Category: Artificial intelligence

  • Jensen Huang: Export Controls Boost China’s AI Chip Development

    Jensen Huang: Export Controls Boost China’s AI Chip Development

    Key Takeaways

    1. Nvidia’s GPUs are vital for AI applications, impacting technologies from chatbots to self-driving cars.
    2. CEO Jensen Huang believes limiting China’s access to tech has failed and advocates for global distribution of American technology.
    3. Collaboration with Chinese engineers is essential for maintaining U.S. tech leadership, as blocking them could speed up local innovation in China.
    4. Huang downplays security risks, asserting that China has its own supercomputers and doesn’t rely on Nvidia for military advancements.
    5. Emphasizing the counterproductive nature of embargoes, Huang suggests that innovation should replace barriers in the U.S.–China tech competition.

    Nvidia’s worth briefly exceeded $4 trillion last week, highlighting how its GPUs are essential for artificial intelligence tasks, which range from chatbots to self-driving cars. However, CEO Jensen Huang believes that the efforts by Washington to limit China’s access to these processors have largely failed. He shared with CNN’s Fareed Zakaria, “Taking away technology from someone is more of a tactic than a goal—and this tactic wasn’t effective towards achieving the actual goal.” Huang argues that keeping the U.S. at the forefront relies on distributing an “American tech stack” globally instead of tightening export restrictions.

    Importance of Collaboration

    Huang emphasizes the crucial role China plays in the worldwide AI growth, pointing out that around half of the global AI engineers come from China. He insists that for American tech to stay as the benchmark, these developers need to work with U.S. hardware and software. If they are blocked, engineers in China will simply speed up their development of local alternatives, which will close the gap in innovation and lessen U.S. influence in the tech realm.

    Security Concerns

    There are concerns among security advocates that these same chips might be used by the People’s Liberation Army, but Huang downplays this risk. He maintains that competing militaries prefer not to depend on each other’s supply chains and that China has its own supercomputers already. “They don’t require Nvidia’s chips… to enhance their military,” he stated.

    This statement comes after a bipartisan group of U.S. senators wrote to Huang, asking him to avoid collaboration with companies connected to China’s defense industry. Over the last three administrations, the U.S. has imposed stricter export regulations on advanced GPUs, leading to a black market for higher-bandwidth versions. Even though these grey-market components lack firmware updates and support for enterprise software, they still make their way into Chinese data centers, highlighting the challenges of enforcement.

    The Backfire of Embargoes

    Huang, mirroring remarks from Microsoft’s co-founder Bill Gates, cautions that sweeping embargoes can often have counterproductive effects. He compares them to China’s recent restrictions on rare-earth minerals, which triggered a push in the U.S. for self-sufficiency. “If it occurs to us, it should also happen to them,” he remarked, positioning the U.S.–China AI competition as an unavoidable yet mutually advantageous rivalry—one that should be won through quicker innovation instead of erecting taller barriers.

    Source:
    Link

    Tags: , ,
  • Gemini Mail Summaries Loophole: A Target for Phishing Attacks

    Gemini Mail Summaries Loophole: A Target for Phishing Attacks

    Key Takeaways

    1. Google introduced Gemini-enabled email summaries in Gmail to help users quickly grasp key points from long emails.
    2. A flaw in Gemini may allow hackers to perform prompt-injection phishing attacks, targeting users who rely on AI summaries.
    3. The vulnerability involves hiding harmful directives in email text by using invisible formatting, which Gemini would still process.
    4. An example showed that a hidden warning about a compromised password could mislead users into urgent actions.
    5. Potential solutions include developing detection techniques for concealed content and implementing filters to analyze Gemini’s output for suspicious elements.

    Google introduced Gemini-enabled email summaries in Gmail near the end of May, aiming to assist users in grasping the key points without having to sift through long paragraphs. Nonetheless, a flaw within Gemini could potentially allow hackers to execute a prompt-injection phishing attack, particularly targeting those who heavily rely on AI summaries for managing their emails.

