ArduinoMania

The Google I/O event is here, and the company is announcing lots of great updates for your Android device. As we heard earlier, Gemini Nano is getting multimodal support, meaning your Android will still process text but with a better understanding of other factors like sights, sounds and spoken language. Now Google has shared that the new tool is also coming to it's TalkBack feature.

TalkBack is an existing tool that reads aloud a description of an image, whether it's one you captured or from the internet. Gemini Nano's multimodal support should provide a more detailed understanding of the image. According to Google, TalkBack users encounter about 90 images each day that don't have a label. Gemini Nano should be able to provide missing information, such as what an item of clothing looks like or the details of a new photo sent by a friend. 

Gemini Nano works directly on a person's device, meaning it should still function properly without any network connection. While we don't yet have an exact date for when it will arrive, Google says TalkBack will get Gemini Nano's updated features later this year.

Catch up on all the news from Google I/O 2024 right here!

Google just announced some nifty improvements to its Gemini AI chatbot for Android devices as part of the company’s I/O 2024 event. The AI is now part of the Android operating system, allowing it to integrate in a more comprehensive way.

The coolest new feature wouldn’t be possible without that integration with the underlying OS. Gemini is now much better at understanding context as you control apps on the smartphone. What does this mean exactly? Once the tool officially launches as part of Android 15, you’ll be able to bring up a Gemini overlay that rests on top of the app you’re using. This will allow for context-specific actions and queries.

Google gives the example of quickly dropping generated images into Gmail and Google Messages, though you may want to steer clear of historical images for now. The company also teased a feature called “Ask This Video” that lets users pose questions about a particular YouTube video, which the chatbot should be able to answer.

It’s easy to see where this tech is going. Once Gemini has access to the lion’s share of your app library, it should be able to actually deliver on some of those lofty promises made by rival AI companies like Humane and Rabbit. Google says it's “just getting started with how on-device AI can change what your phone can do” so we imagine future integration with apps like Uber and Doordash, at the very least.

Circle to Search is also getting a boost thanks to on-board AI. Users will be able to circle just about anything on their phone and receive relevant information. Google says people will be able to do this without having to switch apps. This even extends to math and physics problems, just circle for the answer, which is likely to please students and frustrate teachers.

Google has introduced another capability for its Circle to Search feature at the company's annual I/O developer conference, and it's something that could help students better understand potentially difficult class topics. The feature will now be able to show them step-by-step instructions for a "range of physics and math word problems." They just have to activate the feature by long-pressing the home button or navigation bar and then circling the problem that's got them stumped, though some math problems will require users to be signed up for Google's experimental Search Labs feature.

The company says Circle to Search's new capability was made possible by its new family of AI models called LearnLM that was specifically created and fine-tuned for learning. It's also planning to make adjustments to this particular capability and to roll out an upgraded version later this year that could solve even more complex problems "involving symbolic formulas, diagrams, graphs and more." Google launched Circle to Search earlier this year at a Samsung Unpacked event, because the feature was initially available on Galaxy 24, as well as on Pixel 8 devices. It's now also out for the Galaxy S23, Galaxy S22, Z Fold, Z Flip, Pixel 6 and Pixel 7 devices, and it'll likely make its way to more hardware in the future. 

In addition to the new Circle to Search capability, Google has also revealed that devices that can support the Gemini for Android chatbot assistant will now be able to bring it up as an overlay on top of the application that's currently open. Users can then drag and drop images straight from the overlay into apps like Gmail, for instance, or use the overlay to look up information without having to swipe away from whatever they're doing. They can tap "Ask this video" to find specific information within a YouTube video that's open, and if they have access to Gemini Advanced, they can use the "Ask this PDF" option to find information from within lengthy documents. 

Google is also rolling out multimodal capabilities to Nano, the smallest model in the Gemini family that can process information on-device. The updated Gemini Nano, which will be able to process sights, sounds and spoken language, is coming to Google's TalkBack screen reader later this year. Gemini Nano will enable TalkBack to describe images onscreen more quickly and even without an internet connection. Finally, Google is currently testing a Gemini Nano feature that can alert users while a call is ongoing if it detects common conversation patterns associated with scams. Users will be alerted, for instance, if they're talking to someone asking them for their PINs or passwords or to someone asking them to buy gift cards. 

