Google already unveils Gemini 1.5, here are the new features!

Google already unveils Gemini 1.5, here are the new features!


Gemini (Google Bard)

The long-awaited Gemini (Google Bard), the conversation agent powered by AI and developed by Google, will be deployed in France this Thursday, July 13, 2023, in the form of a free online service.

  • Downloads:
    11212
  • Date of publication :
    02/08/2024
  • Author :
    Googling
  • License :
    Free permit
  • Categories:

    AI

  • Operating system:

    Online service

Google is catching up in the field of generative artificial intelligence. After making available its Gemini Pro model at the beginning of February, renamed Bard “Gemini” and launching an Android/iOS application last week, the American company is formalizing a new generation of its language model: Gemini 1.5.

Showing dramatic improvements in a number of dimensions, 1.5 Pro achieves quality comparable to 1.0 Ultra while requiring less computing power.

Googling

In a blog post published on Thursday, February 15, 2024, Google first explains that the Gemini 1.5 Pro model has a pop-up of 128,000 tokens, compared to 32,000 with Gemini 1.0 Pro.

Advertising, your content continues below

Specifically, AI can process a greater amount of information at once,”including 1 hour of video, 11 hours of audio, databases of over 30,000 lines of code or 700,000+ words”.

The larger a model’s pop-up window, the more information it can absorb and process within a given time frame, making results more consistent, relevant and actionable.

Googling

Gemini 1.5 Pro is therefore at the same level as GPT-4 Turbo (128,000 tokens), while Claude 2.1 remains well in the lead (200,000 tokens).

To achieve this feat, we learn that Google has opted for a Mix of Experts (MoE) architecture. Simply put, the model is divided into several specialized neural networks that are selectively activated to accurately respond to a request. By enabling only the expert areas of the model, Gemini 1.5 Pro significantly gains execution speed while maintaining a high level of precision.

Gemini 1.5 Pro is currently reserved for developers and businesses through AI Studio and Vertex AI. Google emphasizes that it has already made support for pop-ups up to one million tokens available to select testers, and has “successfully tested up to 10 million tokens”. Something to make professionals dream!

Advertising, your content continues below

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *