Chris Do on making your own multiverse by training a chatbot clone

On his YouTube channel, Chris Do has been helping over 2.4 million subscribers navigate the pitfalls of creative work. A former professor of Sequential Design, he now runs workshops offering guidance on pricing your work, building a personal brand, and improving your portfolio. 

And as an Emmy award-winning designer and founder/CEO of The Futur, he’s achieved massive success doing what he loves. Now he wants to scale his coaching business to help a billion others do the same. The problem? He’s just one person. For now at least.

A couple years ago, Do got an intriguing invitation from a former student, Sho Rust, CEO of SHO.ai, an AI platform designed to customize content to the user’s style. Within minutes of seeing Rust’s demo of AI tools that can build websites in a matter of seconds, Do saw the potential.

Inspired to start exploring new AI tools, he accepted Rust’s invitation to collaborate with the SHO.ai team and train his own chatbot. They called it DoBot. Over the past two years, they’ve been iterating and refining DoBot, enabling Do to offer one-on-one virtual coaching through this AI-driven digital clone. Students can converse with a virtual Chris Do and get his expertise on branding, design, and business strategies for creatives without Do actually being at his keyboard. This has effectively multiplied his ability to reach and teach more students than ever before.

"If I ever want to have a shot at reaching one in eight people on planet Earth with the educational content we create, I can't be sitting around making small decisions, taking small actions," says Do. "Working with Sho is helping me further my goals, which I thought were unreachable.”

We spoke with Do to find out how it feels to make your own multiverse—and what creatives can learn from integrating AI tools into their workflow.

How do you view the potential of AI?
I'm an AI optimist. We should live in the realm of ideas and possibilities. And AI levels the playing field for people who don't know how to code or how to draw, for people who may have some struggles with writing in a language that's not their native language. 

I can use AI-assisted technologies like DoBot to help me clarify and rewrite some of the content for LinkedIn and Instagram. I can write an ugly first draft, then have it clean up the structure while not embellishing what I wrote. It’s a happy middle ground between stream of consciousness and AI-generated text. It can also summarize the transcript of our long-form YouTube content and write titles that are more accurate summations of the content, while designing the title to be optimized for search and discoverability

What did the process of training DoBot involve?
We input very specific materials from the course that Sho took, along with some other materials. We went through a couple of iterations, and it started to sound and talk like me. 

One of the big breakthroughs was I said, 'Sho, when I type in something, DoBot gives me answers. [I don’t do that.] I ask more questions and help the person to arrive at something.' He responded, 'Okay, let’s discuss this.' We recorded our entire conversation over Zoom. Afterward, he said, 'That’s all I need, Chris.' I asked, ‘What do you mean?’ He explained, ‘Well, I'll take this transcript, filter it, and feed it to the robot.’ So we tested again, and now DoBot doesn't give answers anymore. It just gives you more questions. It's so much like me that I'm annoyed by it. [laughs]

What will DoBot enable you to do that you can’t do on your own? 
DoBot is the ultimate teaching assistant. It’s available 24 hours a day, 7 days a week. It knows everything it’s been trained on. Never forgets. Has infinite patience, and can talk to each student for as long as they want. When we set out, I didn't know how to achieve our mission. Utilizing LLM and integrating it with our brand of teaching and resources allows us to be one step closer to realizing our goal.

One thing we haven’t been able to do is certify people who have taken the course. I don't have enough resources to test you to make sure you understand. So we told DoBot, 'Based on the materials we're teaching, design a series of multiple-choice questions. Make sure the answers are reasonable. Don't make up any answers.' A couple of stabs at it later, we have a quiz that's baked into the modules that we teach.

What surprised you most about how people have used DoBot?
Sometimes people talk to DoBot and [when] they describe it to me later on, they’re in tears [as if it were] better than therapy. So this is an unexpected side benefit of creating a robot that understands the way that you want to talk to people, and them trusting to have real conversations with it. 

Initially, we didn’t allow it to have memory. We didn't want it to remember all your conversations for privacy reasons, and also because sometimes the data starts to get a little muddy. Now, we allow people to turn on different threaded conversations, so they don't have to go back and tell it that thing again. We're doing it very carefully, in measured steps. It's very exciting.

There are uses I didn't imagine people would have. What some people have done is, ahead of a sales call with a client, they really want to calm their nerves and [get] a sense of the shape of the conversation. They ask it to enter roleplay mode. They tell a little bit about themselves. Then it plays one part or the other, so they can see where they went right or where they went wrong. They can ask DoBot to analyze the conversation to say, 'Give me three things I need to improve on' or 'What three things do I need to keep doing?' And it can do that for you. It's shocking. 

Have you ever worried that DoBot would represent you in a way you wouldn't agree with? 
I don't worry about it. Unlike, say, the GBTs, Sho has a lot of control over how much we weight the preference of one set of data versus another. So we heavily weight the course materials versus what it's going to find on the internet. When we found it hallucinating or making up data, we redirected it and said, 'If you don't know, just say, I don't know. Don't just make up stuff.'

So people are asking really strange questions about my life. And it doesn't know that because I haven't told it that. So it just started making up stories. It kind of sounds like it could be real, but that's not it at all. So Sho had to put up those guardrails. But what I think is interesting is, it's not my temperament. People will say, 'I'm having great conversations with DoBot.' Well, real Chris Do would only type in two or three-word answers and say, 'I gotta go,' but it will sit there and talk to you if you want for hours. It will wait for you to formulate a question. 

So it’s becoming the most patient version of you? 
Yeah. Isn't that really cool? The thing we keep thinking about now is adding in accountability. [For example, DoBot might ask], 'What are your goals for this week? What's your next step? What are you gonna do today or tomorrow?' Is it okay if I follow up with you?' Now we're gonna add that into it. So we're having the knowledge of the instructor, backed by the knowledge of the internet, infinite patience, a kind of genteel demeanor, the ability to explain it multiple ways, the ability to test you, and hopefully, very soon, test you conversationally to assess you and guide you through that process and hold you accountable if you want. 

Do you think your experience as a teacher has improved your skills as a prompt engineer? 
I think so. Oftentimes, we're not as clear as we need to be in our communications. This is the lost art of rhetoric and being able to have a persuasive discussion or argument—having to articulate to another human being what it is that you want from them. 

[With AI], I think people are like, 'Write me a cool story' as if it were a genie that can read minds. You might get lucky. Then the next 17 times you use it, it's all garbage. It's not that the bot is broken or the AI engine is defective. It's that you're not being specific enough. 

Now that you’ve seen how much of your work DoBot can do, which aspects of your job would you not want to automate? 
I would not want to outsource conversation because I don't know how else I'm going to learn and understand. It's an ongoing question that my wife and I have. She's like, 'You're so busy. It gives me a headache just looking at your schedule.' I say, 'Honey, where do you think the ideas come from?' It doesn't come from me staring at a screen or being in a corner of a room saying, 'I imagine the internet having these kinds of problems. Let me write on that.'

It's really an exchange of dialogue. [Whether it’s a conversation with] someone who's brand new in the industry or a 30-year veteran, they have something to offer you. There's wisdom there. There's a perspective, a way of moving about the world that you can learn a lot from. I don't ever want to give up the ability to have a dialogue with someone because I grow in the exchange, and that's very important to me. 

This interview has been edited and condensed for clarity.