The Curious Companion: Ep. 11 – WTF is a Custom GPT?​

Curious Reader!

Welcome to this week’s ChatGPT Curious companion newsletter.

What you came for is below, and you can CLICK HERE to listen to the episode if you decide you prefer earbuds to eyeballs.

Happy reading!

This episode is all about custom GPTs: what they are, what they can do, and how to “build” one. If you’ve ever wished you could clone yourself, this is basically how you do it. In this episode I share how I created my own ChatGP-TA for the Instagram Intensive, the files and instructions I used to make it sound like me, and the simple steps you can take to “clone yourself” without writing a single line of code. Along the way we talk vectors, transformers, input quality, and how to set capabilities so your Custom GPT does exactly what you want.

Refresher: What Does GPT Even Mean?

GPT stands for Generative Pretrained Transformer. We covered this in episode one, so if you want a full deep dive, head back there.

• Generative: It doesn’t just look things up, it generates new text.
• Pretrained: Before being released to the public, the model is trained on massive amounts of (let’s be real, mostly illegally obtained) text. It “learns” the patterns of how words connect.
• Transformer: This is the new word for us today. ‘Transformer’ speaks to what the model is actually doing in order to generate its output.

Here’s the ‘transformer’ breakdown:

• Natural language inputs (what we type into the prompt field) get broken down into tokens (chunks of characters).
• Tokens are represented as vectors (series of numbers).
• The model transforms each vector into another vector that better represents how that “word” relates to the others around it.
  • This is how the word “bank” can mean money in one sentence and riverside in another. The meaning shifts based on context.
• The vector is transformed based on context, and this process repeats across however many transformer layers the model has.
• Finally, the model generates an output by interpreting the transformed vectors and predicting the most likely next token based on probability. (Again, refer back to episode 1.)​

It’s confusing, I know. But when in doubt all I want you to remember is that it’s math. Really complex and cool math.

So… WTF is a Custom GPT?

Now that I’ve broken your brain, let’s talk about Custom GPTs.

A Custom GPT is basically your own version of ChatGPT that works the way you tell it to, based on the instructions you give it.

If you’ve ever said, “I want to clone myself,” this is how you do it.

Now, I titled this episode WTF is a Custom GPT?, not How to Build a Custom GPT. I’ll give you the gist of the build process, but for the step-by-step deep dive, head over to OpenAI’s Help Center FAQ.​

What I Built: ChatGP-TA

Last week, I built a custom GPT for my Instagram Intensive (my 6-week online group coaching program for health and fitness pros who want to learn how to use Instagram for online business).

I call it my ChatGP-TA, because I want folks to think of it as a teaching assistant.

Here’s the idea:

• Students in the program can ask the custom GPT any question they’d normally ask me.
• That could be logistics, feedback on content, help brainstorming, or ideas for creating posts.
• The GPT will give them the same answers I would give them.

And here’s the wild part: building it required zero coding. That’s truly the magic of LLMs ( large language models) like ChatGPT. You can just type or speak the instructions, no special knowledge needed.

The Build Process

OpenAI makes it easy through their GPT Builder, which has two modes:

• Create: you chat with it and it walks you through building.
• Configure: you manually input instructions, files, and details.

The quality of your Custom GPT is fully determined by the inputs you give it. Better input = better output.

For my Intensive, I teach entirely from in-depth outlines. Everything I say on a call is written in those outlines. So when building the Custom GPT, I:

• Uploaded all the call outlines.
• Uploaded the sales page copy (so it has all the logistics in one place).
• Uploaded sales emails (to capture how I teach, what I value, and what I emphasize about the Intensive).

Worth noting: I didn’t actually use the Builder for my custom GPT. Instead, I started with a Project (we talked about Projects back in Episode 6), uploaded all those documents there, and then asked ChatGPT for help creating a Custom GPT. Pro tip: always ask ChatGPT how to use ChatGPT.

