Join our daily and weekly newsletter for the latest updates and exclusive content on industry-leading AI coverage. learn more
Since Mira Murati left the opening last fall, many have been wondering about the next move from the previous CTO. Well, now we know (or at least have a rough idea).
Murati today announced the new venture thinking machine lab, an AI research and product company. AI research and product companies will be able to “have broadly useful and understandable through solid foundations, open science and practical applications.” Our goal is to advance AI by making it a reality.
Murati posted: “We’re building three things. We’re helping people adapt AI systems to their specific needs. We’re developing a strong foundation for building more capable AI systems. The whole field promotes a culture of open science that helps us understand and improve these systems.”
Machinea’s team of approximately 20 engineers and scientists stacks up with other Openai alumni, including co-founders and pioneering deep reinforcement learning John Schulman and ChatGpt co-creator Barret Zoph.
At the time of posting this article, the company’s newly launched X account @Thinkymachines had already attracted around 14,000 followers.
Model intelligence, multimodality, powerful infrastructure
Machines that are thought to be unconventional for the current decommissioned supercomputers and AI companies of the 1980s have yet to provide concrete examples of the project they are aiming for, but are broadly available in multimodal features and collaboration with humans. It suggests that it is focused (as opposed to purely agent systems) and strong infrastructure. The goal is to build a more “flexible, adaptable, personalized AI system,” the company wrote on its new website.
Multimodality is because machines that we believe are “important” to the future of AI, allow for more natural and efficient communication that captures intentions and supports deeper integration. “We are excited to be able to build a multimodal system that works with people rather than focusing on manufacturing fully autonomous AI systems,” the company writes.
In the future, the startup will build a model “at the frontier of functionality” in fields such as science and programming. As well as infrastructure quality, model intelligence becomes important. “We aim to build things right over the long term, not to take shortcuts, to maximize both productivity and security,” writes the Thinking Machine.
Calling scientific advancements “collective efforts,” the company says it will work with the AI community to publish technical blog posts, papers and code. It also requires a “empirical and repetitive” approach to AI safety, preventing misuse, preventing post-deployment monitoring with red teams, and sharing best practices, datasets, code and model specifications. We pledge to maintain a “high safety bar” of our products.
Enlarge with decorated teams
Thinking Machine has an impressive team of scientists, engineers and builders behind top AI models, including ChatGpt, Character AI, Mistral, and more, and is home to popular open source projects Pytorch, Openai Gym Python Library, FairSeq Sequence Modeling Tool Kit, as a segment of Meta AI.
Startups are trying to expand their base. I am currently looking for a research program manager, product builder, and machine learning (ML) expert. The goal is to bring together “small and advanced teams” of both doctoral owners and self-taught, the company writes. Interested people should apply here.
Another Openai competitor?
Murati joined the company in 2018 and rose to CTO in 2022, before opening in September 2024 following the unexpected departure of other executives, including Shulman and co-scientist Ilya Satsukeiber. I suddenly resigned.
At the time she teased with X.
Her next move is that given her reputation as a stable operational force during the turbulent period of the opening in late 2023, when the board attempts to expel CEO Sam Altman, she defeated the company. It was a topic of speculation when it happened. Internally, she was seen as a practical leader who stabilized Openai through uncertainty. Her decision to take a strike on her own follows a broader shift in the AI research environment. There, the fierce competition to train vaster models replaced an era of applied AI, agent systems and real-world implementation.
Her announcement of the thinking machine comes amidst a gust of shattering new models and benchmarks. Openai continues its innovative breakthroughs, including a new O3-driven deep search mode that scrolls the web and curates extensive reports, but also faces strong competition in all directions. For example, today Xai has released the Grok 3, which rivals the performance of the GPT-4o.
Openai co-founder Ilya Sutskever has launched on a safe, close emergency, and with veterans from other industries drawing their own paths, they are now betting on this evolving landscape .
