Table of Contents 64q43
About a month and a half after releasing its language model to the general public, the Google today announced the update for Gemini 1.5. Among the new features, it is faster and has the capacity to read longer content. O Gemini 1.5 will be available to developers with up to one million tokens. Check out what was announced. 1p3x1h
What comes with the update 6x719
The announcement of Google for your language model it is more of an incremental change than a new release. Compared to the previous model, the Gemini 1.5 arrives on the internet with a new architecture called Mixture-of-Experts (MoE). In practice, language models will be divided into small neural networks for better processing and through “specialized” networks, instead of being read by giant parameters.
This new feature directly impacts task processing, which is now faster and ready to perform more complex tasks. O Google also highlights that training Gemini it's faster. It is important to that this AI has three variants: Nano (on Pixel smartphones), Pro (available on the Gemini chatbot) and Ultra (available in Gemini Advanced for developers).
O Google states that the Gemini 1.5 Pro has performance power that resembles the Ultra 1.0, and that it is 87% faster than what was presented at the end of last year. The company claims that, according to tests carried out internally, it is faster and can process more information than the previous generation.
When tested against a comprehensive of text, code, image, audio, and video assessments, 1.5 Pro outperforms 1.0 Pro in 87% of the benchmarks used to develop our LLMs.
Google about new release of Gemini 1.5.
Greater processing capacity 1w2y48
The context window is basically the place where information is processed into tokens. It is important to that a token is not exactly a parameter, but rather blocks of data used to process information. The December 2023 version had for up to 32 thousand tokens, which was already a lot.
However, the Gemini 1.5 now has a context window of up to 1 million tokens. The processing power is so great that, in tests released by the company, artificial intelligence was able to read an entire 402-page PDF file about the Apollo 11 mission, which took man to the Moon, and bring three comical moments, in just a few seconds. . Watch the video below:
Another example shown was the review of a 44-minute silent film. Buster Keaton. Upon receiving the file, the artificial intelligence of the Google managed to identify a specific scene and provide more information about the moment in the film. Check out:
With this maximum amount of 1 million tokens, updating the Gemini will allow artificial intelligence to analyze up to 1 hour of video, 11 hours of audio, 30 thousand lines of code or more than 700 thousand words. And, as you noticed in the examples, this will be done in just a few seconds.
General public availability 6j4g1v
For the first time, only developers should take advantage of the update. Gemini, given that it will only be available via AI Studio ou Vertex A.I in a limited context.
It is also important to that updating the language model of the Google is still under development and the company guarantees developers with early access 128 thousand tokens for testing — batches of 1 million tokens will have restricted access and will be released gradually.
Early testers can try out the 1 million token context window at no cost during the testing period, although they should expect longer latency times with this experimental feature.
Google about new release of Gemini 1.5.
In the not-so-distant future, the language model will be implemented to Google, including personal assistant and chatbot, who stopped being called Bard and now goes by the name Gemini. How soon do you think we will use the new update? Gemini? tell us Comment!
See also other features
How to create images in Bard, Google's AI
With information: Google Blog l Google Brazil l The Verge
reviewed by Glaucon Vital in 15 / 2 / 24.
