AI Safety Setback: Google’s Recent Gemini Model Underperforms in Key Evaluations, Fueling Industry-Wide Alarm