Category: Artificial intelligence

  • MIT Discovers Brain’s Separate Systems for Solids and Fluids

    MIT Discovers Brain’s Separate Systems for Solids and Fluids

    Key Takeaways

    1. The human brain has specialized areas for processing solid objects and non-solid materials, identified in a study published in Current Biology.
    2. The research reveals subregions in the brain’s visual cortex that respond differently to “things” (solids) and “stuff” (liquids).
    3. Researchers used video clips of objects interacting with environments while monitoring participants’ brain activity with fMRI.
    4. Both shape recognition and physical property analysis areas in the brain show distinct reactions to solids and liquids.
    5. These findings could inform the development of advanced AI systems with separate processing models for solids and liquids, enhancing their interaction with the environment.

    In research, neuroscientists found out that the human brain has different specialized areas for processing solid objects compared to non-solid materials. This study, published on July 31 in the Current Biology journal, is the first to identify specific regions in the visual cortex that correspond to this distinction.

    New Insights on Recognition

    Previously, it was known that the brain features specialized areas for recognizing 3D objects. This new research goes further, showing that within the brain’s shape-recognition pathway and the one that analyzes physics, there are subregions that react differently to solid items and flowing materials. The researchers referred to these categories as “things” and “stuff.”

    Research Methodology

    To conduct their study, the team utilized software typically used by visual effects artists to create more than 100 video clips showcasing things and stuff interacting with various environments. Participants watched these videos while their visual cortex was scanned using fMRI (functional magnetic resonance imaging). The results indicated that both the area associated with shape recognition and the one linked to analyzing physical properties reacted to both stuff and things, highlighting specialized subregions for each type of object.

    Implications for AI Development

    This discovery could lay the groundwork for creating more advanced AI robots. Similar to the human brain, AI systems and robotic vision could be designed with distinct computational models for solids and liquids, enabling them to better perceive and engage with their physical environment.

     

    Tags: ,
  • AI Hiring Surges as Applicants Push Back Strongly

    AI Hiring Surges as Applicants Push Back Strongly

    Key Takeaways

    1. Companies are using AI bots for initial job interviews to improve efficiency and manage high volumes of applications.
    2. Many job seekers find automated interviews impersonal and may view them as a negative reflection of company culture.
    3. Candidates often experience confusion during AI interviews due to the lack of human interaction and support.
    4. AI interview vendors claim their tools effectively identify top candidates, but skepticism remains among job seekers.
    5. AI is limited in assessing cultural fit, which still requires human judgment during the hiring process.

    Companies feeling the pressure to sort through a large number of job applications are turning to AI bots for the first round of interviews. Human resources teams mention that these technologies can arrange calls, ask standard questions, and provide lists of qualified candidates, significantly improving efficiency, particularly in high-demand sectors like retail, customer service, and entry-level tech jobs. According to workplace trends expert Priya Rathod, AI “is taking care of that initial stage work that employers need to be more effective and save time.”

    Impersonal Experience for Applicants

    Despite the benefits for employers, many job seekers find the process to be lacking in personal touch or even disrespectful. Debra Borchardt, an experienced editor, decided to exit her first automated interview within minutes, calling the added layer of automation “a step too far” after searching for work for months. Some candidates see the lack of human contact as a warning sign about the company culture, leading them to turn down job offers that involve screening by bots.

    Doubts About AI Interviews

    Allen Rausch, a technical writer who has faced three AI interviews since losing his job, reported that the digital avatars failed to answer fundamental questions about the employers, which left him confused about what to do next and skeptical about the whole process. He stated he might only consider future bot interviews if there was a promise of a follow-up conversation with a real person.

    Vendors of these AI bots maintain that the tools are functioning as they should. Adam Jackson, CEO of Braintrust, a platform that handles numerous automated interviews simultaneously, argues that the rejection of this concept is not as widespread as some might think. He emphasizes that clients are pleased since the software accurately identifies the top 10 percent of candidates for human evaluation.