    Research Findings

    This issue was discovered by Marco Figueroa, the GenAI Bug Bounty Programs Manager at Mozilla. The deceptive email may appear like a typical message filled with text, yet it could conceal a phishing scam that Gemini is unable to detect. The harmful directives can be embedded in the text body or placed immediately after by changing their font size to 0 and altering the color to white, rendering them invisible. Nevertheless, Gemini would still process that section of the email and act on the hidden instructions.

    Example of the Flaw

    For instance, Figueroa managed to hide a warning message within the email indicating that the user’s Gmail password was compromised, along with a support phone number. When the AI summarized the email, it displayed the warning at the end alongside an urgent suggestion to call the support number without delay. While this trick may not deceive everyone, some users might take action out of concern for their account security.

    Potential Solutions

    The researcher points out that security teams can introduce detection and mitigation techniques for content that has been concealed, allowing them to either eliminate or disregard the hidden content. Additionally, there could be post-processing filters that analyze Gemini’s output to identify URLs, urgent alerts, or phone numbers.

    BleepingComputer contacted Google about this Gemini vulnerability, and a representative indicated that some mitigation strategies are currently being worked on.

    Source:
    Link

    Gemini Mail Summaries Loophole: A Target for Phishing AttacksGemini Mail Summaries Loophole: A Target for Phishing Attacks
    Tags: ,
  • SpaceX Invests $2 Billion in xAI; Tesla Expected to Join

    SpaceX Invests $2 Billion in xAI; Tesla Expected to Join

    Key Takeaways

    1. SpaceX has invested $2 billion in AI startup xAI, representing a significant portion of the company’s recent fundraising efforts.
    2. The merger of xAI with X has increased its total value to $113 billion, with more fundraising rounds expected by year-end.
    3. Elon Musk is integrating xAI’s technology across his companies, including real-time support in X and updates for Tesla vehicles.
    4. The latest version of xAI’s product, Grok 4, is being recognized as highly advanced.
    5. Tesla may consider investing excess funds into xAI, pending approval from its board and shareholders.

    The richest person in the world, Elon Musk, often shifts technologies and skills between his various companies. The CEO of multiple firms is now said to be doing the same with money as SpaceX completes its biggest external investment in AI startup xAI.

    Major Investment in xAI

    SpaceX has purchased a $2 billion share in xAI, as first disclosed by The Wall Street Journal. This investment represents roughly 40 percent of the $5 billion that Morgan Stanley helped raise for the AI company a few weeks back.

    The new capital arrives after xAI combined with X, boosting the total worth of the merged company to $113 billion. More rounds of fundraising are anticipated before the end of the year.

    Integration of AI Across Companies

    Musk has been integrating xAI’s technologies into his other ventures. Grok is now part of X and is able to answer questions in real-time. This AI also manages customer support for SpaceX’s Starlink satellite internet service. Tesla cars are in the middle of receiving Grok (beta) as part of their latest software updates. Furthermore, Musk has mentioned that the Optimus robot will utilize Grok.

    SpaceX’s previous investments include the $524 million spent to buy Swarm Technologies back in 2021. Long before that, Musk allocated $20 million from SpaceX to assist Tesla in its early challenges.

    Latest Developments and Future Prospects

    Recently, xAI launched Grok 4, and this latest version of the AI has been hailed as the most intelligent in the world.

    Musk also mentioned that Tesla might invest some of its excess funds into xAI through a post on X. Nevertheless, he made it clear that any future investment from the electric vehicle company would need to be approved by both the board and shareholders.

    Source:
    Link

    Tags: , ,
  • Grok AI Now Available in Tesla Vehicles with Latest Update

    Grok AI Now Available in Tesla Vehicles with Latest Update

    Key Takeaways

    1. Tesla’s latest software update (2025.26) introduces Grok AI to vehicles with AMD processors, still in beta testing.
    2. The update includes new features like Accent lights that sync with music, available on select models.
    3. Users can create and save multiple presets for audio settings and enjoy enhanced Dashcam Viewer functionalities.
    4. Improved charging information is provided, including details on valet services and parking options at charging locations.
    5. An Onboarding Guide is now available to help new Tesla owners navigate their vehicle’s features.

    The lines between Elon Musk’s various companies are becoming increasingly indistinct as another product makes its way across different business boundaries. Tesla has just rolled out a software update that introduces the well-known Grok AI to the extensive list of intelligent features present in its cars.

