VentureBeat

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Advanced Micro Devices announced its Versal Premium Series Gen 2 chip platform for data center customers doing AI processing and other work. The AMD Versal is an adaptive FPGA, or field programmable gate array, platform for system-on-chip customers. It delivers accelerated performance for a wide range of workloads in data centers, communications, test and measurement, and aerospace and defense markets. AMD said the Versal Premium Series Gen 2 will be the FPGA industry’s first devices featuring Compute Express Link 3.1 and PCIe Gen6 as well as LPDDR5X memory support in hard intellectual property. These next-generation interface and memory technologies access and move data rapidly and efficiently between processors and accelerators for tasks such as AI processing. CXL 3.1 and LPDDR5X help unlock more memory resources faster to address the growing real-time processing and storage demands of data-intensive applications across markets. “System architects are constantly looking to pack more data into smaller spaces and move data more efficiently between parts of the system,” said Salil Raje, SVP of adaptive and embedded computing group at AMD, in a statement. “Our latest addition to the Versal Gen 2 portfolio helps customers improve overall system throughput and utilization of memory resources to achieve the highest performance for their most demanding applications from the cloud to the edge.” Using the open-standard interconnect, AMD said the new processors enable high-bandwidth host CPU-to-accelerator connectivity. PCIe Gen6 offers a two times to four times faster line rate compared to competing FPGAs with PCIe Gen4 or Gen5 support, AMD said. CXL 3.1 also offers similar benefits as well as enhanced fabric and coherency. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Leaders of AI projects today may face pressure to deliver quick results to decisively prove a return on investment in the technology. However, impactful and transformative forms of AI adoption require a strategic, measured and intentional approach.  Few understand these requirements better than Dr. Ashley Beecy, Medical Director of Artificial Intelligence Operations at New York-Presbyterian Hospital (NYP), one of the world’s largest hospitals and most prestigious medical research institutions. With a background that spans circuit engineering at IBM, risk management at Citi and practicing cardiology, Dr. Beecy brings a unique blend of technical acumen and clinical expertise to her role. She oversees the governance, development, evaluation and implementation of AI models in clinical systems across NYP, ensuring they are integrated responsibly and effectively to improve patient care. For enterprises thinking about AI adoption in 2025, Beecy highlighted three ways in which AI adoption strategy must be measured and intentional: Good governance for responsible AI development A needs-driven approach driven by feedback Transparency as the key to trust Good governance for responsible AI development Beecy says that effective governance is the backbone of any successful AI initiative, ensuring that models are not only technically sound but also fair, effective and safe. AI leaders need to think about the entire solution’s performance, including how it’s impacting the business, users and even society. To ensure an organization is measuring the right outcomes, they must start by clearly defining success metrics upfront. These metrics should tie directly to business objectives or clinical outcomes, but also consider unintended consequences, like whether the model is reinforcing bias or causing operational inefficiencies. Based on her experience, Dr. Beecy recommends adopting a robust governance framework such as the fair, appropriate, valid, effective and safe (FAVES) model provided by HHS HTI-1. An adequate framework must include 1) mechanisms for bias detection 2) fairness checks and 3) governance policies that require explainability for AI decisions. To implement such a framework, an organization must also have a robust MLOps pipeline for monitoring model drift as models are updated with new data. Building the right team and culture One of the first and most critical steps is assembling a diverse team that brings together technical experts, domain specialists and end-users. “These groups must collaborate from the start, iterating together to refine the project scope,” she says. Regular communication bridges gaps in understanding and keeps everyone aligned with shared goals. For example, to begin a project aiming to better predict and prevent heart failure, one of the leading causes of death in the United States, Dr. Beecy assembled a team of 20 clinical heart failure specialists and 10 technical faculty. This team worked together over three months to define focus areas and ensure alignment between real needs and technological capabilities. Beecy also emphasizes that the role of leadership in defining the direction of a project is crucial: AI leaders need to foster a culture of ethical AI. This means ensuring that the teams building and deploying models are educated about the potential risks, biases and ethical concerns of AI. It is not just about technical excellence, but rather using AI in a way that benefits people and aligns with organizational values. By focusing on the right metrics and ensuring strong governance, organizations can build AI solutions that are both effective and ethically sound. A need-driven approach with continuous feedback Beecy advocates for starting AI projects by identifying high-impact problems that align with core