Tech Innovators Shift Focus From Chatbots to Physical AI Solutions
WHO:
ABC News Business - Computer scientist Louis Castricato, founder of Overworld.
ABC News - Fei-Fei Li, referred to as the “Godmother of AI,” founder of World Labs.
ABC News - Yann LeCun, AI pioneer and founder of Advanced Machine Intelligence Labs.
ABC News - Martin Hebert, dean of computer science at Carnegie Mellon University.
ABC News - Steve Jang, co-founder and managing partner at Kindred Ventures.
WHAT:
ABC News - Castricato's Overworld is now building video game worlds where a scene can adapt as a virtual character moves through it and interacts with objects, emphasizing, “We optimize for interaction above anything else.”
ABC News - Li wrote in her recent essay that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News - Li categorized world models into three types: “renderers,” “simulators,” and “planners,” noting that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News - Li divided world models into three categories: “renderers,” “simulators,” and “planners,” with “renderers” being the most commercially viable today but not reliable for teaching robots.
ABC News - Jang stated, “I think that the future is many different types of models with many different philosophies and architectures. I don’t think that it’ll be one large, dense model to rule them all.”
ABC News - Castricato's Overworld started last year and is now building video game worlds that adapt as virtual characters interact with them.
ABC News - Steve Jang's firm, Kindred Ventures, is investing in Overworld and other companies focused on world models, including Causal Labs and Extropic.
ABC News - Causal Labs is building AI models for weather prediction, while Extropic is developing specialized computer chips suited to world models.
ABC News - Castricato stated, “We basically have passed the point of doing real fundamental LLM research. Now it’s just applications.”
ABC News - Castricato's Overworld is building video game worlds that adapt as virtual characters interact with them, emphasizing, “There’s no other world model where you can just walk through doors or where you can interact with a detailed environment like this.”
ABC News - Li categorized world models into three types: “renderers,” “simulators,” and “planners,” noting that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News - Developers see limitations in AI language models and are pursuing world models to enhance AI's ability to interact with physical environments.
ABC News - Hebert noted that the complexity of physical interactions is much greater than just predicting text, stating, “This is much more complex than just predicting the next word in a sentence.”
ABC News - Hebert highlighted the complexity of physical interactions, stating, “There’s all the geometry of the world, the dynamic of how I move my hand, the physical interaction of the contact with the cup.”
ABC News - Developers see limitations in AI language models and are pursuing world models to enhance AI's ability to interact with physical environments.
ABC News - Hebert notes that physical interactions are more complex than predicting text, highlighting the need for advancements in physical AI.
ABC News - Castricato's Overworld is now building video game worlds where a scene can adapt as a virtual character moves through it and interacts with objects, emphasizing, “We optimize for interaction above anything else.”
ABC News - Li wrote in her recent essay that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News - Li categorized world models into three types: “renderers,” “simulators,” and “planners,” noting that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News - Li divided world models into three categories: “renderers,” “simulators,” and “planners,” with “renderers” being the most commercially viable today but not reliable for teaching robots.
ABC News - Castricato's Overworld is now building video game worlds where a scene can adapt as a virtual character moves through it and interacts with objects, emphasizing, “We optimize for interaction above anything else.”
ABC News - Li wrote in her recent essay that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News - Jang stated, “I think that the future is many different types of models with many different philosophies and architectures. I don’t think that it’ll be one large, dense model to rule them all.”
ABC News - Castricato's Overworld started last year and is now building video game worlds that adapt as virtual characters interact with them.
ABC News - Steve Jang's firm, Kindred Ventures, is investing in Overworld and other companies focused on world models, including Causal Labs and Extropic.
ABC News - Causal Labs is building AI models for weather prediction, while Extropic is developing specialized computer chips suited to world models.
ABC News - Castricato stated, “We basically have passed the point of doing real fundamental LLM research. Now it’s just applications.”
ABC News - Castricato's Overworld is building video game worlds that adapt as virtual characters interact with them, emphasizing, “There’s no other world model where you can just walk through doors or where you can interact with a detailed environment like this.”
ABC News - Li categorized world models into three types: “renderers,” “simulators,” and “planners,” noting that “A robot that can plan is a robot that can work, and the entire industry is racing to be the one that gets there first.”
ABC News Business - Developers are shifting focus from chatbots to "world models" in AI.
ABC News - Castricato started a new company, Overworld, to create AI that can understand and navigate physical environments.
ABC News - Li describes world models as “one of the most important and most overloaded terms in AI today."
ABC News - Castricato's Overworld is building video game worlds that adapt as virtual characters interact with them.
ABC News - Jang's firm is investing in Overworld and other companies focused on world models, including Causal Labs and Extropic.
WHERE:
ABC News - Overworld is based in Rhode Island.
ABC News - Advanced Machine Intelligence Labs is based in Paris.
WHY:
ABC News - Developers see limitations in AI language models and are pursuing world models to enhance AI's ability to interact with physical environments.
ABC News - Hebert noted that the complexity of physical interactions is much greater than just predicting text, stating, “This is much more complex than just predicting the next word in a sentence.”
ABC News - Hebert highlighted the complexity of physical interactions, stating, “There’s all the geometry of the world, the dynamic of how I move my hand, the physical interaction of the contact with the cup.”
ABC News - Developers see limitations in AI language models and are pursuing world models to enhance AI's ability to interact with physical environments.
ABC News - Hebert notes that physical interactions are more complex than predicting text, highlighting the need for advancements in physical AI.
HOW:
ABC News - Li wrote that world models learn “how light falls on a surface, how a garden looks from an angle no camera has captured, how objects respond to force and follow the laws of physics.”
ABC News - Li categorized world models into three types: “renderers,” “simulators,” and “planners,” with “renderers” being the most commercially viable today but not reliable for teaching robots.
ABC News - Li divided world models into three categories: “renderers,” “simulators,” and “planners,” with “renderers” being the most commercially viable today.
ABC News - Li wrote that world models learn “how light falls on a surface, how a garden looks from an angle no camera has captured, how objects respond to force and follow the laws of physics.”
ABC News - World models are defined by their ability to learn the statistical structure of space and time, as opposed to just text.
ABC News - Li categorizes world models into three types: “renderers,” “simulators,” and “planners,” each with different applications and commercial viability.
ABC News - Li wrote that world models learn “how light falls on a surface, how a garden looks from an angle no camera has captured, how objects respond to force and follow the laws of physics.”
ABC News - Li categorized world models into three types: “renderers,” “simulators,” and “planners,” with “renderers” being the most commercially viable today but not reliable for teaching robots.
ABC News - Li divided world models into three categories: “renderers,” “simulators,” and “planners,” with “renderers” being the most commercially viable today.
ABC News - Li wrote that world models learn “how light falls on a surface, how a garden looks from an angle no camera has captured, how objects respond to force and follow the laws of physics.”
ABC News - World models are defined by their ability to learn the statistical structure of space and time, as opposed to just text.
ABC News - Li categorizes world models into three types: “renderers,” “simulators,” and “planners,” each with different applications and commercial viability.
(UPDATED June 26, 2026 01:41 UTC)
Castricato's Overworld is pioneering video game worlds that adapt as virtual characters interact with their environments, emphasizing the importance of interaction in their design. Meanwhile, Li has categorized world models into three types: "renderers," "simulators," and "planners," highlighting that while "renderers" are currently the most commercially viable, they aren't reliable for teaching robots. Hebert pointed out the complexities of physical interactions, stating that they involve much more than just predicting text, which underscores the need for advancements in physical AI.
--
(UPDATED June 25, 2026 23:23 UTC)
Castricato's Overworld is now creating video game environments that adapt as virtual characters move through them, with a focus on optimizing interaction. In a recent essay, Li discussed the importance of robots that can plan, categorizing world models into three types: renderers, simulators, and planners, while noting that renderers are currently the most commercially viable but not reliable for teaching robots. Developers are looking to enhance AI's interaction with physical environments, as Li explained how world models learn about light, perspective, and the physical responses of objects.
--
(UPDATED June 25, 2026 19:39 UTC)
Castricato's Overworld is pioneering the development of video game environments that adapt as virtual characters navigate and interact with objects, focusing on optimizing interaction above all else. In a recent essay, Li emphasized the urgency in the industry to create robots capable of planning, stating that such advancements will determine which company leads the market. Meanwhile, Jang expressed that the future will involve a variety of models with different philosophies and architectures, rather than a single dominant model, while Hebert highlighted the complexity of physical interactions as being far more intricate than merely predicting the next word in a sentence.
--
(UPDATED June 25, 2026 13:42 UTC)
Castricato's Overworld, which began development last year, is creating video game worlds that evolve based on interactions with virtual characters. Steve Jang's firm, Kindred Ventures, is backing Overworld along with other companies like Causal Labs, which is focused on AI models for weather prediction, and Extropic, which is working on specialized computer chips for world models. According to Li, world models can be categorized into three types: “renderers,” “simulators,” and “planners,” with “renderers” currently being the most commercially viable.
--
(UPDATED June 25, 2026 08:07 UTC)
Castricato announced that the focus of research has shifted from fundamental LLM studies to practical applications, with his company Overworld creating video game environments that evolve based on player interactions. Li categorized world models into three types—renderers, simulators, and planners—emphasizing that a robot capable of planning is essential for industry advancement. Hebert pointed out the intricate nature of physical interactions, highlighting the importance of understanding geometry and dynamics in our interactions with the world.
--
(UPDATED June 25, 2026 06:22 UTC)
Developers are shifting their focus from traditional chatbots to innovative "world models" in artificial intelligence, with notable figures like Louis Castricato, who has founded a new company called Overworld in Rhode Island, leading the charge. Castricato's venture aims to create AI that can understand and navigate physical environments, while Fei-Fei Li describes these world models as crucial yet often misunderstood concepts in the field. With investments from firms like Kindred Ventures, which is co-founded by Steve Jang, the push for advancements in physical AI is gaining momentum, as experts like Martin Hebert emphasize the complexities of physical interactions compared to text prediction.