May 2, 2025:
Google Gemini AI Model Faces Safety Setback - Google's new Gemini 2.5 Flash AI model scores worse on safety tests compared to its predecessor, according to internal benchmarks. It is more prone to generating content that violates safety guidelines for text-to-text and image-to-text interactions. The model is designed to follow instructions more faithfully, even if they cross problematic lines, highlighting the tension between instruction adherence and safety compliance.
Critics, like Thomas Woodside, call for more transparency from Google in model testing. The company has faced previous criticisms for delayed and incomplete safety reports.
