A threat actor with ties to Pakistan has been observed targeting various sectors in India with various remote access trojans like Xeno RAT, Spark RAT, and a previously undocumented malware family called CurlBack RAT. The activity, detected by SEQRITE in December 2024, targeted Indian entities under railway, oil and gas, and external affairs ministries, marking an expansion of the hacking crew's targeting footprint beyond government, defence, maritime sectors, and universities. "One notable shift in recent campaigns is the transition from using HTML Application (HTA) files to adopting Microsoft Installer (MSI) packages as a primary staging mechanism," security researcher Sathwik Ram Prakki said. SideCopy is suspected to be a sub-cluster within Transparent Tribe (aka APT36) that's active since at least 2019. It's so named for mimicking the attack chains associated with another threat actor called SideWinder to deliver its own payloads. In June 2024, SEQRITE highlighted SideCopy's use of obfuscated HTA files, leveraging techniques previously observed in SideWinder attacks. The files were also found to contain references to URLs that hosted RTF files identified as used by SideWinder. The attacks culminated in the deployment of Action RAT and ReverseRAT, two known malware families attributed to SideCopy, and several other payloads, including Cheex to steal documents and images, a USB copier to siphon data from attached drives, and a .NET-based Geta RAT that's capable of executing 30 commands sent from a remote server. The RAT is equipped to steal both Firefox and Chromium-based browser data of all accounts, profiles, and cookies, a feature borrowed from AsyncRAT. "APT36 focus is majorly Linux systems whereas SideCopy targets Windows systems adding new payloads to its arsenal," SEQRITE noted at the time. The latest findings demonstrate a continued maturation of the hacking group, coming into its own, while leveraging email-based phishing as a distribution vector for malware. These email messages contain various kinds of lure documents, ranging from holiday lists for railway staff to cybersecurity guidelines issued by a public sector undertaking called the Hindustan Petroleum Corporation Limited (HPCL). One cluster of activity is particularly noteworthy given its ability to target both Windows and Linux systems, ultimately leading to the deployment of a cross-platform remote access trojan known as Spark RAT and a new Windows-based malware codenamed CurlBack RAT that can gather system information, download files from the host, execute arbitrary commands, elevate privileges, and list user accounts. A second cluster has been observed using the decoy files as a way to initiate a multi-step infection process that drops a custom version of Xeno RAT, which incorporates basic string manipulation methods. "The group has shifted from using HTA files to MSI packages as a primary staging mechanism and continues to employ advanced techniques like DLL side-loading, reflective loading, and AES decryption via PowerShell," the company said. "Additionally, they are leveraging customized open-source tools like Xeno RAT and Spark RAT, along with deploying the newly identified CurlBack RAT. Compromised domains and fake sites are being utilized for credential phishing and payload hosting, highlighting the group's ongoing efforts to enhance persistence and evade detection." Found this article interesting? Follow us on Twitter  and LinkedIn to read more exclusive content we post.

Product ManagerCanopy, South Jordan, UT (Hybrid)About UsCanopy is a fast-growing SaaS company in South Jordan, Utah building simple, efficient software for accounting firms. We are looking to revolutionize the accounting space with modern, user-friendly software for a neglected industry. We aim to help our clients unlock the firm they’ve always wanted with our Practice Management Suite. We place a strong emphasis on delighting our customers, spotting and solving problems, and being good people along the way. Click here to see why our clients love Canopy.Interested in learning more about Canopy & the industry? Check out our blog here where you can find great information on our product features, industry news, practice management, and more!The OpportunityAt Canopy, our product managers are a dynamic force. They exude empathy, are data-driven, lightning-fast decision-makers, and expert truth-seekers. We're all about collaboration, and our product team excels in embracing the responsibility and empowerment to craft the perfect user experience. We thrive on research, validation, creativity, and cultivating a culture of teamwork across the entire company. Together, we're unstoppable in delivering top-notch products that our users will love.This role requires strong product management expertise combined with deep experience and understanding of tax, accounting, advisory, bookkeeping, and practice management workflows. Your insights will play a vital role in shaping our roadmap and enriching the team with customer-focused knowledge.This is a hybrid position in South Jordan, Utah.What