Gemma 2, Gemini 1.5 Flash and Pro, powerful AI image generator: What AI products were shown to us at Google I/O 2024 event?

May 15, 2024  12:03

Google announced a slew of exciting products and updates at its annual developer conference, Google I/O 2024, from AI and machine learning initiatives to new TPU processors.

The Tech editors have collected for you all the most interesting AI products presented by Google.

Gemini Updates

One of the most interesting new products is the open-source artificial intelligence model Gemma 2, which includes 27 billion parameters. Its launch is expected in June this year.

27 billion parameters is a significant improvement from the Gemma 2B and Gemma 7B versions released earlier this year, with 2 billion and 7 billion parameters, respectively. According to Google Labs Vice President Josh Woodward, Gemma 2 will offer industry-leading performance in a compact size by being optimized to run on next-generation Nvidia GPUs or a single Google Cloud TPU host on Vertex AI.

The Gemini line of artificial intelligence models has also been expanded with the new Gemini 1.5 Flash model, focused on tasks requiring high speed: the model can process data almost at lightning speed, without delays. The neural network can process text, images and video at high speed and is suitable for applications that require instant responses in real time. It can be used, for example, to communicate with users or clients, or to instantly generate simple images.

And for tasks that do not require very quick answers, the improved Gemini 1.5 Pro model, which can analyze large volumes of text, make generalizations and translations, is better suited. As reported by The Verge, both models use a context window of 1 million tokens, which allows more information to be taken into account when generating responses. By comparison, the GPT-4 context window is 128,000 tokens.

Imagen 3 and other AI tools

Another interesting announcement is a new version of the generative neural network of the Imagen family. Billed as Google's most advanced image generator, the new Imagen 3 understands text queries more accurately, generates more detailed images, makes fewer mistakes and, according to Demis Hassabis, head of Google's AI research unit, creates fewer "distracting artifacts" "

To prevent Imagen 3 from being used to create deepfakes, SynthID technology is used in the image generation process - invisible cryptographic watermarks are applied to media files.

Another interesting innovation is the Veo AI model, with which you can create video clips in 1080p resolution of about a minute, based on a text description. It is possible to use different visual and cinematic styles and edit the generated frames.

Gemini integration into Google services

Google plans to add more AI capabilities to its search engine. Specifically, some search results will have fully AI-generated reviews. And the Ask This Video feature will allow users to use Gemini to search for specific information within a YouTube video.

Gemini will also be integrated into Gmail, allowing users to search, summarize and draft emails. AI is expected to be able to perform more complex tasks, such as processing product returns in an online store.

Android 15 will feature Gemini Live, allowing users to have full voice conversations with an AI assistant that can see and react to the user’s surroundings through the smartphone’s camera.

Gemini Nano, Google's smallest AI model, will be built directly into the Chrome desktop client starting with version 126, allowing developers to use the on-device AI model to create their own features.

In Google Maps, developers will be able to use Gemini's capabilities to create AI descriptions of places and areas based on data from the Google Maps community.

  • Archive