Ethics. Responsibility. Governance. Trust. All big concerns for business leaders right now are with AI. In a recent survey, 76% of CIOs say their organizations do not have an AI-ready corporate policy on operational or ethical use. Businesses also see it as a barrier to defining their AI vision due to concerns about regulatory and ethical risk. We are seeing that attitudes toward ethics and responsibility when planning IT and technology investments have changed. A decade ago, it was more of an afterthought. Today in the AI age, it’s a board-level topic. But it isn’t easy to achieve – let’s explore why. Why Is Responsible and Ethical AI So Tough? The very nature of the technology is a challenge. Generative AI relies on the training data, architecture, and AI engine to produce unique results. If not designed carefully and monitored continuously, you could run into bias. For example, financial data might reflect problematic gender pay gaps or historical data might bring outdated cultural norms to outputs. You need a solution for that. Other challenges include: Legacy governance mechanisms that businesses will need to rework in some capacity. Models that had varying degrees of preparedness even before the Gen AI boom. Skill gaps. AI is a very new field – and companies are concerned about having the right expertise to deliver it quickly and effectively. Getting people on board. People generally fear AI, which will slow adoption. Lack of uniform regulations. There are regulatory gaps in LLM safety, content providence, and risk managements which AI companies are working together to fill. Why Being People-Led Is the Answer Lenovo and NVIDIA have created an AI readiness framework with four pillars in a very intentional order: Security, People, Technology, and Process. “Security” comes first, to make sure you can prevent harm, bias, and unintended or improper use of AI. Then there’s “People” to ensure good change management and that properly trained personnel are involved throughout the AI journey. “Technology” and “Process” come later because, without robust security and onboard people, you won’t release the full value of the AI technology anyway. In short, you should be people-led instead of technology-led. Here are some ways to do that: #1: Ensure Explainability with Constant Human Feedback Explainability is key. You must monitor how and why AI is providing each output and – critically – ensure it stays on track. Explainability usually comes in two forms: 1) White-box solutions: Semantic AI for example, where you can map the logic, training data, inputs, and prompts. As you test it, you can understand where outputs come from and refine from there. 2) Black-box solutions: Closed or open-source systems like ChatGPT. These are less transparent and explainable, so it gets trickier. You always need humans to judge the inputs, infer how reasonable the outputs are, then refine from there. Either way, you need humans to constantly monitor the LLM. It needs to be stress-tested. The model may drift, gain bias, or get different responses wrong. It’s going to learn depending on who’s using it and what data’s put into it – you must consider that in your monitoring framework. The best large-scale generative models will likely hit 80-85% accuracy in benchmark tests. Human feedback is instrumental in bridging that 15%. #2: Base Governance on Transparency and Alignment Companies need a level of governance where transparency is everything. It ensures people are always accountable for their actions. For example, let’s say you’re worried about IP protection. Put a broker in place where, if you use a third-party LLM, you must go through a certain gateway that tracks the prompts and responses. That means people will think twice about what they’re sending. Why? Because transparency is everywhere, placing the onus on to self-regulate and do the right thing. Another essential part is top-to-bottom alignment. Make sure AI initiatives fit the outcomes users, teams, and customers expect. This will help keep everything ethical and responsible – while reducing the risk of skills gaps as the resources you have will match overall business strategy. I’m not saying this is easy. Does corporate strategy always equal what people are doing? Many organizations found that tough even before AI. But it should be a priority here. #3: Get People on Board With a “Show Me” Model When convincing teams to adopt AI, use a “show me” model. Demonstrate clearly how it works and the immediate benefits. How will it make their lives easier and more effective? Here’s an example. Say you’ve got an NVIDIA NIM inference microservice that can accelerate sales pipeline by 30%, you should lead with that to make the benefit immediately clear. If you don’t, people are more likely to distrust or simply not use the AI solution. People Are Everything You need people at the center of your AI adoption strategy as, after all, they’ll be the ones actually using it. A robust governance framework for AI is essential to ensure the safe and responsible deployment of emerging solutions. This will be critical as responsible AI impacts both the AI industry as a whole and industries, such as industrial digitalization, retail, and financial services. source

By Andrew Ceresney, Julie Riewe and Kristin Snyder ( January 9, 2025, 4:48 PM EST) — In 2024, the U.S. Securities and Exchange Commission faced significant legal challenges to its aggressive rulemaking agenda. As discussed below, several key SEC rulemakings were vacated by the U.S. Court of Appeals for the Fifth Circuit after the court found, among other things, that the SEC exceeded its authority or violated the Administrative Procedures Act by acting “arbitrarily and capriciously” in promulgating certain rules…. 