At Google I/O, it was all about AI in Gemini form

Last night, I gave the keynote at Google's developer conference, Google I/O. Last year they told us about them Artificial intelligence plans, but it was a bit scattered. Since then, it has become clearer with the launch of Gemini in December 2023. Development has gone very quickly since then. Now Google has shared its plans for the near future.

Gemini 1.5 Pro

First and foremost, Gemini Advanced subscribers have access to Gemini 1.5 Pro, and is now available in 35 languages. With its 1 million token capacity, it can build its answer for you on up to 1,500 pages of PDF files, 30,000 lines of code, or video in one hour. Google also revealed that the capacity will be increased to 2 million tokens!

Workspace

Gemini comes too Built into applications Such as Gmail, Docs, Sheets, Slides, and Drive. Examples were shown of how Gmail can understand and track multiple concurrent email threads and, for example, summarize offers from multiple senders. But it will also be able to work across apps, so you can, for example, automatically save receipts from emails to a folder on Drive and let Kalkylark record and categorize them.

Android

Since earlier this year, after receiving smarter searches with the help of artificial intelligence, Android in the future will have “AI as a foundation.” The Gemini Nano, with the ability to use text, audio, and images as input and output data, will arrive “later this year,” which we interpret to mean it will launch on the Pixel 9 this fall. It has been shown how the research department can now solve physics problems, explain the causes, and can warn of phone call fraud (at least in English).

Web services

Naturally, Google's growing web services are getting smarter. Before everything comes seek Ability to use real-time information, understand multi-step reasoning, and summarize responses from many different sources. For example, it was explained how to answer “Find the best yoga studios within 10 minutes of my house and what introductory offers they have.”

Google images It will be able to understand the context so that one can, for example, ask “What is my registration number?”

Artificial intelligence tools

Google also showcased new versions of Multimedia generation tool. Their new model Veo creates videos that look more realistic than ever before. The Imagen 3 text-to-image tool understands more detailed descriptions and you shouldn't have as much trouble generating text. Finally, it's explained how Music AI Toolbox can help music creators in the creative process.

It was really impressive though Astra Projectwhere you can ask questions in real time about what the camera is showing:

For more details, you can watch the entire two-hour presentation here:

(Born in 1969): Journalist and news editor. Jonas has been working at Ljud & Bild since 2007 on most things that can be called portable. That is, mobile phones, wireless speakers and headphones. Sometimes the test overlaps with colleagues' areas for multi-room audio, home cinema, and photos. He began his career as a technology journalist at IDG and wrote, among other things, for PC for All, Internetworld, and what later became M3.