Members of Gen Z have grown up hearing that our attention spans are shrinking and that our lives will be defined by the eight-second scroll and an endless stream of content. The narrative is that we’re so addicted to the instant gratification of platforms like TikTok that we’ve lost the capacity for deep, sustained focus.
But what if the next wave of artificial intelligence (AI) isn’t designed to feed that addiction — but to fundamentally change it? What if the future of AI demands young people’s attention, curiosity, and creativity in ways we haven’t experienced before?
As the co-founder of Chima, an applied AI research lab, I’ve spent a lot of time thinking about the emerging concept of “world models,” AI systems that can generate interactive, dynamic environments from a simple prompt. Google DeepMind’s Genie 3, for example, can create a navigable, consistent 3D world from a single sentence, while Fei-Fei Li’s new company, World Labs, is dedicated to building Large World Models (LWMs) that can perceive, generate, and interact with the 3D world, endowing AI with what it calls “spatial intelligence.”
The implications of this technology are staggering — it means the ability to create and explore any world you can imagine, from a historical setting to a fantastical landscape. But what I find most fascinating is the potential impact of world models on Gen Z. We’re digital natives, having grown up with smartphones and social media as constant companions. We’ve also been told we have the attention span of a goldfish. So, what happens when a technology emerges that demands focus? Is this the moment we reclaim our ability to pay attention?
What are world models?
Large language models (LLMs) — the kinds of AI that power chatbots like ChatGPT — learn to generate language by training on vast amounts of text. World models, meanwhile, learn how to simulate worlds by training on the rules of physics, the properties of objects, and the dynamics of physical interactions. Once trained, they can generate persistent worlds for users to explore.
For instance, you can prompt Google DeepMind’s Genie 3 with “A helicopter pilot carefully maneuvering over a coastal cliff with a small waterfall” and then actually fly the helicopter in that world in video game-like fashion. You can explore the “palace of Knossos on Crete as it would have stood in its glorious heyday” or navigate a jet ski through a “festival of lights.” The model can simulate the physics of a hurricane in Florida, with waves splashing over the road and palm trees bending in the wind, or generate fantastical scenarios, like controlling a firefly through an enchanted forest. This represents a fundamental shift in what AI can do. Genie 3 creates not just content, but also context. Its outputs aren’t AI-generated images — they’re experiences.
Li’s World Labs is taking a slightly different approach, focusing on spatial intelligence. The goal is to create AI that can understand and reason about the 3D world, not just generate it. Their LWMs are designed to be persistent, navigable, and controllable, with no time limits or inconsistencies. The lab is also building tools that will allow users to export these generated worlds for use in other applications, like game engines or virtual reality (VR) headsets. This focus on user creation and control is a key differentiator between World Labs and other AI companies, and it hints at a future where more people are actively building AI-generated content — not just consuming it.
In the final weeks of 2025, we saw a flurry of announcements that signaled a major shift in the AI landscape toward world models. AI pioneer Yann LeCun confirmed he’s launching a new startup, AMI Labs, focused entirely on world models, with a staggering reported valuation of over $3.5 billion before even launching. Meanwhile, AI company Runway, known for its work in Hollywood, launched a new family of world models aimed at robotics and gaming. A new generation of world model-focused startups is now emerging, too, including Iconic, which just raised $13 million to build an AI-native game engine for creating “living, adaptive worlds.”
The momentum behind world models is undeniable, and the technology itself is accelerating at an incredible pace. In December 2025, Ryo Lu, head of design at Cursor (the nearly $30 billion startup building an AI-native coding environment), highlighted how the company’s head of engineering built a fully featured city simulator, IsoCity, in just three days using AI tools. We’re talking about creating a world with pedestrians, cars, boats, and even emergencies in a fraction of the time it would have taken just a few years ago.
What’s particularly striking about the AI industry’s focus on world models is that they demand a different kind of engagement from users than today’s popular forms of entertainment. You can’t just passively scroll through a world model; you have to actively explore it. You have to pay attention to your surroundings, build a mental map of the environment, and make conscious decisions about where to go and what to do. In a way, it’s the antithesis of the TikTok experience. It’s a return to the kind of deep, immersive engagement that many of us remember from playing video games like Minecraft or The Legend of Zelda — digital experiences that reward curiosity and exploration, not just fleeting attention.
The attention paradox
This brings us to an interesting paradox about younger people and attention. The popular perception is that members of Gen Z (and now Gen Alpha) can’t focus on anything for more than a few seconds. Researchers, meanwhile, have found evidence that consuming a constant stream of short-form video content on platforms like TikTok can lead to a shortened attention span and a craving for instant gratification. Some studies have even linked heavy TikTok use to cognitive impairment and attention deficits.