    The Limitations of AI in Hiring

    Nonetheless, there is a significant limitation: while AI can confirm skills and measure responses against set criteria, assessing cultural fit is still best done by humans. Jackson does admit that the technology “wouldn’t even try” to determine the cultural fit that becomes important once a candidate is shortlisted.

    Source:
    Link

     

  • AI Automation’s Impact on Knowledge Jobs: Microsoft Copilot Study

    AI Automation’s Impact on Knowledge Jobs: Microsoft Copilot Study

    Key Takeaways

    1. Generative AI has significant implications for jobs involving communication and information processing, impacting professions like journalism, technical writing, and translation.
    2. Jobs that require physical tasks or interpersonal relations, such as caregiving and skilled trades, are less likely to be affected by automation.
    3. The current wave of automation is different from past technological shifts, affecting higher-skilled, knowledge-based professions rather than just simple, routine tasks.
    4. Companies and workers must integrate AI tools into workflows and focus on training to stay competitive in the evolving job market.
    5. A strategic approach to human resources and organizational planning is essential for adapting to the changes brought by generative AI.

    In a new study titled “Working with AI: Measuring the Occupational Implications of Generative AI,” Microsoft Research looked at more than 200,000 anonymized conversations with its AI assistant, Bing Copilot. The study aimed to evaluate the real potential for automation, relying on actual tasks performed with AI support rather than just theoretical scenarios. Findings indicate that tasks involving communication and information processing strongly align with the capabilities of generative AI.

    Impact on Various Professions

    Microsoft’s research points out that this trend particularly impacts certain professional groups, including journalists, technical writers, mathematicians, and translators. In these fields, researchers found a significant link between common job responsibilities and the functionalities offered by AI.

    A notable statement from the publication asserts:

    “Jobs with high-information tasks and intensive communication are more susceptible to AI-driven automation.”

    Jobs Less Affected by Automation

    On the other hand, roles that involve physical tasks or interpersonal relations—like caregiving, skilled trades, and transportation—are seen as being at much lower risk. These jobs demand human interaction, a physical presence, or actions tailored to specific situations, which generative models cannot easily imitate.

    Occupations facing the least threat include nurses, massage therapists, and bricklayers. These professions require physical presence, specialized skills, and direct contact with people.

    The Shift in Automation

    Conversely, translators, interpreters, journalists, and mathematicians are more significantly impacted by the automation brought on by generative AI. Their work mostly revolves around intensive information processing and communication activities, which are increasingly being complemented or even replaced by AI systems.

    The study emphasizes that the current wave of automation is structurally different from past technological shifts. Unlike previous advancements that mainly targeted simple, routine tasks, today’s transformation is affecting higher-skilled, knowledge-based professions.

    Call to Action for Companies and Workers

    There is a pressing call for action among businesses and workers alike. The integration of AI tools into existing workflows, along with training and strategic changes, is becoming crucial for success. Staying competitive now requires not just technical skills but also foresight in organizational planning and a long-term approach to human resources strategies.

    Source:
    Link

     

    Tags: ,
  • Google Signs Deals to Cut AI Energy Use During Peak Hours

    Google Signs Deals to Cut AI Energy Use During Peak Hours

    Key Takeaways

    1. Google has signed agreements with U.S. power companies to reduce electricity use at AI data centers during peak demand times.
    2. The initiative aims to improve grid stability and speed up the connection of new data centers without needing more power infrastructure.
    3. This approach to energy management for AI operations is novel compared to practices in other sectors like manufacturing and cryptocurrency mining.
    4. Google is committed to workforce development, investing $10 million to train electricians and support new energy source approvals.
    5. Energy consumption from AI is expected to triple by the end of the decade, highlighting the need for innovative solutions in energy management.

    Google has made its first agreements with U.S. power companies to reduce electricity consumption at its AI data centers during times when the power grid is heavily loaded. This initiative is part of a broader strategy to tackle increasing energy challenges as AI technologies rapidly evolve.

