Google announces Gemini 1.0, its largest and ‘most capable’ AI model

OpenAI’s dominance in the generative AI field could potentially wane, as Google announced on Wednesday the launch of its most advanced large language model to date, Gemini 1.0. “A new generation of AI models, inspired by how people understand and interact with the world,” is how CEO Sundar Pichai described Gemini in a blog post.

Born from a joint effort between Google’s DeepMind and Research divisions, Gemini boasts a comprehensive suite of cutting-edge generative AI features. Pichai extolled its abilities, stating that it is state-of-the-art in nearly every domain.

Unlike classic models that adopt an integrative approach, Gemini was designed from scratch as a multimodal AI, pre-trained and fine-tuned, to reason about all types of inputs from the outset, thereby outperforming existing multimodal models.

Gemini is also adept at coding, proficient in popular languages like Python, Java, C++, and Go. Google even employed a specialized version of Gemini to develop AlphaCode 2, the successor to last year’s award-winning generativeAI that resolved twice as many challenge questions as its predecessor.

While Google has yet to divulge Gemini’s parameter count, the company praised the model’s operational versatility and its ability to function in various forms from large data centers to local mobile devices. To achieve this conversion, three versions of Gemini will be available: Nano, Pro, and Ultra.

Nano, the smallest version, is primarily designed for on-device tasks. It’s followed by Pro, more versatile than Nano, and set to be incorporated into many of Google’s products including Bard, which on Wednesday, begins using an optimized version of Pro that’s supposedly more advanced in reasoning and understanding.

The improved Bard chatbot will be accessible in the same 170 countries as its regular counterpart, and its roll-out is planned for further expansion throughout 2024. Google also plans to launch Bard Advanced in the coming year with the debut of Gemini Ultra, boasting an even more powerful AI with additional features.

Pro’s capabilities can also be accessed via API calls through Google AI Studio or Google Cloud Vertex AI. Google’s Search, Ads, Chrome, and Duet AI will have Gemini features integrated into their functionalities in the months ahead.

Gemini Ultra, however, won’t be accessible until at least 2024 as it requires further red-team testing before its proposed release to stakeholders for subsequent testing and feedback. Once launched, Ultra is expected to be an extremely potent tool for further AI development.