Keys to a Good Custom GPT

Zoomed out, the goal is to upload files that do two things:

1. Teach it the content → upload documents that contain the actual material.
2. Teach it you → upload documents that capture your voice, values, and approach.

A super-hack for this second piece: Have ChatGPT interview you. Let it generate what it thinks you would say, then correct it. Put all of that into a single “manifesto” file, and upload it.

The other half of the equation is instructions.

• If you always want it to say something a certain way, program that into the instructions.
• If you always want it to reference a specific file, program that too.
• Tone, voice, guardrails, they all go here.

Best practice? Ask ChatGPT to write the instructions for you. And for the advanced folks: store them as Markdown files. They’re faster for ChatGPT to read.

Capabilities and Settings

The final step is setting what your custom GPT can do.

• Can it search the internet?
• Can people upload files?
• These are simple toggles, and you can ask ChatGPT for help deciding what to toggle on or off.

Once it’s ready, you publish it and share the link. I’m gonna suggest that you test it first though, and ask it questions so you can grade the responses and tweak the reference files or instructions as needed.

An example from my test results: I sent my link to my brother, and when he asked it my favorite color (of course he would ask some random shit like that), it responded: “I wasn’t able to find anything in the Intensive resources about Shanté’s favorite color. If you really need to know, the best path is to email the course support team and ask directly.” So I went back and added an “about me” file.

Summary Time

• A Custom GPT is your own version of ChatGPT, built with the instructions and files you give it.
• To make it good, give it strong context (files) and strong instructions.
• Use cases:
  • Cloning yourself
  • Teaching assistant (like my ChatGP-TA)
  • External-facing situations where you want to give other people access to it

For internal use, I still think Projects with project-only memory are the better tool. But custom GPTs are dope for external-facing situations.

Important note: Custom GPTs don’t save individual conversations. If you want continuity, copy your conversations elsewhere or use a Project.

How I Used ChatGPT This Week

Each episode I include a section where I briefly discuss how I used ChatGPT that day/week.

This entire episode has been one long example of how I used ChatGPT, so this week I’d like to highlight how someone from the Curious community (legit just made that name up) has been using ChatGPT.

This week I’m highlighting super dope homie, Kate, a PhD and overall very curious human, who has been using both ChatGPT and Claude.

• For her full-time job as a professor, she uses Claude:
  • Writing tons of curriculum (not her own IP).
  • Feeding it outlines.
  • Asking it to brainstorm.
  • Having it generate documents, templates, guidelines, discussion prompts, and highlight key ideas.
    ​
• For her personal life, she uses ChatGPT:
  • Personal development.
  • Communication with her partner.
    • Attachment style frameworks.
    • Conflict resolution.
    • Understanding each other’s perspectives.

Kate’s usage very much mirrors my response to Rachel’s question in last week’s episode about ChatGPT vs Gemini: “Use whatever you like. At this point, the differences between the models are minimal for most people. It’s a bit like choosing between Google Home and Alexa (I’m team Google btw), and it often just comes down to vibes and which one you like better.”

Alrighty, that’s it for today’s episode. As always, endlessly grateful for you and your curiosity.

Questions, comments, concerns, additions, subtractions, requests? Hit reply or head to the website (​chatgptcurious.com​) and use that contact form. I’d love to hear from you.

Catch you next Thursday.

Maestro out.

AI Disclaimer: In the spirit of transparency (if only we could get that from these tech companies), this email was generated with a very solid alley-oop from ChatGPT. I write super detailed outlines for every podcast episode (proof here), and then use ChatGPT to turn those into succinct, readable recaps that I lightly edit to produce these Curious Companions. Could I “write” it all by hand? Sure. Do I want to? Absolutely not. So instead, I let the robot do the work, so I can focus on the stuff that I actually enjoy doing and you get the content delivered to your digital doorstep, no AirPods required. High fives all around.

Did someone forward you this email?
​Click here to join the ChatGPT Curious newsletter family.​

Stay curious.​