- In a not-so-distant future, every app will have an invisible layer of artificial intelligence.
- Developers are linking arms with OpenAI to access the large language models powering their apps.
OpenAI's artificial intelligence technology is now baked into hundreds of apps, making it one of the key levers that help companies and their developers be more productive.
The latest company utilizing the company's large language models is Superhuman, an email app with a cultish following. It just released a writing assistant built on OpenAI's chatbot ChatGPT, that can draft emails from short prompts, clean up spelling and grammar, and even rewrite messages in the sender's "voice," based on their past emails. Superhuman started testing the offering last week, and it's now available to a private beta group.
"I think we're on the cusp of a huge transformation," Rahul Vohra, the founder and CEO of Superhuman, told Insider. "And in terms of how I would measure this, this is — without exaggeration — on par with the smartphone and the internet itself. We're going to see everything change."
The collaboration hints at a not-too-distant future where much of the software people use has an invisible layer of machine intelligence. Some companies are already doing just that. The language learning app Duolingo lets users role-play with a chatbot. The search engine Perplexity responds to questions in plain language, so the user doesn't have to open eight links to find an answer. The startup Jasper whips up marketing copy in a flash. Each relies on vendors including OpenAI to power its capabilities.
OpenAI's flagship products are large language models, which are, in simplest terms, systems that can learn from processing huge amounts of data and respond to basic prompts. In 2020, the company released its first commercial offering: a tool that lets developers bake its models in their own products. There are now apps for wordsmithing emails, replying to customers, and surfing the web, all usings its models. "For the moment, AI products on the market are essentially just white-labeled versions of ChatGPT," the journalist Casey Newton wrote.
OpenAI has its tentacles in some of the most prominent companies to come out of the Valley, ranging from Snap to Instacart to Shopify and charges a fee for its service.
Software developers could build the infrastructure themselves, but doing so can take years of development, a rare source of engineering talent, and many millions of dollars in computing costs. Instead, many rent a model from a third-party provider, just like they pay Amazon Web Services to store data and Stripe to process payments.
That's what employees of the buzzy workspace app Notion did. Two of the founders, Ivan Zhao and Simon Last, slapped together a prototype for a writing assistant that could help people write, brainstorm, edit, and summarize their notes with one keystroke. It released the tool in a private alpha only a few weeks later. The feature came together so fast, according to Akshay Kothari, the third cofounder and chief operating officer, because Notion hadn't done the grunt work of training a large language model itself. It culled tools from third parties such as OpenAI and Anthropic and folded them into Notion's app. This allowed the company to focus on what it does best: building elegant products that delight people.
"We found ourselves lucky to be able to partner with OpenAI," Kothari told Insider, "partly because they showed the world what's possible and maybe sort of inspired it."