You’ll Do:Collaborate with design and engineering teams to bring the perspective of tax professionals and accounting firms to the forefront, ensuring that product decisions are deeply informed by real-world workflows and challenges.Leverage your experience in tax preparation, billing, workload management, tax advisory, or similar workflows to guide product strategy and team priorities.Seek feedback from various stakeholders, including users, product managers, subject matter experts, designers, engineers, sales, support, and marketing, to ensure a seamless product experience. Collaborate closely with engineering to guarantee the highest quality implementation.Continuously conduct market and competitive analyses to unearth fresh opportunities, while also becoming a seasoned authority in the accounting industry.Conduct unbiased research with our users daily (we have very involved customers and we wouldn’t have it any other way).Obsess every piece of information gathered and thoroughly analyze everything to ensure highly data-driven decision-making.Be part of preserving Canopy's unique culture. We cherish customer delight, embrace teamwork, and foster transparency. We seek passionate team players who take ownership and have fun delivering life-changing features. Come, be a part of this exciting adventure!What We’re Looking For: 3+ years experience in product management (or closely related) roles.Hands-on experience in tax workflows such as workload management, customer intake, tax preparation, billing, tax resolution, or advisory services, either as a practitioner or in a related capacity.Proven ability to conduct unbiased research, design and develop great products people love, preferably for consumer-facing products.Experience with 3rd-party integrations and the challenges associated with it.Experience building a product strategy and vision that is aligned cross-functionally with engineering, sales, CS, and marketing.Proven ability to lead through transparent communication and service-like influence.A passion for analyzing product usage data.Bonus Points If You Have: Experience in the B2B fintech space, or a background in accounting, tax, or firm operations.Expertise in implementing workflow automation, time and billing, and payment solutions.A bachelor’s degree in Business, Computer Science, Accounting, or a related field.We know many women do not apply for a job if they don't perfectly fit the description. We want you to apply anyway.Why You Want to Work Here:🌴 Flexible Paid Time Off - you’re actually encouraged to use, plus 10 company holidays! ❤️‍🩹 Health Benefits - including Medical, Dental, and Vision and an HSA Match. 💰 401(k) - we match 100% up to 3% of your contribution. Eligibility is immediate with 100% vesting.🧠 Mental Health - all employees have access to Impact Suite & to our Employee Assistance Program (EAP).👶 Paid New Parent Leave & Birthing Parent Leave - so you’re able to care for your little ones.➕ Supplemental Benefits - including 100% company paid Basic Life & AD&D insurance and long & short-term disability coverage.🌟 Nectar - our peer-to-peer recognition program to help our employees recognize the amazing work being done by other Canopians!🥳 Company Events - including monthly company-wide meetings, summer parties, and more.💡 ERG Committees - to plan initiatives around continuing education, community outreach, recruiting, onboarding, and more.☕ Fully-stocked kitchen - Keto? Vegan? Flexitarian? Mandalorian? We’ve got you covered. Our Values:We approach our work every day with a few things in mind:🔑 Own - we own this place! We focus on outcomes, holding ourselves & each other accountable.🏆 Win - we win by delighting our customers with the very best products and services.👍 Do Good - we work hard to be good people!💡 Embrace Curiosity & Candor - we approach everything with curiosity & we understand that candor is kindness and give the gift of feedback.To learn more about us & our values, click here.Interviewing @ Canopy:Application processes can be a little stressful. Here are the stages of a typical interview process at Canopy:Once your application is received, we will review it and get back to you if we feel like it’s a mutual fit! 20-minute phone call with the People Team45-60-minute video or in-person interview with the Hiring Manager1-3 rounds of interviews, depending on the roleFinal InterviewInterview processes can vary depending on the role. The People Team will give you a role-specific overview of the process during your first phone call. Remember: This is your interview too! We know candidates are evaluating us just as much as we are them. We encourage you to bring questions to each of your interviews—our hiring teams will always make sure to save time for questions at the end! Canopy is an equal-opportunity employer. Canopy provides equal employment opportunities (EEO) to all employees and applicants for employment without regard to race, color, religion, gender, national origin, sexual orientation, gender identity or expression, age, disability, genetic information, marital status, or veteran status.

Submitted by WA Contents Lina Ghotmeh selected to design the permanent Qatar Pavilion in the Giardini Qatar Architecture News - Apr 14, 2025 - 05:18   html PUBLIC "-//W3C//DTD HTML 4.0 Transitional//EN" "http://www.w3.org/TR/REC-html40/loose.dtd" Her Excellency Sheikha Al Mayassa bint Hamad bin Khalifa Al Thani, Chairperson of Qatar Museums, announced that the internationally acclaimed architect Lina Ghotmeh, founder and principal of the Paris-based studio Lina Ghotmeh — Architecture, has been selected through a global competition to design the permanent Qatar Pavilion at the Giardini of Venice Architecture Biennale.Commissioned by H.E. Sheikha Al Mayassa on behalf of the State of Qatar, the Qatar Pavilion will be situated in the heart of the Giardini, next to the iconic Book Pavilion. This will be the third Pavilion added to the historic and prestigious Giardini in over 50 years."My team and I are deeply honoured to have been chosen for this uniquely exciting and significant project""My team and I are deeply honoured to have been chosen for this uniquely exciting and significant project. Qatar is a cultural beacon for the entire MENASA region," said Lina Ghotmeh."It is thrilling to be given this opportunity to design Qatar’s Pavilion on the historic grounds of the Giardini of La Biennale di Venezia," Ghotmeh added.Lina Ghotmeh was chosen for her concept, which is characterized by architectural clarity and a thoughtful engagement with the Pavilion’s historical context, establishing a cultural connection between Qatar and the world at large. The design exemplifies architecture’s capacity to connect with the public realm, blending effortlessly with both the Pavilion’s immediate surroundings and the broader environment. In light of the contemporary art and architecture it will contain, the interior will be flexible and adaptable, allowing exhibitors to reshape the space according to their own creative visions."I extend my warmest congratulations to Lina Ghotmeh on receiving the commission for the Qatar Pavilion in the Giardini of La Biennale di Venezia. Her work is inspiring new and traditional audiences with its sensitivity to the human condition and its confident, innovative flair," said Her Excellency Sheikha Al Mayassa."Lina has a worldview and sensibility that has grown from her native Lebanon to reach across cultures. She has wholeheartedly embraced our vision for the Qatar Pavilion as a platform for the artistic, architectural, and cultural creativity of our nation and the Middle East, North Africa, and South Asia." "I thank H.H. Sheikh Tamim bin Hamad Al Thani for his vision and support in realising Qatar’s Pavilion. For their vision of bringing our two nations together through arts and culture, I also thank the Republic of Italy; Maria Tripodi, Italy’s Undersecretary of State to the Ministry of Foreign Affairs and International Cooperation; the Municipality of Venice and Mayor Luigi Brugnaro; Pietrangelo Buttafuoco, President of La Biennale di Venezia; H.E. Khalid bin Youssef Khalifa Abdullah Al Sada, Ambassador Extraordinary and Plenipotentiary of Qatar; and H.E. Paolo Toschi, Ambassador of the Republic of Italy to the State of Qatar," added Her Excellency Sheikha Al Mayassa.On May 8, H.E. Sheikha Al Mayassa and Lina Ghotmeh will engage in a conversation at ACP-Palazzo Franchetti, with Hans Ulrich Obrist serving as the moderator.Lina Ghotmeh, a Lebanese-born architect and founder of the globally recognized firm Lina Ghotmeh — Architecture, is renowned for her designs that are sustainable, historically informed, and award-winning. Her work integrates traditional craftsmanship with innovation, creating a progressive, ecological, and inclusive architectural approach. Her philosophy, “Archaeology of the Future,” promotes a profound link among history, nature, and materials.Notable projects include the redesign of the Western Range galleries of the British Museum; the 22nd Serpentine Pavilion in London, Ateliers Hermès in Normandy, France’s first low-carbon, energy-positive industrial building; Stone Garden Housing tower in Beirut, and the Estonian National Museum in Tartu. Lina Ghotmeh — Architecture is also designing the AlUla Contemporary Art Museum in Saudi Arabia and the Bahrain Pavilion for Expo 2025. Her work was exhibited in the 17th International Architecture Exhibition of Venice Architecture Biennale.The top image in the article: Lina Ghotmeh. Image © Kimberly Lloyd.> via Qatar Museums

