PROVIDENCE, R.I. (AP) — Computer scientist Louis Castricato was in his eighth year studying large language models — the artificial intelligence technology behind chatbots like ChatGPT and Claude — when he started to feel like he was hitting a dead end.
“We basically have passed the point of doing real fundamental LLM research,” Castricato said. “Now it’s just applications.”
The researcher quit his doctoral studies at Brown University and started a new company, called Overworld. Its ambition is in its name: AI that can understand and navigate a world, not just words.
There’s still plenty of money to be made from AI chatbots — investors are counting on it as they commit trillions of dollars to leading developers like Anthropic and OpenAI. But a growing number of AI entrepreneurs are dedicating themselves to what they see as the next frontier: “world models” that teach AI systems, and sometimes robots, how to react in a physical environment.
They include some of the field’s most prominent scientists, such as Fei-Fei Li, who describes the concept of a world model as “one of the most important and most overloaded terms in AI today.”
The shift from language-only AI to systems that can perceive and act in the physical world represents a significant reorientation of research priorities within the industry. While chatbots have captured public attention and driven massive investment, some researchers argue that the core scientific challenges of language modeling have largely been solved, leaving incremental improvements and commercial applications as the remaining work.
Castricato’s move from academia to entrepreneurship reflects a broader pattern among AI researchers who see greater opportunity — and greater intellectual challenge — in building systems that must contend with the unpredictability of real-world physics rather than the relatively constrained domain of text.
Overworld and similar startups are working on what the field calls “world models” — AI systems that can simulate or predict how the physical environment will respond to actions, a capability essential for robotics, autonomous vehicles, and any application where an AI must operate outside the controlled realm of text and data.
Li, a Stanford University professor whose work on image recognition helped spark the modern AI boom, has been among the most vocal advocates for research into world models. Her characterization of the term as both critically important and frequently misused underscores the early-stage nature of the field and the lack of consensus about what exactly a world model should be.
The push toward physical AI comes as the broader AI industry continues to attract enormous capital. Investors have committed trillions of dollars to leading developers, betting that the technology will transform industries from health care to transportation. But the entrepreneurs now pivoting toward world models are betting that the next wave of value creation will come not from better chatbots but from AI that can actually interact with the world.