At the time of writing, Google's Gemini AI, introduced just two months ago, is already set to release its next-generation model, known as Gemini 1.5.
The announcement post delves into the intricacies, providing a detailed explanation of all the AI's improvements. Although quite technical, the key point is that Gemini 1.5 is set to deliver "dramatically enhanced performance." Achieving this involved adopting a "Mixture-of-Experts architecture" (MoE), where multiple AI models collaborate seamlessly. The implementation of this structure not only simplified Gemini's training process but also accelerated its ability to learn complex tasks compared to its previous performance.
The rollout of the upgrade will encompass all three major versions of the AI, with the initial release for early testing being Gemini 1.5 Pro, available today.
A noteworthy feature of this model is its "context window of up to 1 million tokens." In the realm of generative AI, tokens serve as the smallest data units used by LLMs (large language models) for processing and text generation. A larger context window enhances the AI's ability to handle more information simultaneously, and the one-million-token capacity of Gemini 1.5 Pro exceeds the capabilities of GPT-4 Turbo, which has a context window cap of 128,000 tokens, for comparison.
Observing Gemini Pro in Operation
Amidst the abundance of numerical data, the real question arises: What does Gemini 1.5 Pro appear like in action? Google has produced several videos that highlight the AI's capabilities. Undeniably intriguing, these videos demonstrate how the enhanced model can effectively analyze and summarize extensive volumes of text in response to a given prompt.
The analytical capabilities extend to various modalities. In a different demonstration, the development team presented the AI with a 44-minute Buster Keaton movie. They uploaded a basic sketch depicting a gushing water tower and inquired about the timestamp of a scene featuring a water tower. Impressively, Gemini 1.5 Pro accurately identified the specific scene, pinpointing it ten minutes into the film. Notably, this accomplishment occurred without any additional information about the drawing or any accompanying text, showcasing the AI's ability to comprehend the context independently.
Cutting-edge Technology
The model is currently not accessible to the general public. It is being offered as an early preview exclusively to "developers and enterprise customers" through Google's AI Studio and Vertex AI platforms at no cost. Google cautions testers about potential long latency times, as the model is still in the experimental phase. There are plans to enhance the speed in the future.
We have reached out to Google to inquire about the anticipated launch dates for Gemini 1.5 and Gemini 1.5 Ultra, as well as the broader release of these next-gen AI models. Updates on this information will be provided at a later time. Meanwhile, you can explore TechRadar's compilation of the best AI content generators for 2024.
