Announced at Google I/O 2023 by CEO Sundar Pichai, Google's upcoming AI system, Gemini, is poised to be a game-changer. Developed by Google DeepMind, it promises to rival and possibly surpass existing language models like OpenAI's ChatGPT. Here's what we've gathered so far:
Gemini aims to be more than just a text-based model. Pichai revealed that it will be multimodal, fusing DeepMind’s AlphaGo capabilities with advanced language modeling. This allows for a richer conversational experience as it will integrate text, images, and other data types.
The system is also expected to feature advanced capabilities like memory and planning, hinting at the possibility of tasks that involve reasoning and decision-making.
Google’s Chief Scientist Jeffrey Dean mentioned that Gemini will leverage Google’s new AI infrastructure, Pathways, for scalable training. The model is anticipated to be larger than GPT-3, exceeding its 175 billion parameters.
DeepMind's CEO, Demis Hassabis, explained that Gemini would not be a one-size-fits-all model. It will come in various sizes and capabilities, incorporating techniques like reinforcement learning and tree search from AlphaGo.
Gemini is expected to employ memory and fact-checking capabilities, possibly using Google Search as a reference to enhance its accuracy and reduce the risk of disseminating false information.
Sundar Pichai envisions Gemini as a stepping stone towards creating "incredible universal personal assistants." These AI systems would be integrated into daily life, assisting in areas such as travel, work, and entertainment.
OpenAI and other competitors are closely monitoring Gemini's development. Google has also reportedly given a select group of external developers early access to the model, signaling that a beta release may be imminent.
Not to be outdone, Meta is also reportedly working on its own LLM to compete with OpenAI, having recently released the Llama 2 model in collaboration with Microsoft.
Gemini's launch could mark a paradigm shift in natural language processing and interactive AI. It combines DeepMind’s cutting-edge research with Google's massive computational resources, potentially fulfilling Google's ambition to responsibly democratize AI for billions.
