“AGI is not a distant theoretical concept. We expect to build it within this decade, possibly within years. The scaling laws have not broken down. If anything, inference-time compute has opened a second frontier we are only beginning to explore.”

“What I call powerful AI could arrive as early as 2026-2027. My primary concern is not whether it will happen but whether the safety infrastructure will be ready when it does.”

“We are likely years, not decades, away from systems that match or exceed human performance across most cognitive domains. The convergence of foundation models, world models, and RL is happening faster than anyone predicted.”