Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Swave Photonics, a holographic display company, has raised $28.27 million in funding as it prepares components for AI-powered smartglasses and heads-up displays. Swave said the Series A investment will catalyze the advancement of its Holographic eXtended Reality (HXR) platform, enabling a reality-first user experience for AI-powered augmented reality (AR) smartglasses and heads-up displays. The company will show its tech at CES 2025. The funding round was co-led by investors Imec.xpand and SFPIM Relaunch, with participation from new investors EIC Fund, IAG Capital Partners, and Murata Electronics North America, as well as existing investors Qbic Fund, PMV, Imec, and Luminate. Leuven, Belgium-based Swave previously raised a $10.47 million seed round in 2023, which propelled the launch of Swave’s HXR technology, as well as the expansion of Swave’s team, which has veterans in photonics and semiconductors. “This round will accelerate Swave’s product introductions as we continue to solve the challenges of today’s AR experiences through true holography,” said Mike Noonen, Swave CEO, in a statement. “We are thrilled with continued support from our existing investors and our new investors. They recognize that Swave uniquely brings together semiconductor, holographic and AI technologies in a way that will deliver cost-effective and truly useful solutions.” Swave is bringing NanoPixel holography to glasses. “AR glasses are set to become the primary interface for AI-powered spatial computing and other applications, and Swave is uniquely positioned to enable this future” said Theo Marescaux, Swave and chief product officer, in a statement. “We are co-designing every element—from our holographic SLMs with cutting-edge nano-pixels, to real-time compute chips, light engines, and AR combiners—delivering the most advanced and integrated solution yet.” “With Swave’s seed funding, we successfully built our team, proved the capabilities of the technology, and completed prototype designs”, said Dmitri Choutov, COO, in a statement. “With Series A funding secured and silicon running at our partner fabs, we are on track to introduce product development kits and soon thereafter production devices.” Swave’s HXR technology uses what it calls the “world’s smallest pixel” to shape light and sculpt high-quality 3D holographic images that create a reality-first user experience, where digital information interacts and adapts to the user’s surroundings. The images allow for the human vision system to process them naturally leveraging patented DynamicDepth technology. AR devices currently being prototyped or on the market are all faced with challenges of high cost, uncomfortable size and weight, significant power usage, and visual phenomena like Vergence-Accomodation Conflict, which cause nausea or fatigue for users. Swave’s unique HXR technology not only solves these issues, but also eliminates the need for the most costly components, such as waveguides or varifocal lenses, inherently required for existing AR devices.  Swave’s technology has been developed for over a decade and the company currently holds 60 core technology patents. Swave announced its HXR platform in April 2024, followed by the achievement of the world’s first true color holographic display, and recently announced that HXR will be recognized at CES 2025 with a CES Innovation Award. source

By Allison Grande ( January 2, 2025, 10:44 PM EST) — A California federal judge Thursday blocked the state from beginning its enforcement of a new law designed to bar online platforms from using algorithms to deliver addictive feeds to children, finding there was “great value” in giving the Ninth Circuit 30 days to consider his decision to largely uphold the measure. … Law360 is on it, so you are, too. A Law360 subscription puts you at the center of fast-moving legal issues, trends and developments so you can act with speed and confidence. Over 200 articles are published daily across more than 60 topics, industries, practice areas and jurisdictions. A Law360 subscription includes features such as Daily newsletters Expert analysis Mobile app Advanced search Judge information Real-time alerts 450K+ searchable archived articles And more! Experience Law360 today with a free 7-day trial. source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Nvidia has completed its acquisition of Run:ai, a software company that makes it easier for customers to orchestrate GPU clouds for AI, and said that it would open-source the software. The purchase price wasn’t disclosed, but was pegged by reports at $700 million when Nvidia first reported its intent to close the deal in April. Run:ai posted the deal news on its website today and also said that Nvidia plans to open-source the software. Run:ai’s software remotely schedules Nvidia GPU resources for AI in the cloud. Neither company explained why Run:ai will open-source its platform, but it’s probably not hard to figure out. Since Nvidia has grown to be the number one maker of AI chips, its stock price has soared to $3.56 trillion, making it the most valuable company in the world. That’s great for Nvidia, but it makes it hard for it to acquire companies because of antitrust oversight. A spokesperson for Nvidia said in a statement only that “We’re delighted to welcome the Run:ai team to Nvidia.” When Microsoft acquired Activision Blizzard for $68.7 billion, it appeased antitrust regulators by licensing Activision’s Call of Duty game to other platforms for a decade to address worries that the company would become too powerful in gaming. The same might be happening here. Run:ai founders Omri Geller and Ronen Dar said in a press release that open-sourcing its software will help the community build better AI, faster. “While Run:ai currently supports only Nvidia GPUs, open-sourcing the software will enable it to extend its availability to the entire AI ecosystem,” Geller and Dar said. They said they will continue to help their customers to get the most out of their AI Infrastructure and offer the ecosystem maximum flexibility, efficiency and utilization for GPU systems, wherever they are: on-prem, in the cloud through native solutions, or on Nvidia DGX Cloud, co-engineered with leading CSPs. The founders also said, “True to our open-platform philosophy, as part of Nvidia, we will keep empowering AI teams with the freedom to choose the tools, platforms, and frameworks that best suit their needs. We will continue to strengthen our partnerships and work alongside the ecosystem todeliver a wide variety of AI solutions and platform choices.” The Israel-based company said its goal when it was founded in 2018 was to be a driving force in the AIrevolution and empower organizations to unlock the full potential