Google Gemini – what kind of beast is this?

Google Gemini – what kind of beast is this?

Google Gemini is the long-promised family of next-generation generative AI models from Google. It has been developed by specialists from DeepMind and Google Research. Google Gemini is presented in four variants: Gemini Ultra, a very large model; Gemini Pro, a model slightly smaller than the previous one; Gemini Flash, a faster, “distilled” version of Pro; and Gemini Nano, which includes two smaller models: Nano-1 and the slightly more powerful Nano-2, which operates in standalone mode. All models are multimodal. In other words, they can work with and analyze not only text. Google claims that they have been pre-trained and fine-tuned on various publicly available datasets: proprietary and licensed audio, images, and videos; codebases; and text in different languages. This distinguishes Gemini from models like LaMDA from Google, which was trained exclusively on text data. LaMDA cannot understand or generate anything beyond text. It is worth noting that the ethics and legality of training models on public data, in some cases without the knowledge or consent of data owners, are unclear. Google has a liability policy for AI to protect certain Google Cloud clients from lawsuits they may encounter, but this policy contains exceptions.

https://xrust.ru/news/310862-google-gemini-chto-za-zver-takoj.html