business or clinical goals. Focus on solving real problems, not just showcasing technology. “The key is to bring stakeholders into the conversation early, so you’re solving real, tangible issues with the aid of AI, not just chasing trends,” she advises. “Ensure the right data, technology and resources are available to support the project. Once you have results, it’s easier to scale what works.” The flexibility to adjust the course is also essential. “Build a feedback loop into your process,” advises Beecy, “this ensures your AI initiatives aren’t static and continue to evolve, providing value over time.” Transparency is the key to trust For AI tools to be effectively utilized, they must be trusted. “Users need to know not just how the AI works, but why it makes certain decisions,” Dr. Beecy emphasizes. In developing an AI tool to predict the risk of falls in hospital patients (which affect 1 million patients per year in U.S. hospitals), her team found it crucial to communicate some of the algorithm’s technical aspects to the nursing staff. The following steps helped to build trust and encourage adoption of the falls risk prediction tool: Developing an Education Module: The team created a comprehensive education module to accompany the rollout of the tool. Making Predictors Transparent: By understanding the most heavily weighted predictors used by the algorithm contributing to a patient’s risk of falling, nurses could better appreciate and trust the AI tool’s recommendations. Feedback and Results Sharing: By sharing how the tool’s integration has impacted patient care—such as reductions in fall rates—nurses saw the tangible benefits of their efforts and the AI tool’s effectiveness. Beecy emphasizes inclusivity in AI education. “Ensuring design and communication are accessible for everyone, even those who are not as comfortable with the technology. If organizations can do this, it is more likely to see broader adoption.” Ethical considerations in AI decision-making At the heart of Dr. Beecy’s approach is the belief that AI should augment human capabilities, not replace them. “In healthcare, the human touch is irreplaceable,” she asserts. The goal is to enhance the doctor-patient interaction, improve patient outcomes and reduce the administrative burden on healthcare workers. “AI can help streamline repetitive tasks, improve decision-making and reduce errors,” she notes, but efficiency should not come at the expense of the human element, especially in decisions with significant impact on users’ lives. AI should provide data and insights, but the final call should involve human decision-makers, according to Dr. Beecy. “These decisions require a level

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More One of the buzzwords of 2024 in AI has been agents, specifically the agentic future. AI agents have become one of the most talked-about trends for enterprises, and as more organizations look to implement agents, the future for agents may look rosy. In the next year, more enterprises could bring AI agents out of sandboxes and into production, making AI agents a big trend for 2025. Steve Lucas, CEO of platform integration company Boomi, said the conversation around AI agents picked up speed this year due to multiple factors in the growth trajectory of generative AI and models.  “I believe there are moments in time in the course of history where there’s convergence, and things come together to create an outcome we didn’t expect so soon,” Lucas said in an interview with VentureBeat. “You have near infinite compute, extraordinarily powerful GPU processing capabilities, data that is not near infinite, sprinkle in a fundamentally new way to take and process inputs and outputs that have all converged at the same time.”  In other words, AI agents became a big deal because we can see a path for these agents to actually work.  When organizations and AI companies talk about an agentic future, they usually mean a time when many tasks within an enterprise are automated. People will either prompt, or do a simple action, and AI agents will begin fulfilling those requests.  In the past few months, large service providers have begun offering access to agents to customers. Salesforce has gone all in on agents with the release of agents called Agentforce. Salesforce chairman and CEO Marc Benioff said during the launch of Agentforce that AI agents represent the “third wave of AI” which the company is very excited about helping to usher in.  It isn’t just Salesforce that is talking up AI agents. Slack will let customers integrate agents from Salesforce, Asana, Workday, Cohere, Writer and Adobe. ServiceNow updated its Now Assist platform with a library of ready-to-use agents, and AWS introduced Agents for Bedrock so clients can build custom agents more quickly.  Lucas and other experts VentureBeat spoke to agreed that 2024 is the year enterprises realize they can bring agents into their technology stack. The following year will bring more agent deployment, but multiple agents working together could still take some time to work well.  The momentum is not slowing down The various platforms available to access a library of agents or low-code ways to build custom agents make it easier for enterprises to consider using agents. The adoption of agents is already growing. A survey from Forum Ventures showed that among 100 senior IT leaders, it spoke to, 48% are ready to bring AI agents into operations. Around 33% said they are very prepared.  As they continue experimenting and figuring out good use cases for their organizations, 2025 will allow companies to test out production in small tasks for agents.  