Esther Mulu, a 69-year-old farmer in drought-stricken Kitui County, Kenya, struggles to feed her seven children and 25 grandchildren, surviving on just one meal a day. Researchers from the University of California, supported by Microsoft, are developing an AI-driven tool to predict food insecurity months in advance. This innovative tool uses a decade’s worth of data, including satellite imagery and weather patterns, to forecast malnutrition cases 3 to 6 months ahead. Since its early 2023 testing, it has expanded from six to cover all of Kenya, allowing for better resource allocation, particularly for children under five. AMREF’s Samuel Mburu emphasizes the importance of this predictive capacity in saving lives. However, scaling these cloud-based AI solutions poses financial challenges, despite initial funding. Optimistically, the tool aims to empower local health officials to intervene earlier with food packs and medical aid, preventing hunger from reaching crisis levels.
Source link
AI Tool Seeks to Anticipate Hunger in Kenya’s Drought-Stricken Areas
Exposing LLM Vulnerabilities: Understanding the Jailbreaking Potential of Major Models
The security of large language models (LLMs) extends beyond jailbreaking; these systems were fundamentally flawed in their design. CyberArk Labs has developed Fuzzy AI, which can jailbreak numerous LLMs, exposing vulnerabilities across models like ChatGPT and Claude. The issue transcends simple hacks; if compromised, LLMs can misinterpret instructions, leading to severe consequences, especially in enterprise settings. The divide between academic AI security research and real-world vulnerabilities exacerbates this problem, as rapid AI development often renders academic findings obsolete. Techniques such as “Operation Grandma” exploit this gap, revealing how easily LLMs can be manipulated. As AI evolves toward agentic systems that execute tasks and make decisions, the risks multiply, necessitating robust security measures that are currently lacking. The opaque nature of AI decision-making compounds these risks, making it difficult to detect compromised systems. Overall, LLMs are not designed with security as a priority, creating a critical need for transparency and proactive security measures in AI development.
Source link
The Advantages of Small Language Models for Enterprises – Computerworld
The rise of AI in enterprises has highlighted the use of large-language models (LLMs), with tools like LiteLLM enabling streamlined access to various LLMs for developers. This allows for unified integration and real-time monitoring. However, there’s growing interest in small language models (SLMs) among InfoWorld readers. Smart Answers, a proprietary LLM, advocates for SLMs based on input from IT leaders, emphasizing their numerous advantages. SLMs are agile, cost-effective, and support rapid prototyping while enhancing security and data privacy. They provide better customization options, making them an optimal choice for organizations looking to maintain control over their data. In summary, while LLMs dominate the conversation, SLMs offer significant benefits that can suit specific organizational needs better.
Source link
BigID Advocates for HR to Train Employees in Effective and Responsible AI Tool Usage
Avi Tzach, Senior Global Technical Recruiter at BigID, emphasizes the permanence and utility of AI tools in the workplace, particularly in HR and recruitment. In CTech’s “HR in the AI Era” series, he discusses how these tools enhance efficiency, streamline processes, and improve candidate sourcing and communication without replacing employees. BigID, founded in 2016, has grown to over 500 employees and raised over $320 million, focusing on AI and data. Despite challenges from regional instability and the impact of the ongoing war, the company continues to expand globally. AI tools, including sourcing tools like Teamable and analytics platforms like Watercooler, are integral to their operations, facilitating improved task completion across various departments. Tzach acknowledges both the benefits and risks of AI, urging careful adoption while prioritizing employee support during turbulent times. BigID is committed to growth, maintaining a strong hiring outlook in diverse international markets.
Source link
NYT’s Amazon Partnership Marks a New Era for Publisher Collaborations
The New York Times has signed an AI licensing deal with Amazon, marking its first such agreement, which allows Amazon products like Alexa to use summaries and excerpts from NYT content. This move signals a shift in the NYT’s stance on AI partnerships, highlighting an openness to collaboration with tech companies under the right conditions. The deal may strengthen NYT’s position in its ongoing legal battle against OpenAI, establishing a market for licensing content for AI training. While Amazon’s AI model, Nova, isn’t as widely recognized as those from competitors like OpenAI and Google, the partnership is expected to focus on enhancing Alexa’s capabilities. Interestingly, the NYT’s decision to ally with Amazon over other tech giants, especially those it is suing, indicates a strategic approach to safeguard its copyright while also exploring new revenue possibilities. Overall, the deal reflects a broader trend of digital publishers engaging in licensing arrangements with AI firms.
