Google is set to launch its most extensive AI model to date, named Gemini.
Introducing Gemini, Google’s largest and most capable AI model. 🧵 #GeminiAI https://t.co/T0tIw9HQyO
— Google (@Google) December 6, 2023
Including three variants – Gemini Ultra, Gemini Pro, and Gemini Nano – the suite aims to accommodate diverse applications, with Gemini Ultra standing out as the first model to surpass human experts in massive multitask language understanding (MMLU). This breakthrough allows Gemini to comprehend and solve problems across 57 subjects, spanning math, physics, history, law, medicine, and ethics.
Gemini will be accessible to developers and enterprise customers starting December 13. Gemini Pro can be accessed via the Gemini API in Google AI Studio or Google Cloud Vertex AI, while Android developers gain access to Gemini Nano. The AI model will also power Google’s Bard chatbot and Search Generative Experience, enhancing conversational-style text responses to search queries.
Enterprises are assured to influence Gemini for advanced customer service engagement, product recommendations, trend identification, and content creation, including marketing campaigns and blog content. Google envisions applications in productivity tools that can summarize meetings or generate code for developers.
Gemini Ultra outperforms human experts on MMLU (massive multitask language understanding): one of the most popular methods of benchmarking AI models.
It involves a combination of 57 test subjects from math to history to law and more. ↓ https://t.co/mwHZTDTBuG pic.twitter.com/587naHIR1Q
— Google DeepMind (@GoogleDeepMind) December 6, 2023
Gemini’s capabilities extend beyond traditional AI models, as illustrated by examples demonstrating its ability to analyze screenshots of charts, interpret complex research documents, and even assist with math homework by identifying correct answers and pointing out errors.
Google CEO Sundar Pichai emphasized the collaborative effort that led to Gemini’s development, highlighting its multimodal design that seamlessly integrates different types of information such as text, code, audio, image, and video. The Gemini Ultra model, in particular, boasts superior performance in understanding nuance and reasoning across varied domains.
While Gemini’s launch was delayed, Google assures users that it has undergone rigorous testing, ensuring comprehensive safety evaluations. Eli Collins, VP of Product at Google DeepMind, acknowledged Gemini’s efficiency, stating that despite being the company’s largest model, it is notably cost-effective to operate.
The introduction of Gemini follows Google’s earlier Bard chatbot launch, with a forthcoming update, “Bard Advanced,” set to utilize Gemini Ultra for enhanced capabilities. Google remains tight-lipped on monetization details for Bard Advanced, emphasizing a focus on delivering a positive user experience.
In addition to Gemini, Google announced its next-generation tensor processing unit, the TPU v5p chip, which promises improved AI model training performance compared to its predecessor. This announcement comes in the wake of similar revelations from cloud rivals Amazon and Microsoft, who showcased custom silicon for AI applications.
Investors have been keenly interested in Google’s plans to turn AI into profitable ventures, prompting the company to experiment with the Search Generative Experience (SGE). Although SGE is still in the early stages, Google plans to incorporate Gemini into it in the coming year, hinting at the vast opportunities these advanced AI models may unlock.
AI-related story: Elon Musk AI Startup Aims High with $1 Billion Bid