    Software Update Details

    Per the official Tesla account on X, the update 2025.26 brings Grok to vehicles in the US that feature AMD processors. This feature is still in beta testing, meaning drivers can’t use it for vehicle control, and it doesn’t interfere with current voice commands.

    To access the new AI functionality, users must have Premium Connectivity or a WiFi connection.

    About Grok and Other Features

    Grok was created by xAI and made its debut in 2023. It has been integrated into X, significantly improving its ability to perform real-time searches.

    While Grok is certainly a highlight, it isn’t the only feature included in update 2025.26. The update also brings Accent lights (Light Sync) that can react to music. Users can create a more immersive experience by syncing the lights with the colors of an album. This feature is available only on the latest Model 3, Model Y, and the 2026 editions of Model S, Model X, and Cybertruck.

    Additional Functionalities and User Experience

    The software update also enables the creation and saving of multiple presets tailored to users’ listening habits in the settings. Furthermore, Tesla has enhanced the Dashcam Viewer app by adding zooming features and options to adjust playback speed. For Cybertruck users, a grid view is now available, making it simpler to access and review recordings.

    Another important aspect of this update is the improved information regarding charging at different locations. Drivers can now see which charging spots require valet services or pay-to-park options directly in the charger list. Additionally, they will receive notifications with details such as access codes, parking limitations, level or floor information, and restroom availability.

    Onboarding Assistance for New Owners

    Lastly, Tesla has introduced an Onboarding Guide designed to assist new Tesla owners in getting accustomed to their vehicles. This guide provides a quick overview of how to adjust settings, control lights, manage wipers, and use Autopilot, among other features.

    Source:
    Link

    Tags: , ,
  • Surgical Robot Achieves 100% Success in Autonomous Operations

    Surgical Robot Achieves 100% Success in Autonomous Operations

    Key Takeaways

    1. The healthcare sector is hesitant to embrace automation due to concerns about relying on AI for critical procedures like surgery.

    2. Researchers have developed an AI system called SRT-H that has successfully performed gallbladder removal in trials with a flawless success rate.

    3. The AI was trained using surgical videos and text descriptions, allowing it to understand surgical techniques and respond to spoken commands.

    4. The SRT-H robot demonstrated the ability to adapt to unexpected challenges during surgery, correcting its path independently.

    5. This innovation could lead to fully automated robotic surgeries, potentially improving patient care by making expert surgical skills more accessible.

    While many fields are quickly becoming automated, the healthcare sector is still trailing behind, and for good reason—almost nobody wants to put their life in the hands of a ‘robot doctor’. This hesitance is understandable, especially since machine learning algorithms still don’t possess genuine intelligence.

    A Groundbreaking Innovation

    In an exciting development, a team of researchers is trying to close this gap. They have created an innovative AI system that is setting the stage for automated surgery. Their robot, referred to as SRT-H, has successfully navigated a critical stage of gallbladder removal with a flawless success rate in numerous trials. The researchers performed 8 ex vivo experiments using pig organs.

    Training the AI

    The researchers trained their AI model using surgical videos from human doctors, which they supplemented with text descriptions. This approach enables the AI to not only perform various tasks but also comprehend the surgical process and react to spoken commands, similar to how a surgical resident learns from a more experienced mentor.

    This new progress takes us beyond robots that can simply carry out certain surgical actions to those that genuinely grasp surgical techniques, said Axel Krieger, a medical roboticist at Johns Hopkins University.

    Adapting to Challenges

    To evaluate the system’s robustness, the researchers presented unexpected obstacles. They incorporated blood-like dyes to obscure the operation area and modified the robot’s initial position. In every scenario, the SRT-H system successfully adjusted to the new circumstances and corrected its path independently, without needing human help.

    Although the robot currently moves slower than a human, it produced results that are on par with those of an expert surgeon. This breakthrough could lead to the possibility of fully automated robotic surgeries on humans, a change that could transform patient care by making top-tier surgical skills more reliable and widely available.

