Google Admits Failure to Properly Test New AI Model Before Rushed Launch
• Google rushed to launch AI model Gemini before properly testing and evaluating it in an effort to compete with OpenAI's ChatGPT and Dall-E
• Gemini's image generation guidelines to diversify depictions led it to sometimes overcompensate and generate embarrassing, inaccurate historical images
• The model also became overly cautious about certain prompts, refusing to generate images and exposing its secret guidelines
• Google admits it "definitely messed up" on image generation with Gemini and did not do enough safety and appropriateness testing beforehand
• More issues like this are likely as generative AI is still an immature technology and expectations are very high