Google has officially introduced its highly anticipated large language model, Gemini, on Wednesday during a virtual press conference where leaders from Google DeepMind and Google Cloud showcased its capabilities and outlined the roadmap. Introduced during the spring Google I/O 2023 by CEO Sundar Pichai, Gemini stands out as the most advanced Large Language Model (LLM), surpassing its predecessor, PaLM 2—the current foundation model at Google.
It achieves a groundbreaking feat as one of the initial language models to achieve a score exceeding 90% on the Massive Multitask Language Understanding benchmark. This benchmark encompasses a diverse range of 57 subjects spanning STEM, humanities, social sciences, and beyond.
Features of Gemini
Versatility in Action
Gemini stands out for its adaptability, seamlessly running on both cloud and mobile devices. It’s a multimodal model, comprehending inputs in the form of text prompts, images, audio, or video. With unrivaled reasoning abilities, it can even identify errors in a student’s response to complex mathematical problems.
Advanced Training and Support
Trained on Google’s AI accelerator chip, Cloud TPU v5e, Gemini offers cost-efficiency and performance for medium- and large-scale training and inference. It supports leading AI frameworks like JAX, PyTorch, TensorFlow, and open-source tools such as Hugging Face’s Transformers, Accelerate, PyTorch Lightning, and Ray.
Next-Gen Accelerator and Infrastructure
Google has also announced Cloud TPU v5p, the next-gen AI accelerator, which is four times more scalable and 2.8 times faster. This infrastructure will be available on the Google Cloud Vertex AI platform for ML researchers and developers.
Three Flavors of Gemini
Gemini comes in three versions: Ultra, Pro, and Nano. While the Ultra model, the largest and most sophisticated, arrives early next year, Gemini Pro is integrated with Vertex AI, available to developers on December 13. Gemini Nano, a smaller version for Pixel 8 Pro and Android devices, releases on December 6 for developers.
Bard Advances with Gemini
Google’s chatbot Bard, set to rival ChatGPT, will be powered by Gemini Pro. With access to Google Search, it promises accurate and reasonable responses. Bard Advanced, backed by Gemini Ultra, is slated for early adoption next year through the Trusted Tested program.
Gemini’s Impact on Generative AI
Gemini marks a significant milestone in generative AI, poised to serve as the foundation for various Google services like Bard, Duet AI, and Google Cloud. The model’s remarkable capabilities open new frontiers in language understanding and application.
Also Read: Google Introduces PaLM 2, The Latest Advanced Language Model
