World Models Will Soon Be Inescapable. Here’s What You Need to Know (www.wired.com)

Niantic, the creator of Pokémon Go, is shifting its focus to developing advanced "world models"—AI systems that use visual inputs to generate and simulate digital environments. Unlike traditional language models, which predict text, world models can analyze visual states and predict subsequent visual scenarios, offering a powerful tool for training robots and driverless cars in realistic simulated settings. This technology, which allows systems to practice in digital recreations of real-world environments, could become integral to the expansion of home robotics and autonomous vehicles by 2027. The significance of this shift is multifaceted. World models promise to enhance the safety and efficiency of autonomous systems, enabling them to react in real-time to their environments through simulations that incorporate an understanding of physics and cause-and-effect relationships. Companies like Wayve demonstrate this by using dashcam footage to create virtual hazardous driving conditions for testing. As major players in the tech industry invest in these capabilities, including AI veterans like Yann LeCun and Fei-Fei Li launching significant projects, the implications extend beyond technology into the labor market, highlighting the growing demand for real-world data collection to train these models. Thus, even those who avoid engaging directly with AI technologies may find their lives increasingly impacted by these developments.