ChatGPT cheat sheet by OpenAI

learn how to get better results from GPTs

Author

Anita Kirkovska
June 09, 2023

Hi folks!👋🏻 This is The Prompt! We're your go-to source for all things AI.

OpenAI released its own cheat sheet & Google is working on a new AI model to help you create consistent brand assets.

Let's talk about it

FEATURED

ChatGPT cheat sheet by OpenAI 🔥

OpenAI released its own cheat sheet on how to get better results from its models.

Here’s a quick summary for you:

1. Write clear instructions

Yes, this is obvious, yet somehow most people miss it. To get really good results your should specify your needs to GPTs and possibly include length, complexity, and format, to get desired outputs. They outlined 6 tactics on how to get better at simple yet effective instructions and you can read about them here.

2. Provide reference text

GPTs might create false responses for complex topics or citations, but supplying them with reference text and instructing them to use it for answers can reduce this risk. Here are 2 tactics.

3. Split complex tasks into simple ones

Complex tasks tend to have higher error rates than simpler tasks. For example: For dialogue applications that require very long conversations, summarize or filter previous dialogue to get better results. Here are 2 more tactics on how to effectively split complex tasks.

4. Give GPTs time to “think”

GPTs, like humans, are more prone to errors when rushed, so instructing them to methodically work out a solution, employ an inner monologue, or review past attempts can improve their reasoning and accuracy. Read about the tactics here.

5. Use external tools

Compensate for the weaknesses of GPTs by feeding them the outputs of other tools. For example, a text retrieval system can tell GPTs about relevant documents. Embeddings are very important here. Here are 2 tactics on how to do this.

6. Test changes systematically

Check if changes improve GPT's performance by testing it broadly, not just on isolated examples. Here’s one tactic for testing your results.

NEW TECH

StyleDrop by Google

StyleDrop is a text-to-image model that can follow a specific style.

But one interesting thing is that StyleDrop is easy to train with your own brand assets and helps you to quickly prototype ideas in your own style.

Imagine providing a set of your branding assets, and then creating more with just a few words.

This is a bit tricky with models like Stable Diffusion and Midjourney, and this might be a move in the right direction for Google.

WHAT ELSE IS GOING ON

🦙 Gen-2 by Runway, is now available to everyone. With Gen-2 you can generate videos using nothing but words.

🏋🏻‍♀️  Zoom has new AI features. ZoomIQ will generate Meeting summaries, thread & email drafts, and meeting catch-ups. It's only available for select plans right now.

👀 AI Spend Set to Hit $1.3 Trillion. According to a Bloomberg report, generative AI is set for huge growth. It’s projected to jump from $40 billion in 2023 to an amazing $1.3 trillion in annual spending by 2032. And the report says that this growth will be led by big companies like Amazon/Google, rather than new innovative startups.

RESOURCES

The best resources we came across lately that will help you become better at writing prompts & building AI apps.

📚 What is Prompt Injection [ quick article ]

👋🏻 Pluginpedia for ChatGPT [ a shared link to search plugins ]

🎥 ControlNet model for QR Code [ Reddit thread ]

TOOLBOX

The latest AI tools to use or get inspiration from.

PROMPT OF THE DAY

TOOL

Midjourney

PROMPT

street style photo of a woman, shot on Kodak gold 200 --style raw --v 5.1

RESULT