    New Agreements and Their Purpose

    The contracts, signed with Indiana Michigan Power and the Tennessee Valley Authority, allow these power providers to request Google to temporarily decrease certain AI operations, such as machine learning processes that require significant electricity. Google states that this adaptability will speed up the connection of new data centers, decrease the necessity for constructing additional power infrastructure, and assist in maintaining grid stability during heatwaves or periods of high power demand.

    “Large electricity loads like data centers can now be interconnected more quickly,” Google mentioned in a blog entry, noting that the program is advantageous for both grid operators and utility consumers.

    A Novel Approach to Energy Management

    While reducing power usage during peak times is a common practice in sectors like manufacturing and cryptocurrency mining, specifically doing this for AI operations is relatively novel. Google’s strategy draws on previous experiments where it shifted tasks to different locations or times based on the availability of cleaner or less expensive energy.

    The tech giant also highlighted its commitment to workforce development and grid policy. In April 2025, Google revealed a $10 million initiative to train more electricians and supported efforts to expedite approvals for new energy sources, including small modular reactors and geothermal energy.

    The Future of AI Energy Use

    As energy consumption from AI is projected to triple by the decade’s end, officials have cautioned that data centers might expand quicker than the power supply in some regions of the U.S. Google’s actions could serve as a model for other cloud service providers, although similar agreements have not yet been confirmed by companies like Microsoft or Amazon.

    Source:
    Link

     

    Tags: ,
  • Tesla Awards Elon Musk $29 Billion Stock for Leadership Retention

    Tesla Awards Elon Musk $29 Billion Stock for Leadership Retention

    Key Takeaways

    1. Tesla has granted Elon Musk $29 billion in restricted shares to encourage his leadership in AI and robotics.
    2. Musk’s previous $56 billion compensation was rejected by the courts, following a legal challenge from investor Richard J. Tornetta.
    3. The new compensation plan requires Musk to hold a senior leadership position for two years and includes a five-year holding period.
    4. If the 2018 CEO Performance Award is reinstated by the courts, Musk will lose the new payment arrangement.
    5. The temporary pay structure aims to increase Musk’s voting rights gradually, addressing his concerns about control and potential ousting by investors.

    Tesla is making significant efforts to keep its CEO, Elon Musk, in charge, and has granted him substantial compensation. A letter shared by the Special Committee of the Boards of Directors on Tesla’s X account reveals that Musk will get $29 billion in restricted shares to motivate him to guide the company as it evolves into “a leader in AI, robotics and related services.”

    Background on Compensation Issues

    Musk was supposed to receive a large compensation in shares back in 2018, which had been approved by a vote from shareholders. Yet, one investor, Richard J. Tornetta, contested the pay package in court, leading to Chancellor Kathaleen McCormick of a Delaware Chancery Court deciding in 2024 that Musk should not get the $56 billion payment.

    In another vote, Tesla shareholders again supported the award, but Chancellor McCormick rejected the decision once more. Interestingly, the court ordered Tesla to pay Tornetta’s legal team $345 million in fees, instead of the roughly $5.5 billion in Tesla shares they had originally sought.

    New Payment Structure

    The recently established payment structure consists of 96 million shares, but it mandates that Musk must hold a senior leadership position continuously for two years. Additionally, there is a required holding period of five years starting from the grant date.

    However, this payment will be lost if the 2018 CEO Performance Award is fully reinstated by the courts in Delaware, as Tesla is currently appealing that ruling.

    Musk has expressed concerns about potentially losing control over Tesla and the possibility of being ousted by activist investors. This temporary pay arrangement aims to alleviate his worries by gradually increasing his voting rights as the shares are granted.

    Source:
    Link

     

    Tags: ,
  • Anthropic Cuts Access to OpenAI’s Claude API Over Terms Breach

    Anthropic Cuts Access to OpenAI’s Claude API Over Terms Breach

    Key Takeaways

    1. Anthropic revoked OpenAI’s access to Claude models due to a breach of commercial agreements.
    2. OpenAI allegedly used Claude for internal testing of its coding tools before GPT-5’s release, violating contract terms.
    3. OpenAI defended its actions as standard industry practice for benchmarking and expressed disappointment over the access suspension.
    4. Anthropic plans to restore limited access for benchmarking and safety evaluations but did not clarify the details.
    5. Withdrawing API access from competitors is a common practice in the tech industry, as seen with previous actions by Facebook and Salesforce.

    Anthropic has taken back OpenAI’s developer access to the Claude family of large-language models, claiming a breach of commercial agreements, as several sources informed Wired on Tuesday.

    Statement from Anthropic

    Christopher Nulty, a spokesperson for Anthropic, mentioned that OpenAI’s internal engineers had been “utilizing our coding tools before the release of GPT-5.” This act is viewed by the company as a clear violation of contract terms that prohibit using Claude to “create a competing product or service” or to reverse-engineer the model.

    Details of the Incident

    The same report highlights that OpenAI linked Claude to exclusive testing tools to evaluate code generation, creative writing, and safety performance against its own systems. The testing allegedly involved sensitive topics, including self-harm and defamation. OpenAI’s communication lead, Hannah Wong, referred to this process as “standard practice” in the industry for benchmarking and showed disappointment over the access suspension, pointing out that OpenAI’s API is still available to Anthropic.

    Future Access and Industry Practices

    Anthropic mentioned that limited access for “benchmarking and safety evaluations” would be reinstated, but did not specify how this restriction would function in real life. The company has a history of limiting competitors’ access: in July, it restricted the start-up Windsurf after rumors suggested the company was linked to an acquisition attempt by OpenAI. At that time, Chief Science Officer Jared Kaplan commented that “selling Claude to OpenAI” would be “strange.”

    Withdrawing APIs from competitors is a common strategy in the tech industry; Facebook notably prevented Vine from using its APIs, and Salesforce recently limited Slack APIs for competing collaboration applications. Anthropic’s decision came just a day after it reduced Claude Code access for all users, citing a surge in demand and violations of service terms.

    Source:
    Link

     

    Tags: , ,
  • Apple Develops Its Own Generative AI Search Engine in Silence

    Apple Develops Its Own Generative AI Search Engine in Silence

    Key Takeaways

    1. Apple has created the Answers, Knowledge and Information (AKI) group to develop an “answer engine” that competes with services like ChatGPT.
    2. Siri currently lacks a conversational search feature and relies on typical Google results, raising concerns about consumer demand for chatbots.
    3. The AKI team, led by Robby Walker, is working on a separate application and enhancing existing services like Siri, Spotlight, and Safari.
    4. Apple faces competition and potential disruption from antitrust issues regarding its deal with Google, while exploring partnerships and acquisitions in AI.
    5. Talent loss within Apple, particularly from the Apple Foundation Models team, raises concerns about the company’s ability to develop its own search engine without third-party models.

    Apple has set up a new group known as Answers, Knowledge and Information (AKI) which aims to develop an “answer engine” that can search the web and provide conversational results. This initiative marks Apple’s first major move towards creating its own competitor to services like ChatGPT.

    Siri’s Limitations

    Currently, Siri can send questions to ChatGPT, but it doesn’t have its own conversational search feature and often resorts to typical Google results. Some executives within Apple have raised doubts about how much consumers really want chatbots. However, the global adoption of services like ChatGPT and Gemini shows that there are risks involved in not innovating.

    Leadership and Development

    The AKI team is headed by Robby Walker, who previously managed Siri. The team is working on both a separate application and new backend systems designed to enhance Siri, Spotlight, and Safari in upcoming software updates. Recent job postings indicate that Apple is looking for engineers skilled in search algorithms, suggesting that the company wants to control the fundamental technology instead of just integrating existing solutions.

    Competitive Landscape

    At the same time, Apple is facing increasing competition. The antitrust case from the U.S. Justice Department could disrupt Apple’s profitable deal that makes Google the default search engine on iOS, which is estimated to be worth around $20 billion each year. Also, generative AI is making it easier for competitors to enter the market: Apple has been looking at partnerships with Perplexity AI and is reportedly very open to acquisitions as it increases its investment in AI infrastructure.