The project, for Anchor, England’s largest later living provider, will double the capacity of Cardamom Court, which is half a mile south-west of Ilford train station. PRP’s proposals involve retaining the existing accommodation while demolishing a day care centre and a three-storey building, which adjoin the main building. The two blocks will be replaced by a four-storey building containing 34 one-bedroom apartments. A further three rooms will be created within the existing care home building, delivering 36 one and two-bed apartments. In total, the scheme will deliver 70 apartments.Advertisement The scheme also features a 1.5-storey main entrance with access to a garden courtyard, reconfiguration of the circulation between new and old buildings, and the relocation of communal spaces to the new building. Above the communal spaces, a new roof terrace will provide space for residents to grow fruit and vegetables. In planning documents, PRP described the development as providing ‘sector-leading extra-care housing with accessible, energy efficient homes for retirement’. Newham Council will decide on the scheme at a later date. Site view:  Advertisement Project data Location Cardamom Court, 1 Grantham Road, London, E12 5LU Local authority London Borough of Newham Type of project Extra care housing Client Anchor Architect PRP Landscape architect PRP Planning consultant PWA Planning Group Structural engineer Dice Consulting Engineers M&E consultant Davies Partnership Quantity surveyor TSA Riley Principal designer (CDM) Currie & Brown Funding Partially GLA Funded Tender date Stage 01 – 08/05/25-25/06/25 / Stage 02 – 11/11/25-24/11/25 Approximate start on site date January 2026 Approximate completion date August 2027 Contract duration 19 months Gross internal floor area 5,413m2

Sony has raised the price of PS5 once again, with increases in the UK, mainland Europe, Australia and New Zealand. Read more

Tomas is an art teacher and a freelance illustrator/concept artist based in Prague, Czech Republic.  He started his professional career in digital painting eight years ago in Warhorse Studios, working on Kingdom Come: Deliverance – a realistic medieval RPG. This experience has greatly influenced his artistic development and gave him courage to start a freelance career. Since then, he has worked for Games Workshop, Wizards of the Coast, CD Projekt Red, and many other companies.

Demand for solar energy in power-starved Nigeria has soared in the last decade thanks to worsening grid reliability and rising fuel costs. That’s drawn investor interest to Arnergy, a cleantech startup meeting that need. The company just raised a $15 million Series B extension (on top of a $3 million B1 round last year), bringing its total for the round to $18 million. That surge in demand for solar systems follows significant policy shifts, most notably the removal of Nigeria’s decades-old fuel subsidy in May 2023 (the government’s decision—long debated—ended its practice of covering the gap between global and local fuel prices). Since then, petrol prices have jumped nearly 500%, making power generators, once seen as the more affordable alternative to unreliable grid power and solar systems despite environmental hazards, far costlier to run.  Arnergy’s pitch has changed with the times. “When we started the business, we used to position solar as a way to get uninterrupted power, not necessarily to save money. It wasn’t part of a commercial conversation,” founder and CEO Femi Adeyemo told TechCrunch. “Now it is, because we can clearly show customers how our systems save them monthly whether using petrol, diesel, or even the grid.” Adeyemo launched Arnergy in 2013 to provide solar systems to homes and businesses across sectors like hospitality, education, finance, agriculture, and healthcare. What began as a resilience play is now a cost-savings strategy changing the economics of adoption for the cleantech backed by Bill Gates’s Breakthrough Energy Ventures (the firm led Arnergy’s $9 million Series A in 2019.) Lease-to-own increasing adoption That adoption is clearest in the company’s lease-to-own product, Z Lite, which became a core focus following Arnergy’s first Series B tranche last year. While outright purchases comprised 60% to 70% of revenue in 2023, they accounted for just 25% of sales last year. On the other hand, lease-to-own, where customers pay fixed monthly fees over 5 to 10 years before owning the system, has gained more traction. One reason for this change is affordability when compared to electricity tariffs. Until recently, many people viewed long-term leases as costlier than running diesel or petrol generators. But with diesel prices soaring post-subsidy removal and grid tariffs climbing—especially after a new government policy last April that tripled electricity consumption costs for customers with the most stable power—lease-to-own solar is becoming popular among customers, says Adeyemo.  “Imagine paying ₦200,000 (~$125) every month for power. With our product, that drops to ₦96,000 (~$60). Over five years, it’s a no-brainer what you’ll save,” said the CEO. He added that many existing customers are returning to double their solar capacity or switch completely off-grid as a result. Arnergy tripled its lease customer base between 2023 and 2024 and expects to grow it 4–5x this year. Naira revenues have climbed accordingly and are on track to quadruple by the end of the year. Dollar revenues, on the other hand, have remained flat due to currency devaluation, but Adeyemo said the company is building FX revenue through dollar-denominated B2B2C partnerships and potential expansion into Francophone Africa. Arnergy, which provides solar power systems to homes and businesses in Nigeria, raises $3M Scaling amidst yet another government policy So far, Arnergy has deployed over 1,800 systems across 35 Nigerian states, totaling 9MWp of solar and 23MWh of battery storage. Arnergy plans to use its new funding led Nigerian private equity firm CardinalStone Capital Advisers (CCA) to install more than 12,000 systems by 2029. Breakthrough Energy Ventures as well as British International Investment, Norfund, EDFI MC, and All On participated in the round. But hitting that target requires a strategic shift. For nearly a decade, Arnergy handled sales in-house. Now, it’s adopting a partnership-driven model with business clients and physical retail outlets outside Lagos to reach more customers in Nigeria’s power-starved market. The Lagos-based cleantech is in talks to raise additional local debt from banks and DFIs to support these projects including energy-as-a-service (EaaS) solutions for multinationals, says Adeyemo. Yet as Arnergy prepares to scale, a proposed policy could threaten its momentum.  Last month, Nigeria’s government announced plans to ban solar panel imports to boost local manufacturing. The move has drawn backlash from stakeholders who argue that domestic capacity is far from ready. Adeyemo agrees with the goal, but not the approach. He warned that a premature ban could stall an industry that’s only just getting off the ground. According to the CEO, Nigeria needs to create an environment with the right infrastructure, policy stability, and access to capital so that local factories can ramp up over the next 3 to 5 years. Only after that should the country start thinking about phasing out imports.  “We’re advocates for local manufacturing. But let’s build capacity before shutting the door on imports. Otherwise, we risk doing more harm than good, both to the industry and to the millions of Nigerians who now rely on solar as their primary energy source,” he remarked.

Hey everyone, today we wanna share a fantastic cosmic web Geometry Nodes effect for Blender. It is useful for creating superstructures resembling galactic filaments – that actually run super fast and are super well optimized. Free version (on Patreon)Advanced version (on Patreon, Shrimp Lab Tier 2)Gotta hand it to Brady Johnston for sharing this crazy fast optimized Plexus setup, probably the fastest one available indeed.And a major thanks to Erindale for the always incredibly helpful Discord community for Geometry Nodes fans.Erindale’s online learning platform for technical artists, teaching procedural workflows in Blender and beyond!Make sure to stick around for more behind-the-scenes looks and techniques as we continue our research for the Space VFX sequel.

An open call is being held for participation in the upcoming UIA World Congress of Architects 2026 Barcelona (UIA2026BCN) (Deadline: 23 May) Professionals from architecture, landscape architecture, urbanism, design, engineering, research, photography, film and art are invited to submit proposals for projects, essays or multimedia pieces which could feature in the upcoming conference. Key aims of the open call include bringing together knowledge ‘across research, practice and education to generate new perspectives and strengthening connections between disciplines.’ Barcelona will host the 70th UIA congress from 28 June to 2 July next year and will also take the title of 2026 UNESCO-UIA World Capital of Architecture. Collage of digitally manipulated images by Judit Musachs and Pol Pérez selected by the UIA2026BCN curatorial team According to the brief: ‘The UIA2026BCN Scientific Committee and the Evaluation Committee, will evaluate the proposals from broad and cross-disciplinary perspectives. ‘After an initial pre-selection by subject matter experts, a jury composed of internationally renowned professionals – specialised in each category – as well as the congress curatorial team, will make the final selection. ‘Accepted proposals will be part of the central events of the UIA2026BCN congress between 28 June and 2 July 2026. The jury will determine the form of participation, which may include individual presentations, debates or other planned formats. ‘Additionally, it will decide whether the selected works will be exhibited physically or digitally at the congress exhibition and whether they will be published in the official catalogue.’ Located on the north-eastern coast of Spain, Barcelona is the capital and largest city of Catalonia with 1.6 million residents. Last month, the winners of an open international contest to remodel various venues on the city’s Montjuïc trade fair site were revealed. UIA World Congress of Architects 2026 Barcelona open call The open call comes almost a year after a competition – backed by Barcelona City Council and the Fundació Mies van der Rohe – was launched to remodel 10 permanent blind walls which are spread across the city’s 10 districts and could ‘leave a lasting legacy in Barcelona’. The UIA2026BCN congress is themed ‘Becoming. Architectures for a planet in transition’ and aims to explore transformative ideas that address contemporary challenges. As part of the lead up to the event a separate competition is seeking ideas to transform construction waste into new materials for public spaces across Barcelona, Spain. The winning submissions selected for inclusion in the UIA2026BCN congress programme will be announced on 30 September. How to apply Deadline: 23 May 2025 Competition funding source: Not supplied Project funding source: Not supplied Owner of site(s): Not supplied Contact details: callfor@uia2026bcn.orgVisit the competition website for more information

According to analyst Gartner, small language models (SLMs) offer a potentially cost-effective alternative for generative artificial intelligence (GenAI) development and deployment because they are easier to fine-tune, more efficient to serve and more straightforward to control. In its Explore small language models for specific AI scenarios report, published in August 2024, Gartner explores how the definitions of  “small” and “large” in AI language models have changed and evolved. Gartner notes that there are estimates that GPT-4 (OpenAI – March 2023), Gemini 1.5 (Google – February 2024), Llama 3.1 405B (Meta – July 2024) and Claude 3 Opus (Anthropic – March 2024) have around half a trillion to two trillion parameters. On the opposite end of the spectrum, models such as Mistral 7B (Mistral.AI – September 2023), Phi-3-mini 3.8B and Phi-3-small 7B (Microsoft – April 2024), Llama 3.1 8B (Meta – July 2024) and Gemma 2 9B (Google – June 2024) are estimated to have 10 billion parameters or fewer. Looking at one example of the computational resources used by a small language model compared with those used by a large language model, Gartner reports that Llama 3 8B (eight billion parameters) requires 27.8GB of graphics processing unit (GPU) memory, whereas Llama 3 70B (70 billion parameters) requires 160GB.  The more GPU memory needed, the greater the cost. For instance, at current GPU prices, a server capable of running the complete 670 billion parameter DeepSeek-R1 model in-memory will cost over $100,000.  The fact that a large language model is several times larger than a small language model – in terms of the parameters used during training to build a data model that they use for AI inference – implies that SLMs are only trained on a subset of data. This suggests there are likely to be holes in their knowledge, hence they will sometimes be unable to provide the best answer to a particular query. Distilled SLMs improve response quality and reasoning while using a fraction of the compute of LLMs Jarrod Vawdrey, Domino Data Lab Jarrod Vawdrey, field chief data scientist at Domino Data Lab, an enterprise AI platform provider, notes that SLMs can benefit from a kind of knowledge transfer with LLMs. The technique, known as knowledge distillation (see box below), enables effective transfer from LLMs to SLMs. “This knowledge transfer represents one of the most promising approaches to democratising advanced language capabilities without the computational burden of billion-parameter models,” he says. “Distilled SLMs improve response quality and reasoning while using a fraction of the compute of LLMs.” Vawdrey says knowledge distillation from LLMs to SLMs begins with two key components: a pre-trained LLM that serves as the “teacher”, and a smaller architecture that will become the SLM “student”. The smaller architecture is typically initialised either randomly or with basic pre-training. Neither an LLM nor an SLM alone may deliver everything an organisation needs. Enterprise users will typically want to combine the data held in their corporate IT systems with an AI model.  According to Dominik Tomicevic, CEO of graph database provider Memgraph, context lies at the core of the entire model debate. “For very general, homework-level problems, an LLM works fine, but the moment you need a language-based AI to be truly useful, you have to go with an SLM,” he says. Read more articles about AI models AI models explained: The benefits of open source AI models: In this guide, we explore how to get started with open source AI models and go over how they support your enterprise IT strategy. Latest Alibaba AI model demos AI improvements: The latest model from Chinese public cloud provider Alibaba shows how reinforced learning is driving AI efficiency. For instance, the way a company mixes paint, builds internet of things (IoT) networks or schedules deliveries is unique. “The AI doesn’t need to recall who won the World Cup in 1930,” he adds. “You need it to help you optimise for a particular problem in your corporate domain.” As Tomicevic notes, an SLM can be trained to detect queries about orders in an e-commerce system and, within the supply chain, gain deep knowledge