Google launches Gemini, a direct challenge to OpenAI

Google launched Gemini, its new natural language model. Google Gemini is a direct challenge to OpenAI.  

Gemini is the company's most flexible LLM (large language model) yet and is optimized for three different sizes: ultra, pro, and nano, which differ in the weight of the model and the type of tasks they can perform.

Google has decided to launch Gemini on Bard in two phases. First, Bard will be updated with the Gemini Pro version adapted to offer the best results to users. It will be released first in English in 170 countries and then move on to include other languages and more countries. Afterwards, it will incorporate the Gemini Ultra version, thus offering the best AI model.

Google took longer than expected to launch Gemini due to the need to conduct various tests and evaluations to test its performance on various tasks. Also, as some sources claim, to be completely sure that the new LLM will be able to comply with the near future legislation regarding the use of AI-driven solutions. 

Some of the tests they have performed include natural image and audio comprehension, as well as mathematical reasoning. The Google team has observed that Gemini outperforms GPT 3.5 on several tasks.

What can Gemini do? 

The new LLM model from Google has a set of distinct functionalities:  

• Multimodal reasoning, enables you to understand complex written and visual data. This ability facilitates the extraction of valuable insights from large amounts of information, providing rapid digital advances in areas such as science and finance by processing hundreds of thousands of documents efficiently.

• Recognize and understand text, images, audio and more at the same time, so you better understand nuanced information and can answer questions related to complicated topics. This makes it especially good at explaining reasoning in complex subjects such as mathematics and physics.

• Multiple programming languages, such as Python, Java, C++ and Go, and stands out as a leading coding model. Gemini Ultra excels in major performance evaluations, such as HumanEval and Natural2Code. It also serves as the basis for advanced coding systems, such as AlphaCode and its evolved version, AlphaCode 2, capable of solving challenges beyond coding, such as mathematical and theoretical computer science problems.