Google has announced the launch of its most advanced and extensive artificial intelligence model to date, Gemini, as the company faces growing pressure to monetise AI technology. This new AI model, available from December 13, is divided into three distinct versions: Gemini Ultra, the most comprehensive and capable; Gemini Pro, adaptable across a wide range of tasks; and Gemini Nano, designed for specific functions and mobile applications.
The plan is to license Gemini to clients through Google Cloud, enabling its integration into their applications. Developers and business clients can access Gemini Pro via the Gemini API in Google AI Studio or Google Cloud Vertex AI starting December 13. Furthermore, Android developers will be granted access to Gemini Nano. Gemini's technology will also be employed in Google's products, such as the Bard chatbot and the Search Generative Experience (SGE), which aims to provide conversational-style answers to search queries.
Applications and innovations
Gemini will transform customer service with advanced chatbots and personalised product recommendations. It also has potential applications in content creation for marketing and blogging, as well as productivity tools for summarising meetings or generating code for developers.
The model has demonstrated impressive capabilities, such as updating a chart from a screenshot after analysing extensive research material, accurately assessing a photograph of a person's maths homework, and identifying correct and incorrect answers.
Remarkably, Gemini Ultra has outperformed human experts in the MMLU (massive multitask language understanding) test, covering various topics from mathematics to ethics. This model exhibits a profound understanding of nuances and complex reasoning.
Collaborative effort and future prospects
Sundar Pichai, CEO of Alphabet Inc., emphasised the collaborative nature of Gemini's development. This AI model is multimodal, meaning it can process and combine different types of data, including text, images, and videos.
Soon, Google's chatbot Bard will incorporate Gemini Pro, enhancing its reasoning and understanding capabilities. Additionally, “Bard Advanced” will utilise Gemini Ultra from early next year, marking a significant advancement in chatbot technology. While Gemini Pro is noted to exceed the performance of GPT-3.5, its comparison with GPT-4 remains ambiguous, although Gemini Ultra has shown superiority in some benchmarks.
The release of Gemini coincides with ongoing discussions about Google's AI monetisation strategies. Eli Collins, vice president of product at Google DeepMind, hinted at Gemini's unique capabilities, which are still being explored.
Despite being Google's largest AI model, Gemini Ultra is noted for its cost-effectiveness and efficiency. Additionally, Google introduced a new tensor processing unit, the TPU v5p chip, enhancing AI model training performance.
As Google continues to push boundaries in AI, anticipation grows around how these advancements will be applied in real-world scenarios and contribute to the company's profits, particularly in AI-enhanced search experiences.