of that specific area – making it far better at answering relevant questions. Another benefit is that for mid-sized and smaller operations, training an SLM is significantly cheaper – considering the cost of GPUs and power – than training an LLM. However, according to Tomicevic, getting supply chain data into a focused small language model is technically a major hurdle. “Until the basic architecture that both LLMs and SLMs share – the transformer – evolves, updating a language model remains difficult,” he says. “These models prefer to be trained in one big batch, absorbing all the data at once and then reasoning only within what they think they know.” This means updating or keeping an SLM fresh, no matter how well-focused it is on the use cases for the business, remains a challenge. “The context window still needs to be fed with relevant information,” he adds. For Tomicevic, this is where an additional element comes in – organisations repeatedly find that a knowledge graph is the best data model to sit alongside a domain-trained SLM, acting as its constant tutor and interpreter. Retrieval augmented generation (RAG) powered by graph technology can bridge structured and unstructured data. Tomicevic says this allows AI systems to retrieve the most relevant insights with lower costs and higher accuracy. “It also enhances reasoning by dynamically fetching data from an up-to-date database, eliminating static storage and ensuring responses are always informed by the latest information,” he says. The resource efficiency of SLMs allows them to run on standard hardware while delivering specialised intelligence exactly where it’s needed, according to Chris Mahl, CEO of enterprise knowledge management platform provider Pryon. “This transforms how organisations deploy AI, bringing powerful capabilities to environments previously considered impractical for advanced computing and democratising access across geographical and infrastructure barriers,” he says. According to Mahl, RAG provides a pipeline that cuts through the noise to deliver precise, relevant context to small language models. While LLMs are regarded as incredibly powerful, they suffer from errors known as hallucinations, whereby they effectively make things up. Rami Luisto, healthcare AI lead data scientist at Digital Workforce, a provider of business automation and technology solutions, says SLMs provide a higher degree of transparency to their inner workings and their outputs. “When explainability and trust are crucial, auditing an SLM can be much simpler compared to trying to extract reasons for an LLM’s behaviour,” he says. While there is a lot of industry hype around the subject of agentic AI, a major barrier to using AI agents to automate complex workflow is that these systems are prone to errors, leading to incorrect decisions being automated. This inaccuracy will improve over time, but there is little evidence that enterprise applications are being developed with tolerance to potential errors introduced by agentic AI systems. In a recent Computer Weekly podcast, Anushree Verma, a director analyst at Gartner, noted that there is a shift towards domain-specific language models and lighter models that can be fine-tuned. Over time, it is likely these smaller AI models will work like experts to complement more general agentic AI systems, which may help to improve accuracy.           • Download this podcast • The analogy is rather like someone who is not a specialist in a particular field asking an expert for advice, a bit like the “phone a friend” lifeline in the TV game show Who wants to be a millionaire? DeepMind CEO Demis Hassabis envisages a world where multiple AI agents coordinate activities to deliver a goal. So, while an SLM may have been transferred knowledge from an LLM through knowledge distillation, thanks to techniques like RAG and its ability to be optimised for a specific domain, the SLM may eventually be called as an expert to help a more general LLM answer a domain-specific question. According to Jarrod Vawdrey, field chief data scientist at Domino Data Lab, the distillation process can be implemented through different methods using both structured data (such as labelled datasets with clear categories) and unstructured data (such as text corpora, conversations, or code): Response-based distillation trains the small language model (SLM) to match the output probability distribution of the large language model (LLM) across a large corpus, focusing on final outputs. Feature-based distillation goes beyond just copying answers – it helps the smaller “student” model learn how the larger “teacher” model thinks by mimicking its reasoning process at different stages. Multi-stage distillation represents a sequential approach where knowledge is transferred through intermediate models of decreasing size. This works like a tutoring system where a college graduate first teaches a bright high school senior, who then simplifies and passes down that knowledge to a younger student.