Today is the day you’ve been waiting all year for: the Google I/O address! Every year, Google starts its developer meeting with a lot of announcements, including many shows of new things it’s been working on. Brian started things off by telling us what to expect.
You might not have had time to watch the whole two-hour presentation on Tuesday, so we did it for you. We gave you quick hits of the most important news from the keynote as they were released in a list that was easy to read and skim. Let’s go!
Genkit For Firebase
Firebase Genkit is a new part of the Firebase platform that makes it easy for developers to make AI-powered apps in JavaScript and TypeScript. Soon, Go will also be able to use Firebase Genkit. Using the Apache 2.0 license, it’s an open source framework that lets writers quickly add AI to both new and old apps.
There are a lot of common uses for Genkit that the company is talking about on Tuesday. These include creating and summarizing material, translating text, and making images.
AI Over And Over
Of course, Google used Gemini to keep track of how many times AI was mentioned at #GoogleIO today. There was even one more after this.
Google I/O on Tuesday lasted 110 minutes, but the company said 121 times that AI was important during that time. At the end of the talk, CEO Sundar Pichai brought up the number and joked that the company was “doing the hard work” of counting for us. We were ready for it, so it wasn’t a surprise.
Generative AI To Help People Learn
Google also showed off LearnLM today, a new group of creative AI models that have been “fine-tuned” for learning. Google’s DeepMind AI development team and Google development are working together on it. Google says that LearnLM models are made to teach kids a wide range of subjects in a “conversational” way.
Google Classroom is being used as a test bed for LearnLM, even though it is already available on some of the company’s other sites. Along with that, it works with teachers to find ways that LearnLM can make planning lessons easier and better. Google says LearnLM could help teachers find new ideas, content, and activities, or it could help them find things that are perfect for a certain group of students.
When we talk about learning, YouTube now has tests made by AI. People can “raise their” hands when watching educational movies with this new conversational AI tool. People can get helpful answers, ask questions to get more information, or take a quiz on the subject.
Given that Gemini model can handle long-context, this will be helpful for people who have to watch longer educational films like lectures or seminars. A small group of Android users in the U.S. are now getting these new tools.
Updates On Gemma 2
Google heard that devs wanted a bigger Gemma model a lot, so they are adding a new 27-billion-parameter model to Gemma 2. The next version of Google’s Gemma phones will come out in June. Google said that this size was chosen by Nvidia to work best with next-generation GPUs and can also work well with a single TPU host and vertex AI.
Google Play
People are interested in Google Play because it has a new feature that lets you find new apps, new ways to get new users, updates to Play Points, and other improvements to tools for developers like the Google Play SDK Console and Play Integrity API.
A new tool called the Engage SDK is very interesting to developers because it will let app makers show off their content to users in a full-screen, engaging experience that is tailored to each user. However, Google says that people can’t see this surface right now.
How To Spot Scams On The Phone
During the call on Tuesday, Google showed off a tool that it thinks will help users avoid falling for scams.
The feature will be added to a later version of Android. It uses Gemini Nano, which is the smallest version of Google’s generative AI software and can be run completely on the device. In real time, the system does a good job of listening for “conversation patterns commonly associated with scams.”
Google gives the example of someone telling lies about being a “bank representative.” The system will also react to common scams like asking for passwords and gift cards. The world is still full of people who can fall for these types of scams, even though most people know they are attempts to steal your money. Once it’s set off, it will show a warning that the user might be falling for bad people.
Look At Photos
With the launch of an experimental feature called “Ask Photos,” which is driven by Google’s Gemini AI model, Google Photos is getting some AI help. When the update comes out later this summer, users will be able to look through all of their Google Photos by using natural language queries that use an AI’s knowledge of the photos’ content and other metadata.
Before the AI update, users could look for specific people, places, or things in their photos. Now, with natural language processing, it will be easier to find the right content without having to do as much searching by hand.
The example was also cute. “Golden Stripes” is the name of a band made up of a tiger toy animal and a Golden Retriever.
Everything About Gemini
The Gemini In Gmail
Thanks to its Gemini AI technology, Gmail users will be able to find, summarize, and write their emails. More advanced chores will also be possible with emails. For example, it can help you process an e-commerce return by searching your inbox, finding the receipt, and filling out an online form.
The Gemini 1.5 Pro
Gemini can now look at longer documents, codebases, videos, and audio records than it could before. This is another improvement to the generative AI.
A secret look at a new version of Gemini 1.5 Pro, the company’s top-of-the-line model right now, showed that it can hold up to 2 million tokens. That’s twice the most that could be spent before. At that level, the new version of Gemini 1.5 Pro can handle the most data of any model on the market.
Be A Gemini
The company showed off a new feature in Gemini called Gemini Live. This feature lets Gemini users have “in-depth” voice talks on their phones. Gemini is a robot that users can ask questions of while it’s talking, and it will change based on how people talk in real time. Gemini can also see and react to what’s going on around its users by using photos or videos taken by their phones.