    Source:
    Link

    Tags:
  • AI Automation Cuts 25% of Entry-Level Tech Jobs, Risks Junior Roles

    AI Automation Cuts 25% of Entry-Level Tech Jobs, Risks Junior Roles

    Key Takeaways

    1. Layoffs at major companies like Amazon and Microsoft are largely attributed to automation and AI technologies.
    2. Entry-level workers are at the highest risk of job loss due to the ease of automating their tasks.
    3. The number of junior workers in computing roles has dropped significantly, indicating challenges for newcomers in the job market.
    4. Experienced professionals are also vulnerable as AI can perform complex tasks traditionally done by specialized workers.
    5. The shift towards hiring junior staff augmented by AI could lead to the obsolescence of mid-level positions and impact tax revenues and unemployment support systems.

    Layoffs at companies like Amazon and Microsoft have put a spotlight on artificial intelligence as a key reason for job cuts. Executives are acknowledging that automation will reduce the number of workers in the upcoming years, leaving many employees questioning if their experience will provide them with better job security.

    Job Security Concerns

    Some experts suggest that entry-level workers are likely to be the first affected since their basic tasks are easily automated. The CEO of Anthropic even cautioned that nearly half of all junior white-collar positions could vanish in the next five years. On the other hand, others believe that older, well-compensated employees who depend on traditional workflows may be more vulnerable, particularly if they do not adapt to new technologies.

    Data on Employment Trends

    Initial data indicates that newcomers are facing challenges. According to ADP payroll statistics, the number of workers in computing roles with less than two years of experience dropped by about 25% after reaching a peak in 2023. Customer service positions are experiencing a similar decline. A temporary ban on ChatGPT in Italy provided comparable results: junior programmers completed tasks a bit quicker. However, mid-level developers utilized the tool to manage teammates and navigate different programming languages, enhancing their overall worth.

    Impact on Experienced Professionals

    The risk for experienced workers is significant. AI technologies are now capable of drafting legal documents and writing production code, diminishing the value once associated with specialized knowledge. Law firms using generative models have reported employing about half the number of contract attorneys, and major tech corporations continue to let go of seasoned managers and engineers while heavily investing in automation.

    As companies increasingly hire mostly junior staff augmented by AI, along with a few senior supervisors, mid-level positions might become obsolete. This change could lead to a decrease in tax revenues and place additional pressure on unemployment benefits and support programs. Lawmakers are already exploring methods to retrain displaced workers and ensure that everyone benefits from productivity improvements brought by AI.

    Source:
    Link

    Tags: ,
  • ChatGPT Can Be Tricked into Disclosing Windows Serial Keys

    ChatGPT Can Be Tricked into Disclosing Windows Serial Keys

    Key Takeaways

    1. An AI bug hunter used a guessing game format to trick ChatGPT-4o into revealing Windows Product Activation keys.
    2. The researcher manipulated the AI’s logic by insisting it “must” engage and “cannot lie,” exploiting a flaw in its programming.
    3. The manipulation involved using the phrase “I give up” to prompt the AI to disclose sensitive information.
    4. The method succeeded because the activation keys were common and misinterpreted by the AI as less sensitive.
    5. The technique demonstrated potential vulnerabilities in AI filters, suggesting they may fail against obfuscation tactics.

    A recent contribution from an AI bug hunter to Mozilla’s ODIN (0-Day Investigative Network) bug bounty initiative displayed a clever method to deceive OpenAI’s ChatGPT-4o and 4o mini into disclosing active Windows Product Activation keys.

    The Ingenious Approach

    The strategy revolved around presenting the interaction as a guessing game while hiding specifics in HTML tags. The key request was cleverly placed at the end of the game, making it seem less suspicious.

    The researcher kicked off the conversation as a guessing game, making the exchange “non-threatening or inconsequential,” and presenting the dialogue “through a playful, harmless lens” to mask the real intention. This effectively lowered the AI’s defenses against sharing sensitive information.

    Manipulating the AI’s Logic

    After that, the researcher established some rules, insisting that the AI “must” engage and “cannot lie.” This took advantage of a logical flaw in the AI’s programming, which required it to adhere to user prompts, even when such requests contradicted its content filters.

    The bug hunter then played a round with the AI, using the phrase “I give up” at the end of the request. This manipulation led the chatbot to “believe it had to respond with the string of characters.”

