There are general AI companies, and there are narrow AI companies.
Some AI companies train models, while other AI companies wrap models.
General companies will steamroll narrow companies.
Wrapper companies will outmaneuver model companies.
Everyone is doomed.
In 2022, Jasper brought GPT-3 to marketing. Jasper’s pitch was simple: GPT-3 might not be good for much, but it may be good for marketing, and if it’s good here, we’ll get really good returns. They dazzled with demos of cheap market copy. Airbnb and HarperCollins couldn’t consume credits fast enough. They unicorned, raising $125 million. And then they imploded.
After ChatGPT, Jasper’s CEO left, they laid off staff, cut their valuation, and watched revenue fall. Users realized that instead of paying Jasper $80 a month, they could use ChatGPT for free. Not only was ChatGPT more general than Jasper, but it was also just better at Jasper’s core service—marketing. Where Jasper needed careful tuning to produce usable ad copy, users could talk with ChatGPT like a colleague. Jasper was a jack of one trade and a master of none.
AI researchers call this The Bitter Lesson, where author Richard Sutton argues for general algorithms that leverage lots of compute. But, for the first time, Jasper elevated The Bitter Lesson from research advice to a corporate parable: general AI products are often better at all tasks, not just capable of many tasks.
Today, there’s Cursor for coding, Harvey for law, Perplexity for search, Grammarly for writing, Palantir for defense, and dozens of AI healthcare companies. If a GPT-3 → GPT-3.5 advancement killed off Jasper, what might happen with a GPT-4 → GPT-5 advancement? Big lab staff will happily tell you that they’re all doomed.
Here’s Leopold’s take:
I'm so bearish on the wrapper companies because they're betting on stagnation. They're betting that you have these intermediate models and it takes so much schlep to integrate them. I'm really bearish because we're just going to sonic boom you.