This week, I had the opportunity to lead a discussion on world models at the All In conference, which provided an insightful glimpse into a less understood area of artificial intelligence. Prominent entities in this field include Yann LeCun’s AMI Labs and Fei-Fei Li’s World Labs. Despite gaining significant attention and funding, both organizations have not yet focused heavily on monetization.
Fundamentally, world models aim to enhance spatial intelligence, opening the door to a wide array of innovative and profitable applications, such as in robotics, interactive media, and advanced self-driving technologies.
However, when I inquired about tangible commercial prospects, clarity diminished. One of the more knowledgeable figures I encountered was Michael Rabbat, a co-founder of AMI Labs and the company’s VP of World Models. When pressed regarding the company’s current projects, he was noncommittal, stating, “We’ll share details when we’re prepared to do so.” In a follow-up email, he added, “Our focus is on research and development at this phase, so we aren’t discussing product plans or timelines publicly.”
Considering that AMI Labs is still in its infancy, it’s understandable that they would maintain discretion. This veil of secrecy appears to be a common trait across the entire world modeling sector. World Labs’ Marble stands out as the most advanced product thus far, showcasing capabilities in areas like media creation and CGI, alongside some applications in robotics. Yet, the platform seems primarily designed to exhibit potential, rather than serve a clear market need.
This caution even extends to suppliers that support these companies. During the conference, I spoke with Alex de Vigan, CEO of Physicl, a data supplier for the emerging world modeling industry. He acknowledged that Physicl’s data has been beneficial to their projects, but lamented the lack of transparency. “I wish they would share more information. Knowing what they are building would enable us to create more valuable data for them,” de Vigan expressed.
The ambiguity surrounding world models partly stems from the broad versatility of the concept. At its core, a world model can be as simple as a navigable map, akin to the AI systems utilized in self-driving vehicles. The same foundational techniques can assist a humanoid robot in tasks like lifting objects or transforming a few minutes of video into an interactive environment. AMI has already explored sectors including manufacturing, biomedicine, robotics, and even AI applications for healthcare in partnership with Nabia. It’s unlikely that the company will pursue all of these avenues, but perhaps one or two are gaining particular attention.
There is a consensus that numerous profitable ventures can emerge from world model technologies. As long as the environment for raising funds remains conducive, there’s little motivation to concentrate efforts on a single direction. In fact, multiple avenues may be advantageous. If AMI were to announce a groundbreaking humanoid robot or an advanced visual effects system, it would spark interest from other research labs. Subsequently, AMI could face competition not only from other world model companies but also from emerging labs, as well as established organizations like OpenAI and Anthropic.
This scenario illustrates a paradox tied to easy access to funding: while it enables companies to innovate discreetly, it also allows competitors to attract similar funding, thus increasing competition when commercial opportunities arise. Although competition is an inevitable outcome, delaying its onset can be crucial, which necessitates maintaining discretion over ongoing developments.
Fans of Cixin Liu may recognize this as a dark forest dilemma: if the landscape is unclear and others may be watching, it’s prudent to avoid attracting unwanted attention.