of their AI infrastructures. “Over the years, our world-class team has achieved milestones that we could only dream of back then. Together, we’ve built innovative technology, an amazing product, and an incredible go-to-market engine,” the founders said. Run:ai helps customers to orchestrate their AI Infrastructure, increase efficiency and utilization, and boost the productivity of their AI teams. “We are thrilled to build on this momentum, now as part of Nvidia. AI and accelerated computing are transforming the world at an unprecedented pace, and we believe this is just the beginning,” the Run:ai founders said. “GPUs and AI infrastructure will remain at the forefront of driving these transformative innovations and joining Nvidia provides us an extraordinary opportunity to carry forward a joint mission of helping humanity solve the world’s greatest challenges.” Nvidia has been a longtime maker of graphics chips, and those chips have become a lot more useful in recent years in running AI software. Now the company is also emphasizing software, and this acquisition is aimed at giving customers maximum choice, efficiency and flexibility for GPU orchestration software. Nvidia and Run:ai have been working together since 2020 and they have joint customers. TLV Partners led the seed round for Run:ai in 2018. Rona Segev, managing director of TLV, said in a statement, “The AI market in early 2018 seemed like a different world. OpenAI was still a research company and Nvidia’s market cap was ‘only’ around $100 billion. We met Omri and Ronen who painted a picture for us of what the future of AI would look like. In their vision of the future, AI was ubiquitous.” Segev added, “Everyone on the planet would be interacting with AI daily, and it would be obvious that every company would be leveraging AI in one way or another. The only thing preventing that vision from becoming a reality, according to them, was the lack of efficiency and [the] costs associated with training AI models and running them in production on multiple GPU clusters. To solve this problem, Omri and Ronen pitched an idea of creating an orchestration layer between AI models and GPUs that would enable a much more efficient use of the underlying compute resources leading to faster training times and significantly reduced costs.” And Segev said, “Of course, this was all theoretical at the time as they hadn’t yet incorporated a company, let alone a product. We didn’t know much about the industry at the time. But there was something special about Omri and Ronen. They had a unique combination of intellect, charm, craziness and humility that created the perfect recipe for the type of founders we’re looking to back.” source

Interactive Voice Response (IVR) systems are often the first point of contact between a business and its customers. A well-crafted IVR script sets the tone for the interaction, helping callers navigate quickly and efficiently to the support they need. Clear, intuitive scripts not only save time for both customers and agents but also reduce frustration, leaving a positive impression of your brand. Conversely, a poorly designed IVR can feel like a maze, driving callers to hang up or escalate their frustration to your agents. This can harm your reputation and increase operational costs. Below, I’ve curated 21 IVR script examples that you can adapt to your needs. After the list, I’ve included a brief tip section to help you develop scripts with the best chance of delivering an excellent customer experience. 1 RingCentral RingEx Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Medium (250-999 Employees), Large (1,000-4,999 Employees), Enterprise (5,000+ Employees) Medium, Large, Enterprise Features Hosted PBX, Managed PBX, Remote User Ability, and more 2 Talkroute Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Any Company Size Any Company Size Features Call Management/Monitoring, Call Routing, Mobile Capabilities, and more How to use this list of IVR scripts My goal is to offer you script examples for your entire IVR call flow, so I have broken down the process into seven common interactions where an IVR script is typically required: Greeting scripts. Menu options scripts. On-hold scripts. After-hours scripts. Maximum wait time scripts. IVR payment scripts. IVR survey scripts. For each of these types of IVR scripts, I have offered three variations. My hope is that you will be able to find one that aligns with your business needs. Along the way, I’ve included guidance and links to relevant content. Improving scripts will help you maximize containment rate for IVR call flows by ensuring customers get clear information and take full advantage of self-service options. 21 ready-to-use scripts and examples Greeting IVR scripts It’s critical to provide a straightforward greeting that lets callers know they have the correct number and what to expect moving forward. Even if you have never done it before, it’s not hard to create an IVR recording that sounds professional, on-brand, and contributes to a positive first impression. 1. General greeting Welcome to [Your Business Name]! We’re happy to help. Please choose from the following options, or press 0 to speak with a customer support specialist. 2. Promotional greeting Welcome to [Your Business Name], where [insert your tagline here]. Press 1 to learn more about [the promotion currently being offered] or choose from the following options. 3. Customer service greeting We’re sorry you’re experiencing difficulties with our product. Our team is here to make it right. Please choose from the following options, or press 0 at any time to speak with a representative. Scripts for different menu options With menu scripts, the rule of thumb is to say the option first, followed by the number selection. This is a simple way to make it easier for callers to navigate IVR phone trees — they hear the word they are looking for and then the correct number to press. 4. General menu script Welcome to [Your Business Name]. For information about our products and services, press 1. For billing and payment inquiries, press 2. For technical support, press 3. To speak with a member of our sales team, press 4. For business hours and location information, press 5. To speak with a representative, press 0. To hear these menu options again, press #. 5. Promotion menu script Welcome to [Your Business Name]! We’re excited to offer [promotion] throughout the month of [month]. To take advantage of this promotion, press 1. To learn about new products, press 2. To inquire about discounts and offers, press 3. For assistance with placing an order, press 4. To speak with a sales representative, press 0. To hear these menu options again, press #. 