Google jumps into the world of artificial intelligence and shows his model Twins, a technology that can process various forms of information such as video, audio and text. Alphabet CEO Sundar Pichaihas spoken of the “effort” put into creating this model, which will have three versions: Nano, Pro and Ultra.
Each of these versions is designed to consume a different amount of computing power. According to Google, the more powerful version is designed for operation in data centers, while the smaller version is intended to work efficiently on mobile devices.
The Pro version is now available in the wizard Google Bardwhich will receive the Ultra version in 2024. In addition, the Nano version will soon be released on the market Pixel 8 Proand promises to reach other company terminals throughout 2024.
Google claims that Gemini is “significantly” cheaper than other models and argues that it is “not only more powerful, but also more powerful.” much more efficient” said Eli Collins, Vice President of Product at DeepMind. Developing the Ultra model requires a lot of computing power, which Google is working on.
Alphabet also announced a new generation of its custom AI chips, also known as Tensor Processing Units (TPUs). The new version of its customer processors can train large language models almost three times faster than previous generations. The new chips will be available to developers in “preview” starting this Wednesday.
In its presentation, Google showed several videos demonstrating how Gemini works can tell what a person is painting (in this case a duck). The person paints this duck blue while Gemini tells him that it is not a common color for ducks, but that there may be some of that color.
He is also able to see and solve a mathematical problem correct exercisesexplain whether they are well solved or not and describe in detail any errors that may occur.