Google Releases Supercharged Version of Flagship AI Model Gemini
Google has released a supercharged version of its flagship AI model, Gemini 1.5 Pro. The new model delivers dramatically enhanced performance compared to the previous Gemini 1.0 Ultra version.
Gemini 1.5 Pro is a mid-size multimodal model that performs at a similar level to the larger 1.0 Ultra model, but with improved efficiency. It uses a new Mixture-of-Experts (MoE) architecture, which divides the neural network into smaller "expert" pathways that can be selectively activated based on the input. This specialization makes the model more efficient to train and serve.
A key breakthrough in Gemini 1.5 Pro is its ability to process up to 1 million tokens, achieving the longest context window of any large-scale foundation model to date. This long-context understanding will enable entirely new AI capabilities and help developers build more useful applications. However, this experimental 1 million token feature is currently only available in a limited preview for select developers and enterprise customers.
Google is also emphasizing extensive ethics and safety testing for Gemini 1.5 Pro, in line with the company's AI Principles and robust safety policies. The company is working to optimize the model's latency, computational requirements, and user experience as it rolls out the full 1 million token context window.