If you’re not excited about ChatGPT, then you’re not being creative

9 mins

The hype around AI in general and ChatGPT, in particular, is so intense

that it’s very understandable to assume the hype train is driving

straight toward the trough of disillusionment.

But I want to challenge your instinct toward skepticism: If you’re not

excited about ChatGPT, you’re probably not being creative enough. And if

you are excited about ChatGPT, you’re probably still not excited enough.

In this article, I’m going to make the case for that excitement. My

argument is this: ChatGPT, as it stands, is cool but the potential is

most visible when you identify its two primary blockers:

Your creativity and the ecosystem’s integrations.

ChatGPT is a window

The simplicity of the ChatGPT interface is welcoming but its simplicity

makes the whole project easy to underestimate.

Take AI-assisted writing, for example. Many of the first headlines when

ChatGPT debuted were about how we’d either be automating all writing or

about how bad ChatGPT is at writing and how there’s no way we could

automate writers. Both angles miss the point.

The problem is that prompts are hard — both because it takes some

practice to learn how to get ChatGPT to create what you want and because

the most exciting prompts requires creativity.

Nat Eliason provides a good example.

In an article on using ChatGPT as a writing coach

he demonstrates the process of using ChatGPT to help him write a section

of his novel.

The first prompt — “write a vivid description of someone getting out of

bed” — falls flat, but it includes a couple of details he can use in

another draft. Next, he asks ChatGPT to rewrite his attempt at writing

the section.

This isn’t impressive either. What Nat does next though is incredible.

He asks ChatGPT to write the section from scratch like the famous

novelist David Foster Wallace.

Next, he asks ChaptGPT to critique his original draft as if it were

David Foster Wallace:

Nat does a few other cool things, such as asking ChatGPT to increase the

suspense and add more detail, but the point of these examples is to

demonstrate how high the ceiling is if you’re willing to be creative.

You don’t need to be a writer to be excited. The excitement comes from

looking at the early use cases and imagining parallel creativity. For

every use case that emerges, there will likely be a similar David Foster

Wallace technique you can deploy.

I know it’s cliche, but it’s a cliche because it’s true and I’m going

to say it anyway: This is only the beginning.

Patrick McKenzie, better know as Patio11,

writes that

“most people who see ChatGPT and think ‘Huh, neat, but not really more

than a toy’ aren\’t playing forward the 6-18 months it takes scaled

product teams to start integrating LLMs into a few pilot projects and

then 2\~4 product cycles, after which it will be in \*so much.\*”

The David Foster Wallace technique is cool but it’s the result of one

person’s brainstorming. Patio11 goes on to say that “there is an

adoption cycle among developers / product people / companies just like

there is an adoption cycle among consumers.” As builders adopt AI,

implement AI, iterate on the AI products they build, and develop and

share best practices, the possibilities will stretch beyond what we can

realistically imagine.

Integrations will be an exponential multiplier

Andrew Ng, computer scientist and co-founder of Coursera,

compares AI to electricity:

“Just as electricity transformed almost everything 100 years ago,

today I actually have a hard time thinking of an industry that I don’t

think AI will transform in the next several years.” This is another

comparison that builds hype and invites skepticism but it\’s valid not

because it’s literally true but because it’s directionally true.

If you focus only on what you can do within the ChatGPT chat interface,

you\’re limiting your imagination. Like with electricity, your

predictions would be entirely inaccurate if you focused only on powering

lightbulbs.

As Patio11 said above, there’s an adoption cycle amongst developers and

companies just as there is amongst consumers. But the result of that

work won’t just be pure AI apps — there will be a multitude of

integrations small, large, and transformative.

At present, Notion has implemented an AI writing tool,

Quizlet has created an AI tutor,

Shopify has made an AI shopping assistant and

if you’re hooked into the AI community on Twitter, you’ll see amazing

stuff every day, such as people using AI to make instant call notes and

to use documentation to automatically answer instant call notes

These early integrations are exciting, but what’s even

more exciting is the seeds of exponential progress being sewn —

progress that will come from the nature of AI development and the

results of collective creativity.

As Evan Armstrong writes,

AI product development doesn’t play by the same rules as traditional

product development. Traditionally, product development involves a

steady and mostly linear refinement of materials and features. With AI

development, in Evan’s colorful worlds, “AI is a snake eating its own

tail because the various components feed off of each other.” As

companies build integrations and data stores, progress can take

unpredictable leaps forward if, for example, improving training data

also improves access points.