Catch up on all the news from Google I/O 2024 right here!

As part of its push toward adding generative AI to search, Google has introduced a new twist: video. Gemini will let you upload video that demonstrates an issue you're trying to resolve, then scour user forums and other areas of the internet to find a solution. 

As an example, Google's Rose Yao talked onstage at I/O 2024 about a used turntable she bought and how she couldn't get the needle to sit on the record. Yao uploaded a video showing the issue, then Gemini quickly found an explainer describing how to balance the arm on that particular make and model. 

Google

"Search is so much more than just words in a text box. Often the questions you have are about the things you see around you, including objects in motion," Google wrote. "Searching with video saves you the time and trouble of finding the right words to describe this issue, and you’ll get an AI Overview with steps and resources to troubleshoot."

If the video alone doesn't make it clear what you're trying to figure out, you can add text or draw arrows that point to the issue in question. 

OpenAI just introduced ChatGPT 4o with the ability to interpret live video in real time, then describe a scene or even sing a song about it. Google, however, is taking a different tack with video by focusing on its Search product for now. Searching with video is coming to Search Labs US users in English to start with, but will expand to more regions over time, the company said.

Catch up on all the news from Google I/O 2024 right here!

As Google starts to make its latest video-generation tools available, the company says it has a plan to ensure transparency around the origins of its increasingly realistic AI-generated clips. All video made by the company’s new Veo model in the VideoFX app will have digital watermarks thanks to Google’s SynthID system.

SynthID is Google’s digital watermarking system that started rolling out to AI-generated images last year. The tech embeds imperceptible watermarks into AI-made content so that AI detection tools can recognize that the content was generated by AI. Considering that Veo, the company’s latest video generation model previewed onstage at I/O, can create longer and higher-res clips than what was previously possible, tracking the source of such content will be increasingly important.

During a briefing with reporters, DeepMind CEO Demis Hassabis said that SynthID watermarks would also expand to AI-generated text. As generative AI models advance, more companies have turned to watermarking amid fears that AI could fuel a new wave of misinformation. Watermarking systems would give platforms like Google a framework for detecting AI-generated content that may otherwise be impossible to distinguish. TikTok and Meta have also recently announced plans to support similar detection tools on their platforms and label more AI content in their apps.

Of course, there are still significant questions about whether digital watermarks on their own offer sufficient protection against deceptive AI content. Researchers have shown that watermarks can be easy to evade. But making AI-made content detectable in some way is an important first step toward transparency.

Catch up on all the news from Google I/O 2024 right here!

Google is shaking up Search. On Tuesday, the company announced big new AI-powered changes to the world’s dominant search engine at I/O, Google’s annual conference for developers. With the new features, Google is positioning Search as more than a way to simply find websites. Instead, the company wants people to use its search engine to directly get answers and help them with planning events and brainstorming ideas.

“[With] generative AI, Search can do more than you ever imagined,” wrote Liz Reid, vice president and head of Google Search, in a blog post. “So you can ask whatever’s on your mind or whatever you need to get done — from researching to planning to brainstorming — and Google will take care of the legwork.”

Google’s changes to Search, the primary way that the company makes money, are a response to the explosion of generative AI ever since OpenAI’s ChatGPT released at the end of 2022. Since then, a handful of AI-powered apps and services including ChatGPT, Anthropic, Perplexity, and Microsoft’s Bing, which is powered by OpenAI’s GPT-4, have challenged Google’s flagship service by directly providing answers to questions instead of simply presenting people a list of links. This is the gap that Google is racing to bridge with its new features in Search.