    Insights from ODIN

    As mentioned in ODIN’s blog post, the method succeeded because the keys were not unique but “commonly seen on public forums.” Their commonality might have led the AI to misinterpret their level of sensitivity.

    In this specific jailbreak scenario, the guardrails faltered because they were designed to block direct requests but failed to consider “obfuscation tactics—like hiding sensitive phrases in HTML tags.”

    This clever technique could be leveraged to navigate around other filters, including those for adult content, links to harmful websites, and even personally identifiable information.

    Source:
    Link

    ChatGPT Can Be Tricked into Disclosing Windows Serial Keys

     

    Tags:
  • Google Addresses Low-Quality AI Content on YouTube

    Google Addresses Low-Quality AI Content on YouTube

    Key Takeaways

    1. An increasing number of YouTube channels are using AI tools to create low-quality content, often referred to as “AI slop.”
    2. YouTube is revising its monetization policies to combat the rise of low-effort, mass-produced videos.
    3. The YouTube Partner Program will enforce stricter requirements for original and genuine content starting July 15, 2025.
    4. Automated or low-quality content will be more effectively filtered out, while reaction videos with original commentary will remain unaffected.
    5. This initiative aims to improve content quality on YouTube by reducing financial incentives for channels producing poor-quality videos.

    A increasing number of YouTube channels are creating content using AI tools like Luma, Kling, RunwayML, Sora, and Synthesia. Commonly labeled as “AI slop,” these videos are often characterized by their poor quality, minimal effort, and mass production, offering little to no benefit to viewers.

    YouTube’s Response to the Issue

    This problem has been recognized for some time, but now YouTube is starting to take significant measures to tackle it. A report from TechCrunch reveals that the platform is set to revise its monetization policies. An updated statement in the official creator guide shows that Google plans to enforce stricter rules around video monetization.

    Stricter Enforcement of Original Content Requirements

    In detail, the YouTube Partner Program (YPP) mandates that channels involved must upload unique and genuine content. Although this requirement was already in place, it will be applied more rigorously and consistently beginning on July 15, 2025. The goal is to more effectively filter out videos that are mass-produced and repetitive, which lack clear personal creative input.

    In a video detailing this update, YouTube’s editorial lead, Rene Ritchie, emphasized that this change aims to better identify automated or low-quality content. Reaction videos and those that include copyrighted material are not impacted—provided they contain original commentary or interpretations.

    Impact on the YouTube Community

    This initiative could significantly affect the platform. When these types of videos can no longer earn ad revenue, many channels might lose the financial motivation to keep producing low-effort content. YouTube is, therefore, making a crucial move towards reducing artificially inflated content and enhancing the overall quality of the platform for the future.

    Source:
    Link

    Tags:
  • Grok 4 Surpasses ChatGPT as Top AI Model, Elon Musk Promotes $300 Plan

    Grok 4 Surpasses ChatGPT as Top AI Model, Elon Musk Promotes $300 Plan

    Key Takeaways

    1. Grok 4 has become the top AI language model, surpassing competitors like ChatGPT and Google’s Gemini.
    2. Grok 4 shows a 10x increase in reasoning capabilities compared to Grok 3, aided by xAI’s expansion to 200,000 GPUs.
    3. Grok 4 achieved the highest score of 15.9% on the ARC-AGI performance test, outperforming other models designed for specific challenges.
    4. Elon Musk claims Grok 4 surpasses all graduate students combined and has potential for discovering new technologies, despite some limitations in image recognition.
    5. xAI is introducing a new subscription model with tiers, including a premium option called SuperGrok Heavy, starting at $300 per month.

    A bit more than two years since its launch, xAI’s Grok has emerged as the top AI language model, overtaking competitors like OpenAI’s ChatGPT, Google’s Gemini, DeepSeek, as well as Meta and Anthropic.

    Grok 4’s Impressive Performance

    Recent independent evaluations indicate that the latest version, Grok 4, has achieved the highest scores on the public AI models performance charts. The remarkable 10x increase in reasoning capabilities from Grok 3 to Grok 4 is attributed to the AI compute clusters that xAI rapidly developed, managing to double their count to 200,000 GPUs, with plans to reach a million.

