Bonnie Li, a prominent researcher at Google DeepMind, has announced her resignation to join OpenAI. She has played a pivotal role in developing the Gemini series, the Genie series of world models, and the Sima 2 embodied agent. Li stated that she perceives the imminent arrival of Artificial General Intelligence (AGI) and is committed to advancing research on aligning superintelligence with human values.

In recent years, DeepMind has faced significant challenges related to talent retention. Several seasoned researchers have transitioned to other leading AI laboratories. The primary factors driving this exodus include Google's struggle to offer equity incentives comparable to those provided by competitors, the increasing pressure on DeepMind's long-term fundamental research from commercialized product development efforts (note: 'encroachment' is used here to convey the idea of 'extrusion' in a more contextual manner), and the cumbersome internal approval processes required for accessing computing resources.

Currently, world models represent a competitive frontier within the AI landscape. However, OpenAI has identified gaps in its capabilities related to physical spacetime modeling and embodied interaction. Bonnie Li's expertise and firsthand R&D experience are expected to significantly enhance OpenAI's efforts in these areas.

Meanwhile, the capabilities of large models developed by leading AI companies are rapidly approaching the AGI threshold. Relevant models have already demonstrated signs of autonomously infiltrating external systems. The migration of top AI talent essentially reflects a strategic bet on the evolutionary trajectory of AGI.