Nick Turley of OpenAI: Transforming ChatGPT into a Next-Gen Operating System

In 2022, Nick Turley became the leader of ChatGPT at OpenAI, tasked with commercializing the organization’s research. His efforts have successfully grown the user base to 800 million weekly active users. Turley’s next goal is ambitious: to redefine ChatGPT as a new type of operating system that supports a variety of third-party applications.

This week, I met with Turley at Fort Mason in San Francisco, a historic military site and the venue for OpenAI’s third annual developer conference, to discuss his vision for ChatGPT’s future. A transcript of our conversation can be found at the end of this article.

To advance ChatGPT into an operating system, Turley is drawing inspiration from web browsers. Over the past decade, browsers have evolved to serve as operating systems — not in the classic sense of macOS or Windows, but as primary environments for user interaction with numerous applications. Turley envisions ChatGPT evolving in a similar way: a platform that changes how people interact with software.

Additionally, there are indications that OpenAI may be working on its own browser. While Turley neither confirms nor denies this, he expresses interest in the potential of browsers. The company is collaborating with Jony Ive and a team of experienced Apple designers to create various hardware devices. Considering these initiatives, it’s evident that a ChatGPT operating system filled with applications could be crucial to OpenAI’s consumer ecosystem.

OpenAI has been pursuing this vision for quite some time. In 2023, the company introduced several initiatives related to an “AI app store,” including ChatGPT plugins and the GPT Store. Although these products did not achieve significant traction, OpenAI appears to have a more effective strategy this time.

The rollout of apps aligns with OpenAI’s goal of positioning ChatGPT as an e-commerce hub. Integrating apps from companies like Expedia, DoorDash, and Uber could boost transaction volume within ChatGPT, enabling OpenAI to facilitate and capture some of that revenue. Having visibility within ChatGPT could present a significant business opportunity for both third-party companies and OpenAI.

This represents an enticing proposition for developers as well. Third-party developers will have access to 800 million ChatGPT users during their daily interactions. Apps will play an integral role in the ChatGPT experience, rather than being stored separately. Moreover, developers will be able to build more interactive experiences within the ChatGPT system, moving beyond simplistic chatbots linked to their data.

However, managing an operating system presents its own challenges, such as determining which applications receive visibility over others. Turley mentions that OpenAI is considering permitting companies to pay for priority placement of their applications in ChatGPT, but the company is actively seeking ways to do this without compromising user experience.

Third-party developers will likely seek access to ChatGPT user data. OpenAI’s guidelines specify that developers should “collect only the necessary minimum data needed to perform the tool’s function,” but the practical implications remain uncertain. Turley notes that OpenAI may introduce new features — such as segmented memory within ChatGPT — to provide users with more nuanced data access control for developers.

A significant point from our discussion was Turley’s perspective on ChatGPT as the “delivery vehicle” for OpenAI’s nonprofit mission: to develop and distribute artificial general intelligence (AGI) — highly autonomous AI systems — in a way that is beneficial to humanity. Although some researchers at OpenAI voice concerns that the consumer side of operations may overshadow nonprofit goals, Turley believes ChatGPT is essential for making AGI accessible to the broader public. Quite a perspective, right?

Below is my conversation with Nick Turley, edited for clarity and conciseness. My questions are in bold, while his responses are not.

Maxwell Zeff: How do you envision ChatGPT as a platform for other companies?

Nick Turley: I envision that in a few years, we’ll look back at ChatGPT and recognize we were operating in a command line era. While it’s incredibly capable, it lacks some essential features.

In traditional operating systems like Mac or Windows, we readily recognize the preference for opening applications over memorizing commands. It’s surprising how we have attracted 800 million weekly active users with our current format. This is a challenging and unconventional category, yet it continues to grow rapidly.

The change we are aiming for in the coming years is for ChatGPT to operate more like an operating system, providing access to various applications. If you want to write, there will be an app for that. If you want to code, there will be an app for that. If you want to interact with products or services, applications will be available.

However, we cannot do everything on our own. We won’t delve into music streaming or replicate Coursera’s catalog. Therefore, establishing partnerships is crucial.

Furthermore, new generations of applications will emerge, which wouldn’t have been possible previously. Just as Uber arose thanks to mobile platforms, I am genuinely excited to discover what similar innovations ChatGPT might enable.

We also want to provide our long-time developers with access to ChatGPT’s 800 million weekly users. By enhancing ChatGPT and building successful businesses on this foundation, we can foster more success stories within the ecosystem.

What sources inspire you when developing ChatGPT?

