Google has unveiled its latest large language AI model called Gemini. It was first teased during this year’s I/O developer conference and it is being launched to the public in three different sizes. The tech giant seems to be positioning it as a competitor to OpenAI’s GPT-4 model, comparing the two in a number of benchmarks with the former coming out on top in most cases.
The first AI model is a light version dubbed Gemini Nano, which is built to run natively and offline on Android devices. This version is coming to the Pixel 8 Pro with two features, namely auto-summarisation in the Recorder app and Smart Reply on Gboard. While its compatibility is limited to the Tensor 3 chip for now, Google mentioned that Qualcomm, Samsung, and MediaTek are making chips capable of running Nano, meaning that it will eventually be available for other phones.
Next up is Gemini Pro, a more powerful but still relatively power efficient model which will power several of the company’s AI services and already serves as the backbone of the Bard chatbot. As for availability, developers and enterprise users will be able to access Gemini Pro through Google Cloud’s Generative AI Studio or Vertex AI starting on 13 December.
The top-tier AI model is Gemini Ultra, which is capable of multimodality, meaning instead of being limited to text input and outputs like the lower-end models, Ultra works with video, images, and audio. Google DeepMind CEO Demis Hassabis saying that it will become even more general in the future as it gains more senses. This particular model will be coming early next year to developers and enterprise customers along with Bard Advanced.
But how does it compare to GPT-4? Well, according to Google’s test, Gemini Ultra is capable of beating OpenAI’s model in 30 out of 32 [PDF] academic benchmarks used in large language model research. Moreover, the company claims that its new model is much more efficient due to it being trained on Google’s in-house Tensor Processing Units.
Unlike GPT-4, Gemini is currently only available in English but it will support other languages soon. CEO Sundar Pichai added that the new AI will even be integrated in its search engine to make it faster as well as the Chrome browser, ad products, and more globally, so it’s not a far stretch to say that Gemini will reach most internet users one way or another.
