In the race to achieve artificial general intelligence (AGI), current large language models (LLMs) still fall short due to their limitations in understanding spatial and temporal movements. This is why General Intuition, a New York-based startup backed by Jeff Bezos, is betting on video game data to train the next generation of AI. According to CEO Pim de Witte, world models trained on gaming data will drive the next big leap in physical AI, filling the gap left by text-centric models like ChatGPT and Claude.
Diễn biến chi tiết
During an episode of TechCrunch's Equity podcast, CEO Pim de Witte dived deep into the company's new direction. General Intuition was originally spun out of the gaming platform Medal TV and quickly garnered significant attention from the investment community. The startup recently closed a $320 million funding round, bringing its valuation to $2.3 billion. This round saw participation from major players, including Coatue, former Google CEO Eric Schmidt, and researchers from MIT and Google DeepMind.
Phân tích kỹ thuật & Công nghệ
According to General Intuition, current LLMs are exceptionally good at processing text but lack intuitive awareness of the physical world. Video games provide rich 3D simulated environments with pre-established physics, collisions, and real-time motion sequences. Training world models on gaming data helps AI learn how objects interact and move in 3D space, serving as an ideal stepping stone before deploying them onto real-world robots.
Ý kiến chuyên gia & Nhận định
Many tech experts note that General Intuition's approach is a clever alternative as high-quality text data on the internet begins to deplete. However, transferring simulated data from games to the real world still faces a significant "reality gap" in precision. Additionally, CEO Pim de Witte openly discussed the ethical boundaries that must be considered, especially when these physical AI models could eventually be applied to defense applications.
Tác động & Tương lai
If General Intuition's methodology succeeds, video game data will become an invaluable resource for the global robotics industry. The ability to accurately simulate physical behavior will shorten development cycles for autonomous robots and industrial automation. This could mark a crucial paradigm shift, moving AI beyond computer screens to interact safely and intelligently with the physical world around us.