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 Story, an intellectual property blockchain, believes that creators, developers and artists should be able to be rewarded for what they contribute to AI. But it’s not easy to trace their contributions, and Story, which has raised $140 million to date (formerly Story Protocol), is doing something about that. Story, the world’s intellectual property blockchain has announced its use of Stability AI’s cutting-edge models to usher in a new era of open-source AI development, which allows contributors – creators, developers, artists – to capture the value they create by contributing to the AI ecosystem. With the use of Stability AI’s technology, Story aims to address the critical challenge of properly attributing, tracking, and monetizing creative work generated via AI. Story is focused on addressing the lack of a clear path for creators to monetize their derivative works in the open-source ecosystem. Despite the incredible progress in AI, proper attribution and monetization for creators’ IP has not kept up with the rate of innovation. Story’s ecosystem partners. “We’re thrilled to leverage Stability AI’s models to tackle the most pressing challenges we face with the rapid rise of AI,” said Jason Zhao, chief protocol officer at PIP Labs, Story’s initial core contributor, in a statement. “The combination of AI and blockchain is not only incredibly powerful, but necessary. Blockchains secure digital property rights in the era of AI driven creative abundance. By leveraging Stability AI’s technology and Story’s technology, we’re showcasing how the proper incentive structures can ensure attribution and empower creators, driving AI development forward.” Jason Zhao, chief protocol officer at PIP Labs. Story and its ecosystem applications will use Stability AI’s leading foundational image models to build AI applications that embed tracking of contributions across the AI development life cycle to enable fair compensation to all creators involved with a monetized output.  Mahojin and ABLO are two AI applications building on Story that leverage Stability AI’s foundation models and Story’s blockchain technology. Mahojin, a search-to-generate AI remixing platform and ABLO, a collaborative AI platform that allows creators to design physical goods with leading brand IPs use Stability AI’s models to allow users to easily bring their creative vision to life and Story’s technology to enable better provenance and attribution across the AI stack. These two projects showcase real-world use cases and illustrate how to unlock new ways for creators to safeguard their IP and earn from their contributions in a dynamic, shared creative economy. With this kind of tracking, it means that artists and other creators will be able to get paid for their work more fairly, quickly and easily. It also means that the work of those artists, creators and developers could be used more widely. Stability AI, maker of Stable Diffusion, is used by more than five million people to generate images and media through its generative AI model. It focuses on open source models and has expanded into generation of video, audio and more. “Empowering creators is at the core of everything we do at Stability AI. We are thrilled to see our models used in Story’s blockchain technology to ensure proper attribution and rewards contributors,” said Scott Trowbridge, vice president of Stability AI, in a statement. He said a decentralized model for the creator industry is going to become increasingly important. Credit: Image generated by VentureBeat with Stable Diffusion Large 3.5 Story said it is committed to exploring different use cases for how AI and blockchain can come together to meet the evolving needs of creators and developers in the age of generative AI. For example, one area of exploration is registering training data like an artist’s unique style or voice as IP with transparent usage terms on Story. Anyone can then train and fine-tune their own model using this IP. If a creator uses this model to generate an output that is monetized, everyone in this chain of creation wins and benefits together. By leveraging Stability AI’s cutting-edge models, Story is taking a key step toward creating a sustainable and fair internet in the age of AI. PIP Labs, an initial core contributor to the Story Network, is backed by investors, including A16z crypto, Endeavor, and Polychain. PIP Labs was cofounded by a serial entrepreneur with a $440 million exit and Deepmind’s youngest PM with the veteran founding executive team with a diverse background in consumer tech, generative AI, and Web3 infrastructure. source

Preventing waste and abuse has long eclipsed innovation as a focus for the US government. Despite this, innovation is still happening. In fact, the US government doesn’t have an innovation problem; it has an adoption problem. The theory of constraints tells us that aiming our efforts anywhere except the source of a bottleneck is unlikely to deliver value. That’s why some of the most innovative initiatives in government focus on accelerating procurement, streamlining compliance, and platform engineering. What do these things have in common? They all aim organizational effort at attacking the most impactful bottlenecks in modern service delivery in government programs. It’s critical, as government organizations continue to work to improve their delivery models and reduce time lines around procurement and compliance, that they don’t lose sight of where the rubber meets the road: modern service delivery and product development. Our research shows that two of the common threads among organizations that do this well are in embracing agile development practices from the executive down to the program office level and leveraging open-source projects where possible instead of reinventing the wheel. While these two concepts are rarely paired together explicitly, both play a key role in accelerating development and fostering innovation. Let’s explore what makes open-source and agile adoption a unique proposition for government organizations. Open-Source Adoption In Government Is Viable But Challenging Open-source software is a powerful alternative to commercial or customized solutions, allowing development teams to build on community efforts and inherit their collective innovation. But for government organizations, adopting open-source projects comes with its own set of hurdles: The stringent regulatory environment, national security considerations, and complex acquisition processes make assessing the viability of an open-source project a critical precursor to adoption. The Top Challenges In Government Open-Source Adoption Security, compliance, and privacy: Government security, compliance, and privacy laws might not be top of mind for open-source contributors, leading to gaps that must be filled during implementation. Influence over project direction: Government program offices likely won’t have significant influence over the direction of large open-source projects, making alignment with specific needs challenging if you are expecting a full solution out of the box. Licensing and legal considerations: Licensing terms and legal considerations can change, leaving programs scrambling to adapt. For example, when HashiCorp changed its licensing from the Mozilla Public License to the Business Source License, organizations had to quickly reassess their use of HashiCorp products to ensure compliance with the new terms. Interoperability and integration risks: Open-source projects are often a great starting point for new applications, but legacy modernization projects bring additional complexity and