OpenAI is split into three divisions. There is Research, where the frontier models are trained. There is Scaling and Infrastructure, which handles the massive computers this training relies on. And then there is Applications, which takes the product of these two groups’ work, “and then brings it to the people,” says Vijaye Raji, who is chief technology officer of the Applications division.
Raji leads key technical teams—such as data, experimentation, and growth—that help power ChatGPT and Codex, OpenAI's coding tool, the use of which he says has made the company's engineers “10 times more productive.” He is also responsible for the tech behind ChatGPT’s new ad program, which launched in February, and which reached $1 billion in annualized revenue in late August. Raji says OpenAI built the system in such a way that one model supplies answers, while a separate model supplies clearly-delineated ads. “We make sure that users never mistake what the model says [for] an ad,” he says. “We will always optimize for and prioritize user trust over anything else.”
Correction, September 22
The original version of this story misstated elements of Raji's responsibilities as CTO. He does not lead engineering for ChatGPT and Codex; he leads key technical teams that help power those tools.
Disclosure: Investors in OpenAI include Salesforce, where TIME co-chair and owner Marc Benioff is CEO. OpenAI and TIME also have a licensing and technology agreement that allows OpenAI to access TIME’s archives.