6. Customer feedback menu script Thank you for calling [Your Business Name]. We value your business and appreciate your feedback. To participate in a customer satisfaction survey, press 1. To provide feedback on your recent experience, press 2. To leave a testimonial, press 3. For general inquiries, press 4. To speak with a customer feedback representative, press 5. To hear these menu options again, press #. Scripts for callers on hold Dealing with high call queuing times is hard on call center staff, but it’s also annoying for customers who are waiting to speak to an agent. Providing respectful messaging to customers on hold is crucial. Tip: consider implementing a queue callback option and adding it to your on-hold IVR scripts. This lets callers hang up and receive a callback once an agent is available, which is a win-win. Callers can get off hold and agents have the flexibility to deal with inquiries at a more manageable rate. 7. General on-hold script Thank you for choosing [Your Business Name]. Your call is important to us, we look forward to being able to assist you shortly. Please stay on the line to speak to a representative. You can also say, “Call me back” to schedule a call back from our next available agent. You will not lose your place in line. 8. Estimated wait time script At [Your Business Name], we understand your time is valuable. The current estimated wait time is approximately [X] minutes. We appreciate your patience. In the meantime, consider asking our system for help with simple inquiries. You can say things like, “Make an appointment,” “Check my balance,” or “What are your opening hours?” 9. On-hold promo script Thank you for holding. To show our appreciation, we’re offering you [details about the promotion]. Press 1 to claim this offer, or mention it to our customer service representative after connecting. We’ll be with you shortly. After-hours IVR scripts Unless you have agents available to answer calls 24/7, you’ll want to have an after hours

The Digital Markets, Competition and Consumers Act is designed to regulate the behaviour of major digital firms with significant market power in the U.K. The legislation grants the Competition and Markets Authority new powers to impose requirements on tech companies with “Strategic Market Status,” reminiscent of the “gatekeeper” organisations that must abide by the E.U.’s Digital Markets Act. However, while these laws share similarities, the new legislation is less one-size-fits-all: Under the DMCCA, the CMA can apply bespoke regulations, so-called “Conduct Requirements,” to companies with SMS to address their specific issues. Issues might relate to: The DMCCA was created in response to a report published in 2019 by the Digital Competition Expert panel, chaired by Jason Furman an economic policy professor at Harvard University and former chief economist to U.S. President Barack Obama. It contained recommendations to open up digital markets in the U.K. While the DMCCA was greenlit in 2020, due to various delays, it was only passed by Parliament in April 2023 and given Royal assent in May 2024. It is expected to come into force in January 2025. SEE: Google Abusing Dominant Position in Ad Tech Sector, Says U.K. Government What is the aim of the Act? The DMCCA aims to improve competitive conditions in digital markets by enabling interventions that encourage investment, innovation, and growth in all U.K. tech companies, ultimately providing consumers with access to the best possible technologies for them. The CMA stated in a press release that the legislation will enable it to investigate Google, Apple, and other large tech firms’ potentially anticompetitive practices “more holistically.” The rules “will build on and leverage its experience in areas it has already studied, such as mobile ecosystems, which includes app stores.” In a January 2024 interview, Sarah Cardell, CEO of the CMA, said: “The new regime is specifically designed to keep pace with developments in fast-moving digital markets, complementing our existing competition and consumer protection powers. “The DMCC Bill will establish a very targeted approach to address the substantial and entrenched market power of a small number of firms. This will ensure that challenger firms can bring forward genuinely disruptive and exciting new innovations that will create great new products for consumers.” What’s hot at TechRepublic Who will the Act impact? After a formal investigation, the CMA will designate “a very small number” of firms as having SMS and, therefore will be subject to the DMCCA. These firms must have: A “substantial and entrenched market power in a digital activity that is linked to the UK.” A “position of strategic significance.” A global turnover of more than £25 billion or a U.K. turnover of more than £1 billion. SMS designations will be reviewed a minimum of every five years. According to Cardell, between three and four will be launched in the first year of the new regime. In August, the CMA rejected Google’s proposed policy changes regarding purchases made within apps listed on its Play Store, which had spurred an investigation. This suggested that the company would be one of the first to achieve SMS because, if the CMA accepted the changes, it would be limited in what actions it could take under the DMCCA. What will the Act enable the U.K. government to do? The DMCCA gives new enforcement powers to a new group established inside the CMA called the Digital Markets Unit. It counts existing CMA directors and a former Ofcom exec as members. The DMU will draft a unique set of “conduct requirements” for each company that has SMS. They must abide by these behaviours even before exhibiting anti-competitive practices to prevent them from ever occurring. This approach differs from other competition laws, where remedies are delivered after an investigation uncovers a violation. SEE: Regulator CMA to Scrutinize Microsoft and Other Cloud Service Providers in the UK As well as conduct requirements, the DMU can make “pro-competition interventions” that will actively address a company’s adverse effects on competition that stem from its disproportionate market power. Examples of how the DMU might support healthy competition in digital markets include: Preventing bundling or tying of products or services. Preventing self-preferencing of products or services. Mandating competitor access to data or functionality. Requiring interoperability of products or services. Requiring “choice screens” that allow users to select their preferred default apps or services over the company’s own. Requiring transparency concerning aspects of companies’ algorithms. Requiring fairer trading terms. Furthermore, the DMU will require SMS companies to report any merger valued at least £25 million and a U.K. connection. What are the penalties for non-compliance? Under the DMCCA, the CMA