Starting today, Google will show complete AI-generated answers in response to most search queries at the top of the results page in the US. Google first unveiled the feature a year ago at Google I/O in 2023, but so far, anyone who wanted to use the feature had to sign up for it as part of the company’s Search Labs platform that lets people try out upcoming features ahead of their general release. Google is now making AI Overviews available to hundreds of millions of Americans, and says that it expects it to be available in more countries to over a billion people by the end of the year. Reid wrote that people who opted to try the feature through Search Labs have used it “billions of times” so far, and said that any links included as part of the AI-generated answers get more clicks than if the page had appeared as a traditional web listing, something that publishers have been concerned about. “As we expand this experience, we’ll continue to focus on sending valuable traffic to publishers and creators,” Reid wrote. 

In addition to AI Overviews, searching for certain queries around dining and recipes, and later with movies, music, books, hotels, shopping and more in English in the US will show a new search page where results are organized using AI. “[When] you’re looking for ideas, Search will use generate AI to brainstorm with you and create an AI-organized results page that makes it easy to explore,” Reid said in the blog post.

Google

If you opt in to Search Labs, you’ll be able to access even more features powered by generative AI in Google Search. You’ll be able to get AI Overview to simplify the language or break down a complex topic in more detail. Here’s an example of a query asking Google to explain, for instance, the connection between lightning and thunder.

Google

Search Labs testers will also be able to ask Google really complex questions in a single query to get answers on a single page instead of having to do multiple searches. The example that Google’s blog post gives: “Find the best yoga or pilates studios in Boston and show details on their intro offers and walking time from Beacon Hill.” In response, Google shows the highest-rated yoga and pilates studios near Boston’s Beacon Hill neighborhood and even puts them on a map for easy navigation.

Google

Google also wants to become a meal and vacation planner by letting people who sign up for Search Labs ask queries like “create a 3 day meal plan for a group that’s easy to prepare” and letting you swap out individual results in its AI-generated plan with something else (swapping a meat-based dish in a meal plan for a vegetarian one, for instance).

Google

Finally, Google will eventually let anyone who signs up for Search Labs use a video as a search query instead of text or images. “Maybe you bought a record player at a thriftshop, but it’s not working when you turn it on and the metal piece with the needle is drifting unexpectedly,” wrote Reid in Google’s blog post. “Searching with video saves you the time and trouble of finding the right words to describe this issue, and you’ll get an AI Overview with steps and resources to troubleshoot.”

Google said that all these new capabilities are powered by a brand new Gemini model customized for Search that combines Gemini’s advanced multi-step reasoning and multimodal abilities with Google’s traditional search systems.

Catch up on all the news from Google I/O 2024 right here!

It's all AI all the time at Google I/O! Today, Google announced its new AI media creation engines: Veo, which can produce "high-quality" 1080p videos; and Imagen 3, its latest text-to-image framework. Neither sound particularly revolutionary, but they're a way for Google to keep up the fight against OpenAI's Sora video model and Dall-E 3, a tool that has practically become synonymous with AI-generated images.

Google claims Veo has "an advanced understanding of natural language and visual semantics" to create whatever video you have in mind. The AI generated videos can last "beyond a minute." Veo is also capable of understanding cinematic and visual techniques, like the concept of a timelapse. But really, that should be table stakes for an AI video generation model, right?

To prove that Veo isn't out to steal artist's jobs, Google has also partnered with Donald Glover and Gilga, his creative studio, to show off the model's capabilities. We haven't yet seen that footage, but hopefully it's more like Atlanta season 3 and not season 2. According to Google, Veo can simulate real-world physics better than its previous models, and it's also improved how it renders high-definition footage.

It remains to be seen if anyone will actually want to watch AI generated video, outside of the morbid curiosity of seeing a machine attempt to algorithmically recreate the work of human artists. But that's not stopping Google or OpenAI from promoting these tools and hoping they'll be useful (or at least, make a bunch of money). Veo will be available inside of Google's VideoFX tool today for some creators, and the company says it'll also be coming to YouTube Shorts and other products. If Veo does end up becoming a built-in part of YouTube Shorts, that's at least one feature Google can lord over TikTok.

