When I first entered the workforce, every business had a server closet where it kept the computers that ran the applications used daily by employees to run the business. Cloud computing changed that, and it is much less common now. But it didn’t stop application development, or innovation, and it certainly wasn’t anti-competitive in any way.
I think about GPT-3 from OpenAI in the same vein. Nothing about GPT-3 makes it difficult to build a startup in the NLP space. In fact, GPT-3 allows you to abstract away the NLP and focus on other parts of your application while treating the NLP as an API integration.
So what should get built? Here are 3 ideas I expect to emerge.
- Business composition tools. Tools like Niches will help users create content of all types in a partially automated fashion, saving time and money. There will be several companies in this space that will be successful, and they will be differentiated by the workflows they offer on top of GPT-3, and the UX of those workflows. They will also focus on integrations. Expect some of the players to integrate with other content creation platforms and view themselves as a content creation tool that sits across your content creation platforms. Expect other players to look at how enterprise tool adoption clusters, and sit at the middle of those common clusters. And finally, expect a few to focus on functional verticals, most likely sales related messaging like cold emails, battle cards, and collateral.
- Real time language tools. The one thing a model like GPT-3 can’t do so well is deal with language that is happening today. What I mean is that all the text for today’s news, for example, hasn’t been trained into GPT-3, so it can’t answer a question about breaking news. Expect to see some startups build news related tools that harness GPT-3 but add to it. These may answer questions, generate content, and more.
- Decision analysis tools. There is an opportunity to use GPT-3 to aid in making business decisions using natural language. These startups will connect to analytics systems, and function sort of like NLIDB — natural language interfaces to databases. But the functionality will go beyond just information retrieval to more complex analysis.
But keep in mind, there are problems building on GPT-3. At the moment, scalability is not automation. Growing requires you to meet with the OpenAI team and regularly explain your use cases, which requires their approval. Over time, other offerings will compete with GPT-3 and that issue will go away, but startups today building on this platform need strong business development and partnership skills to manage the OpenAI relationship.
At PJC, we are big fans of the emerging NLP stack and believe it unlocks a whole new world of startups that previously were difficult to build. If you are working on something in this space, we’d love to talk with you about it.