Google announced a new version of Gemini artificial intelligence: Gemini 1.5 Flash is higher and more efficient for multimodal applications, in line with a giant technology. This is not a unique novelty for Google I/O 2024, even carried out from the third day (05/14) in the United States.
Many conversations were held with journalists who participated, CEO Sundar Pichai said that Google is investing in artificial intelligence over a decade ago. “We have a lot of opportunities going for us,” said the number one executive at a time when market analysts are convinced of the business conditions of competing with ChatGPT and other recent technologies.
No event, Google only announced one you will have to respond to the AI of our United States e presentor o Projeto Astra, an AI that has been everything and has seen itself in its corners.
Gemini 1.5 Flash
Google I/O 2024 is proof that Google is moving to meet high expectations.
Both Gemini 1.5 Flash and Gemini 1.5 Pro contain a context pair of 1 million tokens. This number corresponds to the capacity dimension of the lidar model with complex prompts and pricing. The comparison title, Claude bought 200 million tokens, while GPT-4 goes to 128 million tokens and the Gemini app goes to 32 million tokens.
It was said that Gemini 1.5 Pro would earn a mark of 2 million tokens until the end of the year. The executive does not specify any data.
The different models have lidar capabilities with rates for translation, dialogue, programming, logic and writing. In the case of the Flash version, it is proposed to create CVs, conversations (like chatbot), photo and video captions, and the extraction of long documents or tables. “This is possible because we learned Flash from Gemini 1.5 for the distillation process” because knowledge from a major model (professor) is passed back to a minor model (aluminum) preserving the most important information .
Other Gemini Activities
The Gemini 1.5 Pro model has also been improved. Google offers a particular programming capability, or logical logic, as well as a capability for managing long-term conversations with ideas and business. He will be released first for the murderers of Advanced Geminiplan integrated with Google One with more AI resources.
Assassins can send Google Drive files or attachments from the device, for the AI to consume the content and send the requested responses. Google says the archives are sealed and are not used to train artificial intelligence models.
From the Gemini 1.0 Nano, considered the company's highest model, you will also be able to understand the images. Today it is limited to pure text. A novelty that begins with the mobile phones of the Google Pixel line. We support you precisely to discover the large-scale manufacturers, such as Samsung and Motorola, who will embark on their smartphones.
Gemma 2
The Gemini line models are proprietary. This means that interested companies and investors are not exactly on board with Google, which they normally move to using APIs with a new service. So you know, o Google mantém o Gemma, model aberto, nosmos moldes do Llama 3 (meta), Phi-3 (Microsoft) e Grok (X/Twitter).
Ultimately, Google reveals an update for Gemma 3, which creates a new architecture. We have now made the LLM faster and more efficient. He will be released from several tamanhos, as details are not initially presented.
Image 3 and Veo
For other language models, Google is introducing the Veo video generation process and a new version of the image generation process.
The video is capable of creating videos with high resolution (Full HD) and a duration of more than 1 minute. Depending on the company, ferramenta follow varied visual styles. Google promises impressive proficiency in understanding the director's creative vision. The disso part must also be prompted for longer.
Then the company began using “timelapse” or “aerial landscape footage” instructions and making people, animals and objects “move realistically in their gravitations.”
A ferramenta Imagen já é conhecida nossa. It is now a matter of generating 3 quality and fidelity improvements for image generation. Google confirms that users have images identical to reality. Image 3 will also be able to colocar speak phrases and sentences in the figures, where they are if a gold of the head is now (which I use Dall-3 to do that is or falando).
