Jason Calacanis, the tech company founder, angel investor, and industry commentator revealed last week that OpenAI is building an iOS app for ChatGPT. In the long-running This Week in Startups podcast and YouTube Show, Calacanis said he was invited to the beta, which is currently available to a select few in the Testflight app. He went on to show a few screenshots and how it worked.
Calacanis was particularly interested in the fact that as he was creating queries, the app was automatically categorizing the threads by topic. This feature is also available in the ChatGPT client on the web.
A link to the segment where he and his co-host, Molly Wood, discuss the app is immediately below. Calacanis said:
“What’s interesting about this, is it keeps your threads as you can see. There is a nice history here. And you can start to see how this app could be very powerful over time because you could search your threads.”
Calacanis also said, “Now, it doesn’ have an audio interface yet, except for the one built into iOS, but you could start to imagine this having more and more features that would be very interesting.”
The features he is talking about are those you would typically find in a voice assistant. One of those is search, and he shared a short video using the app, which is included below.
You can see that he employs conversational search to refine his intent. He wanted to know the top restaurants in Yountville, California, and which have the best duck. He envisioned using the app to then book restaurant reservations or check on availability. A quick reference to Google Duplex was mentioned by Wood. Cerence recently demonstrated a similar feature for drivers.
ChatGPT as an Assistant
There are already iOS apps that connect to the GPT-3 API and provide writing assistant features. We featured AI21’s Wordtune iOS app in Synthedia in December and had coverage of ParagraphAI’s launch way back in August.
However, AI writing assistant features are not what we saw in the demonstration. It was the same ChatGPT that is available through a web browser. In your browser, you can easily use ChatGPT as a writing assistant by cutting and pasting the response segments you like into a word processing or publishing application. That is doable on mobile but is not quite as convenient. What Calacanis demonstrates is something else.
The interface on mobile is useful for asking questions and receiving responses. That is one of the primary functions of a voice assistant. Granted, Sam Altman, OpenAI’s CEO, commented in December that search was not necessarily a high priority of the next round of feature enhancements.
An earlier Synthedia post stated, “OpenAI’s Sam Altman indicated in a Tweet that task execution was likely to be the next major area of GTP-3’s development. ChatGTP’s system name is Assistant. Task execution is an obvious evolution of the service.” We are likely to see something more along the lines of Siri features than Google Search in the near term.
There are also technical challenges associated with using OpenAI’s GPT-3 model for search. It does not have a retrieval model architecture. That means it cannot give you source information for your searches today. Connecting it to other retrieval model services could fill this gap, but I am not expecting this soon based on Altman’s comments.
Application or Platform
Finally, there is the question of OpenAI’s long-term business model. It was introduced to the world as a platform with APIs for developers to build applications. Companies such as Jasper AI, Rally, and Microsoft layered end-user application functionality on top of the GPT-3 API.
OpenAI’s Playground can do some of what Jasper AI does in terms of a writing assistant, but it seems more like a demonstration environment than a mass-market application. Then DALL-E came along. It has APIs and an interface for users to engage directly and buy credits for generating images. ChatGPT is looking more and more like an end-user application, and the iOS app is only going to reinforce this direction.
Reports suggest that Jasper AI and maybe even Microsoft with Copilot generated more revenue than OpenAI did in 2022. So, it is unsurprising that OpenAI is experimenting in the application category. In-house applications could soon represent a significant revenue source and provide the company with more direct knowledge of user adoption and preferences.
Long-term, I still expect OpenAI to be platform-first. However, we are likely to see more applications in the interim. Mobile apps are just the start.