Join our daily and weekly newsletters for the latest updates and the exclusive content on AI coverage. Learn more
Since Mira Murati left Openai Last FallMany wondered about the next movement of the old CTO. Well, we know now (or at least a difficult idea).
Murati went to X today to announce his new business Thinking laboratoryA research and product company of AI in order to “advance AI by making it largely useful and understandable by solid foundations, open sciences and practical applications”.
Murati published: “We are building three things: helping people adapt AI systems to work for their specific needs; Develop solid foundations to build more capable AI systems; Promote an open science culture that helps the whole field to understand and improve these systems. »»
The Thinke team.
Since the publication of this article, the new launch of the company X @Thinkymachines had already accumulated around 14,000 subscribers.
Model, multimodality, strong infrastructure intelligence
Reflection machines – not to be mistaken with the now missing supercomputer and the company of the IA of the 1980s – do not yet offer specific examples of planned projects, but suggests a broad accent on multimodal capacities, collaboration Human-Ai (as opposed to purely agentic systems) and a strong infrastructure. The objective is to build more “flexible, adaptable and personalized AI systems”, writes the company on its New website.
Multimodality is “essential” in the future of AI, according to thinking machines, because it allows a more natural and effective communication which captures intention and supports more in -depth integration. “Instead of focusing only on the manufacture of fully autonomous AI systems, we are delighted to build multimodal systems that work with people in collaboration,” writes the company.
In the future, the startup will create “at the capacity border” models in fields such as science and programming. Model intelligence will be essential, as is the quality of infrastructure. “We aim to build things properly for the long term, to maximize both productivity and security, rather than taking shortcuts,” writes machines.
Calling scientific progress a “collective effort”, the company affirms that it will collaborate with the AI community and will publish technical blog articles, articles and code. It will also take an “empirical and iterative” approach to IA security, and undertake to maintain a “high safety bar” to prevent improper use, monitor red team and after deployment and share Best practices, data sets, code and model specifications.
Develop a decorated team
Reflection machines have an impressive team of scientists, engineers and manufacturers behind the best AI models, including the Chatppt, the AI and Mistral character, as well as the Popular Pytorch Open-Source projects, the library OPENAI Gym Python, the FAIFSEQ sequence modeling toolbox and the Meta AI segment.
The startup seeks to develop this base; He is currently looking for a research program manager, as well as experts from product manufacturers and automatic learning (ML). The objective is to constitute a “small high caliber team” of doctoral and self -taught holders, writes the company. Interested people should apply here.
Another competitor of Openai?
Murati suddenly resigned from Openai in September 2024 – after the unexpected departure of other leaders, notably Schulman and co -founder of former chief scientist Ilya Sutskever – after joining the company in 2018 and mounted in CTO in 2022 ( The year that brought us the revolutionary chactgpt) and CTO in 2022 (the year that brought us the revolutionary chatgpt). .
At the time, She teased on x: “I go away because I want to create time and space to make my own exploration.”
His next move had been a subject of speculation given his reputation of regular operational force during the tumultuous period of Openai at the end of 2023, when the attempt to attempt the attempt by CEO Sam Altman briefly upset the company. Internally, it was considered a pragmatic leader who kept the OpenAi stable by uncertainty. Her decision to get rid of herself follows a broader change in the AI research landscape – where the frantic breed to form ever heavier models gives way to an applied AI era, of agency systems and execution of the real world.
The announcement of his thinking machines occurs in the middle of a burst of new model capacities and broken references. OPENAI continues to make innovative breakthroughs – including its new in -depth research mode fueled by O3, which scrolls the web and organizes in -depth reports – but it also faces strong competition in all directions. Today, for example, Xai released Grok 3which competes with the performance of GPT-4O.
With the co-founder of Openai Ilya Sutskever Launch Sure superintelligence And other veterans of the industry stretching their own paths, the question is now where the machines of reflection will place its bets in this evolving landscape.