is empowered to impose penalties for failures such as non-compliance with enforcement or final orders. Businesses and individuals may be required to produce testimonies or other documentation to aid the DMU’s work. Cardell said the CMA has legal obligations to maintain confidentiality regarding information and whistleblowers. For fixed penalty amounts, businesses can face fines of up to £30,000 or 1% of their turnover. For penalties calculated at a daily rate incurred by individuals, the maximum is £15,000 or 5% of total turnover per day of non-compliance. SEE: Data (Use and Access) Bill: What Is It and How Does It Impact UK Businesses? Cardell said that the CMA “hope(s) that in many cases we will be able to secure timely and beneficial changes without resorting to formal action,” and instead resolve disputes by engaging directly with SMS firms. What are critics saying about the Act? Despite mostly positive feedback in a government consultation in late 2024, the Act has not been met with universal acclaim. Critics are concerned that, instead of complying with the CMA, tech companies will simply exclude the U.K. from the rollout of new products. Evidence of exclusion has already taken place in Europe. For example, Apple will not be making its new suite of generative AI capabilities, Apple Intelligence, available on devices in the E.U. initially, citing “regulatory uncertainties brought about by the Digital Markets Act,” according to Bloomberg. SEE: Apple Intelligence EU:

From data storage to business applications and beyond, companies of all sizes rely on the cloud for day-to-day operations and critical business processes. Protecting cloud-based infrastructures with robust security standards is crucial for modern organizations. Cloud network security is a popular approach. But is it right for your business? Read on to find out. 1 RingCentral RingEx Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Medium (250-999 Employees), Large (1,000-4,999 Employees), Enterprise (5,000+ Employees) Medium, Large, Enterprise Features Hosted PBX, Managed PBX, Remote User Ability, and more 2 Talkroute Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Any Company Size Any Company Size Features Call Management/Monitoring, Call Routing, Mobile Capabilities, and more What is cloud network security? Cloud network security is a broad term that covers all security measures a company uses to protect its private cloud network, public cloud network, or hybrid cloud network. It includes everything from the technology used to internal policies, processes, and controls. It helps businesses defend against data breaches, cyber attacks, unauthorized access, service interruptions, and other threats to their infrastructure. Network security (regardless of how it’s implemented) is just one of the many security layers that businesses use to protect themselves from vulnerabilities. But it’s arguably the most important, as your network is often the first line of defense against attacks. Deploying cloud network security the right way can be the foundation of your company’s entire approach to IT security. SEE: How your business can benefit from a network security policy. How does cloud network security work? Cloud network security uses multiple defense layers between infrastructure components and devices on your network. First, software helps set security policies and pre-defined rules for the network. From there, the software inspects all of the data packets and traffic on the network to enforce those policies. For example, approved users can be granted access to digital assets through an application on the cloud network while unauthorized users are blocked. It can also integrate with other security protocols, such as gateways and firewalls, to provide organization-wide control over the network. With APIs and other integrations, IT security admins can use cloud network security processes to monitor networks in real time, segment networks, and detect threats based on network patterns. Many modern cloud security systems depend on AI and machine learning to help detect and block threats, which is something that might not always work with a rules-based security system. SEE: Check out the best threat protection solutions.  Pros and cons of cloud network security Like any IT security framework or methodology, cloud security has its pros and cons. For most, the positives outweigh the negative. Benefits and advantages Centralized management — Cloud network security gives IT admins a single place to configure and monitor security policies, including the ability to integrate with on-premises solutions. Automated security monitoring — Once configured, cloud security systems automatically protect against threats without straining IT resources. Data protection — Deploying a cloud network security system helps protect data stored in cloud servers and applications on your network (both in transit and at rest). Compliance — You can set up your network security systems to comply with regulatory standards, like GDPR, PCI DSS, HIPAA, and more. Data encryption — While encrypted data doesn’t prevent breaches or attacks, most cloud network security companies include encryption, which makes it more challenging for bad actors to access data if they breach your network. Real-time threat detection and prevention — When working properly, cloud network security systems automatically detect and block threats to your network as they happen. Scalability — Robust cloud security allows organizations to confidently scale processes and applications using cloud resources, knowing that they’ll have reliable access. Policy-based enforcement — System admins have a more granular level of control based on custom policies that scale with your organization. Reduce risk of breaches and attacks — A cloud network security solution can drastically reduce security vulnerabilities while preventing hacks, malware, ransomware, and other malicious incidents. Potential drawbacks and challenges to consider Misconfigurations — It can easily be misconfigured and it’s prone to human error. Speed of change — As cloud resources change alongside access controls of different employees, malicious users can exploit vulnerabilities before your policies are updated. DDoS attacks — Advanced DDoS attacks, which can overwhelm servers and disrupt cloud-based services, could prevent authorized users from accessing your system. Accuracy — At times, cloud systems can yield false positives. This can be dangerous if policies are changed due as a result, opening the door for real threats to slip through the cracks. Cost — Advanced cloud systems are expensive to deploy and maintain at scale, especially those using AI technology to monitor network traffic and detect threats in real time. Insider threats — Someone with privileged access could unknowingly (or intentionally) attack systems from the inside. When it makes sense to use cloud network security for your business Any business that has heavily invested in cloud infrastructure is a good fit. This is especially true if you have a lot of data or run numerous applications in the cloud. It also makes sense for hybrid cloud environments. Because you have a combination of on-premises and cloud infrastructure, a cloud-based security system can help you centralize everything across your network. Another common reason why businesses use it is to comply with industry-specific or location-specific compliance standards. You can set up your cloud network security policies to adhere to security protocols for GDPR in Europe, PCI compliance for payment acceptance, HIPAA compliance in the medical industry, and more. If your organization has remote employees who access your network through an encrypted connection, you can also use cloud security to authenticate them and their devices. When you should avoid cloud network security Cloud network security is a necessity for most, but it’s not for everyone. It may not be enough if you’re dealing with sensitive data that requires the strictest security standards. Organizations working

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Meta — parent company of Facebook, Instagram, WhatsApp, Threads and more — runs one of the biggest recommendation systems in the world. In two recently released papers, its researchers have revealed how generative models can be used to better understand and respond to user intent.  By looking at recommendations as a generative problem, you can tackle it in new ways that are richer in content and more efficient than classic approaches. This approach can have important uses for any application that requires retrieving documents, products or other kinds of objects. Dense vs generative retrieval The standard approach to creating recommendation systems is to compute, store and retrieve dense representations of documents. For example, to recommend items to users, an application must train a model that can compute embeddings for the users’ requests and embeddings for a large store of items.  At inference time, the recommendation system tries to understand the user’s intent by finding one or more items whose embeddings are similar to the user’s. This approach requires an increasing amount of storage and computation capacity as the number of items grows because every item embedding must be stored and every recommendation operation requires comparing the user embedding against the entire item store. Dense retrieval (source: arXiv) Generative retrieval is a more recent approach that tries to understand user intent and make recommendations not by searching a database but by simply predicting the next item in a sequence of things it knows about a user’s interactions. Here’s how it works: The key to making generative retrieval work is to compute “semantic IDs” (SIDs) which contain the contextual information about each item. Generative retrieval systems like TIGER work in two phases. First, an encoder model is trained to create a unique embedding value for each item based on its description and properties. These embedding values become the SIDs and are stored along with the item.  Generative retrieval (source: arXiv) In the second stage, a transformer model is trained to predict the next SID in an input sequence. The list of input SIDs represents the user’s interactions with past items, and the model’s prediction is the SID of the item to recommend. Generative retrieval reduces the need for storing and searching across individual item embeddings. So its inference and storage costs remain constant as the list of items grows. It also enhances the ability to capture deeper semantic relationships within the data, and provides other benefits of generative models, such as modifying the temperature to adjust the diversity of recommendations.  Advanced generative retrieval Despite its lower storage and inference costs, generative retrieval suffers from some limitations. For example, it tends to overfit to the items it has seen during training, which means it has trouble dealing with items that were added to the catalog after the model was trained. In recommendation systems, this is often referred to as “the cold start problem,” which pertains to users and items that are new and have no interaction history.  To address these shortcomings, Meta has developed a hybrid recommendation system called LIGER, which combines the computational and storage efficiencies of generative retrieval with the robust embedding quality and ranking capabilities of dense retrieval. During training, LIGER uses both similarity score and next-token goals to improve the model’s recommendations. During inference, LIGER selects several candidates based on the generative mechanism and supplements them with a few cold-start items, which are then ranked based on the embeddings of the generated candidates.  LIGER combines generative and dense retrieval (source: arXiv) The researchers note that “the fusion of dense and generative retrieval methods holds tremendous potential for advancing recommendation systems,” and as the models evolve “they will become increasingly practical for real-world applications, enabling more personalized and responsive user experiences.” In a separate paper, the researchers introduce a novel multimodal generative retrieval method named Multimodal preference discerner (Mender), a technique that can enable generative models to pick up implicit preferences from users’ interactions with different items. Mender builds on top of the generative retrieval methods based on SIDs and adds a few components that can enrich recommendations with user preferences. Mender uses a large language model (LLM) to translate user interactions into specific preferences. For example, if the user has praised or complained about a specific item in a review, the model will summarize it into a preference about that product category.  The main recommender model is trained to be conditioned both on the sequence of user interactions and the user preferences when predicting the next semantic ID in the input sequence. This gives the recommender model the ability to generalize and perform in-context learning and to adapt to user preferences without being explicitly trained on them. “Our contributions pave the way for a new class of generative retrieval models that unlock the ability to utilize organic data for steering recommendation via textual user preferences,” the researchers write. Mender recommendation framework (source: arXiv) Implications for enterprise applications The efficiency provided by generative retrieval systems can have important implications for enterprise applications. These advancements translate into immediate practical benefits, including reduced infrastructure costs and faster inference. The technology’s ability to maintain constant storage and inference costs regardless of catalog size makes it particularly valuable for growing businesses. The benefits extend across industries, from ecommerce to enterprise search. Generative retrieval is still in its early stages and we can expect applications and frameworks to emerge as it matures. source

