Former OpenAI CTO returns after 18 months with warning for AI industry

Mira Murati has spent roughly 18 months building a frontier AI company largely out of public view. When she finally sat down with Bloomberg in San Francisco on Thursday, her first major media appearance since leaving OpenAI, she was characteristically careful.

Murati showed off an appearance of what Thinking Machines Lab calls "interaction models", which is totally different from traditional AI interfaces. Instead of the usual back-and-forth prompts, their system processes audio, text, and video streams super fast in just 200-millisecond chunks.

Murati said this was just the start, not some complete package. They didn't announce any release dates either. The company did launch one thing the public could check out: Tinker, an API for tweaking open-source AI models.

In November 2023, when OpenAI's board fired Sam Altman and Murati stepped in as interim CEO during what staff called "the blip", it defined her public image. When questioned, she said her decisions felt clear in the moment – protecting both the mission and the team. That was what made her choices seem right, even as everything fell apart.

Murati claimed the organisation would've "imploded" without her input during those chaotic five days. However, she admitted there were limits to her certainty. Looking back, she wishes she had pushed for more info, a smoother transition, and more transparency.

When asked if she still trusted her ex-boss, Murati avoided the question. Instead, she kept coming back to one main issue: the concentration of big AI decisions in the hands of just a few folks. Adding more on this, she shared that it's not really about the character of anyone in charge but more about the lack of structural checks in the industry.

Murati thinks we've been focusing way too much on discussing individual virtues rather than ensuring institutions are actually accountable.

Chang pressed Murati about the recent departure of some big-name researchers from Thinking Machines. Although she'd been avoiding this topic publicly, Murati played it down. She said that when you start a new lab from scratch, you get all that typical startup volatility crammed into a shorter time frame.

Regarding AI causing job loss, weapons, and other risks, she shot down both doom and glory scenarios. She believes we aren't doomed and can shape things positively, so it's not set in stone.