AI MEASUREMENT : WHY NEUTRAL BENCHMARKS ARE CRUCIAL FOR TRUSTING AI PROGRESS | ZIHAN WANG SOLUTION

What happened: Zihan Wang, co-founder of an AI research company and a non-profit foundation, argues that AI benchmarks built and scored by the labs being measured cannot be trusted, and that neutral, reproducible benchmarks are needed. Key numbers: Industry produced more than 90% of notable frontier AI models in 2025, up from Why it matters: Wang says benchmarks must be built across competing institutions so rivals can reproduce results, and he treats convening more than 50 institutions as the core work, since a benchmark holds only when parties who do not trust each other all sign it.

Zihan Wang, co-founder of an AI research company and a non-profit foundation, argues that AI benchmarks built and scored by the labs being measured cannot be trusted, and that neutral, reproducible benchmarks are needed.

Wang says benchmarks must be built across competing institutions so rivals can reproduce results, and he treats convening more than 50 institutions as the core work, since a benchmark holds only when parties who do not trust each other all sign it.