Google unveils Gemini, its largest AI model, to take on OpenAI
Gemini is Alphabet’s first AI model after the merger of its AI research units
Google parent Alphabet has introduced Gemini, its largest and most advanced AI model to date, as the tech giant aims to compete with rivals OpenAI’s GPT-4 and Meta’s Llama 2 in the burgeoning artificial intelligence (AI) space. Gemini is the first AI model from Alphabet following the merger of its AI research units, DeepMind and Google Brain, into a single division called Google DeepMind, led by DeepMind CEO Demis Hassabis.
Gemini is a groundbreaking AI model that has been developed from scratch and is unique in its multimodal capabilities. This means that it can comprehend and process various types of information simultaneously, including text, code, audio, images, and videos.
Alphabet CEO Sundar Pichai described Gemini as the result of one of the company’s most significant science and engineering endeavors, emphasizing that this new era of models marks the realization of the vision behind the establishment of Google DeepMind earlier this year.
Gemini offers three different sizes: Ultra, Pro, and Nano. The Ultra version is designed for highly complex tasks, the Pro version for a wide range of tasks, and the Nano version for on-device tasks.
Gemini Pro will soon be accessible to developers through the Gemini API in Google AI Studio and Google Cloud Vertex AI. Starting December 13, interested developers can leverage Gemini Pro for their projects. Additionally, Gemini Nano will be available to Android developers through AICore, a new system capability being introduced in Android 14. Initially, this capability will be available on Pixel 8 Pro devices starting December 6, with plans to expand support to other Android devices in the future.
Google aims to integrate Gemini into all of its products, with the Bard language model utilizing a fine-tuned version of Gemini Pro for enhanced reasoning, planning, and understanding. Gemini Nano will introduce new features to Pixel 8 Pro smartphones, including the ‘Summarise’ function in the Recorder app, and will soon be integrated into Smart Reply in Gboard, initially accessible on messaging app WhatsApp.
Gemini’s introduction will significantly improve Google’s generative AI search offering, Search Generative Experience (SGE). The company reports a 40 percent reduction in latency and improvements in quality in English-language searches conducted in the United States.
Alphabet CEO Sundar Pichai believes the ongoing transition to AI will be more impactful than the shift to mobile or the web. He anticipates that AI will create opportunities across all aspects of life, drive innovation, and propel economic progress. Pichai asserts that the potential of AI is just beginning to be realized and that its possibilities are far-reaching.
Alphabet first gave a preview of Gemini at its annual developer conference, Google I/O, held in May 2023. This announcement coincides with Google’s efforts to catch up with OpenAI, which recently unveiled its latest AI model, GPT-4 Turbo. GPT-4 Turbo is an advanced version of OpenAI’s flagship GPT-4 model and was released last month with support from Microsoft.
Gemini Ultra surpasses other AI models in terms of performance, outperforming the current state-of-the-art results on 30 out of 32 widely-used academic benchmarks for language models. The Gemini Ultra AI model also surpasses human experts on the massive multitask language understanding (MMLU) benchmark, which tests knowledge and problem-solving abilities across 57 subjects.
Gemini Pro outperforms its predecessor, GPT-3.5, in six out of eight benchmarks, including MMLU and GSM8K (Grade School Math 8K), which assesses grade school math reasoning.
According to Google Assistant and Bard Vice-President Sissie Hsiao, Gemini represents a significant milestone in AI development and a new era for Google. Demis Hassabis, CEO of Google DeepMind, highlights Gemini’s flexibility, as it can efficiently operate on a range of devices, from data centers to mobile devices. Gemini’s capabilities will greatly enhance the opportunities for developers and enterprise customers to build and scale with AI.
Gemini’s first version excels in multimodal reasoning, enabling it to understand complex written and visual information. It can extract insights from hundreds of thousands of documents by reading, filtering, and comprehending information. Gemini also demonstrates proficiency in nuanced understanding and can tackle questions relating to intricate subjects, making it adept at explaining reasoning in fields like math and physics.
The AI model is even capable of comprehending, explaining, and generating high-quality code across multiple popular programming languages, including Python, Java, C++, and Go.
Alphabet has been diligent in enhancing safety measures considering Gemini’s capabilities. The company has engaged in groundbreaking research on potential risk areas such as cyber-offense, persuasion, and autonomy. It has also implemented adversarial testing techniques to identify critical safety issues in advance of deploying Gemini. Alphabet collaborates with a diverse group of external experts and partners to rigorously evaluate the models across a range of issues.
With the advent of Gemini, Google seeks to transform AI into an expert helper or assistant, making it more intuitive and useful. This latest AI model represents an important step towards achieving that vision. As Google integrates Gemini into its various products and services, including Search, Ads, Chrome, and Duet AI, users can anticipate an even more informed and intelligent experience in the months to come.
The article is optimized for search engines.