Deloitte Head of AI Jim Rowan said clients who’ve started limited tests of agents see the potential of agents “as skilled collaborators that enterprises have been searching for that understands personal preferences.” Boomi’s Lucas said his company is anticipating the number of customers using its agents “should go up 10x next year.” He said around 2,000 clients actively use Boomi’s agents.  However, while 2025 could see a boom in agents, some enterprises may also consider the cost of using agents widely. Paul van der Boor, vice president for AI at investment company Prosus, told VentureBeat that agentic use will only keep growing, but companies have to remember there is a cost inherent to this technology.  “The trajectory is not going change because I think the direction is clear,” van der Boor said. “Keep in mind that there’s also a lot of practical considerations because one of the things agents do is they require multiple calls to various elements, and they require more tokens, so they’re more expensive.”  AI agents will see evolution, too Lucas said the best use of agents is when they move from solitary actors to digital employees working with each other and human workers to complete tasks. But we won’t see multi-agents in production early in 2025. Lucas said what is most likely is the rise of agent islands. “You’ll have islands, like the Salesforce island, the Boomi island, the Oracle island. Over time, these agents will talk to each other,” he said.  The next few years could see the rise of agents taking more of a proactive role in the enterprise. Deloitte’s Rowan said some AI agents could become multipurpose agents that anticipate users’ needs. For example, the agent could proactively scan someone’s inbox, categorize inbound emails from clients, reference those with a list of priorities, tailor responses and flag any information to the employee.  “Over time, agents will level up on the cognitive nature of the task they’re performing. I don’t think we’re there yet because agents now are still operating more at the behest of the employee,” he said.  One future AI agent evolution could be a conductor or orchestration agent. Meta agents, one of the many terms for this concept, is an AI agent that directs traffic or actions of other agents.  Paul Tether, CEO of market intelligence firm Amplyfi, said the so-called Meta Agents is the ultimate next step for enterprise AI agents.  “By the end of next year, we’ll start to see meta agents emerge,” he predicted.  source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Puzzle, a San Francisco-based fintech startup, has launched an AI-powered accounting platform designed to automate up to 90% of routine tasks, allowing accountants to focus on more strategic work. In an exclusive interview with VentureBeat, Puzzle CEO Sasha Orloff outlined how the company’s new general ledger software integrates complex accounting policies directly into the platform, aiming to eliminate the need for manual spreadsheet processes. “What we’re launching now is effectively taking the general ledger, the backbone of accounting, and bringing complicated accounting logic from spreadsheets into the core accounting software,” Orloff said. The platform supports both cash and accrual accounting, offering a solution for businesses of all sizes. Orloff emphasized that the system is designed to provide real-time, accurate accounting tailored to the increasing demands of today’s fast-paced business environment, especially as the accounting industry faces a shortage of talent and growing workloads. Automating complex accounting tasks with Puzzle’s AI general ledger Puzzle’s platform addresses the challenges of manual accounting by automating processes like revenue recognition, asset depreciation, and prepaid expenses. Traditionally, these tasks require spreadsheets, which must then be reconciled with accounting software such as QuickBooks. “In QuickBooks, you typically have to calculate things like revenue recognition, fixed assets, and prepaid expenses manually in spreadsheets,” Orloff explained. “You’ll have QuickBooks open on one half of the screen and a spreadsheet on the other. With Puzzle, all of that logic and calculation is handled inside the software.” Puzzle allows users to set up accounting rules—referred to as “software-driven accounting policies”—for different types of transactions, such as SaaS subscriptions or prepaid contracts. “You save it, and then it just gets applied when an invoice or a bill comes in,” Orloff said. This automation reduces the risk of errors and eliminates much of the manual, time-consuming work accountants typically face. Ensuring accuracy with human-in-the-loop AI A key concern with AI-driven automation is ensuring accuracy, particularly in fields like accounting where precision is critical. Puzzle addresses this issue by allowing accountants to control the level of automation they use. Orloff described this flexibility, saying, “You can create a rule in our system that says, ‘Let the system take its best guess, and I’ll review it later,’ or ‘I want to do it manually.’ The accountant is always in control.” Puzzle tags each transaction with information about how it was processed, providing transparency. “Everything is tagged, so you know whether something was drafted by AI or if it’s a high-confidence transaction the system has handled before,” Orloff said. This feature allows accountants to trace transactions and verify their accuracy. By maintaining human oversight, Puzzle mitigates the risk of AI errors, or “hallucinations,” as Orloff called them. “AI can hallucinate, but humans make mistakes too,” Orloff said. “That’s why we designed a system where AI suggests things, but the accountant can verify and control everything.” Addressing the talent shortage in accounting with AI Puzzle’s launch comes at a critical time for the accounting profession. The industry is facing a severe talent shortage, with 75% of accountants nearing retirement, 300,000 having left the workforce, and CPA applications are down nearly 30%. Burnout rates are also high, with 99% of accountants reporting feeling overworked due to the repetitive nature of their jobs. Orloff sees Puzzle as a way to alleviate some of these pressures. “We’re seeing a massive transformation in accounting with the introduction of AI,” he said. Unlike competitors such as QuickBooks, which recently ran a campaign encouraging businesses to “fire your accountant,” Puzzle’s approach is to support rather than replace accountants. “We’re here to take accountants and accounting firms and make them the heroes of their companies,” Orloff said. He envisions AI-driven tools like Puzzle enabling accountants to play more strategic roles in businesses. “If we can move accountants from the back office to a seat at the table for the most important financial decisions, that’s a win for everyone,” Orloff said. “The role of an accountant will become higher paid and more impactful, with a focus on big-picture decisions instead of routine tasks.” Rapid Growth for Puzzle as AI Transforms Accounting Since Puzzle’s public launch less than a year ago, the platform has processed more than $30 billion in transactions for over 3,000 businesses, ranging from startups to small businesses using tools like Stripe, Gusto, and Brex. According to Orloff, Puzzle’s growth has been largely driven by word of mouth, with the company experiencing 15-20% month-over-month growth, 70% of which has been organic. While Puzzle initially gained traction with startups, demand from small businesses and accounting firms has grown significantly. “We started working with startup communities because they use modern tools and were eager to adopt new accounting solutions,” Orloff said. “But we began to see inbound interest from small businesses like doctors’ offices, law firms, and retail stores.” Accounting firms, in particular, are turning to Puzzle to manage more clients without increasing staff. “There’s been a shortage of accountants, and accounting firms are turning away clients,” Orloff explained. “With our automation, they can handle more business at higher margins, with greater customer satisfaction.” AI as a strategic advantage for the Future of Accounting Orloff believes that Puzzle’s platform represents the next step in the evolution of accounting. “When Excel came out, 1 million bookkeeping jobs were eliminated, but 1.2 million higher-paying advisory roles were created,” he said. “We’re going to see a similar shift today. The boring, repetitive work will be automated, and accountants will spend more time helping businesses devise tax strategies and improve their financial health.” Orloff sees Puzzle as a tool that not only benefits accountants but also the businesses they serve. “We’re building a system that makes accounting easier and more enjoyable, and that strengthens the relationship between the accountant and the business owner,” he said. “It’s a win-win.” As more businesses adopt Puzzle, the platform’s automation capabilities will continue to improve, creating a self-reinforcing cycle of efficiency and

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More When large language models (LLMs) emerged, enterprises quickly brought them into their workflows. They developed LLMs applications using Retrieval-Augmented Generation (RAG), a technique that tapped internal datasets to ensure models provide answers with relevant business context and reduced hallucinations. The approach worked like a charm, leading to the rise of functional chatbots and search products that helped users instantly find the information they needed, be it a specific clause in a policy or questions about an ongoing project. However, even as RAG continues to thrive across multiple domains, enterprises have run into instances where it fails to deliver the expected results. This is the case of agentic RAG, where a series of AI agents enhance the RAG pipeline. It is still new and can run into occasional issues but it promises to be a game-changer in how LLM-powered applications process and retrieve data to handle complex user queries. “Agentic RAG… incorporates AI agents into the RAG pipeline to orchestrate its components and perform additional actions beyond simple information retrieval and generation to overcome the limitations of the non-agentic pipeline,” vector database company Weaviate’s technology partner manager Erika Cardenas and ML engineer Leonie Monigatti wrote in a joint blog post describing the potential of agentic RAG. The problem of ‘vanilla’ RAG While widely used across use cases, traditional RAG is often impacted due to the inherent nature of how it works. At the core, a vanilla RAG pipeline consists of two main components—a retriever and a generator. The retriever component uses a vector database and embedding model to take the user query and run a similarity search over the indexed documents to retrieve the most similar documents to the query. Meanwhile, the generator grounds the connected LLM with the retrieved data to generate responses with relevant business context. The architecture helps organizations deliver fairly accurate answers, but the problem begins when the need is to go beyond one source of knowledge (vector database). Traditional pipelines just can’t ground LLMs with two or more sources, restricting the capabilities of downstream products and keeping them limited to select applications only.  