    Talent Challenges

    Moreover, Apple’s internal capabilities are being challenged by the loss of talent. In the past month, four important members of the Apple Foundation Models team have moved to Meta’s new super-intelligence lab, attracted by better pay and the chance to work on more advanced technologies. Their exit raises questions about whether Apple might need to use third-party large-language models for Siri while it continues to develop its own search engine.

    Future Outlook

    All these factors suggest that Apple is gearing up to combine on-device privacy with a proprietary generative search experience. This strategy aims to decrease reliance on Google, keep AI talent within the company, and offer a unique Apple-branded alternative to ChatGPT and Gemini in the future.

    Source:
    Link

     

    Tags: , ,
  • AI-Powered Ransomware: Chatbots Negotiate with Victims

    AI-Powered Ransomware: Chatbots Negotiate with Victims

    Key Takeaways

    1. The Global Group is using chatbots to communicate directly with victims for the first time, automating ransom requests based on encrypted file analysis.

    2. Automation allows for continuous negotiations with multiple victims simultaneously, requiring human intervention only in complex situations.

    3. The chatbot adjusts its messaging based on the victim’s profile to apply psychological pressure, aiming to hasten ransom payments.

    4. The Global Group operates a ransomware-as-a-service model, providing infrastructure and tools while partners carry out attacks, affecting at least 17 companies worldwide.

    5. The use of AI by cybercriminals is changing the threat landscape, requiring IT security experts to develop AI-driven solutions to counter automated communication.

    For the first time ever, the Global Group has started using chatbots to talk directly to its victims. As per Axios, this AI system examines encrypted sample files, checks for successful encryption, and automatically sends out ransom requests.

    Automation in Communication

    Human intervention is only required when a situation becomes more complicated or escalates. This automated messaging enables continuous negotiations with several affected parties at once, providing a significant level of scalability.

    Tailored Psychological Pressure

    As reported by Cybersecuritynews, the chatbot customizes the language, tone, and frequency of its messages based on the victim’s profile to increase psychological pressure. When combined with traditional extortion methods, this automated system aims to disturb targets and push them to pay quickly.

    The Global Group’s platform has been operational since June 2025. Picus Security suggests that the organization has created a ransomware-as-a-service model, offering the necessary infrastructure, encryption tools, and chatbots while its partners are responsible for infecting target systems. A mobile dashboard is available for operatives to oversee and manage ongoing attacks. Axios highlights that at least 17 companies in the US, UK, Australia, and Brazil have been affected.

    The Shift in AI Usage

    The use of AI by cybercriminals marks a significant change: artificial intelligence is not just a defensive tool anymore. IT security experts now face the task of quickly recognizing automated communication patterns and responding with their own AI-driven solutions. This situation underscores how dynamic and technologically intricate today’s threat landscape has become.

    Source:
    Link

     

    Tags:
  • Humanoid Robot Assists with Household Chores: Laundry Skills Shown

    Humanoid Robot Assists with Household Chores: Laundry Skills Shown

    Key Takeaways

    1. Humanoid robots like Figure’s Model 02 are advancing in technology, showcasing new capabilities for household chores.
    2. The Model 02 can load a washing machine, demonstrating impressive accuracy and smooth movements.
    3. There are requests for more comprehensive video footage to better understand the robot’s laundry process.
    4. Concerns about the authenticity of the robot’s smooth movements have been addressed, attributing it to the company’s in-house AI model, Helix.
    5. Figure is targeting industrial customers for the Model 02, with expected pricing higher than some existing models, making it uncertain if it will be affordable for average consumers.

    Humanoid robots have made great strides in technology in the recent times. Clips of these androids have been shared widely on the internet, showcasing them boxing, engaging in football, or executing impressive kung fu stunts. While these performances are certainly captivating, they don’t provide much real value for everyday life. However, the Model 02 by Figure is a notable exception.

