Google announced the launch of Gemini, its largest and most advanced artificial intelligence model yet, as it seeks to reestablish itself as an AI leader after the meteoric rise of chatbot ChatGPT developed by OpenAI.
Gemini is a multimodal model, meaning it can understand and interact with text, images, audio and video. This sets it apart from other popular AI systems like ChatGPT that can only process text. Gemini will be integrated into Google’s existing products over the next year, including the Bard conversational AI service, Google Cloud platform, Pixel phones, and potentially Search and advertising.
There are three versions of Gemini targeted at different applications. Gemini Nano is optimized for on-device processing on smartphones and other consumer devices. Gemini Pro delivers strong language understanding capabilities across a wide range of tasks. Gemini Ultra, launching next year after extensive testing, is the most powerful version designed to tackle Google’s toughest AI challenges.
Executives stated Gemini Pro outperformed GPT-3.5, the model that originally powered ChatGPT. They believe Gemini Ultra exceeds the capabilities of OpenAI’s latest model GPT-4 based on internal benchmark testing, particularly in areas like mathematical reasoning. However, they did not provide the full technical details to allow independent verification.
The launch of Gemini reflects Google’s renewed prioritization of AI throughout its products and services after largely watching from the sidelines as OpenAI seized the spotlight in recent years. Google CEO Sundar Pichai called it the beginning of the “Gemini era” and claimed the technology will have a bigger impact than fire or electricity.
Google hopes Gemini will help it catch up in advanced conversational AI. The current Bard chatbot is receiving an upgrade to Gemini starting today to improve its contextual reasoning and planning abilities. A more powerful Bard integration powered by Gemini Ultra is slated for next year.
Gemini took several months to develop utilizing resources and researchers across Google and sister-company DeepMind. Google believes Gemini is much more cost-efficient than predecessor models like PaLM and can run on its latest TPU v5 chips tailored for AI workloads.
As with all large language models, reliability and bias issues remain a concern. Google states Gemini underwent extensive safety testing and external red-team penetration testing will continue as capabilities progress. Responsible development will be critical with technology edging closer towards artificial general intelligence.