Google has finally showcased its formidable competitor to ChatGPT: Gemini. The highly anticipated unveiling of Google’s new Language Model (LLM) is now a reality, promising to be the most advanced AI to date. Surprisingly, contrary to speculations, Google has decided to launch its new AI model earlier than anticipated. However, it comes in three different sizes: Nano, Pro, and Ultra. Unfortunately, the most powerful iteration, Gemini Ultra, won’t debut until early next year.
Revolutionary Multimodal Capabilities
Gemini stands tall as an AI model surpassing all its rivals in major tests. It’s a multimodal model, capable of comprehending information from various sources—text, images, videos, audio, and even code. According to Google, it’s their “most flexible model yet.”
Please follow us on Twitter and Facebook
Outperforming GPT-4
Google reports that Gemini Ultra’s results outshine GPT-4 in 30 out of 32 widely used academic tests, slightly surpassing the percentages achieved by OpenAI’s GPT-4.
Unparalleled Language Mastery
Scoring an impressive 90.04% in Massively Multitask Language Understanding (MMLU), Gemini becomes the first model to surpass human experts in a test that amalgamates 57 subjects, including physics, history, medicine, ethics, and problem-solving capabilities.
A Novel Approach to Reasoning
Gemini is uniquely designed from scratch, adopting a distinct problem-solving approach. Its native multimodal capacity means it’s pretrained to seamlessly combine different modalities. In a demonstration video, Gemini showcases its real-time interpretation of drawings, object relations, and personalized song suggestions based on user instructions.
Unveiling AlphaCode 2
Introducing AlphaCode 2, a new code generation system. Google describes this system as excelling in complex mathematics and theoretical understanding of computer science. According to data, AlphaCode 2 outperforms 85% of participants, an upgrade from the 50% performance of AlphaCode 1.
Efficiency Revolution
While Google hasn’t officially shared the number of parameters, Gemini Ultra stands as the most efficient model ever created, delivering high performance while consuming minimal energy.
Advanced Integration and Future Rollouts
Gemini’s integration begins with Google Bard, as it transitions to Gemini Pro. The Google chatbot will start using the medium version of Gemini, available in English across 180 countries, with a European rollout in the coming months. Next year, Google plans to launch Bard Advanced, an upgraded version integrating Gemini Ultra, currently undergoing rigorous trust and security checks.
Integration into Google Ecosystem and Beyond
Apart from Bard, Gemini will feature in services like Search, Ads, Chrome, and Duet AI. Developers gain access to Gemini Pro through the API on Google AI Studio or Vertex AI starting December 13. Additionally, Gemini will be available in Google Pixel 8 Pro. AICore, a new service, allows app creators to harness AI capabilities, starting with Gemini Nano, the lightweight version.
Expanding Accessibility
Google hints at extending Gemini’s availability to other Android 14 devices in the future, though details remain unspecified.
Focused on Safety and Innovation
In Google’s pursuit of AI dominance, Gemini not only showcases impressive capabilities but also promises the most comprehensive safety evaluations. Google has collaborated with external experts to identify blind spots and implemented specific security classifiers to flag content involving violence or negative stereotypes.
A New Era for Google
Sundar Pichai, Google’s CEO, heralds Gemini’s arrival as the dawn of a new era. While Gemini’s arrival has been eagerly anticipated, the upcoming year also holds the promise of GPT-5. These advancements mark a continuous evolution in the field of AI, characterized by significant breakthroughs seemingly every month.
