Google Launches Gemini 1.5, Surpassing GPT
Google has launched a new version of its AI model family, a Gemini update which is the Gemini 1.5, equipped with high capabilities, enabled to perform up to 10 times better than the GPT-4 Turbo, developed by its rival, OpenAI.
The tech giant said in a statement that the new model it is testing within the new version is Gemini 1.5 Pro, with a capacity of 128.000 tokens, a maximum measurement unit available for commands requested by users.
The Alphabet-owned company is testing a capacity of up to one million tokens, making this model the greatest intelligent model ever. The company pointed out that these tokens can be available for trial to several researchers and companies within its AI Studio and Vertex AI services.
Google also explained that Gemini update to the 1.5 version will allow users to input commands consisting of a one-hour video, 11 hours of audio clips, more than 30.000 lines of code, or a text command exceeding 700.000 words.
Currently, Google’s researchers are testing an initial version with a capacity of up to ten million tokens. The new sophisticated model successfully analyzed a document of 402 pages related to detailed discussion between astronauts of the Appollo 11 mission to the moon.
Gemini update 1.5 also analyzed a 44-minute silent film to understand its context and details, so it answered detailed questions about it.
The California-based company explained that its new foundation model makes it easier to understand and analyze code, even more than 100.000 code line in one command, since Gemini 1.5 can suggest modifications and clarify the way these codes work as well as their importance.
According to Google, the Gemini 1.5 model has demonstrated highly efficient handling of commands that include diverse content such as code, text, images, audio clips, and videos, reaching an effectiveness of 87%, compared to Gemini 1.0 Pro. Its performance was comparable to Gemini 1.0 Ultra, but with a significant decrease in energy consumption.
An Efficient Architecture
To build Gemini update to the 1.5 version, Google used an advanced architecture, a combination of Transformers models and the Mixture of Experts. The latter is the latest component was used by the American company to develop its smarts models.
This approach achieves higher efficiency and speed of performance and lowers the level of energy and resources required to operate this model.
In parallel, Google has indicated that it will soon start offering paid packages for using the Gemini 1.5 Pro model to developers and the corporate sector in its AI Studio and Vertex AI services, with capacities ranging from 128,000 tokens to one million tokens.
Inside Telecom provides you with an extensive list of content covering all aspects of the tech industry. Keep an eye on our Intelligent Tech sections to stay informed and up-to-date with our daily articles.