Google held the keynote session of its annual developer-focused Google I/O occasion on Tuesday. Through the session, the tech large closely targeted on new developments within the synthetic intelligence (AI) entrance and launched numerous new AI fashions in addition to new options for the present infrastructure. A significant spotlight was the introduction of a two million token context window for Gemini 1.5 Professional, which is at the moment obtainable for builders. A sooner variant of Gemini in addition to the following era of Google’s small language fashions (SML) Gemma 2 was additionally launched.
The occasion was kickstarted by CEO Sundar Pichai who made one of many greatest bulletins of the evening — the provision of a two million token context window for Gemini 1.5 Professional. The corporate launched a a million token context window earlier this yr however thus far, it was solely obtainable to builders. Google has now made it typically obtainable in public preview and could be accessed by way of Google AI Studio and Vertex AI. As an alternative, the 2 million token context window is completely obtainable by way of waitlist to builders utilizing the API and to Google Cloud clients.
With a context window of two million, Google claims, the AI mannequin can course of two hours of video, 22 hours of audio, greater than 60,000 strains of codes, or greater than 1.four million phrases in a single go. Apart from enhancing contextual understanding, the tech large has additionally improved Gemini 1.5 Professional’s code era, logical reasoning, planning, multi-turn dialog, in addition to the understanding of pictures and audio. The tech large can also be integrating the AI mannequin into Gemini Superior and Workspace apps.
Google has additionally launched a brand new addition to the household of Gemini AI fashions. The brand new AI mannequin, dubbed Gemini 1.5 Flash is a lightweight mannequin that’s designed to be sooner, extra responsive, and cost-efficient. The tech large mentioned that it has labored on enhancing its latency to enhance its velocity. Whereas fixing complicated duties wouldn’t be its power, it might probably do duties comparable to summarisation, chat purposes, picture and video captioning, knowledge extraction from lengthy paperwork and tables, and extra.
Lastly, the tech large introduced the following era of its smaller AI fashions, Gemma 2. The mannequin comes with 27 billion parameters however can run effectively on GPUs or a single TPU. Google claims that Gemma 2 outperforms fashions twice its measurement. The corporate is but to launch its benchmark scores.
