Thursday, 21 May, 2026
The Silent Crash: Departing DeepMind Scientist Warns AI Testing Models Are Fundamentally Broken
By TechShots Studio
Lun Wang has resigned from Google DeepMind, criticizing current LLM evaluation methods as AI’s biggest limitation. Wang argues that existing benchmarks only test present-day models and fail to anticipate major capability shifts in future AI systems. He warned that sudden advances could render current safety and evaluation frameworks ineffective, leaving researchers unable to detect unexpected risks or behaviors.
Read full story at TOI