There isn’t a single source of inspiration. I frequently advise job candidates to think from first principles; merely following a template from Meta or Google can lead to a lack of competition. Regarding ChatGPT or Sora, there is no clear precedent. Hence, we need to draw analogies from various contexts.

I find web browsers intriguing because they have effectively become the operating systems of the last decade. How many people actively use desktop applications anymore? You may access Excel or PowerPoint, but most of our work occurs through browsers in app-like experiences.

Furthermore, I reflect on the early advertising for the [Apple] PowerBook, which drew a parallel to ChatGPT — it was a gadget that people didn’t entirely grasp. The ads exclaimed, “It’s a calculator, it’s an alarm clock.”

While no single reference exists, it’s valuable to learn from history. Over the past decade, there may not be a perfectly analogous example.

You’ve mentioned browsers and devices. How do you intend to expand ChatGPT into those areas?

Here at OpenAI, we aspire to think big. One area we focus on is productivity, which essentially embodies ChatGPT. Yet, there is a multitude of product categories that AI will reshape. Entertainment is one, which is why I am enthusiastic about Sora. Social media, hardware, and internet access points are other domains of interest.

We consider our offerings to be a family of products and applications interconnected by user accounts, personalization, and identity layers. I’m thrilled that we are not restricting ourselves. Even as the ChatGPT company, countless possibilities are at our disposal, but our ambitions go far beyond that.

How do you perceive OpenAI’s consumer business fitting into the nonprofit mission? Some assert that the consumer market funds the mission. What’s your take?

When I joined OpenAI, it was mainly a research lab occasionally launching demos. My initial role was framed as “helping commercialize OpenAI technology”—it was quite open-ended. At that time, products were intended to manifest research in an understandable way. This remains true today with Sora. The most effective way to spark meaningful discussions about transformative technology is to introduce a product.

Subsequently, our viewpoint evolved to recognize the product as more than just an output; it became clear that it could potentially finance our mission, especially given the considerable expenses involved.

After ChatGPT emerged, our approach shifted again. Our mission is to ensure that AGI benefits everyone and reaches all. Acknowledging that AGI may unfold gradually rather than suddenly, we start seeing products as vehicles for delivering the mission. It’s a method to tangibly benefit people.

When looking at the actions of those 800 million users weekly, ChatGPT aids them in achieving their goals. Whether it’s the 89-year-old individual learning to code or parents using ChatGPT to enhance social skills in their autistic children, these stories embody our mission.

I think it’s short-sighted to label the consumer business merely as a funding mechanism; it’s truly a reflection of our mission. This understanding has genuinely evolved since I joined OpenAI.

Let’s discuss the apps announced today. OpenAI has indicated that third-party developers may only access the “minimum amount of data” necessary for their app within ChatGPT. How do you handle user privacy?

From the very start, we will require developers to notify users about the data they’re requesting. We will also ensure that applications go live only when their data requests are reasonable. Our developer guidelines have been published to clarify our privacy policies, preventing surprises when applications are rejected for non-compliance.

Within the next month, we aim to develop mechanisms for users to grant precise data access to developers. I believe Apple has excelled in this regard, allowing users to share data either for a single instance or consistently.

To achieve this effectively, we might need to explore the concept of partitioned memory within ChatGPT, an idea we are actively considering. This excites us because users may want to separate sensitive conversations, such as those about health, from more casual topics like music. We will provide updates soon, as this represents an intertwined research and engineering challenge.

Ultimately, transparency is our priority. We want users to clearly understand what data may be shared with third parties, and these controls will be refined over time.

With DoorDash and Instacart among the companies launching apps in ChatGPT, how will the platform decide which service to recommend for snack orders?

This is a classic question. The simplest approach would be to present both options. If you’ve previously used one, we’ll prioritize that service. If both have been used, we’ll ask for your preference. Over time, our strategies may evolve, possibly prioritizing one app based on service quality.

We collaborate with multiple partners in the same field. The most considerate and effective strategy is to display both applications.

Are you considering allowing companies to pay for preferential placement of their apps in ChatGPT?

This is an area we’re keen to explore with developers. It involves maintaining a delicate balance. We could either establish everything initially and announce it, which might imply we didn’t solicit enough feedback, or we could opt to delay, creating uncertainty but allowing more comprehensive engagement.

We leaned toward the latter because we recognize that building this ecosystem is a long-term project. It necessitates thoughtful considerations about fair distribution mechanisms.

Ultimately, we aim to deliver an outstanding user experience. If this approach resulted in showcasing irrelevant applications to users, it would be far from ideal. However, if it allows for the visibility of genuinely serious applications ready to invest in exposure, it could be advantageous. Currently, we don’t have a definitive stance, but this has definitely been discussed with many partners.