(Image source from: Nytimes.com)
Amazon's creator Jeff Bezos' AI company Prometheus has successfully secured $12 billion, achieving a valuation of $41 billion. Co-founded by Bezos and Vik Bajaj, the previous co-founder of Verily, which is Google's life science division, Prometheus aims to create an "artificial general engineer" focused on physical AI. The company is fundamentally changing the strategies employed by AI firms like OpenAI, known for ChatGPT. Instead of simply providing software with digital text, Prometheus is instructing its AI in physics, material science, and actual industrial limits. "If I studied gymnastics from a thousand books, I still wouldn’t become a good gymnast. There are various types of knowledge necessary to train these models on appropriate data," Bezos mentioned in a recent discussion. The main goal of the startup is to boost the effectiveness of businesses engaged in designing and constructing computers, cars, spacecraft, and other tangible products.
The wealthy founder of Amazon, who resigned as CEO in 2021 to dedicate himself to new projects like space exploration and AI, stated: "The LLMs (large language models), while quite amazing and beneficial, have been trained on a vast collection of humanity's existing knowledge from the internet. " Bajaj, Bezos' co-founder, expressed to NYT: "Building something like a jet engine can't be done with words alone - not even with mathematical language. It involves multidimensional forces and how they evolve over time. " Prometheus is developing software that can automate the creation and design of a variety of intricate physical systems, from jet engines to pharmaceuticals. Nonetheless, the company is currently keeping details confidential.
Prometheus employs 150 staff members across its locations in San Francisco, London, and Zurich. Bezos, who mainly resides in Miami, Florida, has been spending more time at the San Francisco headquarters of the startup, where he shares the role of co-CEO with Bajaj. To grasp what Prometheus is truly developing, it's useful to examine the existing AI environment. Envision today's well-known generative AI tools like ChatGPT or Google Gemini as exceptionally well-educated scholars. They have absorbed nearly every book, article, and website available online. They possess the ability to compose poetry, translate languages, write code, and create images in mere moments.
However, you cannot hand them a wrench and expect them to fix a leaking pipe underneath your sink. They can only interpret the world through language and digital patterns.
Physical AI seeks to change this situation. It involves providing AI with a comprehension of reality, beyond just language. Instead of training a model with sentences, companies in this domain educate software on the principles of gravity, thermodynamics, fluid dynamics, and the strength of materials. If traditional AI is like a brain in a jar, Physical AI represents a brain that recognizes texture, mass, resistance, and friction. For Prometheus, this does not necessarily entail creating humanoid robots that walk.
Instead, the focus is on developing an "artificial general engineer" - a kind of software that inherently knows that you can't create a rocket booster using a material that will melt at a certain temperature. It basically connects the huge difference between computer code and real materials like concrete, steel, and carbon fiber.
Making this change is much more complex than just training a chatbot to compose an essay. Discussing this technical challenge, Dr. Srinivas Padmanabhuni, a specialist in AI and CTO of AiEnsured, mentioned to NDTV that while physical AI is clearly the upcoming area of progress, its intricacy shouldn't be overlooked. "Understanding a messy, real physical world and dealing with those hurdles is much harder and much more intricate than predicting the next word," Padmanabhuni stated, suggesting that the industry requires a significant shift in resource allocation to achieve this. "Actually, a large portion of the funding that is currently going into large language models should be redirected towards developing these types of models. " Bezos is not the only one recognizing that the next significant opportunity in artificial intelligence lies beyond chatbots. A vast and well-funded environment is quickly forming around physical AI, with several main players pursuing different strategies to excel in the physical realm.
Consider Physical Intelligence (Pi), a startup based in San Francisco that is aiming to create a universal "brain" for robots. Founded by well-known AI researchers and former Google DeepMind employees, Pi develops core software that can be integrated into any robot, whether it's an industrial arm in an automobile factory or a folding machine in a laundry service. The market is progressing extremely quickly; after obtaining $600 million in a Series B funding led by Alphabet's CapitalG towards the end of last year, Pi has been in discussions to raise another $1 billion, which would send its value soaring past $11 billion. Interestingly, Bezos was one of the early individual investors in this company. Then there’s the hardware-focused strategy led by Figure AI. Valued at around $2.6 billion after significant investments from Microsoft, Nvidia, and Amazon, Figure is dedicated to creating autonomous humanoid robots. They have already used their hardware in BMW manufacturing plants to perform dangerous, repetitive warehouse tasks, believing that the future of physical AI shapes up to look very human.
Elon Musk is guiding Tesla along a similar route, significantly investing billions in creating the Optimus Gen 3 humanoid robot with the aim of putting them directly onto Tesla’s automotive assembly lines to address labor shortages. However, before a robot can manufacture a machine, an AI must predict how that machine will operate. That's where companies like PhysicsX come into play. Established by former Formula 1 engineers, this London-based startup creates the basic simulation technology that underlies heavy manufacturing. Their AI enables industries such as aerospace and defense to model stress, aerodynamics, and physics before any metal part is made. Underlining the strong interest from investors in this vital technology, PhysicsX secured $300 million in a Series C funding round led by Temasek, more than doubling its valuation to $2.4 billion.
Prometheus' most recent investment round was backed by Bezos personally, along with JPMorgan Chase, Goldman Sachs, and BlackRock, among others. While the initial surge in AI was marked by programs capable of writing and speaking, the upcoming stage may focus on technology that can create and transport.
