- Inclined.ai
- Posts
- 🤖 ChatGPT Needs Better Prompting
What's up? You're reading Inclined AI. If you’re reading this, help me. I’m stuck in my new Apple Vision Pro headset.
Here's what we’ve got our eye on:
OpenAI spills the beans on what makes a good ChatGPT prompt
Apple steers clear of AI hype during their conference
Learn to embrace ChatGPT for students
Marc Andreessen shares his thoughts on AI
OPENAI THINKS YOU NEED HELP WRITING PROMPTS
You expect the best from ChatGPT. It’s only natural to think the AI is malfunctioning, but OpenAI feels the loss in performance is coming from the other end.
That’s right. The company argues poor prompts are causing ChatGPT to suck. It’s not the model; it’s you.
To remedy this failure, they offer six strategies to improve your prompt engineering.
1/ Write clear instructions
That feels obvious, but people get this confused. Give the chatbot more to work with and talk like you’re working with a toddler.
Provide examples, give them personas, and add the steps necessary to complete the task. All that helps
2/ Provide reference text
ChatGPT will make sh*t up if you let it. Ground their answers in references. Give it the text you want it to cite.
Pretend they’re a student that needs a study guide for an exam. The answers are more likely to be accurate, then, right?
3/ Split complex tasks into simpler tasks
Patience is a virtue. Complicated tasks like coding need multiple prompts to get the most out of ChatGPT. Split the big goal into smaller concepts.
Summarize the flow as you go to keep ChatGPT’s context window engaged so it doesn’t miss details it may forget to reference.
4/ Give GPTs time to “think”
This one is simple. Ask ChatGPT to explain its reasoning and encourage it to work out its solution.
You might remember this from other pieces we wrote. That framing slows ChatGPT down and makes it compile the entire process with more logic.
5/ Use external tools
OpenAI is admitting weakness here. These Large Language Models cannot do everything independently, but other tools can help empower them to complete specific tasks.
Wolfram helps ChatGPT solve math problems; other plugins help the model read files. Mixing and matching the best tools for the job makes everything more manageable.
6/ Test changes systemically
I won’t lie; this is the hardest for anyone to do. OpenAI’s advice encourages you to increase your sample size when testing a new prompt.
It’s OpenAI’s response to prompt drift, the idea that a prompt becomes weaker over time. The issue is that to test the systematic benefits of a prompt, you need time or a computer science degree.
Otherwise, you’re better off trying prompts till they work.
The full write-up is here; you can even check out the OpenAI Cookbook, which includes resources for prompts & tools.
Regardless, the fact their team pushed out this guide shows their content team is catching up to the demands of their users. We all know we can improve at talking with ChatGPT and other models.
The best teachers are the ones who built it and use it every day.
You’re staring at the drab layout of ChatGPT, begging your mind to devise a clever prompt. Why does this keep happening?
You need GasbyAI—an AI personal assistant that looks better and works better than plain ole’ ChatGPT.
GasbyAI has an easy-to-use interface, robust prompt library, and unique character options to help you get the right results from your written results. Plus, it’s multimodal, so you can transcribe audio and feed it pictures to your heart’s content.
But the fun doesn’t stop there. Their tool reads content from URLs (even YouTube), lets you upload PDFs and DocX files, search content from the web, and can even extract keywords.
All you need is an OpenAI API key, and you’re off to the races.
In a literal instant—no login required—you get access to over 600 premade prompts and 170 AI personas optimized for your use cases. You don’t need to struggle in the sea of possibility that is ChatGPT.
Quick Nuggets
🍎 Tim Cook admits that he uses ChatGPT and that the company is watching its progress
🦠 A new deep-learning model hopes to address public health concerns and vaccine hesitancy
📝 Encourage your students to try ChatGPT the logic is that they’ll find they don’t need it
📓 AI threatens newsrooms, and the leaders in the industry need to respond
☝️ This is why CNET is revising its AI policy and updating past stories
🎵 The AI era of music: will it suck?
🔒 LlamaIndex wants to help you add private data to large language models
👀 Marc Andreessen, the founder of Netscape and co-head of a16z, writes his view of AI
🍿 Watch these 5 films to learn more about the themes that surround the future of AI
📱 Apple’s WWDC event steers away from the AI hype and focuses on the mission at hand
🔥 Fresh Products
Sivi AI - is a text-to-visual design generator (link)
Argil - adds AI automation to their no-code platform (link)
blobr - helps you chat w/ any API using ChatGPT (link)
Trolly AI - craft SEO-optimized articles 2X faster than normal (link)
Beam - updates their MacOS access to ChatGPT (link)
Dante AI - GPT models trained on your private data (link)
Local AI - now lets you run models on Mac M1s (link)
Maya - plan trips from start to finish (link)
HeyGen - adds Discord bot w/ talking photo (link)
Postys - buy a poster you generate w/ AI (link)
Good Content, Meowy Night
Just one of the few cat-themed remasters in the course, but it’s my favorite. The cat gives me major Nyan energy.
That’s it for today. I hope you enjoyed the latest edition of inclined.ai - Davis.
Check Out Our Sunday Edition
You can dive into more AI news and topics with us every week by subscribing to our premium edition.
We’ve written about the following:
If you’re not already subscribed, that’s okay. We’re offering a free 7-day trial so that you can read this one. That’s how excited I am to post it.