In a tightly‑controlled interview with Bloomberg’s Emily Chang in San Francisco, Mira Murati broke a year‑and‑a‑half silence to discuss the state of artificial intelligence and the future of her new venture, Thinking Machines Lab. The former OpenAI chief technology officer, who helped launch ChatGPT, DALL‑E and Codex, now serves as the lab’s chief executive officer. Her appearance was timed to coincide with a series of milestones the company has quietly achieved since Murati left OpenAI in September 2024.
Thinking Machines has raised $2 billion in capital and secured a gigawatt of Nvidia’s Vera Rubin compute. It has shipped one product—Tinker, an API that lets developers fine‑tune open‑source models—while working on a second offering that Murati described as a “first step” toward a new class of AI interfaces.
Interaction models aim for real‑time collaboration
The forthcoming product, branded “interaction models,” departs from the prompt‑and‑response paradigm that dominates most AI services. Murati explained that the models ingest continuous streams of audio, text and video, breaking the input into 200‑millisecond intervals. The technical term is “full duplex,” and the company claims its TML‑Interaction‑Small variant can generate a response in 0.40 seconds—roughly the speed of natural conversation. By capturing interruptions, mid‑thought corrections and pauses, the models are designed to mirror the texture of human communication, aligning with Thinking Machines’ thesis that powerful AI should augment, not replace, human collaboration.
Murati declined to set a release date, emphasizing that the technology is still in early stages. She positioned the work alongside Tinker, noting that the API remains the lab’s only shipping product to date.
Beyond the technology, Murati used the interview to flag a deeper concern: the AI industry lacks structural governance checks. She argued that concentrating consequential decisions in a handful of leaders creates systemic risk, a lesson she drew from the chaotic five‑day boardroom showdown in November 2023 when OpenAI’s board ousted Sam Altman and briefly installed Murati as interim CEO. While she said she felt clear about protecting the mission during that period, she admitted the episode exposed a need for better transition planning and greater transparency.
The conversation also turned to talent churn at Thinking Machines. Co‑founder and CTO Barret Zoph, co‑founder Luke Metz and founding member Sam Schoenholz all returned to OpenAI in January, and five other founding members have departed for Meta, reportedly lured by nine‑figure compensation packages. Murati downplayed the exits, describing the rapid scaling of a frontier AI lab as “compressing years of normal organisational volatility into months.” She acknowledged the allure of massive paychecks but suggested that compensation alone does not capture the full story behind talent moves.
When pressed about her trust in Sam Altman, Murati sidestepped personal judgment, instead reiterating that the real issue is the concentration of power across the industry. Good people can make bad calls; without structural safeguards, well‑intentioned organisations may drift toward harmful outcomes.
Murati’s remarks paint a picture of an AI landscape that is both fiercely competitive and increasingly scrutinized. OpenAI continues to dominate headlines, Anthropic has raised $30 billion and is courting investors at an $800 billion valuation, and Elon Musk’s xAI is poised for a record‑breaking IPO under the SpaceX umbrella. In that environment, the quiet conviction Murati has maintained is beginning to wane.
Whether Thinking Machines can sustain its vision of human‑AI collaboration amid market pressures that reward speed, scale and billions in capital remains an open question. For now, Murati’s return signals a willingness to engage publicly, warn about governance gaps, and showcase a technology that could reshape how humans interact with machines.
This article was written with the assistance of AI.
News Factory APP - agentic news to boost your SEO & AEO.