Fast-paced markets, shrinking budgets, and increasing shareholder scrutiny are just a few of the factors piling pressure onto today’s project managers. Projects are also becoming more complex, thanks to the increasing integration of emerging technologies like AI, heightened regulatory requirements, and the necessity for adaptability in volatile economies. It is imperative that project managers keep an eye on what is around the corner in the industry so they are prepared to take advantage of new processes or avoid pitfalls. TechRepublic spoke to industry experts to find out the top trends to watch in 2025. 1 monday.com Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Any Company Size Any Company Size Features Agile Development, Analytics / Reports, API, and more 2 ClickUp Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Any Company Size Any Company Size Features Agile Development, Budget / Expense Tracking, Document Management / Sharing, and more 3 Quickbase Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Small (50-249 Employees), Medium (250-999 Employees), Large (1,000-4,999 Employees), Enterprise (5,000+ Employees) Small, Medium, Large, Enterprise Features Agile Development, Analytics / Reports, API, and more Trend 1: Increase in hybrid project teams Alan Zucker, Founding Principal of consulting firm Project Management Essentials “I see interest in hybrid project management continuing to increase in 2025. Interest in agile is waning, and many agilists say it is dead. Hybrid projects combine elements of one or more project approaches. For decades, project managers have pragmatically blended processes and practices based on context and specific project needs. Hybrid allows project managers to move away from the binary waterfall-agile world to one where patterns of practice include lean, kanban, and DevOps. Successful project managers must make intentional choices when deciding how to execute their projects.” SEE: Explore the key features and benefits of hybrid project management. Trend 2: Squad-based teams reduce bureaucracy Jack Skeels, CEO of training firm AgencyAgile “The shift to squad-based teams reflects growing dissatisfaction with traditional approaches and the perceived ineffectiveness of agile as it’s often implemented. Leaders and stakeholders are weary of overly complex management structures and expensive project management software tools that fail to deliver meaningful results. Instead, leaders are embracing simpler, more effective models: small, self-managed teams—typically small squads of five to 15 people. These small teams embody the original spirit of agile by collaborating and getting things done quickly with innovation. The model encourages a culture of innovation, collaboration, and responsiveness to change—key traits for businesses competing in fast-paced markets.” More project management coverage Trend 3: Move to decentralized project management Molly Beran, founder of project management consultancy Projects By Molly “I expect to see a lot of organizations re-thinking their approach to creating centralized Project Management Offices. In the past few years, PMOs have been all the rage—companies rush to set them up, build templates and processes, and then usually start to see them slowly wither. Why? There are many reasons, of course, but I find that one of the most prevalent reasons is that while there is a rush to stand-up tools and processes, there are rarely enough people skilled in project management to actually use the tools and get the work done. Also, in the rush to set up an office, it’s really typical for companies to lose sight of their larger strategic or organizational priorities. In a sense, they get so caught up in creating a centralized PMO that they forget why it exists in the first place—to get the work done that best aligns with the strategic priorities of an organization. I predict that in 2025 and beyond, companies will start pulling back on centralized PMOs and go back to more decentralized project management, where each department or area has in place experts who understand the core business processes, and also get asked to manage projects.” SEE: Read TechRepublic’s guide to the top project management certifications. Trend 4: Focus on AI literacy among project managers Cornelius Fichtner, president of Project Management PrepCast and host of The Project Management Podcast “Project managers should experience a ‘rude awakening’ as they recognize the limitations of their current generative AI interactions. The difference between successful and struggling projects hinges primarily on the project manager’s depth of AI understanding. Many project managers forget that they not only need to use AI on their projects, but they will also be asked to lead projects intended to bring AI capabilities to various departments in their company. They need a really broad and solid understanding of what AI is and can do in order to serve stakeholders from marketing and finance as these departments are augmented with AI.” SEE: 9 Best AI Project Management Tools for 2024 Trend 5: Accelerated job training through AI Justin Tan, IT Project Management Office leader at Thermo Fisher Scientific “Imagine AI systems that can instantly generate comprehensive project plans based on the context and conditions, predict potential risks with greater accuracy, optimize resource allocation, and provide contextual decision support that historically required years of professional experience. Junior professionals without extensive traditional experience will leverage such AI-powered platforms to access institutional knowledge and best practices, effectively compressing years of learning into actionable recommendations—accelerating learning and project execution capabilities. From my experience playing a key role in leading digital transformation initiatives, the most successful organizations will be those that strategically integrate AI not as a replacement for human intelligence, but as a collaborative tool that amplifies human potential.” SEE: Read more artificial intelligence coverage from TechRepublic. Trend 6: Resource management software grows in importance Michele Badie, Professional Development Strategist at Skills Recharged “We’ll continue skills-centric resource management discussions, and action plans to connect the right skills to the right tasks while ensuring teams thrive and stay on task. Real-time tools—for example, resource planning software, AI-driven allocation, employee well-being, and collaboration and communication tools—will support making resource allocation seamless. At the same time, project managers, as integral parts of the process,