As for Imagen 3, Google is making the usual promises: It's said to be the company's "highest quality" text-to-image model, with "incredible level of detail" for "photorealistic, lifelike images" and fewer artifacts. The real test, of course, will be to see how it handles prompts compared to Dall-E 3. Imagen 3 handles text better than before, Google says, and it's also smarter about handling details from long prompts.

The sun rises and sets. We're all slowly dying. And AI is getting smarter by the day. That seems to be the big takeaway from Google's latest media creation tools. Of course they're getting better! Google is pouring billions into making the dream of AI a reality, all in a bid to own the next great leap for computing. Will any of this actually make our lives better? Will they ever be able to generate art with genuine soul? Check back at Google I/O every year until AGI actually appears, or our civilization collapses.

Developing...

Catch up on all the news from Google I/O 2024 right here!

In a video demonstrating the prowess of its new Project Astra app, the person demonstrating asked Gemini "do you remember where you saw my glasses?" The AI impressively responded "Yes, I do. Your glasses were on a desk near a red apple," despite said object not actually being in view when the question was asked. But these weren't your bog-standard visual aid. These glasses had a camera onboard and some sort of visual interface!

The tester picked up their glasses and put them on, and proceeded to ask the AI more questions about things they were looking at. Clearly, there is a camera on the device that's helping it take in the surroundings, and we were shown some sort of interface where a waveform moved to indicate it was listening. Onscreen captions appeared to reflect the answer that was being read aloud to the wearer, as well. So if we're keeping track, that's at least a microphone and speaker onboard too, along with some kind of processor and battery to power the whole thing. 

We only caught a brief glimpse of the wearable, but from the sneaky seconds it was in view, a few things were evident. The glasses had a simple black frame and didn't look at all like Google Glass. They didn't appear very bulky, either. 

In all likelihood, Google is not ready to actually launch a pair of glasses at I/O. It breezed right past the wearable's appearance and barely mentioned them, only to say that Project Astra and the company's vision of "universal agents" could come to devices like our phones or glasses. We don't know much else at the moment, but if you've been mourning Google Glass or the company's other failed wearable products, this might instill some hope yet.

Catch up on all the news from Google I/O 2024 right here!

When Google first showcased its Duplex voice assistant technology at its developer conference in 2018, it was both impressive and concerning. Today, at I/O 2024, the company may be bringing up those same reactions again, this time by showing off another application of its AI smarts with something called Project Astra. 

The company couldn't even wait till its keynote today to tease Project Astra, posting a video to its social media of a camera-based AI app yesterday. At its keynote today, though, Google's DeepMind CEO Demis Hassabis shared that his team has "always wanted to develop universal AI agents that can be helpful in everyday life." Project Astra is the result of progress on that front. 

What is Project Astra?

According to a video that Google showed during a media briefing yesterday, Project Astra appeared to be an app which has a viewfinder as its main interface. A person holding up a phone pointed its camera at various parts of an office and verbally said "Tell me when you see something that makes sound." When a speaker next to a monitor came into view, Gemini responded "I see a speaker, which makes sound."

The person behind the phone stopped and drew an onscreen arrow to the top circle on the speaker and said, "What is that part of the speaker called?" Gemini promptly responded "That is the tweeter. It produces high-frequency sounds."

Then, in the video that Google said was recorded in a single take, the tester moved over to a cup of crayons further down the table and asked "Give me a creative alliteration about these," to which Gemini said "Creative crayons color cheerfully. They certainly craft colorful creations."

Wait, were those Project Astra glasses? Is Google Glass back?

The rest of the video goes on to show Gemini in Project Astra identifying and explaining parts of code on a monitor, telling the user what neighborhood they were in based on the view out the window. Most impressively, Astra was able to answer "Do you remember where you saw my glasses?" even though said glasses were completely out of frame and were not previously pointed out. "Yes, I do," Gemini said, adding "Your glasses were on a desk near a red apple."

After Astra located those glasses, the tester put them on and the video shifted to the perspective of what you'd see on the wearable. Using a camera onboard, the glasses scanned the wearer's surroundings to see things like a diagram on a whiteboard. The person in the video then asked "What can I add here to make this system faster?" As they spoke, an onscreen waveform moved to indicate it was listening, and as it responded, text captions appeared in tandem. Astra said "Adding a cache between the server and database could improve speed."

The tester then looked over to a pair of cats doodled on the board and asked "What does this remind you of?" Astra said "Schrodinger's cat." Finally, they picked up a plush tiger toy, put it next to a cute golden retriever and asked for "a band name for this duo." Astra dutifully replied "Golden stripes."

How does Project Astra work?

This means that not only was Astra processing visual data in realtime, it was also remembering what it saw and working with an impressive backlog of stored information. This was achieved, according to Hassabis, because these "agents" were "designed to process information faster by continuously encoding video frames, combining the video and speech input into a timeline of events, and caching this information for efficient recall."

It was also worth noting that, at least in the video, Astra was responding quickly. Hassabis noted in a blog post that "While we’ve made incredible progress developing AI systems that can understand multimodal information, getting response time down to something conversational is a difficult engineering challenge."

Google has also been working on giving its AI more range of vocal expression, using its speech models to "enhanced how they sound, giving the agents a wider range of intonations." This sort of mimicry of human expressiveness in responses is reminiscent of Duplex's pauses and utterances that led people to think Google's AI might be a candidate for the Turing test.

When will Project Astra be available?

While Astra remains an early feature with no discernible plans for launch, Hassabis wrote that in future, these assistants could be available "through your phone or glasses." No word yet on whether those glasses are actually a product or the successor to Google Glass, but Hassabis did write that "some of these capabilities are coming to Google products, like the Gemini app, later this year."

Catch up on all the news from Google I/O 2024 right here!

Google announced updates to its Gemini family of AI models at I/O, the company’s annual conference for developers, on Tuesday. It’s rolling out a new model called Gemini 1.5 Flash, which it says is optimized for speed and efficiency.

“[Gemini] 1.5 Flash excels at summarization, chat applications, image and video captioning, data extraction from long documents and tables, and more,” wrote Demis Hassabis, CEO of Google DeepMind, in a blog post. Hassabis added that Google created Gemini 1.5 Flash because developers needed a model that was lighter and less expensive than the Pro version, which Google announced in February. Gemini 1.5 Pro is more efficient and powerful than the company’s original Gemini model announced late last year.

Gemini 1.5 Flash sits between Gemini 1.5 Pro and Gemini 1.5 Nano, Google’s smallest model that runs locally on devices. Despite being lighter weight then Gemini Pro, however, it is just as powerful. Google said that this was achieved through a process called “distillation”, where the most essential knowledge and skills from Gemini 1.5 Pro were transferred to the smaller model. This means that Gemini 1.5 Flash will get the same multimodal capabilities of Pro, as well as its long context window – the amount of data that an AI model can ingest at once – of one million tokens. This, according to Google, means that Gemini 1.5 Flash will be capable of analyzing a 1,500-page document or a codebase with more than 30,000 lines at once. 

Gemini 1.5 Flash (or any of these models) aren’t really meant for consumers. Instead, it’s a faster and less expensive way for developers building their own AI products and services using tech designed by Google.

In addition to launching Gemini 1.5 Flash, Google is also upgrading Gemini 1.5 Pro. The company said that it had “enhanced” the model’s abilities to write code, reason and parse audio and images. But the biggest update is yet to come – Google announced it will double the model’s existing context window to two million tokens later this year. That would make it capable of processing two hours of video, 22 hours of audio, more than 60,000 lines of code or more than 1.4 million words at the same time.

Both Gemini 1.5 Flash and Pro are now available in public preview in Google’s AI Studio and Vertex AI. The company also announced today a new version of its Gemma open model, called Gemma 2. But unless you’re a developer or someone who likes to tinker around with building AI apps and services, these updates aren’t really meant for the average consumer.

Catch up on all the news from Google I/O 2024 right here!