Why Frontier Models Fail as Buy-Side Analysts - Janie Gu, Google DeepMind & Suyeol Yun, LinqAlpha
Kaggle joined Google DeepMind in February 2026 as its official public AI benchmarking platform
“the state of benchmarks is fundamentally broken”
Google DeepMind absorbed Kaggle in February 2026 and repositioned it as GDM's open, neutral public benchmarking platform, shifting from evaluating humans to evaluating advanced AI models. The talk introduces 'Mandate Bench,' a domain-specific benchmark built with fintech firm LinqAlpha to test whether frontier models can perform as buy-side securities investment analysts—and finds they fall short. The broader thesis is that the current benchmark ecosystem is fundamentally broken, unable to assess real-world workflow performance, and Kaggle aims to become the industry standard fix.