At first glance, Live doesn’t look like a big step up from what’s already out there. But Google says it uses newer techniques from generative AI to make picture analysis better and less likely to make mistakes. It also says it combines these techniques with a better speech engine to make multi-turn dialogue more consistent, emotionally expressive, and realistic. Read more
No. 2 Gemini
Now for a short message. The smallest of Google’s AI models, Gemini Nano, is also being built right into the Chrome desktop viewer, starting with Chrome 126. The business says that this will let devs use the on-device model to power their own AI projects. Google wants to use this new feature to power things like the “help me write” tool from Workspace Lab that is already in Gmail.
Gemini On Your Phone
Google’s AI replacement for Google Assistant, Gemini on Android, will soon be able to fully utilize its ability to seamlessly connect with Google’s apps and the Android operating system. People will be able to drag and drop pictures made by AI right into Gmail, Google Messages, and other apps. Google says that YouTube users will be able to tap “Ask this video” to get specific details from a YouTube video.
Find Gemini On Google Maps
Starting with the Places API, developers will soon be able to use the Gemini approach on the Google Maps platform. Place and area highlights made by generative AI can be shown in the apps and websites of developers. The reports are made by Gemini using information from the more than 300 million people who use Google Maps and contribute to the community. Which is better? No longer will developers have to write their own accounts of places.
Tensor Processing Units Work Faster Now
Tensor Processing Units (TPU) chips from Google were shown off. They are the sixth version of these chips. Later this year, they will go on sale under the name Trillium. As you may remember, it’s become a tradition to talk about the next version of TPUs at I/O, even though the chips don’t come out until later in the year.
The compute speed of each of these new TPUs will be 4.7 times better than that of the fifth generation. Trillium has the third version of SparseCore, which Google calls “a specialized accelerator for processing ultra-large embeddings common in advanced ranking and recommendation workloads.” This may be even more important.
AI In Search
People who think Google is losing market share to companies like ChatGPT and Perplexity are wrong. The company is putting more AI into its search engine to calm people down. It is giving users in the U.S. overviews that are driven by AI this week. The company also wants to use Gemini as an agency to help with things like planning trips.
Generative AI is what Google wants to use to organize the whole search results page for some queries. That’s on top of the AI Overview feature that already exists and gives you a short summary of information about a subject you were looking for. After being part of Google’s AI Labs program, the AI Overview feature is now open to everyone as of Tuesday.
Generative AI Gets Better
Google unveiled Imagen 3, the newest member of its Imagen family of creative AI models.
Demis Hassabis, CEO of DeepMind, Google’s AI research group, said that Imagen 3 gets the text prompts that it turns into images better than Imagen 2, and that it is “more creative and detailed” than earlier generations. He also said that the model makes fewer mistakes and “distracting artifacts.”
Hassabis also said, “This is our best model yet for rendering text, which has been hard for models that make images.”
Project IDX
The next generation of the company’s browser-based creation environment for AI, called Project IDX, is now in open beta. With this update, the IDE now works with the Google Maps Platform, which lets developers add geolocation features to their apps. It also works with the Chrome Dev Tools and Lighthouse, which help developers fix their apps. Google is soon going to make it possible to launch apps to Cloud Run, which is Google Cloud’s serverless platform for running front- and back-end services.
Veo
Google is going after OpenAI’s Sora with Veo, an AI model that can use text to make 1080p videos about a minute long. Veo can record different types of visual and cinematic styles, such as time lapses and shots of scenery. It can also edit and fix footage that has already been made.
It also builds on Google’s early business work in video generation, which was shown off in April and used the Imagen 2 family of image-generating models to make video clips that play over and over again.
Click The Circle To Look
More difficult psychic and math word problems can now be solved by the AI-powered Circle to Search tool on Android. This feature lets users get answers right away by making gestures like circles. That way, you can use Google Search from anywhere on your phone by doing something, like clicking, circling, highlighting, or scribbling. Also, it’s better to help kids with their homework from Android phones and computers that can do it.
Pixel 8a Google couldn’t wait until I/O to show off the newest Pixel phone, so they announced the Pixel 8a last week. The phone starts at $499 and will be sent out on Tuesday. The changes are also what we’ve come to expect from these updates. Adding the Tensor G3 chip is at the top of the list.
Pixel Slate
The Pixel Tablet from Google, which is now called Slate, can be bought. Back in August of last year, Brian reviewed the Pixel Tablet around this time. All he talked about was the base. It’s interesting that you can buy the computer without it.
We will be adding to this post all day…
Also Read: Google Fires People From the Flutter, Dart, and Python Teams Just Weeks Before Its Developer Meeting
We are starting an AI blog! If you sign up here, you’ll start getting it in your email on June 5.
What do you say about this story? Visit Parhlo World For more.
