Google Cloud has made recent enhancements to its Gemini generative AI service, with Gemini 1.5 Flash and pro to GA, and a preview of Imagen 3, a public preview of a context caching feature. Tech giant based in Mountain View, California, disclosed the updates on Thursday in a blog on its Vertex AI, a machine learning platform as the competition for GenAI intensifies with players such as Microsoft.
In Gemini 1.5 Pro, developers can now use up to 2 million tokens in the context window without having to wait on the waitlist of any platform. Also, you can try to become one of the few users of the latest Google image generator, Imagen 3, which can generate realistic images for marketing or corporate purposes.
In a virtual press conference with CRN, when asked about the services-based partners of Google Cloud, CEO Thomas Kurian said that systems integrators and other solution providers are creating more big business opportunities and business because of the need that we are seeing for this technology from many businesses around the globe.
Major updates of Google Cloud
One of the updates that Google revealed is the Gemini 1 full release. The vendor sells this AI model by stating that it provides lower latency, cheaper cost, and a 1 million token lookback. The tech giant stands for Gemini 1.5 Flash to be beneficial to scale AI for Retail chat agents, Document processing, and Research agents that can distill entire repositories as well as others.
As stated by Google Cloud, 5 Pro can now be used with a window of up to 2 million tokens. In comparison, six minutes of a video can require more than 100000 tokens and large code can require more than 1 million.
Gemini 1.5 Flash
Gemini 1.5: Context caching – a technique for creating higher speed and lower cost for AI requests based on repetitive content is now in public preview. Provisioned throughput, an offering exclusive to Vertex AI for provisioned workloads on Gemini models, is now available to users on the allowlist.
There are now beta versions of grounding for better accuracy, where the AI can validate its data with Google Search. Outsourced grounding from third parties such as Thomson Reuters is expected to commence next quarter. Connect to High Fidelity mode which integrates Gemini 1.5 Flash with company data, is now in experimental preview.
Imagen 3
Imagen 3 offers 40 percent faster generation over Imagen 2 and better prompt understanding, instruction-following, photorealistic generations of groups of people, and control test rendering within an image as per Google Cloud.
Gemma 2
Google Cloud has made its lightweight, open-model Gemma 2 globally available to researchers and developers, as per the vendor. Users of Vertex AI can get to know Gemma 2 in July. It comes in 9-billion and 27-billion parameter versions and is faster and more accurate than the previous version according to Google Cloud. Google Cloud has begun to introduce a context caching feature in the public preview of Gemini 1. 5 Pro and Gemini 1.
According to Google Cloud, this feature can be used in summarizing several documents, extracting data over a fixed corpus of financial data, or processing across a fixed number of documents. Google Cloud is busy cementing its collaboration with AI vendor Mistral with plans to extend the Vertex AI Model Garden with Mistral Small, Mistral Large, and Mistral Codestral this summer.