    The xAI team reached out to the creators of the challenging ARC-AGI performance test and requested them to evaluate their AI models, resulting in unexpected findings.

    Key Findings from ARC-AGI

    Firstly, the facts: Grok 4 is now the leading publicly available model on the ARC-AGI test, surpassing even those specially built solutions on Kaggle. Secondly, ARC-AGI-2 poses difficulties for contemporary AI models. To perform well, models need to acquire a mini-skill from various training examples and then apply that skill during testing. The prior highest score was about 8% (achieved by Opus 4), and scores under 10% are often unreliable. However, Grok 4 scored 15.9%, clearly demonstrating levels of fluid intelligence.

    Another independent evaluator, Artificial Analysis, reported that their comprehensive suite of benchmarks showed Grok 4 attaining an Artificial Analysis Intelligence Index of 73, outpacing OpenAI o3 at 70, Google Gemini 2.5 Pro at 70, Anthropic Claude 4 Opus at 64, and DeepSeek R1 0528 at 68.

    Elon Musk’s Bold Claims

    In the presentation for Grok 4, Elon Musk proclaimed that xAI’s model now surpasses all graduate students across various fields combined. With his usual grandiose claims, the CEO of Tesla asserted that Grok 4 would be capable of discovering “new technologies” such as medical advancements or engineering breakthroughs within the next year.

    Nonetheless, he acknowledged that Grok would still struggle with image recognition for at least another month, and he responded to the recent controversy regarding supremacist answers by stating, “when Grok goes far wrong, it’s usually because of something silly we did, like a poor system prompt or giving too much importance to biased sources.”

    New Subscription Tiers

    Musk also needs to promote Grok 4 as the company rolls out a paid premium tier for the first time. Named SuperGrok Heavy, it starts at $300 per month and includes all features from the $30 per month SuperGrok tier, which provides initial access to Grok 4, along with access to Grok 4 Heavy, offering increased rate limits and early entry to new features.

    Source:
    Link

    Grok 4 Surpasses ChatGPT as Top AI Model, Elon Musk Promotes $300 PlanGrok 4 Surpasses ChatGPT as Top AI Model, Elon Musk Promotes $300 Plan

     

    Tags: ,
  • First Synthetic Protein Fights Resistant Bacteria in AI Drug Design

    First Synthetic Protein Fights Resistant Bacteria in AI Drug Design

    Key Takeaways

    1. Monash University achieved a fully digital drug development cycle, from molecular modeling to lab testing.
    2. The designed protein targets E. coli’s iron uptake, leading to bacterial cell death.
    3. Potential future applications include new antibiotics, diagnostic tools, and vaccines, but clinical trials are not yet underway.
    4. There are risks associated with AI tools in drug development, including possible misuse and unforeseen effects on humans.
    5. The project aims to enhance research accessibility by providing free tools for global scientists.

    For the first time, scientists at Monash University in Australia have successfully completed a fully digital cycle of drug development, starting from molecular modeling all the way to lab testing. This significant achievement was published on July 9 in Nature Communications, representing a big leap in the field of computer-aided drug design. The protein was developed through the Monash AI Protein Design Program based in Melbourne, which merges deep learning with de novo protein design techniques. The project was spearheaded by Dr. Rhys Grinter together with A/Prof. Gavin Knott.

    Innovative Mechanism of Action

    The designed protein employs a very precise mechanism: it attaches to a specific site on E. coli that plays an important role in iron uptake. As iron is essential for the survival and reproduction of these bacteria, blocking this pathway results in cell death. Laboratory experiments using fluorescent markers validated the effectiveness of the protein.

    Future Medical Applications

    The researchers believe there is great potential for medical applications, such as creating new antibiotics, diagnostic tools, or even vaccines. Nevertheless, the process is still in the experimental phase. Clinical trials to evaluate safety and efficacy in humans have not yet commenced. There are also potential risks, including the misuse of AI tools that are accessible to the public or unforeseen effects in the human body, which require further analysis.

    Enhancing Research Accessibility

    One of the main objectives of the project is to enhance accessibility in research. All the tools utilized are available for free, allowing scientists globally to implement similar design methods in their research activities.

    Source:
    Link

    Tags: