help me plan my vacation to Europe?” break the prompt up into small segment, such as “I’m interested in visiting Italy, France, and Spain,” “I’m traveling with with my family,” and “We have a budget of $5,000 for the trip.”
- “What’s the best way to learn a new language?” breaking the prompt up into small segments, such as “I want to learn Spanish,” “I have a busy schedule and limited time to study,” and “I’ve tried using using apps like Duolingo but haven’t seen much progress.”
- “What’s the best way to get in shape?” break the prompt up to smaller segments, such as “I want to lose weight,” “I don’t have access to gym,” and “I have a knee injury that limits my exercise options.”
4. Use formatting and structure
My calculations were right:
Skimability = readability.
Even for robots. So I will say…
- Use bullet points for structure
- Label sections of the prompt
- Make important stuff bold
(bonus: this also makes your prompt more readable for humans)
- Instead of writing a long paragraph about a problem, use bullet points to break down the key issues and provide a clear structure for the information.
- Use headings to break up different sections of prompt, such as “Background,” “Problem Statement,” and “Desired Outcome.”
- Use bold or italicized text to highlight key terms or phrases that are particularly important relevant to the task at hand.
Lastly, ChatGPT uses algorithms to generate a response in form of numbers, which then are translated back into text. Isn’t it mind-blowing how language is just numbers for AI? I learned a lot going down this rabbit hole and I hope you did too.
How ChatGPT ACTUALLY works:
- Yerba Mate
- Nicolas Cole
- More Magic
Thank you for reading! If you enjoyed this thread:
Follow @hybridcloudtech on Twitter for more content on writing & AI and comment if any of this was new to you!
- Artificial Intelligence AI – Meaning, Benefits, Types, Movie, Future
- Fraud Detection Algorithms Using Machine Learning and AI
- 7 Steps To Get Started in Machine Learning | Data Science
- How AI is Building the Robo-Advisor Landscape