&w=3840&q=100)
Gemma 3n: All about Google's open model for on-device AI on phones, laptops
At its annual Google I/O conference, Google unveiled the Gemma 3n, a new addition to its Gemma 3 series of open AI models. The company said that the model is designed to run efficiently on everyday devices like smartphones, laptops, and tablets. Gemma 3n shares its architecture with the upcoming generation of Gemini Nano, the lightweight AI model that already powers several on-device AI features on Android devices such as voice recorder summaries on Pixel smartphones.
Gemma 3n model: Details
Google says Gemma 3n makes use of a new technique called Per-Layer Embeddings (PLE), which allows the model to consume much less RAM than similarly sized models. Although the model has 5 billion and 8 billion parameters (5B and 8B), this new memory optimisation brings its RAM usage closer to that of a 2B or 4B model. In practical terms, this means Gemma 3n can run with just 2GB to 3GB of RAM, making it viable for a much wider range of devices.
Gemma 3n model: Key capabilities
Audio input: The model can process sound-based data, enabling applications like speech recognition, language translation, and audio analysis.
Multimodal input: With support for visual, text, and audio inputs, the model can handle complex tasks that involve combining different types of data.
Broad language support: Google said that the model is trained in over 140 languages.
32K token context window: Gemma 3n supports input sequences up to 32,000 tokens, allowing it to handle large chunks of data in one go—useful for summarising long documents or performing multi-step reasoning.
PLE caching: The model's internal components (embeddings) can be stored temporarily in fast local storage (like the device's SSD), helping reduce the RAM needed during repeated use.
Conditional parameter loading: If a task doesn't require audio or visual capabilities, the model can skip loading those parts, saving memory and speeding up performance.
Gemma 3n model: Availability
As part of the Gemma open model family, Gemma 3n is provided with accessible weights and licensed for commercial use, allowing developers to tune, adapt, and deploy it across a variety of applications. Gemma 3n is now available as a preview in Google AI Studio.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
India Today
an hour ago
- India Today
AI's dirty secret: How your chatbot is draining Earth's water and energy
Every time you ask a chatbot for help or generate an AI image, somewhere a vast facility hums to life. It's easy to forget that each 'effortless' AI interaction leans on gigantic data centres -- buildings full of servers running nonstop, devouring electricity and gulping is World Environment Day, so it feels fitting to pull back the curtain on AI's unseen carbon footprint, even as we keep creating content with the help of AI, like the images used in this very POWER BEHIND THE MAGIC: WHY AI NEEDS HUGE DATA CENTRESAI models like ChatGPT and sophisticated image generators don't float on dreams. They rely on colossal data centres packed with racks of servers that rarely sleep. IAD71 Amazon Web Services data centre (as on July 17, 2024 in Ashburn, Virginia) () In 2024, global data centres -- including those powering AI -- used around 460 terawatt-hours (TWh) of electricity. That's roughly equal to Sweden's entire annual power consumption (International Energy Agency, 2024).Forecasts suggest this figure could double to 1,000 TWh by 2026, which would almost match Germany's yearly electricity per the US Department of Energy, a single data centre can draw over 100 megawatts (MW) of power -- enough to run 80,000 average American homes at now you can begin to see why AI has become a 'sizable' contributor to global emissions. AI's rapid expansion is a big driver behind this surge. CARBON COST OF CHATBOTS: ENERGY USED PER PROMPTTraining a large language model isn't a weekend project -- it's an electricity instance, powering up OpenAI's GPT-3 used about 1.3 gigawatt-hours (GWh) of electricity, enough to supply roughly 120 average American homes for a year (ScienceDirect, 2023).Once the model is live, every single prompt you send to GPT-4 still sips power -- about 0.0003 kilowatt-hours (kWh). That's like leaving a 60-watt light bulb on for 20 seconds (Epoch AI, 2025).Now imagine this at scale: ChatGPT answers over 1 billion queries a day. If every Google search worldwide -- around 9 billion a day -- were replaced with a ChatGPT query, global energy use would spike by an extra 10 TWh per year. That's enough to power all the homes in a city the size of San Francisco for a year (Hugging Face and Carnegie Mellon University, 2024).advertisementGenerating a single AI-created image isn't free either. It can use as much energy as fully charging your smartphone once. THE WATER-COOLING FACTOR: HOW AI IS DRAINING RESOURCESElectricity tells only half the story. Those thousands of servers run hot, and cooling them eats up alarming amounts of 2024, Google's data centres consumed nearly 6 billion gallons (22.7 billion litres) of water -- about one-third of Turkey's drinking water for an entire year, as per its 2024 Environment model training, every 10-20 prompts to Google's Bard chatbot reportedly required about 500 ml of water -- mostly for average, data centres use about 1.8 litres of water per kWh of electricity consumed (Meta Platforms, 2023). Put another way, a single large data centre can guzzle as much water in 24 hours as a small that thirst is rising -- industry water consumption grows by 8-20% each year, thanks to AI's surging role and stiff competition (University of California, Riverside, and University of Texas at Arlington, 2025). advertisementWHO'S DOING IT BETTER: GREEN AI INITIATIVES AND THEIR LIMITSSome big names in tech are trying to curb AI's environmental chew. Microsoft, for instance, has cut its water use per computing unit by 39% since 2021, saving about 125 million litres per data centre annually by using reclaimed water and smarter cooling methods (Meta Platforms, 2023).Alternative cooling strategies -- like air cooling or immersing servers in liquid -- are also under the microscope. But these options come with their own cooling cuts water use but can drive up energy needs for pumps or special fluids. Air cooling can be less water-dependent, yet efficiency dips when temperatures climb (Meta Platforms, 2023).Switching to renewables helps, but even 'green' electricity can't keep pace with AI's relentless demand. A recent MIT Lincoln Laboratory report warned that global demand for clean power is lagging far behind AI's growth many companies still don't fully share their energy or water metrics, making it near impossible to verify progress. advertisementTHE BOTTOM LINEAI's magic comes at a genuine cost. Data centres powering our digital conversations and image creations now rival entire nations in electricity use. They swig water comparable to small as AI continues its meteoric rise, this environmental toll only grows World Environment Day, it's worth asking: how can we keep enjoying the convenience of AI without letting it swallow our planet?'Green AI' efforts -- more efficient hardware, smarter cooling, renewable electricity -- are steps in the right direction. But the gap between AI's hunger and our ability to supply green power remains we don't face this challenge head-on, our planet will bear the brunt. Every prompt you send, every image you generate, leaves an unseen footprint that's getting harder to Watch
Time of India
an hour ago
- Time of India
Google takes a gamble in class action jury trial over cell phone data use
HighlightsGoogle is facing an $800 million lawsuit in Santa Clara County, California, from Android smartphone users who claim the company misappropriates their cellphone data, affecting an estimated 14 million Californians. The plaintiffs allege that Google secretly transmits data over cellular networks even when devices are turned off, which they argue improperly consumes purchased data from mobile carriers without user consent. Despite Google's history of settling class actions, the company is opting for a trial, disputing the plaintiffs' claims that they have a property interest in cellular data allowances and arguing that no actual losses were incurred. Class actions rarely go to trial, which is why a case against Google is proving to be an outlier. The tech giant is defending itself before a jury in Santa Clara County, California, superior court in an $800 million lawsuit by Android smartphone users who say Google misappropriates their cellphone data. A jury of eight women and four men was seated on Tuesday in what lawyers say is expected to be a three-to-four-week trial, with opening statements kicking off on Wednesday. The stakes are high, but the class, which includes an estimated 14 million Californians whose mobile devices use Google's Android operating system, is in some ways just an appetizer. The same plaintiffs lawyers from Korein Tillery; Bartlit Beck and McManis Faulkner are litigating a parallel case in San Jose federal court covering Android users in the other 49 states, with billions of dollars in alleged damages. The plaintiffs in court papers say that even when their phones are turned off, Google causes Android devices to surreptitiously send information over cellular networks "for Google's own purposes," including targeted digital advertising. These transfers improperly eat up data that users purchase from their mobile carriers, the plaintiffs allege. Google spokesperson Jose Castaneda said the claims "mischaracterize standard industry practices that help protect users and make phones more reliable," he told me. "We look forward to making our case in court." A unit of Mountain View, California-based Alphabet, Google has a well-used playbook for settling class actions. Earlier this week, for example, the company agreed to pay $500 million to resolve shareholder litigation - a move that comes on the heels of a $50 million deal in May to resolve class-wide allegations of racial bias against Black employees and a $100 million payout in March to a proposed class of advertisers who claimed they were overcharged for clicks on ads. So why is Google taking this case to trial? In court papers, Google's outside counsel from Cooley argue that Android users incurred no actual losses, and that consumers consented to Google's so-called "passive" data transfers via terms of service agreements and device settings. The lawyers also dispute the fundamental premise of the case: that cellular data allowances can be considered "property" under California law and subject to conversion, a civil cause of action that involves taking a person's property without permission. When the "rhetoric and hyperbole are set aside, Plaintiffs' theory is revealed as little more than a (misguided) product design claim - not wrongful conversion," defense counsel wrote. The Cooley team, which includes Whitty Somvichian, Michael Attanasio, Max Bernstein and Carrie Lebel, declined comment. The plaintiffs sued Google in Santa Clara County Superior Court in 2019, asserting that they have a property interest in their cellular plans' data allowances, and that each quantum they pay for has a market value. They don't object to data transmissions when they're actively engaged with Google's apps and properties, like checking email or playing a game. But they say Google never told them it would avail itself of their cellular data when they weren't using their phones to send and receive a range of information on their usage. "The upshot is that these phone users unknowingly subsidize the same Google advertising business that earns over $200 billion a year," plaintiffs lawyer George Zelcs of Korein Tillery said via email. In addition to injunctive relief, the plaintiffs want Google to reimburse them for the value of the cellular data the company consumed. Per person, the amount is modest - 1 to 1.5 megabytes of data each day, the plaintiffs estimate. To put that in context, Americans used just over 100 trillion megabytes of wireless data in 2023, my Reuters colleagues reported. But with a class period dating back to 2016, the totals add up quickly. In court papers, Google lawyers sound almost incredulous at the amount of the claimed nationwide damages, which they say runs in the tens of billions - more than the $7.4 billion Perdue Pharma settlement for the opioid crisis, they note. "Plaintiffs cannot show remotely commensurate harm to the class," they wrote. In denying Google's motion for summary judgment in May, Judge Charles Adams allowed the plaintiffs' claim for conversion to go forward, ruling there are triable issues of material fact for jurors to decide. While Adams said no direct state law precedent exists as to whether cell phone data is property, he pointed to a decision by the 9th U.S. Circuit Court of Appeals last year in the parallel federal class action, Taylor v Google. In that case, U.S. Magistrate Judge Virginia DeMarchi in San Jose sided with Google and dismissed the complaint with prejudice in 2022, only to be reversed and remanded on appeal. The appellate panel in an unpublished decision ruled that the plaintiffs plausibly alleged they incurred damages when Google used their cellular data. Adams in a pre-trial order set some limits on what the lawyers will be allowed to argue to the jury. Plaintiffs may not suggest Google engages in "surveillance" of Android users, he wrote, or that the data transfers are a privacy violation. As for Google, Adams said, it "must not present evidence or argument suggesting that this case is 'lawyer driven' or was 'invented' by Plaintiffs' counsel."
The Print
an hour ago
- The Print
Govt jobs aspirants can apply for SSC recruitment tests through dedicated mobile app
'All candidates applying for recruitment exams conducted by SSC can fill the entire application form through the application itself,' SSC Chairman S Gopalakrishnan told PTI. He said the upgraded 'mySSC app' will be of great help to job aspirants, especially those from rural areas. New Delhi, Jun 5 (PTI) Government jobs aspirants can now apply for recruitment tests conducted by the Staff Selection Commission (SSC) through an upgraded mobile application developed by it, a top official said on Thursday. He said there is no specific need to go to any computer centre or depend on any other operator in order to fill up and submit applications. 'The candidates can also identify themselves through Aadhaar OTP and face authentication on the same app,' Gopalakrishnan said. He said the application developed for SSC by Cubastion Consulting firm is going to be a big boon for the candidates and would add to ease of dealing with the Commission throughout the exam cycle from applying till joining. 'We are happy to be a part of this change with SSC,' said Cubastion co-founder and CEO Ravi Kumar. He said India's digital expansion requires trusted platforms and by bringing together biometric identity verification and Aadhaar-enabled services 'we are making mass hiring fairer, more secure, and accessible, particularly for rural candidates'. Until now, many of these candidates relied heavily on common service centres and cyber cafes to fill out their application forms, often at a cost, Kumar said. 'With the new mySSC app, candidates can now complete the entire process – from registration to application – directly on their mobile phones. It removes private middlemen from the system and puts power back into the hands of the applicants. Our future vision is to make this app a one-stop solution for the entire recruitment lifecycle,' he added. In the past, candidates could apply online through the Commission's website and take a live photo via the mySSC mobile app, according to a notice issued by the SSC. 'Starting from June 2025, the entire application process can now be completed seamlessly within the app itself,' said the notice dated June 2, 2025. Candidates will be able to apply for SSC examination using specific android mobile phones through the mySSC mobile app which can be downloaded from Google Play Store, it said. 'This will be applicable for all the exams being conducted June 2025 onwards,' the Commission said. The mySSC mobile app of the Commission has been significantly upgraded which is aimed at enhancing the user experience and expanding the range of services available, it said. The update would include integration of the Aadhaar authentication services, the Commission said. In order to apply through the mySSC mobile app, candidates must first register using their Aadhaar number on the OTR page. The mobile app works only on Android phones with version 11 or higher, the SSC said. Before applying for the examinations candidates must also install the Aadhaar Face RD app on their phone to complete the face authentication form, it said and shared a link to it. In another notice issued on Wednesday, the Commission said that candidates are seeking clarification whether demographic details such as name, date of birth, address etc. from Aadhaar will prevail over the data filled by the candidates in the OTR form. 'In this regard, it is clarified that the information entered by the candidates in the OTR shall be treated as final for all recruitment related purposes and details mentioned in Aadhaar will not override the candidate-provided information in OTR,' the notice said. The SSC had in April this year decided to implement Aadhaar-based biometric authentication in its all examinations in order to verify candidates' identity on a voluntary basis. The Commission is one of the largest recruitment agencies of the central government with its main mandate to conduct selections for non-gazetted posts in various central ministries and departments, among others. The SSC conducts seven mandated all India open competitive examinations, including the Combined Graduate Level Examination (CGLE) besides three limited departmental competitive examinations, according to the officials. PTI AKV ZMN This report is auto-generated from PTI news service. ThePrint holds no responsibility for its content.
