Today’s article is brought to you by Lemon.io.

Are you tired of the endless search for top engineering talent for your startup?

Look no further than Lemon.io.

Lemon.io provides a trusted source of pre-vetted candidates to make your search easier and faster. They offer tested and verified devs, affordable rates, and a zero-risk replacement guarantee.

With Lemon.io, you can have an engineer start working on your project within a week instead of months, and you don't have to waste your time on candidates that are not qualified.

Request a free quote. Get a match in 48 hours.

Last October, an AI-powered content creation tool called Jasper, was valued at $1.5B.

It was built on top of GPT-3, OpenAI's large language model (LLM) and was making waves as one of the first handful of AI startups to find a practical use case for AI in our day-to-day lives. Since GPT-3 was a natural language AI model, it made sense that the first wave of startups to strike gold would be copywriting companies.

But all of that was about to change.

A month later in November, 2022, OpenAI released ChatGPT, built on top of OpenAI's still unreleased LLM. Everyone and their mom could send the chatbot a prompt and receive AI-generated response quickly.

Within days, there were endless articles and Twitter threads on how you could generate lists of blog ideas, write entire articles, craft the perfect Tweet or whip up a cold email. All things that Jasper was great at. Except ChatGPT was free.

Presumably, OpenAI will eventually paywall ChatGPT and things will stabilize for their partners. But I guarantee you, it was a sobering reminder for the team at Jasper that their core business is built on someone else's land.

I use Jasper as an example, but a similar fate will befall the hundreds of startups that are building the next wave of hot, new, tech startups "powered by AI"…

…unless they find a way to differentiate immediately.

Let’s get into it.

P.S. — Stick around to the end and I’ll break down what Jasper’s been building over the last 3 months to ensure their survival.

Welcome to Making Product Sense
Join thousands of others like you, learning how to build great products and companies from world-class builders.

Subscribe now

The “AI Moat”

Startups that build on top of Large Language Models like GPT-3, send the prompts from their users to the LLM's API and receive a response in return. That response is then served up to their users in the product. So while hundreds of startups are excitedly proclaiming that they’re “powered by AI,” they’re all powered by THE SAME AI.

If two different products sent the same queries to GPT-3, they would likely get very similar results which, unfortunately, means that the AI moat isn't much of a moat after all.

GPT-3 is probably the most well-known model out there thanks to ChatGPT’s meteoric rise to 100,000,000 users in 2 months, but there are a handful of competitors like Google's LaMDA, PaLM, and GLaM, DeepMind's Gopher and Chinchilla, Meta's OPT, Amazon's AlexaTM and Hugging Face's BLOOM. So theoretically, you could choose a different model from a competitor and get a different set of results.

Despite that, it doesn't take a lot of Googling to realize that the list of quality LLM's is short. Which makes total sense, because it takes an enormous amount of money, time and resources to create a model from scratch which naturally disqualifies most players from the get-go.

So at the end of the day, you're either:

“Scrooge McDuck rich” and get to build your own differentiated LLM yourself
OR you build on top of someone else's and you're forced to find a new way to stand out from the competition.

The Holy Trinity of Differentiation

If the APIs serve up the raw content from the prompts you send it, then the options for a unique product really boil down to three things:

Prompt Engineering: The way you fetch the information
Response Engineering: The way you remix the information
Feature Engineering: The way you use the information

Ideally, you're doing all three.

Here's why:

In the long-run, possibly even after layering on the Holy Trinity of Differentiation, value will flow to either the AI platforms (ie. OpenAI) or the companies with the greatest distribution (ie. Microsoft). And a lot of times, those will end up being the same (ie. Google). All they have to do is roll up the best ideas from the scrappy startups and offer the exact same value to their broader user base, extinguishing the little guys.

Tyler Tringas @tylertringas

1/ this market map will become an iconic case study of "was a feature not a product" with all of the value ultimately being split between the actual core AI platforms in the top left corner and incumbent software tools with distribution (mostly not shown) who will add AI features

Copywriting is a perfect example. Jasper gives you a powerful AI assistant who helps you write copy. So when Microsoft builds the same set of features into their Office 365 suite, suddenly Word, Excel, Powerpoint and the literal dozens of other products that they offer to their user base of over a billion people is supercharged with the power of OpenAI. Except, as they did with “The New Bing,” it'll likely be an even newer and more powerful version than what Jasper has access to.

Checkmate.

3 Paths for AI Startups

I don't enjoy being all doom and gloom. I really just want to paint a realistic picture of the landscape of AI startups so we can be more strategic in how we think about the opportunities available.

I THINK it really comes down to a few options:

Build your own AI model with a unique data set and offer an API for others to build on top of, thus establishing yourself as the invaluable "picks and shovels."
Build in a niche that isn't currently dominated by an incumbent with outsized distribution, build brand recognition within that niche (and thus capture a distribution advantage), and pray that Google or Microsoft doesn't decide they want a piece of the pie.
Move very quickly to build an attractive suite of unique features with the Holy Trinity of Differentiation in such a way that it would be easier for the incumbents to acquire you than to build the features themselves and exit for a healthy sum.

If I missed something, drop a comment below. I really am curious to see which strategies are available in this space.

Jasper seems to be pursuing #3 with a light dusting of #2. Let’s break it down.

How Jasper Wins

Like any good company, Jasper is working hard to shore up its foundation. They understand the inherent instability of a product that's drawing water from the same well as another competitor and they're making intentional strides toward building (and educating about) their moat.

Here's an excerpt from their recent "Jasper for Business" announcement they released last Friday.

"We’re often asked what’s behind Jasper’s AI outputs. Or how Jasper is different from just using an LLM. The secret to Jasper sits in our AI Engine.
When a creator submits a prompt, Jasper’s AI Engine selects the right model for that job across a collection of LLMs, we then overlay that with reliable citations and recent searches from Google, infuse the output with what you’ve taught us about your brand and products, and release a result that is tailored, current, and optimized for marketing.
This new model is enriched beyond that which you’d get from tapping straight into an LLM. And, with the latest releases, it can accompany you wherever you create online."
—Jasper.ai, ”Announcing Jasper for Business”

The Jasper team addressed the concern head-on and, in my opinion, they nailed it. Let's break down their response in relation to the Holy Trinity of Differentiation:

Prompt Engineering: "When a creator submits a prompt, Jasper’s AI Engine selects the right model for that job across a collection of LLMs..."
Response engineering: "...infuse the output with what you’ve taught us about your brand and products, and release a result that is tailored, current, and optimized for marketing."
Feature Engineering: "And, with the latest releases, it can accompany you wherever you create online."

It's clear that the talented team over at Jasper has been thinking a lot about how to navigate the turbulent waters of the AI landscape and they're executing on it incredibly quickly. Time will tell whether or not they get snatched up by one of the incumbents who'd rather buy their feature set than build it or if they have carved out enough of a niche that they'll be able to build their own empire.

Thank you for reading :) This post is public and free. If you found it interesting, please share!

Share

That’s all for this one - I’ll catch ya next week.

—Jacob ✌️

Subscribe now

❤️ Smash that heart!

If you enjoyed this article, smash that heart icon to show some love! 🙏

Making Product Sense

Jasper.ai | The AI Startup's Survival Guide