Further, there can also be certain complex cases where the apps built with traditional RAG can suffer from reliability issues due to the lack of follow-up reasoning or validation of the retrieved data. Whatever the retriever component pulls in one shot ends up forming the basis of the answer given by the model. Agentic RAG to the rescue As enterprises continue to level up their RAG applications, these issues are becoming more prominent, forcing users to explore additional capabilities. One such capability is agentic AI, where LLM-driven AI agents with memory and reasoning capabilities plan a series of steps and take action across different external tools to handle a task. It is particularly being used for use cases like customer service but can also orchestrate different components of the RAG pipeline, starting with the retriever component. According to the Weaviate team, AI agents can access a wide range of tools – like web search, calculator or a software API (like Slack/Gmail/CRM) – to retrieve data, going beyond fetching information from just one knowledge source.  As a result, depending on the user query, the reasoning and memory-enabled AI agent can decide whether it should fetch information, which is the most appropriate tool to fetch the required information and whether the retrieved context is relevant (and if it should re-retrieve) before pushing the fetched data to the generator component to produce an answer.  The approach expands the knowledge base powering downstream LLM applications, enabling them to produce more accurate, grounded and validated responses to complex user queries.  For instance, if a user has a vector database full of support tickets and the query is “What was the most commonly raised issue today?” the agentic experience would be able to run a web search to determine the day of the query and combine that with the vector database information to provide a complete answer. “By adding agents with access to tool use, the retrieval agent can route queries to specialized knowledge sources. Furthermore, the reasoning capabilities of the agent enable a layer of validation of the retrieved context before it is used for further processing. As a result, agentic RAG pipelines can lead to more robust and accurate responses,” the Weaviate team noted. Easy implementation but challenges remain Organizations have already started upgrading from vanilla RAG pipelines to agentic RAG, thanks to the wide availability of large language models with function calling capabilities. There’s also been the rise of agent frameworks like DSPy, LangChain, CrewAI, LlamaIndex and Letta that simplify building agentic RAG systems by plugging pre-built templates together.  There are two main ways to set up these pipelines. One is by incorporating a single agent system that works through multiple knowledge sources to retrieve and validate data. The other is a multi-agent system, where a series of specialized agents, run by a master agent, work across their respective sources to retrieve data. The master agent then works through the retrieved information to pass it ahead to the generator.  However, regardless of the approach used, it is pertinent to note that the agentic RAG is still new and can run into occasional issues, including latencies stemming from multi-step processing and unreliability. “Depending on the reasoning capabilities of the underlying LLM, an agent may fail to complete a task sufficiently (or even at all). It is important to incorporate proper failure modes to help an AI agent get unstuck when they are unable to complete a task,” the Weaviate team pointed out.  The company’s CEO, Bob van Luijt, also told VentureBeat that the agentic RAG pipeline could also be expensive, as the more requests the LLM agent makes, the higher the computational costs. However, he also noted that how the whole architecture is set up could make a difference in costs in the long run. “Agentic architectures

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More A team of computer scientists has developed a method that helps artificial intelligence understand when to use tools versus relying on built-in knowledge, mimicking how human experts solve complex problems. The research from the University of California San Diego and Tsinghua University demonstrates a 28% improvement in accuracy when AI systems learn to balance internal knowledge with external tools — a critical capability for deploying AI in scientific work. How scientists taught AI to make better decisions “While integrating LLMs with tools can increase reliability, this approach typically results in over-reliance on tools, diminishing the model’s ability to solve simple problems through basic reasoning,” the researchers write in their paper. “In contrast, human experts first assess problem complexity using domain knowledge before choosing an appropriate solution approach.” The new method, called “Adapting While Learning,” uses a two-step process to train AI systems. First, the model learns directly from solutions generated using external tools, helping it internalize domain knowledge. Then, it learns to categorize problems as either “easy” or “hard” and decides whether to use tools accordingly. The two-step process researchers developed to teach AI systems when to use tools versus rely on internal knowledge, mirroring how human experts approach problem-solving. (Credit: UC San Diego / Tsinghua University) Small AI model outperforms larger systems on