    Robot’s New Capabilities

    A new video posted by Figure’s founder Brett Adcock on X features the 02 as it takes on the task of laundry. Although it can’t yet run the machine by itself, it shows it can load it up – giving us a peek into how humanoid robots may help with chores at home in the future.

    To be honest, the Figure 02 does take a bit of time to load the washing machine, but it demonstrates an impressive level of accuracy. For example, it notices when laundry items are sticking out of the drum and adjusts its actions accordingly. Many viewers have expressed their admiration for the 02’s abilities, especially its smooth movements. However, there’s a frequent request for more comprehensive video footage that captures the entire process.

    Questions About Authenticity

    Some viewers have raised doubts about the video’s genuineness, pointing to the robot’s unusually seamless movements. In response, Adcock clarified that this smoothness comes from Helix, the in-house AI model developed by the company. Figure intends to offer Helix to other robot makers down the line, which could pave the way for more humanoid robots taking on household duties. Nonetheless, it’s still unclear if these robots will be affordable for the average consumer.

    While no pricing details have been revealed yet, Figure is targeting industrial customers with the 02. Its price is anticipated to be much higher than Unitree’s R1, which may attract individual users, although likely more for fun rather than practical purposes.

    Source:
    Link

     

    Tags: ,
  • Tencent Launches 4 Compact Open-Source Hunyuan Models

    Tencent Launches 4 Compact Open-Source Hunyuan Models

    Key Takeaways

    1. Tencent has launched new compact Hunyuan models with 0.5 billion, 1.8 billion, 4 billion, and 7 billion parameters, optimized for low-power devices and edge deployment.

    2. These models excel in language comprehension, math, and reasoning tasks, thanks to their innovative “fusion reasoning” architecture.

    3. They feature a native 256K token context window, enabling them to process large amounts of text in a single inference, useful for applications like meeting transcript analysis.

    4. The models are compatible with popular inference frameworks and have early support from companies like Arm, Qualcomm, Intel, and MediaTek for tailored deployment packages.

    5. Practical applications demonstrate their effectiveness, such as rapid spam filtering in Tencent Mobile Manager and improved conversation quality in smart-cabin assistants.

    Tencent has unveiled a fresh lineup of compact Hunyuan models: 0.5 billion, 1.8 billion, 4 billion, and 7 billion parameters. These models are designed for low-power and edge deployment. All four versions can now be found on GitHub and Hugging Face and can perform inference using a single consumer-grade graphics card. This makes them ideal for devices with limited resources like laptops, smartphones, and smart-cabin systems.

    Impressive Performance

    Even though they are small, these models deliver top scores in tasks such as language comprehension, math, and reasoning on various public benchmarks. Tencent credits a unique “fusion reasoning” architecture for these impressive results, which allows users to switch between a quick-response mode for straightforward answers and a slower mode for detailed, multi-step reasoning.

    Technical Highlights

    A standout feature of these models is their native 256K token context window, capable of processing about 500,000 English words in one go. Tencent points out that their in-house tools like Tencent Meeting and WeChat Reading utilize these models to analyze complete meeting transcripts or entire books at once, preserving character relationships and plot elements for further inquiries.

    Integration and Endorsements

    The four compact LLMs are compatible with popular inference frameworks, such as SGLang, vLLM, and TensorRT-LLM, and they support a variety of quantization formats. Early support from companies like Arm, Qualcomm, Intel, and MediaTek suggests that deployment packages tailored for their specific processors are on the way.

    Practical applications from early users highlight the release’s focus on real-world utility. Tencent Mobile Manager has achieved rapid spam filtering in milliseconds without needing to send data off the device. Additionally, a dual-model approach in Tencent’s smart-cabin assistant optimizes power consumption while enhancing conversation quality. Tencent believes these examples show that smaller models can provide enterprise-level capabilities when designed thoughtfully.

    Source:
    Link

    Tencent Launches 4 Compact Open-Source Hunyuan ModelsTencent Launches 4 Compact Open-Source Hunyuan Models

     

    Tags: