Category: AI

IBM has announced the release of its Granite 4.0 Nano AI models, as reported by VentureBeat. These Nano AI models, ranging from 350 million to 1.5 billion parameters, represent a shift towards efficient and accessible AI solutions.

The key feature of these models is their ability to run locally, even directly in web browsers, showcasing advancements in edge computing capabilities. Unlike traditional AI models that require substantial compute resources, IBM’s Nano models are designed to operate efficiently on consumer hardware or edge devices without relying on cloud infrastructure.

By prioritizing strategic scaling over raw scale, these compact models demonstrate impressive performance comparable to larger counterparts, challenging the notion that bigger models equate to better outcomes. IBM’s approach towards smaller, more efficient AI models signifies a significant advancement in the industry, emphasizing the importance of practical deployment and accessibility.

Moreover, the release of the Granite 4.0 Nano models under the Apache 2.0 license opens up opportunities for researchers and developers, enabling them to leverage these advanced AI capabilities for various applications, including commercial usage.

IBM’s innovative approach not only reshapes the AI landscape but also sets a new standard for compact, high-performance models that cater to the evolving needs of developers and businesses in an increasingly edge-centric computing environment.

Source: VentureBeat

Microsoft is expanding the capabilities of its Copilot AI assistant to empower non-technical workers in app development and automation. The new tools, App Builder and Workflows, allow users to create applications, automate tasks, and develop specialized AI agents through simple conversations, eliminating the need for coding.

The integration of natural language prompts enables Copilot to generate full-stack business applications, complete with databases and security controls, based on user descriptions. App Builder utilizes Microsoft Lists for data storage and facilitates easy sharing of applications. Workflows automate tasks across Microsoft’s suite of products, enhancing productivity. Additionally, a simplified version of Copilot Studio allows users to create specialized AI assistants tailored to specific tasks.

These new capabilities are included in the existing Microsoft 365 Copilot subscription at no extra cost, aligning with Microsoft’s strategy of offering substantial value at a low price point. By integrating development tools into the conversational window of Copilot, Microsoft aims to democratize software development and transform how users interact with AI-powered productivity tools.

While Microsoft positions the tools as accessible to all office workers, professional developers remain crucial for systems interacting with external parties due to security and risk considerations. The democratization of software development raises governance and complexity concerns, which Microsoft addresses through administrative controls in the Microsoft 365 admin center.

Source: VentureBeat

Alibaba’s Qwen team has unveiled a significant enhancement to its Qwen Deep Research tool, allowing users to effortlessly create research reports, interactive web pages, and multi-speaker podcasts with just a few clicks. This update, a proprietary release distinct from Qwen’s previous open-source models, leverages Qwen3-Coder, Qwen-Image, and Qwen3-TTS to power its functionalities.

By enabling users to convert research into diverse formats seamlessly, Qwen aims to streamline the content creation process. The core workflow involves user collaboration within the Qwen Chat interface, data extraction from various sources, and customized code generation when necessary.

Users can choose to generate professional web pages or podcasts, all hosted by Qwen for easy sharing and playback. This multifaceted approach caters to different consumption preferences, providing written, visual, and audible content options.

While the tool’s focus on new content creation differs from Google’s NotebookLM, which is more oriented towards organizing existing data, the initial feedback on Qwen Deep Research has been positive. The tool’s availability through the Qwen Chat app emphasizes its accessibility for users seeking a comprehensive research-to-publish solution.

Qwen’s foray into multi-format content creation signifies a shift towards holistic research tools that combine guidance, analysis, and output creation. As the platform continues to evolve, it will be interesting to see how it balances convenience and depth in research production.

Source: VentureBeat

Google Cloud and DeepMind have introduced a framework, Watch & Learn (W&L), to address the challenge of training computer-use agents (CUAs) efficiently. This framework aims to alleviate the data bottleneck in CUA development by automatically generating high-quality training examples without human annotation, using raw videos for demonstrations.

By utilizing the data produced by Watch & Learn, companies can enhance their existing computer-use models and create custom CUAs for internal tasks, eliminating the need for expensive specialized training models. The framework’s approach not only improves performance on computer-use tasks but also enables in-context learning examples for CUAs, facilitating real-world application without extensive manual intervention.

Watch & Learn’s methodology revolves around redefining the creation of CUA demonstrations by focusing on the ‘inverse dynamics objective.’ This strategy involves predicting intermediate actions between consecutive observations rather than generating trajectories directly, leading to more robust and generalized outcomes across applications.

The framework’s three key stages include training an inverse dynamics model, retrieving raw videos, and training CUA agents. By employing this process, the researchers were able to generate a substantial corpus of state transitions and produce annotated trajectories with high-accuracy action labels.

In testing, Watch & Learn demonstrated improvements in fine-tuning open-source models and enhancing the performance of general-purpose multimodal models for in-context learning. These advancements showcase the scalability and practicality of utilizing web-scale human workflows to advance CUAs towards real-world deployment.

This framework streamlines CUA development and enables enterprises to leverage existing video resources for training data, paving the way for more efficient and cost-effective CUA implementations.

Source: VentureBeat

Businesses are witnessing a transformation in the way they leverage technology, particularly with the integration of AI-powered cameras redefining business intelligence. As reported by VentureBeat, these intelligent devices are no longer just tools for safety but have become crucial sources of real-time data and operational insights.

IP cameras, coupled with artificial intelligence, are empowering companies to extract valuable business intelligence, enhance operational efficiency, and gain a competitive edge. By treating cameras as vision sensors and operational insight sources, businesses can convert everyday visibility into measurable business value.

The integration of embedded AI in these intelligent devices has opened avenues for generating actionable insights that feed directly into intelligence platforms, ERP systems, and real-time dashboards, leading to significant improvements in various sectors. For instance, in manufacturing, AI-enabled cameras are detecting production line defects early, preventing costly errors. In retail, these cameras are optimizing product placement based on customer journey mapping. Even in healthcare, facilities are using these solutions to enhance patient care and reduce operational costs.

Companies like BMW, Google Cloud, A.C. Camargo Cancer Center, and Vanderbilt University are already leveraging the benefits of AI-powered cameras to enhance business intelligence and operational efficiencies across different industries.

The future of AI in video intelligence holds promises of predictive operations, versatile analytics, technological collaboration, and sustainability initiatives. Axis Communications stands at the forefront of these advancements, offering open-source, scalable systems that address current challenges while paving the way for future opportunities.

As AI continues to evolve, traditional views of IP cameras and edge devices are giving way to innovative approaches that promise not only stronger security but also improved business outcomes. The value of these technologies lies in their capabilities, making them indispensable for business success in an era of advancing artificial intelligence.

Source: VentureBeat

Google Cloud is expanding its offerings in the AI training space with the launch of Vertex AI Training, a service designed to cater to enterprises seeking to train their own models at scale. This move positions Google Cloud as a competitor to providers like CoreWeave and AWS, offering a managed Slurm environment, data science tooling, and access to chips capable of large-scale model training.

Jaime de Guerre, senior director of product management at Google Cloud, highlighted the growing demand from organizations of various sizes for better compute optimization in a reliable environment. The Vertex AI Training service targets companies engaged in large-scale model training rather than simple fine-tuning, focusing on longer-running jobs spanning hundreds or even thousands of chips.

Customizing AI models is becoming more prevalent as enterprises seek to create models tailored to their specific needs. Google Cloud’s Vertex AI Training differentiates itself by providing access to a wide range of chips, training management services, and expertise gained from training Gemini models.

Early adopters of Vertex AI Training include AI Singapore and Salesforce’s AI research team, demonstrating the service’s appeal to organizations building customized models. While training AI models can be expensive and complex, services like Vertex AI Training aim to simplify the process by offering managed environments for efficient model training.

Source: VentureBeat

Anthropic, a San Francisco-based AI company, has launched an integration of its Claude AI system with Microsoft Excel. This move aims to transform the financial services industry by enabling financial analysts to utilize the AI system directly within Excel, enhancing efficiency and accuracy in financial modeling and analysis. Anthropic’s expansion into Microsoft Copilot Studio and Researcher agent demonstrates its commitment to becoming a leading AI platform for banks and asset managers. The company’s strategic Excel integration targets the core of modern finance, offering a transparent and collaborative tool to navigate complex financial models.

Moreover, Anthropic’s data partnerships with major financial information providers like Aiera, Third Bridge, and LSEG further solidify its position by providing access to crucial market data and proprietary research. By building data moats around its financial AI platform, Anthropic ensures high-quality inputs for its AI outputs, enhancing its competitive edge in the industry.

The introduction of pre-configured workflows known as Agent Skills streamlines common financial tasks, offering automation for financial analysts. Anthropic’s success in the financial services sector is evident through partnerships with prominent clients like AIA Labs at Bridgewater and Commonwealth Bank of Australia, showcasing significant productivity gains and efficiency improvements.

While the AI deployment landscape in finance is rapidly evolving, regulatory uncertainty poses challenges and opportunities for AI adoption. Anthropic’s emphasis on responsible AI use and human oversight addresses concerns surrounding AI in finance, ensuring trust and reliability in decision-making processes.

As competition intensifies among major tech companies targeting finance AI, Anthropic’s strategic Excel integration and data partnerships position it as a formidable player in the industry.

Source: VentureBeat

MiniMax-M2, the latest open-source large language model (LLM) from the Chinese startup MiniMax, has emerged as a capable solution for enterprise AI applications. This model, available under the permissive MIT License, demonstrates impressive performance in reasoning, coding, and task-execution benchmarks, rivaling proprietary systems like GPT-5 and Claude Sonnet 4.5. MiniMax-M2’s efficient Mixture-of-Experts architecture enables high-end agentic and developer workflows suitable for enterprise deployment, while maintaining a manageable activation footprint.

With a focus on practicality and cost-efficiency, MiniMax-M2 offers scalable performance through its sparse model design, allowing for faster execution and reduced compute requirements. The model’s benchmark leadership across agentic and coding workflows positions it as a compelling option in the AI landscape, catering to tasks like automated support, R&D, and data analysis within enterprise environments.

MiniMax-M2’s competitive pricing further enhances its appeal, making it an attractive option for organizations seeking high-performance AI models at a reasonable cost. Additionally, MiniMax’s emphasis on structured tool calling and interleaved thinking format underscores its suitability for autonomous developer agents and AI-augmented operational tools, adding a new dimension to enterprise AI capabilities.

As MiniMax continues to innovate and expand its offerings, the company is emerging as a key player in the open-source AI space, providing accessible and efficient models for real-world applications. MiniMax-M2’s release marks a significant milestone in open-source AI development, offering enterprises a reliable and transparent solution for intelligent systems that prioritize controllable reasoning and practical utility.

Source: VentureBeat