Google has officially introduced its cutting-edge artificial intelligence model, Gemini AI Model, capable of processing diverse forms of information, including video, audio, and text.
In a significant move, Google presented Gemini as its “largest and most capable AI model,” marking the advent of a “Gemini era.” The company envisions deploying this technology across various settings, from major corporations to consumer devices like the Google Pixel 8 Pro.
What sets Gemini apart is its “multimodal” functionality, distinguishing it from existing AI models that typically specialize in only one type of user prompt, such as images or text. Google asserts that Gemini is equipped for more advanced reasoning, demonstrating a nuanced understanding of information compared to its predecessor.
In a blog post, Alphabet CEO Sundar Pichai expressed, “This new era of models represents one of the biggest science and engineering efforts we’ve undertaken as a company.”
Since the launch of OpenAI’s ChatGPT about a year ago, Google has been in a race to develop AI software that can compete with the innovations introduced by the Microsoft-backed company.
Google has already incorporated a segment of the new Gemini model technology into its AI assistant, Bard, and plans to release the most advanced version of Gemini through Bard early next year.
Alphabet has designed three versions of Gemini, each tailored to utilize varying amounts of processing power. The most potent version is intended for data center deployment, while the smallest is optimized for efficient operation on mobile devices.
Although Gemini is the largest AI model developed by Google’s DeepMind AI unit, it is notably more cost-effective for users compared to the company’s previous larger models. Eli Collins, Vice President of Product at DeepMind, emphasized that Gemini is not only more capable but also far more efficient, addressing concerns about computing power requirements during training.
Additionally, Alphabet unveiled a new generation of its custom-built AI chips, tensor processing units (TPUs). The Cloud TPU v5p, designed for training large AI models, is configured in pods of 8,960 chips. The latest version of these custom processors can train large language models almost three times faster than previous generations. The chips are now available for developers in a “preview” phase, starting from Wednesday, as announced by the company.