Mira Murati carefully returns to the spotlight

Mira Murati is not a natural being on the conference stage. As a CTO of OpenAI she was present, but rarely the public face of the company. As CEO of her own company, Thinking Machines Lab, she was even more difficult to find. When she met Bloomberg in San Francisco on Thursday – her first major media appearance for about 18 months – it was worth the attention, even though she was looking for not to say too much. The time makes sense. Thinking Machines spent most of the year and a half to act mostly in the background: capital acquisition, recruitment of researchers and provision of a product, Tinker, one API to fine tune open source AI models. In the meantime, companies competing for the same talents, customers and headlines have become more and more ubiquitous. OpenAI, where Murati worked as CTO for six years, is constantly in the news. The dynamics of Anthropic is everything everyone can talk about right now. And xAI, Elon Musks AI company, was launched in advance of its presumably massive stock exchange SpaceX which generates its own attraction to attention and investment. In this environment it is less profitable to stay with the head down; At some point you have to make some noise just to remind the market that you exist. Murati used the Bloomberg appearance exactly for it and not much more. She gave a view of what Thinking Machines calls “interaction models” and described them as a fundamentally different kind of AI interface. Instead of the round-based Prompt-and-Response Dynamics, which today accounts for most AI products, the company's models are designed to process continuous audio, text and video streams in 200 million intervals, she said the interviewer Emily Chang. The idea is that they can capture the texture of human communication – the interruptions, the corrections in the middle of thought, even thought breaks – in a little closer to real time. But Murati was looking at it as a first step and not as a finished product, and she refused to provide a specific release date for anything. She also answered questions about the episode, which she focused on the public for the first time: the chaotic week in November 2023, when the Executive Board of OpenAI Sam Altman released and became Interims CEO. Within OpenAI it was called “Blip”. Murati said that she was aware of her decisions at any moment – that the protection of the mission and the team was the continuous line that made the decisions seem obvious, even though the situation seemed to fall apart from the outside. She said that the company would be “imploded” if it had not participated in this strange five-day phase and the immediate consequences. However, it admitted that clarity on intention is not the same as clarity on the consequences. In retrospect, she said that she had pushed more information, a better transition plan and more transparency. But what she didn't say, at least not directly, is whether she thinks everything went well. When asked if she still trusted her former boss, she left the question and directed the conversation to a greater concern that she came back several times: the concentration of following decisions in a few hands – not only at OpenAI, but throughout the industry. She was worried, she said, less about the character of a single leader (although she admitted that it was important) than the lack of structural controls. GoodPeople make bad calls. Well-meaning organizations are deviating. The virtue had been given too much attention and the governance too little, she said. Chang also politely pushed her to the departures of several top-class researchers from Thinking Machines in recent months, a topic that Murati has largely avoided in the public and that she has played down on Thursday. First, she said, compressed the construction of an AI research laboratory from scratch to years of normal organizational volatility to months. She also admitted that compensation – the nine-digit sums that have become a standard currency in the fight for AI talent – stimulates the imagination of people, but she pointed out that this is usually not the whole story. To the laughter of some spectators, she said about her own competitive instinct: “When I wake up in the morning, I don’t think about how to kill the competitors.”

Of course, Chang asked what comes next for the AI in the big and the whole, also for the people who once said that AI companies were strengthened by AI, but who have recently been afraid because there are rumors about mass displacements of jobs, not to mention a future in which AI is used to produce chemical weapons. Murati, born in Albania and speaking with a slight Eastern European accent, reacted to scale. She rejected the formulation of an unavoidable dystopia or an unavoidable utopia and argued that none of the two results was predetermined and that the time we are currently in is that which will determine the further course of things. Nevertheless, she said – and not for the first time during the interview – that the future will look very different and not better when people take their hands off the steering wheel too soon.