Google announced the launch of Gemini, its largest and most capable AI model, designed to generalise and understand, operate across, and combine different types of information. Gemini can run on everything from data centers to mobile devices.
Key features:
State-of-the-art performance. Gemini is the first model to outperform human experts on MMLU (massive multitask language understanding) with a score of 90.0%. It uses a combination of 57 subjects for testing both world knowledge and problem-solving abilities. Instead of just using its first impression, Gemini can use reasoning capabilities to evaluate data more carefully before answering difficult questions.
Next-generation capabilities. Gemini 1.0 was trained to recognize and understand text, images, audio, and more at the same time. With the use of multimodal reasoning capabilities, it can extract insights from a huge volume of documents at digital speeds.
Scalable and efficient. Gemini 1.0 was trained at scale on Google’s AI-optimized infrastructure using Tensor Processing Units (TPUs) v4 and v5e.
Responsibility and safety. Gemini has the most comprehensive safety evaluations of any Google AI model to date through novel research into potential risk areas like cyber offense, persuasion, and autonomy. With Google Research’s adversarial testing techniques, critical safety issues are identified in advance of Gemini’s deployment. Google also worked with a group of external experts and partners to test the models.