But that’s not the whole story. As a 2022 McKinsey report pointed out, the idea that Gen Z has an eight-second attention span is a myth. It’s not that members of my generation are incapable of deep focus; we’re just more selective about where we direct our attention. In a world of infinite content, we’ve become expert curators, quickly filtering out the noise to find the things that truly interest us — and when we find something we care about, we’re more than willing to invest our time and attention into it. The same McKinsey report notes that Gen Z has helped catalyze a revival in book sales, particularly young-adult fiction. We’re also avid consumers of long-form video content, from deep-dive documentaries to multi-hour video game streams. “Fifty-nine percent of Gen Zers use short-form video to discover content that they’ll then watch longer videos about,” wrote McKinsey.
The recent launch of Sora, OpenAI’s video generation app, illustrates this tension in Gen Z’s consumption habits. Sora hit over 1 million downloads in fewer than five days, outpacing ChatGPT’s initial growth despite being invite-only. But here’s what’s interesting: The app is designed as a social feed where users scroll through AI-generated videos, complete with memes and what critics have called “AI-generated slop.” It’s essentially TikTok for AI-generated content. This suggests that even when social media users have access to powerful generative AI tools, our default mode is still the endless scroll and passive consumption of bite-sized content.
This makes the challenge for world models particularly interesting. If developers can incorporate even sophisticated AI video generation into a scrollable feed, what does that say about how Gen Z will engage with the slower, more deliberate experience of world exploration? Sora’s rapid adoption shows that AI tools that fit existing consumption patterns are in demand. But the question isn’t necessarily whether world models can bridge the gap between instant gratification and deep engagement — it’s whether they’ll need to in order to succeed.
Three possible futures
I see three possible scenarios for how the future of world models could play out, each with different implications for how younger generations develop cognitively and creatively.
A cognitive renaissance
Younger generations embrace world models as a new form of creative expression and experiential learning. We use them to build games, create movies, and explore historical events in a way that was previously impossible. We become a generation of world-builders, not just content consumers. There’s reason to believe this optimistic vision of the future, where AI is a tool for human development, could become reality, too: The rise of the “creator economy” and the popularity of games like Minecraft and Roblox show that younger generations are already deeply engaged in world-building and user-generated content.
The great rejection
Younger generations find world models to be too slow, demanding, and unrewarding. The instant gratification of TikTok and other short-form content platforms proves to be too powerful a draw for their attention. World models become a niche technology used by a small group of professionals and enthusiasts, but they fail to capture the mainstream imagination. This scenario suggests that the dopamine-driven feedback loops of social media are powerful enough to outcompete even the most sophisticated immersive experiences.
The hybrid future
I think a third scenario is the most likely outcome. In this future, younger generations don’t choose between short-form content and AI-generated worlds — they engage with both. Short-form content provides entertainment and social connection, but we also turn to world models for education, creative fulfillment, and professional development. We become adept at switching between different modes of digital consumption, from the shallow and fast to the deep and slow.
However, this shifts attention to how the benefits of world models will be distributed. Those who have the time, resources, and motivation needed to create world models and understand their underlying systems could develop significant advantages in spatial reasoning, systems thinking, and creative problem-solving over those who primarily consume the models. This isn’t about world models being good or bad; it’s about recognizing that different modes of engagement produce different cognitive outcomes, and we need to be mindful of the potential to create a new kind of digital divide.
Agency and ownership
As AI-generated worlds become more immersive and persistent — and populated by human-like AI agents — they surface new issues about ownership, control, and the nature of digital experiences.
Consider what companies like Iconic are building: “living, adaptive worlds” where AI characters “retain memory across sessions and evolve over time.” These are worlds designed to feel alive — they respond to us and remember us. The potential applications for these models are remarkable, from education to entertainment, but they also raise questions about the relationship between users and digital environments: When the most engaging experiences in a young person’s life take place within digital platforms owned by tech companies, what does that mean for autonomy and agency? When AI characters are designed to be perfectly responsive and endlessly patient, how does that shape expectations for real-world relationships?
These aren’t questions with simple answers, but they’re worth considering given that AI experts as noteworthy as LeCun are predicting that “world models will be the dominant model for AI architectures” within three to five years. If he’s right, we’re not just looking at a new technology platform — we’re looking at a fundamental shift in the way humans relate to the digital environment and, potentially, reality itself.
Ultimately, the impact of world models on younger generations will depend on how we choose to engage with them. We can approach them as tools for learning and creation, using them to develop new skills and explore new possibilities. We can consciously recognize and attempt to mitigate their downsides, including the possibility that these experiences may supplant real-world engagement and that their cognitive benefits might be unevenly distributed. Or we can settle for the path of least resistance, wandering worlds that others create.
The technology is here. The investment is flowing. The world models are being built. The question now is: What role will we play in them?
This article From infinite scroll to infinite worlds: How AI could rewire Gen Z’s attention span is featured on Big Think.