Piloting the first double-blind AI evaluations

Google has initiated the first double-blind evaluation of its proprietary Gemini Flash Lite AI model to prevent benchmark contamination. This evaluation uses cryptographic methods to keep test data confidential, ensuring unbiased assessment of the model's capabilities. The effort involves partners like the Singapore AI Safety Institute and MLCommons to enhance evaluation integrity.