complex tasks What makes this development significant is its efficiency-first approach. Using a language model with just 8 billion parameters — far smaller than industry giants like GPT-4 — the researchers achieved a 28.18% improvement in answer accuracy and a 13.89% increase in tool usage precision across their test datasets. The model demonstrated particular strength in specialized scientific tasks, outperforming larger models in specific domains. This success challenges a fundamental assumption in AI development: that bigger models necessarily yield better results. Instead, the research suggests that teaching AI when to use tools versus rely on internal knowledge — much like training a junior scientist to know when to trust their calculations versus consult specialized equipment — may be more important than raw computational power. Examples of how the AI system handles different types of climate science problems: a simple temperature calculation (top) and a complex maritime routing challenge (bottom). (Credit: UC San Diego / Tsinghua University) The rise of smaller, smarter AI models This research aligns with a broader industry shift toward more efficient AI models in 2024. Major players including Hugging Face, Nvidia, OpenAI, Meta, Anthropic, and H2O.ai have all released smaller but highly capable models this year. Hugging Face’s SmolLM2, with versions as small as 135 million parameters, can run directly on smartphones. H2O.ai’s compact document analysis models have outperformed tech giants’ larger systems on specialized tasks. Even OpenAI entered the small model arena with GPT-4o Mini, offering similar capabilities at a fraction of the cost. This trend toward “AI downsizing” reflects growing recognition that bigger isn’t always better — specialized, efficient models can often match or exceed the performance of their larger counterparts while using far fewer computational resources. The technical approach involves two distinct learning phases. During training, the model first undergoes what the researchers call “World Knowledge Distillation” (WKD), where it learns from solutions generated using external tools. This helps it build up internal expertise. The second phase, “Tool Usage Adaptation” (TUA), teaches the system to classify problems based on its own confidence and accuracy in solving them directly. For simpler problems, it maintains the same approach as in WKD. But for more challenging problems, it learns to switch to using external tools. Business impact: More efficient AI systems for complex scientific work For enterprises deploying AI systems, this research addresses a fundamental challenge that has long plagued the industry. Current AI systems represent two extremes: they either constantly reach for external tools — driving up computational costs and slowing down simple operations — or dangerously attempt to solve everything internally, leading to potential errors on complex problems that require specialized tools. This inefficiency isn’t just a technical issue — it’s a significant business problem. Companies implementing AI solutions often find themselves paying premium prices for cloud computing resources to run external tools, even for basic tasks their AI should handle internally. On the flip side, organizations that opt for standalone AI systems risk costly mistakes when these systems attempt complex calculations without proper verification tools. The researchers’ approach offers a promising middle ground. By teaching AI to make human-like decisions about when to use tools, organizations could potentially reduce their computational costs while maintaining or even improving accuracy. This is particularly valuable in fields like scientific research, financial modeling, or medical diagnosis, where both efficiency and precision are crucial. Moreover, this development suggests a future where AI systems could be more cost-effective and reliable partners in scientific work, capable of making nuanced decisions about when to leverage external resources — much like a seasoned professional who knows exactly when to consult specialized tools versus rely on their expertise. The power of knowing when to ask for help Beyond the immediate technical achievements, this research challenges the bigger-is-better paradigm that has dominated AI development. In demonstrating that a relatively small model can outperform its larger cousins by making smarter decisions about tool use, the team points toward a more sustainable and practical future for AI. The implications extend far beyond academic research. As AI increasingly enters domains where mistakes carry real consequences – from medical diagnosis to climate modeling – the ability to know when to seek help becomes crucial. This work suggests a future where AI systems won’t just be powerful, but prudent – knowing their limitations just as skilled professionals do. In essence, the researchers have taught AI something fundamentally human: sometimes the smartest decision is knowing when to ask for help. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More To many enterprises, Box is a well-known file sharing and data collaboration application.  