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More The end of the year 2024 has brought reckonings for artificial intelligence, as industry insiders feared progress toward even more intelligent AI was slowing down. But OpenAI’s o3 model, announced just last week, has sparked a fresh wave of excitement and debate, and suggests big improvements are still to come in 2025 and beyond. This model, announced for safety testing among researchers, but not yet released publicly, achieved an impressive score on the important ARC metric. The benchmark was created by François Chollet, a renowned AI researcher and creator of the Keras deep learning framework, and is specifically designed to measure a model’s ability to handle novel, intelligent tasks. As such, it provides a meaningful gauge of progress toward truly intelligent AI systems. Notably, o3 scored 75.7% on the ARC benchmark under standard compute conditions and 87.5% using high compute, significantly surpassing previous state-of-the-art results, such as the 53% scored by Claude 3.5. This achievement by o3 represents a surprising advancement, according to Chollet, who had been a critic of the ability of large language models (LLMs) to achieve this sort of intelligence. It highlights innovations that could accelerate progress toward superior intelligence, whether we call it artificial general intelligence (AGI) or not. AGI is a hyped term, and ill-defined, but it signals a goal: intelligence capable of adapting to novel challenges or questions in ways that surpass human abilities. OpenAI’s o3 tackles specific hurdles in reasoning and adaptability that have long stymied large language models. At the same time, it exposes challenges, including the high costs and efficiency bottlenecks inherent in pushing these systems to their limits. This article will explore five key innovations behind the o3 model, many of which are underpinned by advancements in reinforcement learning (RL). It will draw on insights from industry leaders, OpenAI’s claims, and above all Chollet’s important analysis, to unpack what this breakthrough means for the future of AI as we move into 2025. The five core innovations of o3 1. “Program synthesis” for task adaptation OpenAI’s o3 model introduces a new capability called “program synthesis,” which enables it to dynamically combine things that it learned during pre-training — specific patterns, algorithms, or methods — into new configurations. These things might include mathematical operations, code snippets, or logical procedures that the model has encountered and generalized during its extensive training on diverse datasets. Most significantly, program synthesis allows o3 to address tasks it has never directly seen in training, such as solving advanced coding challenges or tackling novel logic puzzles that require reasoning beyond rote application of learned information. François Chollet describes program synthesis as a system’s ability to recombine known tools in innovative ways — like a chef crafting a unique dish using familiar ingredients. This feature marks a departure from earlier models, which primarily retrieve and apply pre-learned knowledge without reconfiguration — and it’s also one that Chollet had advocated for months ago as the only viable way forward to better intelligence.  2. Natural language program search At the heart of o3’s adaptability is its use of chains of thought (CoTs) and a sophisticated search process that takes place during inference — when the model is actively generating answers in a real-world or deployed setting. These CoTs are step-by-step natural language instructions the model generates to explore solutions. Guided by an evaluator model, o3 actively generates multiple solution paths and evaluates them to determine the most promising option. This approach mirrors human problem-solving, where we brainstorm different methods before choosing the best fit. For example, in mathematical reasoning tasks, o3 generates and evaluates alternative strategies to arrive at accurate solutions. Competitors like Anthropic and Google have experimented with similar approaches, but OpenAI’s implementation sets a new standard. 3. Evaluator model: A new kind of reasoning O3 actively generates multiple solution paths during inference, evaluating each with the help of an integrated evaluator model to determine the most promising option. By training the evaluator on expert-labeled data, OpenAI ensures that o3 develops a strong capacity to reason through complex, multi-step problems. This feature enables the model to act as a judge of its own reasoning, moving large language models closer to being able to “think” rather than simply respond. 4. Executing Its own programs One of o3’s most groundbreaking features is its ability to execute its own CoTs as tools for adaptive problem-solving. Traditionally, CoTs have been used as step-by-step reasoning frameworks to solve specific problems. OpenAI’s o3 extends this concept by leveraging CoTs as reusable building blocks, allowing the model to approach novel challenges with greater adaptability. Over time, these CoTs become structured records of problem-solving strategies, akin to how humans document and refine their learning through experience. This ability demonstrates how o3 is pushing the frontier in adaptive reasoning. According to OpenAI engineer Nat McAleese, o3’s performance on unseen programming challenges, such as achieving a CodeForces rating above 2700, showcases its innovative use of CoTs to rival top competitive programmers. This 2700 rating places the model at “Grandmaster” level, among the top echelon of competitive programmers globally. 5. Deep learning-guided program search O3 uses a deep learning-driven approach during inference to evaluate and refine potential solutions to complex problems. This process involves generating multiple solution paths and using patterns learned during training to assess their viability. François Chollet and other experts have noted that this reliance on “indirect evaluations” — where solutions are judged based on internal metrics rather than tested in real-world scenarios — can limit the model’s robustness when applied to unpredictable or enterprise-specific contexts. Additionally, o3’s dependence on expert-labeled datasets for training its evaluator model raises concerns about scalability. While these datasets enhance precision, they also require significant human oversight, which can restrict the system’s adaptability and cost-efficiency. Chollet highlights that these trade-offs illustrate the challenges of scaling reasoning systems beyond controlled benchmarks like ARC-AGI. Ultimately, this approach demonstrates both the potential and the limitations of integrating

