1. Building and deploying custom GPTs
Welcome back!
2. Custom GPTs
While custom instructions let you set some preferences, custom GPTs provide even more flexibility. They act as specialized versions of ChatGPT designed for particular use cases.
Some of the best GPTs include:
Canva — Design presentations or social posts
Consensus — Search and synthesize results from 200M academic papers
Khan Academy’s Code Tutor — Expand your coding skills
3. Custom GPTs
And they span the domains of writing, research, programming, education, and lifestyle.
4. Example without custom GPTs
Let's start with an example.
“Tell me about the recent advancements in generative AI.” ChatGPT provides an overview, but it can be fallible and hallucinate. How can we ensure trustworthiness while accessing real research without trawling through thousands of papers?
5. Example using custom GPTs
Using the Consensus GPT, we can ask the same question:
“Tell me about the recent developments in generative AI.”
6. Example using custom GPTs
The result is transformed, providing explanations and references to the original papers.
Let’s dive into how we can build a Custom GPT of our own.
7. Building a custom GPT
Log into ChatGPT and click on “Explore GPTs” in the top left corner.
8. Building a custom GPT
You’re presented with the GPT store. Here, you can find useful and popular GPTs created by others. This is where ours will soon sit. First things first, click ‘Create’.
9. Building a custom GPT
You’ll be presented with two options to build your GPT: “Create” and “Configure.” “Create” allows you to simplify the build of your GPT by using natural language. You’re conversing with ChatGPT in plain English to determine its instructions and capabilities. The other option, “Configure,” provides more control over how you build your GPT. You’re able to customize each step with precision. Let’s use the “Configure”.
10. Building a custom GPT: Name
We’ll begin by naming our GPT: AI Insight Bot.
11. Building a custom GPT: Knowledge base
We’ll upload Stanford University’s latest AI Index Report to the custom GPT’s knowledge base. This will extend ChatGPT’s knowledge and provide up-to-date information.
12. Building a custom GPT: Description
Description: Ask questions about the latest trends in AI with Stanford's AI Index Report.
We can even add an image to our GPT with DALL•E, ChatGPT’s image generation tool.
13. Building a custom GPT: Instructions
Instead of creating custom instructions that reside inside our personal account, the instructions we create here are tailored to our AI insight bot.
14. Building a custom GPT: Instructions
Remember the PFSET prompt framework we explored in Chapter 2? We’re going to condense this to “PFS,” just adding the persona, framework, and specifications to our GPT prompt.
15. Building a custom GPT: Instructions
Let’s set the persona of our AI Insight bot:
16. Building a custom GPT: Instructions
Define the framework:
17. Building a custom GPT: Instructions
And finally add our specifications for how we want our GPT to respond.
18. Building a custom GPT: Instructions
Now, our instructions are set. We’ve prepared our GPT to provide concise, direct answers. Now let’s add some conversation starters.
19. Building a custom GPT: Conversation starters
Conversation starters are a great way to kick off the conversation with our GPT. Think of them as sparks for discussion, igniting our train of thought.
20. Building a custom GPT: Capabilities
It’s time to upload our report. This will form the foundation of our AI Insight Bot’s knowledge base. Finally, we’ll turn off “Web browsing” and “DALL·E Image Generation.” These tools aren’t necessary for the GPT we’re creating today and could distract our AI Insight Bot when creating an answer.
21. Building a custom GPT: Sharing
Now all that’s left to do is hit “Create.” We can choose to share on the GPT store for public access or save it for private use.
22. Building a custom GPT: Testing
Let’s now put our AI Insight bot to the test. We’ll click on the conversation starter we set: “What are the trends in generative AI investment?”.
Our GPT searches its knowledge base and provides a beautifully succinct and relevant response. It even includes that “AI Fun Fact” we specified in our instructions. Way to go
23. Wrap up
OpenAI has made it easier to build custom GPTs without needing technical skills.
24. Wrap up
As we’ve explored, you can configure GPTs using the most native thing to humans: natural language.
25. Wrap up
GPTs are great for exploring contexts requiring domain-specific knowledge. Today, we created a custom GPT using an AI industry report.
26. Wrap up
Tomorrow, you could create a custom GPT for company FAQs to help employees with questions like “How many days of holiday do I get this year?” The opportunities are endless.
27. Let's practice!
It’s time to dive into the exercises to put your custom GPT knowledge to the test.