Over the course of the last year in particular Box has become a lot more, thanks to its generative AI efforts. Today those efforts are getting a huge boost with technologies that will remake how enterprise users can benefit from their own data. Box AI was announced in May 2023 as the company’s initial foray into using AI to help enable more utility from data and documents. Since then Box has added Microsoft 365 Copilot integration and AI-focussed hubs for curated search. Today at the company’s BoxWorks event, Box is pushing significantly forward with its new Box AI Studio and Box Apps technologies. Instead of just using AI to query and better understand data, the two new applications will enable organizations to use enterprise AI to build agent-driven workflows as well as applications. The announcement marks a transformative moment for the company, which has evolved from a secure file-sharing platform to an intelligent content management solution provider. “If we think about our path, we got to over a billion in revenue on that core foundation of secure sharing, collaboration and content management,” Aaron Levie, co-founder and CEO of Box, told VentureBeat. “Our path to 2 billion is much more going to be driven by these advanced set of content management and intelligent content management use cases.” New Box AI Studio introduces custom agents to drive advanced Enterprise AI workflows Leading the announcement is Box AI Studio, a new platform that enables enterprises to create and deploy custom AI agents.  The studio allows organizations to select AI models, implement custom instructions and deploy specialized agents across their enterprise environments. Built on existing partnerships with Anthropic, Google and OpenAI, the AI Studio platform is designed to support various business scenarios. For example, sales enablement teams can create custom agents that understand company-specific language and protocols while accessing consolidated content through Box Hubs. “Imagine if you’re the head of sales enablement at a company and you want people to be able to go and ask questions about any sales process,” Levie explained. “You might want to create a custom agent that knows how to use your business’s language and is effectively instructed to only answer questions in a way that conforms to your sales process or policies.” While Box AI Studio enables the creation of agents, Levie emphasized that it’s not yet a fully agentic AI capability. The concept of agentic AI typically also involves AI agents being able to act autonomously on behalf of users. “This is the initial, kind of foundational component for eventually doing agentic AI,” Levie explained. “We are letting you create agents, and those agents have a basic set of tools within the platform that they can use, and custom instructions and guardrails you can set up.” Box Apps will let enterprises build simple no code applications  With the new Box Apps technology, the company is set to solve a long standing challenge that enterprises have faced. That is, making data usable in a specific interface. Levie said that in the past enterprises would come to him and say that they have all their content in Box and they wanted to use Box as part of a contract management system for example. The challenge was that Box didn’t have the user interface component to enable an enterprise to easily build a contract management solution. That now changes with Box Apps. Box Apps allows users to build instant, intelligent applications directly within the Box platform, without the need for extensive custom development. The Box Apps provide a user interface and dashboard that gives users access to structured data and metadata extracted from the content stored in Box, which has not been possible before without building separate applications. Box Apps eliminates the need for external development or hosting, allowing organizations to build intelligent workflows directly within the Box environment. Box Apps is built on technology that Box acquired in January, called Crooze. It enables rapid development of applications for contract management, digital asset management, invoice processing, and other business-critical workflows. The new Box AI Studio and Box Apps capabilities will be part of a new subscription tier called Enterprise Advanced.  “This is the biggest set of new product enhancements that we’ve ever had as a company,” Levie said.  “It opens up a much broader set of use cases that we can solve for customers and this gets us into almost every line of business within the enterprise.” source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More AMD said today it will lay off 4% of its global staff, or perhaps somewhat less than 1,000 of its estimated 26,000-person workforce. The cuts come at a time when AMD has been soundly beating Intel in the x86 processor market. But AMD has also been second in the transition from graphics processing units (GPUs) to AI accelerators in competition with AI chip market giant Nvidia. AMD said through a spokesperson, ″As a part of aligning our resources with our largest growth opportunities, we are taking a number of targeted steps that will unfortunately result in reducing our global workforce by approximately 4%. We are committed to treating impacted employees with respect and helping them through this transition.” In an SEC filing last year, AMD said it had 26,000 employees. Today, AMD said only that the number of layoffs would be under 1,000. AMD’s stock has fallen this year while Nvidia’s is up around 200%, turning it into the most valuable public company in the world with a market cap of $3.6 trillion. AMD’s market value is $227 billion. AMD said in October it expects $5 billion in AI chip sales this year, about a fifth of the $25.7 billion. AMD has a stronghold in processors/GPUs for game consoles, but that market has been weaker than expected in this generation, partly due to the pandemic supply shortages for the PlayStation 5 and Xbox Series X/S. But Mercury Research reports that AMD’s share of processors against Intel is 34% now, up dramatically from years ago. source