Evan also points out that the AI industry is “rooted in academia,”

which means that “whenever a company launches a new product, they

typically publish an accompanying article in a major scientific journal

outlining the math, data set, and process they used to create the

model.” Advancements in AI have been and will likely remain

extraordinarily accessible to companies and lone hackers alike.

That kind of collective creativity can produce incredible results. At a

large scale, we’ll have companies building AI products that can

integrate with and feed into other AI products; at a smaller scale,

we’ll have examples like the one I cited in the previous section —

using novel techniques, such as calling upon AI David Foster Wallace, to

use existing tools better.

At first glance, this is the kind of creativity that already feels

familiar. In the tech industry, especially, we’ve long shared new tools,

tips, and tricks with each other. But that first glance is missing the

new player: AI. Already, ChatGPT produces novel, unpredictable results.

At scale, across a multitude of integrations, and after different kinds

of training and fine-tuning, no single person will be able to keep up

with all.

As I was writing this article, OpenAI introduced the ability for

developers to integrate ChatGPT and Whisper models via API.

Maybe even more significant is that OpenAI slashed prices with

ChatGPT now costing \$.002/1k tokens (ten times cheaper than GPT-3.5)

and Whisper costing \$0.006 / minute.

As costs drop, momentum builds: More people can try and more companies

can integrate and as everyone experiments, more people can see the

potential and buy in.

Examples are everything

The biggest temper on AI hype is the sheer overload of it all. It’s

already effectively impossible to keep up with all the progress being

made, especially given the fact that every new product and integration

will have undiscovered, creative techniques beyond the use cases even

the creators might have imagined.

My recommendation is to pay attention to the builders and not the

growing cottage industry of hype. You’ll inevitably hear endless stories

about how AI will be our salvation or our doom and about products that

promise more than they deliver.

Watching the builders is more exciting because the practical realities

of building actual products will lead to more innovation than people

prognosticating from the sidelines.

I humbly submit our work here at MermaidChart as an illustrative

example.

At Mermaid Chart, we\’ve made it so you can use OpenAI, at

the click of a button, to either generate a diagram from text or

generate a text that summarizes a Mermaid Chart diagram.

In the below image, our

Projects view, you can press the highlighted button to

open a dialog where you can enter text that the AI can use to generate a

diagram.

And in the next image, you can see the dialog for generating the diagram.

In the next image, you can see the generated diagram.

You can then click the highlighted button in the below image to

generate a summary.

If you were only looking at the flashy images produced by models like

Stable Diffusion, you might reasonably assume AI can’t

produce a precise diagram. But because Mermaid Chart offers

a way to transform text into diagrams, an integration has made

AI-generated diagrams possible.

With new integrations emerging every day, from new companies and

established companies, the possibilities are both endless and

unpredictable. Excitement is warranted, but you’ll want to

ground it in watching the builders, not the tweeters.

Winter isn’t coming

There’s precedent for the collapse of AI hype — there’s even a name for

it: AI winter

The first AI winter occurred between 1974 and 1980. The second AI winter

occurred between 1987 and 1993.But just as important is the AI spring

that happened after each winter.

From 1980 to 1987, after the first winter, expert systems emerged while

Deep Thought and Deep Blue took the chess-playing stage. From 1993 to

2011, after the second winter, intelligent agents emerged while a

Stanford robot drove autonomously for 131 miles and Watson defeated two

Jeopardy! champions.

From 2011 on, we saw the rise of deep learning and big data. And now,

OpenAI, ChatGPT, and the veritable Cambrian explosion of ChatGPT

integrations.

Even now, skepticism still isn’t entirely unfounded. You could point to

salient counterexamples, such as the continually hyped and punted era of

autonomous cars, or to the

stat that

85% of analytics, AI, and big data projects fail. But you could also

point to companies like UiPath, which has made billions producing

robotic process automation software informed by ML and AI, and

scientific advancements like

Alphafold,

which can predict a protein’s shape better than biologists can.

The failure of AI is possible, but so is its success.

And that’s ultimately what excites me the most. It’s not that there

aren’t problems with AI. It’s that even the problems are

exciting. When I saw the screenshots of the Bing AI chatbot going

around — you know, like the one where the bot said it wanted to be alive) —

I felt a mixture of things.

A little Eliezer Yudkowsky-inspired

concern is

probably warranted, but so is a little laughter and amusement. What I

come back to though is the fact that AI did something we couldn’t

predict. Given the creativity of human inputs and the extensibility of

product integrations, the only thing we can reasonably predict is that

we’ll be surprised.

Author
Knut Sveidqvist
Creator of MermaidJS and founder of Mermaid Chart