[Yang Zhilin Unveils Kimi's Technical Roadmap at GTC: Betting on Token Efficiency, Long Context, and Agent Swarms] At NVIDIA's GTC conference in 2026, Yang Zhilin, founder of Dark Side of the Moon's Kimi, delivered a public speech. He stated that to continuously push the upper limits of large model intelligence, it is necessary to reconstruct foundational elements such as optimizers, attention mechanisms, and residual connections. Following the official release of Kimi K2.5 at the end of January this year, Yang Zhilin systematically disclosed the technical roadmap behind the model for the first time during his speech. He summarized Kimi's evolutionary logic as a resonance across three dimensions: Token efficiency, long context, and intelligent agent swarms. "Current scaling is no longer simply about resource stacking; instead, we must simultaneously seek scale effects in computational efficiency, long-term memory, and automated collaboration. If the technological gains across these three dimensions can be multiplied, the model will exhibit intelligence levels far beyond the current state," he said. Furthermore, he predicted that future intelligent forms will evolve from single agents to dynamically generated swarms. (Sci-Tech Innovation Board Daily)