Ex-OpenAI CTO Murati’s New Team Takes Shape

A month ago, we reported that former OpenAI chief technology officer Mira Murati had begun to pitch current OpenAI staffers on joining her new venture. It seems like at least one OpenAI researcher has taken her up on her offer.

Mianna Chen, a research program manager at OpenAI responsible for preparing AI models for release, has recently left and plans to join Murati at her new company, according to a person familiar with the move. (It isn’t clear whether the hiring is finalized.)

We’ve also got clarity on who else Murati is working with: she is teaming up with ex-head of post-training Barret Zoph and former senior researcher Luke Metz, both of whom we’ve reported have left the model developer in recent months, according to two people with direct knowledge of the new venture.

The moves are a reminder of the threats that OpenAI continues to face from big tech companies and other startups—including those launched by recently-departed former employees. A representative of Murati didn’t want to say anything when we contacted them.

Chen’s loss is particularly significant given that she led the launches of a number of OpenAI’s models and products, including GPT-4o, its o1 reasoning models and Advanced Voice Mode. She worked in a team called post-training, which helps to specialize AI models for specific tasks and provide them with feedback on the quality of their outputs. OpenAI is likely hoping to hold onto its post-training talent even tighter because, as we reported in this story over the weekend, gains from pre-training, or the process of giving models a broad understanding of the world by showing them lots of diverse data, seem to be slowing down.

It’s still not clear what Murati’s new startup will focus on, and whether it will develop foundation models or products built on top of already-existing models. Even so, we’d bet that some VCs are chomping at the bit to back the four researchers, given their pedigree—and we wouldn’t be surprised if more OpenAI staffers leave to join them in the upcoming weeks and months.

OpenAI’s head of safety systems Lilian Weng announced on Friday that she would be leaving the firm, adding onto the spree of departures from the company’s safety team, though she didn’t elaborate on what she’s doing next.

OpenAI still has lots of core talent and big product launches on the way, though. A lot of researchers—within and outside the company—are excited about the advances underlying its o1 reasoning models, despite rumblings within the company that simply throwing more data and compute at models won’t be enough to keep up the rate of improvement we’ve seen thus far.

Specifically, the ability to improve model responses by giving it more compute (i.e. time to “think”) when answering user questions could represent a new way to keep the AI party going.

We may get more data on this soon, as we’ve been hearing that OpenAI is hoping to release the full version of its o1 reasoning model by the end of this year, according to a person with knowledge of the product. We’ll be keeping our eyes peeled!