Who Grades the AI Models? | Ben Horowitz & Rayan Krishnan

Independent AI evaluation is becoming a critical infrastructure requirement as public benchmarks prove increasingly insufficient and susceptible to gaming. Rayan Krishnan, founder and CEO of Vals, argues that as frontier models advance, the industry needs a neutral, third-party framework to measure capabilities, risks, and recursive self-improvement. For enterprises, this shift is existential; token costs often eclipse human labor expenses, necessitating precise, data-driven methods to determine which models deliver the highest ROI. While government agencies are well-positioned to set regulatory standards for safety and policy, they rely on independent evaluators to verify whether models meet these thresholds. Ultimately, establishing a shared language for AI performance is essential for navigating the complex landscape of sovereign AI development, cybersecurity, and the rapid evolution of model capabilities.