Google’s AI chatbot for web "Bard" has been renamed "Gemini" after the underlying multimodal large language model (MMLLM) it is based on. Google also announced that the Gemini app will soon be rolling out for Android, while iOS users will be able to access it through the Google app.
Bard was initially based on the Pathways Language Model (PaLM), until December 2023 when Google announced Gemini.
“To reflect the advanced tech at its core, Bard will now simply be called Gemini,”CEO Sundar Pichai said in a blog post published February 8.
Pichai added that Gemini will soon be available on Google Cloud and Google Workspace apps such as Gmail, Docs, Sheets, Slides and Meet.
Google claims that Gemini is its most “capable” AI model yet and is the first to outperform human experts on MMLU (massive multitask language understanding), a benchmark test used to evaluate the knowledge and problem solving abilities of AI models.
Though PaLM 2 can handle multimodal inputs like images and code, it was primarily designed and trained for text-based tasks such as text summarization, text generation, question answering, and translation. On the other hand, Gemini is built as a multimodal LLM, which makes it much better at understanding and processing information from text, images, code, audio, and video.
Google has classified Gemini into three variants based on their capabilities. The most advanced variant, Gemini Ultra, is the largest and most capable model and is designed for complex tasks such as code generation, scientific document processing, and creative collaboration. It requires a significant amount of computational power.
The mid-sized version, Gemini Pro, is meant for tasks such as advanced reasoning. It is expected to be used in servers and high-end Pixel smartphones for handling real-time translation and question answering.
The third model, Gemini Nano, is the most efficient and is optimized for low resource tasks such as text generation and summarization on smartphones and smart speakers.
In addition to renaming Bard, Google also announced that users can access Gemini Ultra by subscribing to a premium version called Gemini Advanced. It has been bundled with Google One subscription and will cost Rs 1,950 per month with 2TB of cloud storage.
Though Gemini has been optimized only for English, the Advanced version can answer queries in up to 40 languages that Gemini is available in.
OpenAI's release of ChatGPT and Microsoft's subsequent integration of GPT-4 into its consumer and enterprise products including Bing search and Azure significantly impacted the AI landscape and prompted Google to rush its work on generative AI and launch Bard in March 2023.