Best for managing finances and cash flow: Pipedrive Best CRM with a Starter Bundle for small businesses: HubSpot Best for managing projects: monday CRM Best for product development: ClickUp Best for straightforward integrations: Insightly Best white label CRM for small businesses: Bitrix24 Best offering advanced features with security: Capsule Best for advanced sales and marketing tools: EngageBay Best for integrating with Google Workspace: Copper CRM Customer relationship management software that caters to small businesses helps organizations manage and track leads and customer interactions. While offering a mix of core and advanced CRM features for small businesses to streamline their current sales processes, these CRM tools are also scalable to grow with it. CRM software like Insightly, HubSpot, and EngageBay have free plans, which is great for small businesses with a limited budget. But for small businesses with the funds and personnel, tools like Pipedrive, Capsule, and monday CRM are also worth considering. 1 Pipedrive CRM Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Any Company Size Any Company Size Features Calendar, Collaboration Tools, Contact Management, and more 2 monday CRM Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Any Company Size Any Company Size Features Calendar, Collaboration Tools, Contact Management, and more 3 Creatio CRM Employees per Company Size Micro (0-49), Small (50-249), Medium (250-999), Large (1,000-4,999), Enterprise (5,000+) Medium (250-999 Employees), Large (1,000-4,999 Employees), Enterprise (5,000+ Employees) Medium, Large, Enterprise Features Dashboard, Document Management / Sharing, Email / Marketing Automation, and more Top CRM software for small businesses comparison Small and midsize businesses investing in CRM software want a tool that is accessible throughout both the sales and marketing departments for visibility and collaboration between teams. This allows processes to be automated into a CRM cycle and deals to be brought in and closed faster. CRM software with advanced features such as marketing tools helps these small businesses house all of their customer relationships, from generation to nurturing, within one platform. Another important feature organizations should confirm before committing to a software solution is how a CRM solution integrates with the existing tech stack, like calendars or email providers. Software Our star rating (out of 5) Pipeline management Marketing tools Lead nurturing Integrations Starting price* Pipedrive 4.4 Yes Limited Yes Yes $14 per user per month HubSpot 4 Yes Yes Yes Yes Free starting price monday CRM 3.5 Yes Yes Limited Yes $12 per user per month ClickUp 4.1 Yes Yes Limited Yes Free starting price Insightly 4.4 Yes Yes Yes Yes Free starting price Bitrix24 4.1 Yes Yes Yes Yes Free starting price Capsule 4 Yes Limited Yes Yes Free starting price EngageBay 4.1 Yes Yes Yes Yes Free starting price Copper CRM 3.8 Yes Yes Yes Yes $9 per user per month *Price when billed annually. Pipedrive: Best for managing finances and cash flow Image: Pipedrive In addition to Pipedrive’s advanced pipeline management tools, users can automate and streamline their finances with AI and app integrations. Small and midsize companies rely on predictive analytics around finances and cash flow, and the good news is Pipedrive can integrate with different accounting systems and software to automate invoices, schedule payments and even forecast revenue. Why I chose Pipedrive Pipedrive is a popular CRM solution for small businesses because of its intuitive pipeline building. Users can build out new or existing sales pipelines and then manage customers from beginning to end with helpful automations and reports. Pipedrive can also connect and integrate with a variety of other business tools in your tech stack, such as Slack, Zoom, Gmail, Trello, Microsoft Teams and more. While Pipedrive’s premium plans are priced at average rates, I recognize that small businesses might prioritize a free CRM over a paid one to get started; in that case, I recommend looking into  HubSpot. For more information on this provider, check out our full Pipedrive review. Pricing Essential: $14 per user per month, billed annually, or $24 per user when billed monthly. This tier comes with customizable pipelines, deal import and export, file attachments, calendar views and more. Advanced: $34 per user per month, billed annually, or $44 per user when billed monthly. This plan offers all Essential features, plus email templates, group emailing, meeting scheduler and more. Professional: $49 per user per month, billed annually, or $64 per user when billed monthly. All features mentioned are included in this plan, in addition to unlimited visual dashboards, custom permission sets, 24/5 live chat support and more. Power: $64 per user per month, billed annually, or $79 per user when billed monthly. This tier offers all Professional functionality, plus phone support, 200,000 open deals and more. Enterprise: $99 per user per month, billed annually, or $129 per user when billed monthly. This plan comes with all available features, including unlimited open deals, custom fields, reports and security alerts. Features Email campaigns: Personalize messages to customers that are scheduled to engage with them at the most ideal time. Workflow automation: Spend less time working on administrative tasks and more time selling with automated emails, touchpoints and more. Conversation tracking: View contact history with every contact, including calls, emails and in-person meetings. Business forecast dashboard in Pipedrive. Image: Pipedrive Pipedrive pros and cons Pros Cons 14-day free trial. No free-for-life version. 24/7 customer support. Reports of limited reporting and analytics. Mobile app available. Limited marketing tools. HubSpot: Best CRM with a Starter Bundle for small businesses Image: HubSpot HubSpot offers a special Starter Bundle built for small businesses and startups. The bundle gives organizations access to every HubSpot starter product at a discounted price—this includes the Smart CRM starter, Marketing, Sales, CMS, Service, Operations and Commerce hubs. This suite gives users access to a variety of advanced tools for everything pre- and post-sale in one sales pipeline. Why I chose HubSpot HubSpot is another CRM provider popular for its robust free version. CRMs that offer a free tier typically have a maximum of two or three users.