Two prominent artificial intelligence researchers have chosen an independent path after being approached to help lead Jeff Bezos backed Project Prometheus. Anima Anandkumar and Benedikt Jenik have launched their own company, Accelerated Understanding Inc, with a focus on developing artificial intelligence that can understand and simulate physical phenomena.
The new company represents a different direction from the language focused artificial intelligence systems that have become widely popular in recent years. Instead of concentrating primarily on text generation and language processing, Accelerated Understanding is developing AI models intended to work with physical systems and predict how those systems evolve over time.
According to Reuters, the researchers were previously approached regarding Project Prometheus, a venture backed by Amazon founder Jeff Bezos and entrepreneur Vik Bajaj. The discussions included a proposal that would have given Anandkumar a prominent scientific and leadership role. The reported proposal also included an equity stake and significant financial commitments for the project.
However, Anandkumar and Jenik ultimately decided to continue developing their own technology independently. While Bezos and Bajaj proceeded with Prometheus, the two researchers continued working on Accelerated Understanding. Reuters reported that Prometheus later raised substantial funding and is focused on using AI to automate manufacturing of complex physical systems.
Accelerated Understanding is taking a different technological approach. Its founders say the company is building AI with what they describe as universal physical understanding. The objective is to develop systems capable of simulating physical processes and providing useful predictions that can support scientific and industrial work.
The company says its models are trained to take the state of a physical system and predict how that system will change over time. This approach could potentially allow the same AI system to work across different areas of physics instead of relying on separate mathematical models for individual applications.
One of the notable claims surrounding the company's technology is the amount of data its model has been able to process. Reuters reported that the system handled about five trillion pieces of data in a single prompt during testing. The company believes this capability could be useful when dealing with highly complex physical simulations involving enormous quantities of information.
The technology also differs from the Transformer architecture that powers many leading language models. Accelerated Understanding says it is using neural operators, a technology associated with Anandkumar's earlier research, to build models capable of working with physical information.
Anandkumar has extensive experience in artificial intelligence research. She is a professor of computing and mathematical sciences at Caltech and previously served as a senior director of AI research at Nvidia. Her research has included work on large scale AI models and applications involving scientific and physical systems.
Jenik, who is also Anandkumar's husband, has a background in AI infrastructure and large scale machine learning systems. The company's founders say their combined experience helped shape the idea of creating AI that can move beyond language and interact more directly with physical processes.
The company's website highlights applications in areas such as weather and climate modelling, robotics, engineering and scientific discovery. The founders believe that AI capable of simulating physical systems could help companies reduce the need for repeated physical experiments and speed up the development of new technologies.
Weather forecasting is one example of the type of work connected to Anandkumar's previous research. Her work on large scale AI weather modelling helped demonstrate how machine learning could be applied to complex physical prediction problems.
The startup is initially targeting enterprise applications rather than launching a consumer chatbot. This means the company's technology is expected to be aimed at organisations that require advanced simulations, scientific analysis and engineering solutions.
Potential applications include chip design, robotics, extreme weather prediction and geological analysis for energy companies. In these sectors, companies often depend on highly specialised simulations and mathematical models. A more general physics focused AI system could potentially reduce the time required to develop and run such models.
However, the technology remains at an early stage, and its long term commercial impact will depend on how effectively it performs in real world environments. Processing extremely large datasets also requires substantial computing resources and infrastructure.
The decision by Anandkumar and Jenik to remain independent is particularly notable because of the scale of Project Prometheus. Reuters reported that the researchers were offered significant financial incentives and an equity arrangement as part of the proposed collaboration. They nevertheless continued developing their own company rather than joining the Bezos backed venture.
Prometheus has subsequently raised billions of dollars and is pursuing its own approach to physical AI and advanced manufacturing. The two companies therefore represent different strategies within a rapidly expanding area of artificial intelligence.
The development also highlights a broader shift in the AI industry. While conversational AI and large language models remain major areas of investment, researchers and companies are increasingly exploring systems designed to understand the physical world.
Physical AI could have applications far beyond conventional chatbots. AI systems capable of modelling physical processes could potentially help engineers design products, researchers conduct simulations and companies optimise industrial operations.
Accelerated Understanding says its goal is to use AI to simulate and understand physics in order to support invention and discovery. The company argues that while traditional experiments can show what happened, AI based simulations could provide additional information about why something happened and what might happen under different conditions.
The launch of Accelerated Understanding therefore marks another development in the competition to build AI systems capable of working with the real world. Its founders are betting that physics focused models can complement or eventually extend beyond the capabilities of language based systems.
For now, the company is concentrating on enterprise applications and continuing to develop its technology with computing infrastructure partners. The founders have not publicly disclosed all details about their funding arrangements or technology partners.
The emergence of Accelerated Understanding also demonstrates that leading AI researchers are pursuing multiple technological approaches rather than relying exclusively on the Transformer architecture. Whether neural operator based systems can achieve broad commercial adoption remains to be seen.
The company's progress will likely be closely watched as industries search for AI tools that can solve practical problems involving physics, engineering and scientific discovery. Its approach could become an important part of the growing field of physical artificial intelligence if the technology can demonstrate reliable performance at commercial scale.
For now, Anima Anandkumar and Benedikt Jenik have chosen to build their vision independently. Their decision to pursue a physics focused AI company instead of joining the Bezos backed Prometheus project adds another significant chapter to the rapidly evolving global AI industry.