Source link
Samsung Galaxy Devices Set to Welcome a New Gemini Rival – Sammy Fans
Samsung is set to introduce a new artificial intelligence (AI) feature called “Gemini” on its Galaxy devices, competing with Google’s Gemini AI. This innovation aims to enhance user experience by integrating advanced AI functionalities across various applications and services. The Gemini AI is designed to facilitate more intuitive interactions, improve productivity, and provide personalized responses based on user behavior. Samsung’s effort reflects a growing trend among tech companies to incorporate AI capabilities into their products, ensuring competitiveness in the evolving digital landscape. With this move, Samsung reaffirms its commitment to integrating cutting-edge technology in its devices, potentially offering a seamless and smarter user experience that rivals existing AI models. This initiative could also position Samsung favorably within the AI market, as it seeks to leverage its vast ecosystem of devices and services to maximize the effectiveness and reach of its new AI features.
Source link
Grammarly Raises $1 Billion in Funding Round
Grammarly has secured $1 billion in financing from General Catalyst’s Customer Value Fund to enhance product innovation, expand sales and marketing, and pursue strategic acquisitions. This funding follows Grammarly’s recent acquisition of AI productivity tool maker Coda, transforming it into an AI-driven productivity platform. CEO Shishir Mehrotra expressed confidence that this partnership will accelerate Grammarly’s growth, allowing it to better serve its 40 million users and generate over $700 million in annual revenue. Grammarly’s AI assistant supports users in various writing tasks and promotes improved communication. General Catalyst’s managing director, Pranav Singhvi, sees this investment as crucial for Grammarly’s expansion, particularly in enterprise adoption. The integration of Coda is viewed as a significant milestone toward realizing Grammarly’s vision of AI collaboration across work environments, aiming to optimize writing and communication for businesses.
Source link
Introducing Enhanced Fine-Tuning Models and Techniques in Azure AI Foundry
Azure AI Foundry has introduced significant enhancements in model fine-tuning with Reinforcement Fine-Tuning (RFT) for the upcoming o4-mini model and Supervised Fine-Tuning (SFT) for the GPT-4.1-nano model. RFT, geared for complex decision-making environments, allows models to adaptively respond to nuanced business logic through feedback mechanisms, enhancing their contextual understanding and reasoning capabilities. This method has already benefitted legal tech startup DraftWise, improving contract generation accuracy by 30%.
Simultaneously, SFT for GPT-4.1-nano enables organizations to align model outputs with specific terminologies and workflows, particularly beneficial for high-volume scenarios like customer support. Additionally, Llama 4 Scout fine-tuning has been made available, offering a robust model infrastructure with advanced customization options. These upgrades reflect Azure AI Foundry’s commitment to providing tools for developing tailored, efficient AI systems that meet diverse organizational needs.
Source link
Zing Coach Review: Is This AI Fitness App The Key to Achieving Your Goals?
Zing Coach is an innovative AI fitness app designed to provide personalized training experiences, making it accessible for users who may face barriers in accessing traditional personal trainers due to cost, scheduling, or location. The app features a chatbot for workout advice, personalized plans, and a Body Scan function for body composition analysis. With over 1 million users and more than 5,600 five-star reviews, it offers various subscription plans, including a free version with limited features. Key functionalities include an AI Coach that adapts workouts based on user feedback, real-time exercise monitoring, diverse workout options, and community support for motivation. While it excels in user-friendliness and variety, some drawbacks include inconsistencies in workout equipment suggestions and technology limitations. At around $9.99 per month, Zing Coach is a cost-effective alternative to human trainers, making it valuable for anyone looking to enhance their fitness journey at home.
Source link
Inside Uber’s Innovation: How the Product Chief Leverages AI for Success
Uber’s Chief Product Officer, Sachin Kansal, revealed his use of AI tools in a recent podcast episode. He utilizes OpenAI’s ChatGPT and Google’s Gemini to summarize lengthy reports, often ranging from 50 to 100 pages, which he lacks time to read. This helps him stay informed about rider sentiments in different markets like South Africa and Brazil. Additionally, he employs AI as a research assistant; for instance, he consulted ChatGPT regarding driver feedback on a potential new feature, finding it valuable for brainstorming. Kansal also mentioned his intention to use NotebookLM, Google Lab’s note-taking tool, particularly for its podcasting feature that summarizes content through AI discussions. NotebookLM, launched in mid-2023, has gained popularity among researchers and AI enthusiasts. Uber’s CEO, Dara Khosrowshahi, is pushing for more employees to adopt AI tools for coding, emphasizing their importance for the company’s future.
Source link