StartupTalky

Tag: Google AI Overviews

  • Google Offers up to INR 26 Lakh Reward for Finding Security Flaws in its AI Systems

    To identify and address security vulnerabilities in its artificial intelligence (AI) systems, Google has started a new incentive programme. The organisation is rewarding those who find significant flaws that have the potential to cause actual harm with incentives of up to $30,000 (about INR 26 lakh).

    Rogue actions—situations in which an AI system is deceived into performing an action it shouldn’t—are the focus of this new AI bug reward programme. Examples include a secret command that compels an AI to summarise a user’s private emails and forward them to an attacker, or an AI question that might cause Google Home to unlock a door.

    Google has given precise examples of what constitutes an AI bug. These comprise any flaw that allows a huge language model or other generative AI tool to be exploited to get around security, change data, or do undesirable behaviours. For example, in the past, researchers discovered vulnerabilities that made it possible to manipulate smart home equipment by manipulating calendar events, opening shutters, or turning off lights without authorisation. Keep in mind that not all AI problems will result in compensation.

    It isn’t enough to just make Gemini make a mistake or produce unpleasant text. Instead, these kinds of problems ought to be reported via Google’s AI products’ regular feedback features, which allow safety teams to examine and correct model behaviour over time.

    CodeMender by Google

    In addition to the recently launched bug bounty programme, Google also unveiled CodeMender, an AI agent that automatically fixes security vulnerabilities in code. According to the business, 72 vulnerabilities in open-source projects have already been fixed by CodeMender after it was reviewed by human specialists.

    Serious rogue action defects in Google’s main products, including Search, Gemini Apps, Gmail, and Drive, are eligible for the top award of $20,000. The sum can reach $30,000 with bonuses for exceptionally creative or high-quality reports. Smaller problems or faults in other products, such as NotebookLM or Jules, are eligible for lower awards.

    Researchers have already made over $400,000 from Google’s AI-related reports in the past two years. Simply put, this new initiative makes things more competitive and official. Our daily lives are increasingly involving AI technologies.

    They can be found in home appliances, laptops, phones, and even the instruments we use at work. Attackers can become more inventive as these systems become more powerful. In essence, Google is promising that we will compensate someone who can break it before the bad people do.

    Quick
    Shots

    •Bugs include AI being tricked to perform harmful or
    unintended actions, e.g., exposing private data or controlling smart devices.

    •Flaws that allow large language models or
    generative AI tools to bypass security, alter data, or behave undesirably.

    •Google’s AI agent automatically fixes
    vulnerabilities; 72 open-source issues resolved so far.

    •Serious defects in Search, Gemini, Gmail, Drive
    eligible for $20k–$30k; smaller flaws in other products get lower awards.

  • Nvidia Spends $900 Million to Hire Enfabrica CEO and License Startup’s Technology

    According to a report by CNBC, Nvidia recently spent more than $900 million to license Enfabrica’s technology and hire Rochan Sankar, the CEO of the artificial intelligence hardware business, along with other staff members. According to the report, Nvidia is paying cash and stock in a deal that is reminiscent of recent AI talent acquisitions done by Meta and Google.

    The report further stated that the acquisition was finalised last week, and Rochan Sankar, the CEO of Enfabrica, has joined Nvidia. Since the introduction of OpenAI’s ChatGPT in late 2022, Nvidia has been the mainstay of the AI boom. Large language models are trained using the company’s graphics processing units (GPUs), which are typically purchased in large clusters and enable giant cloud providers to provide AI services to customers.

    Nvidia Building a Strong AI Network

    Nvidia’s latest AI chips, such as the A100, came in towering racks with 72 GPUs installed and cooperating, whereas its older models were single processors that slid into servers. Microsoft stated on September 18 that the $4 billion data centre in Wisconsin would house that kind of technology.

    The 2019-founded company Enfabrica claims that its technology can link over 100,000 GPUs. Nvidia may be able to provide integrated systems around its processors with the aid of this technology, enabling clusters to function as a single computer. In 2023, Nvidia made a $125 million Series B investment in Enfabrica under the leadership of Atreides Management.

    The business claimed that the valuation was five times higher than its Series A capital, but it did not reveal it at the time. Spark Capital, Arm, Samsung, and Cisco were among the investors who helped Enfabrica raise an additional $115 million towards the end of last year. The post-money valuation was approximately $600 million, according to PitchBook.

    Nvidia’s Competitors also Hiring Elite AI Talent

    Through agreements that resemble acquisitions, tech powerhouse Meta, Google, Microsoft, and Amazon have all heavily invested in hiring elite AI talent. Because of the agreements, the companies are able to hire outstanding researchers and engineers without having to worry about the regulatory complexities associated with acquisitions.

    The largest of these transactions occurred in June, when Meta acquired a 49% share in Scale AI and paid $14.3 billion to Alexandr Wang and other Scale AI founders. A month later, Google revealed that it had reached a $2.4 billion deal, including licence costs, to hire Varun Mohan, co-founder and CEO of Windsurf, a business that specialises in artificial intelligence code, along with other R&D staff. Google struck a similar agreement to hire Character’s founders last year. AI. Amazon did the same for Adept, while Microsoft did the same for Inflection.

    Quick
    Shots

    •Nvidia’s deal mirrors AI talent
    acquisitions by Meta, Google, Microsoft & Amazon.

    •Nvidia GPUs power large language
    model training and dominate AI infrastructure.

    •Nvidia had already invested $125M in
    Enfabrica’s Series B in 2023.

    •Enfabrica raised funds from Spark
    Capital, Arm, Samsung, Cisco, hitting around $600M valuation.

  • Inside Google’s AI Revolution: $85B Bet, Productivity Drive, and a $4 Trillion Milestone

    In what he described as a period of “extraordinary investment,” Google CEO Sundar Pichai has challenged staff to increase productivity and efficiency as the company doubles down on artificial intelligence. Pichai stated that the corporation must adjust to the AI transition without a corresponding increase in staff during an internal meeting, the audio recording of which CNBC was able to acquire. Pichai informed staff members that employing a large number of personnel is the appropriate response to any period of exceptional investment.

    Pichai Pushes for Productivity in the Age of AI

    However, Pichai believes his team must do more in this AI era by utilising this shift to promote increased productivity. He underlined that we must be “frugal with our resources” and that he is still “very optimistic” about Google’s present state of affairs. Google is actively creating tools for software engineers to speed up development cycles and promote wider AI usage across teams, according to Brian Saluzzo, a top executive at the business.

    New Internal Tools: AI Savvy Google & Building with Gemini

    Saluzzo unveiled AI Savvy Google, a new internal platform that provides staff members with product-specific training, toolkits, and carefully selected learning resources to hone their AI abilities. He also revealed Building with Gemini, a joint training initiative with DeepMind that aims to give engineers cutting-edge skills by utilising Google’s Gemini AI model.

    Google Surpasses $4 Trillion: AI and Cloud Demand Soar

    Last week, Google became the second firm after Nvidia to reach the historic $4 trillion market cap milestone. Following an impressive earnings report in the AI sector, the market capitalisation experienced a significant spike.

    Thanks to new AI features and a stable digital advertising market, Google exceeded Wall Street’s quarterly revenue and profit projections. Additionally, due to the strong demand for cloud computing services, Google raised its capital spending estimate to $85 billion.

    Chief Financial Officer Anat Ashkenazi stated during the call that demand and expansion prospects are anticipated to drive significant increases in capital spending in 2026. Although the rate of server deployment has sped up, Ashkenazi continued, Alphabet still has more customers requesting its cloud services than it can provide.

    Windsurf Acquisition: Supercharging Code Generation

    Last month, Google revealed that it has hired a number of senior employees from AI code generation start-up Windsurf. According to a person familiar with the agreement, Google is paying $2.4 billion in license fees as part of the agreement to use parts of Windsurf’s technology on non-exclusive terms. According to a media report, Google will not acquire a stock or any kind of controlling position in Windsurf.

    Co-founder Douglas Chen, Windsurf CEO Varun Mohan, and a few members of the coding tool’s R&D team will join Google’s DeepMind AI group.

    According to media reports from June, the deal came after months of talks between Windsurf and OpenAI to sell the company for a price that may reach $3 billion. This deal demonstrated the interest in the code-generation field, which has become one of the fastest-growing AI applications.

  • Google AI Overviews Expands to India, Bringing New Features and Enhancements

    As part of its ongoing effort to rethink search using generative AI, Google has taken a giant leap forward by extending its AI-powered Search Overviews to six additional countries, including India. The goal of this feature is to improve the user experience and expose individuals to other viewpoints and knowledge by providing brief explanations of complicated subjects.

    Using generative AI, AI Overviews synthesise data from multiple sources and display it in an approachable manner; they were introduced in Search Labs last year. This facilitates the rapid understanding of difficult concepts, the investigation of other points of view, and the exploration of particular areas of interest.

    Google is expanding AI Overviews to other countries after receiving strong feedback from consumers in the US and Search Labs worldwide. These countries include Brazil, Mexico, Indonesia, India, and the UK. Every country will have local language support for this function.

    What Google is Offering to India?

    Both English and Hindi versions of AI Overviews will be made accessible in India, and they will include features that are tailored to the needs of Indian users. A number of the most notable features are available, including the ability to switch between English and Hindi results with ease, the ability to listen to responses from AI Overviews with only a push of the ‘Listen’ button, and many others.

    Google has reaffirmed its commitment to connecting consumers with a wide variety of sources. The purpose of AI Overviews is to inspire users to explore and interact with the internet, hence generating higher-quality traffic for businesses and publishers.

    In a blog post, Hema Budaraju, Senior Director of Product, Search (AI) Generative Experiences, Google, announced that the company is introducing popular features that were well-received during their Search Labs experiment: the ability to easily switch between English and Hindi results using a language toggle button and the ability to listen to responses with Text-To-Speech by tapping the ‘Listen’ button.

    According to Budaraju, the expansion will be introduced slightly over the period of a few weeks.

    Google is also launching a new right-hand link display for AI Overviews on desktop, which can be accessed on mobile by pressing the site icons on the upper right. This is all part of an effort to help consumers navigate more relevant websites and discover what they need.

    Controversy Regarding AI Overview

    Google had to limit the AI-generated feature’s use just fifteen days after it debuted due to user complaints that the responses were harmful and unreliable. The AI was found to be recommending bizarre things to users, such as swallowing a small pebble with their vitamins, drinking urine to pass a kidney stone, fixing pizza cheese with Elmer’s glue, and many more.

    How Does Generative AI Search Work?
    Generative AI search would certainly revolutionize web search, but what matters most is accurate and unbiased AI, as the results can only be as good as the training data set.
    StartupTalkyRishi Mundra