considerations. Interoperability with legacy architectures can lead to costly integration challenges. Agile: Commitment Vs. Execution Federal leaders are committed to agile development, but the results at the team level often lag. Despite strong support for modern service delivery practices, few projects are fully agile in their execution. This paradox highlights the challenges and opportunities in adopting agile methodologies within government agencies. The Key Impediments To Agile Adoption Cultural resistance and security concerns: A blend of cultural resistance and legitimate security and compliance concerns can hinder agile adoption. Waterfall-oriented frameworks: Acquisition, budgeting, and contracting frameworks still promote waterfall approaches, making agile implementation challenging. Federated organizational structure: The inherently federated structure of government organizations complicates the shift to agile. Mission-critical systems: Mission-critical systems and infrastructure cannot be replaced by half-baked experiments, necessitating a cautious approach. Unlock The Full Reports To dive deeper into how to build a foundation for success and accelerating adoption, read the full reports on assessing open-source viability in government projects and accelerating agile in the US government. source

By Andrew Karpan ( January 9, 2025, 9:36 PM EST) — OpenAI and its backers at Microsoft will try persuading a New York judge to dismiss one of the major copyright suits against them, with arguments that using news stories to train the startup’s artificial intelligence model is a “transformative” use. … 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

Customer service expectations have changed dramatically over the past several years, with more and more people expecting to get help faster than ever before. A natural language IVR (Interactive Voice Response) solution is an automated system that allows callers to speak in a conversational, free-form way to interact with the system, using speech recognition and AI to understand and process requests. This technology relies on Automated Speech Recognition (ASR) and Natural Language Processing (NLP) to interpret what callers need — from there the natural language IVR can provide relevant responses and route calls based on the caller’s intent. You may have heard it described as a conversational IVR — it’s the same thing. Unlike traditional IVRs, which rely on rigid menu options and keypad inputs, natural language IVRs enable a more intuitive and flexible user experience. People were hesitant to adopt this at first, but today, it is the new normal in the customer service industry as more and more call center software providers are offering the feature. In theory, it’s a win-win. On the one hand, customers get to express their needs in words that come naturally to them rather than navigating a menu, and on the other, agents get to save valuable time by not having to talk to people who don’t actually need a human’s help. Let’s take a closer look, though, because the initial setup and ongoing training is bound to be more costly than a regular IVR. 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 3 CloudTalk 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 24/7 Customer Support, Call Management/Monitoring, Contact Center, and more Natural language IVR vs a regular IVR Here’s a simplified breakdown of traditional IVR technology, and where a natural language IVR goes further: Traditional IVR Relies on predefined prompts and menu options. Requires users to press buttons or speak specific phrases to navigate. Requires users to follow a fixed set of options. Uses scripted responses and basic speech recognition. Natural Language IVR Allows users to speak in natural language. Recognizes, interprets, and responds to a wide variety of conversational inputs. Allows users to engage in more open-ended dialogues. Adapts to different user responses based on context. Prompts users with clarification questions instead of starting over. Traditional IVR systems are incredibly useful — but no matter how complex you make them, they are essentially pre-recorded navigation menus. Customers call in, listen to a series of menu options, and then press a number that corresponds to their choice. Natural language IVR allows customers to interact by using their natural way of speaking rather than having to say a bunch of pre-determined phrases or punch in a series of numbers. This helps improve customer satisfaction — since no one likes fighting with robo-menus — and it gives phone system administrators a much greater degree of freedom to set up IVR call flows. How natural language IVR works (in detail) Natural language IVR works by combining complex speech recognition and pattern-spotting. When a customer says something to the IVR, the IVR recognizes some of the words or phrases they said and knows (or guesses) how to respond based on decision parameters you can configure ahead of time. This process relies on several key technologies, including ASR, NLP, Natural Language Understanding (NLU), and Natural Language Generation (NLG). First, the system uses ASR to detect that speech is happening and convert it into text. Next, the NLU component analyzes the transcribed text, identifying the intent behind the words — whether the caller wants to make an appointment, ask a question, or request information. This step is crucial, as it involves extracting meaning from the speech and understanding the context of the request. Finally, NLG is employed to generate a human-like response, crafting a reply that sounds natural and relevant to the conversation, based on patterns the system has learned through training data. Learn more about how AI in call centers is revolutionizing conversational technology like natural language IVR. It’s transforming the customer service experience by providing more efficient, intuitive, and personalized interactions. Natural language IVR example Let’s use a simple real world example of a caller greeted by an IVR who says, “I wanna (sic) make an appointment.” The natural language IVR uses ASR and NLP to interpret the request. The system recognizes the intent behind the phrase— wanting to make an appointment — and asks the caller if they are correct. “So you would like to make an appointment, do I have that right?” Once the request is confirmed, the IVR directs the call to the appropriate next step, such as scheduling with an available representative, or offering options for time slots (if your IVR is integrated with your appointment scheduling software). This is an improvement for callers, who would ordinarily have to listen to a pre-recorded greeting with basic information and a menu of options. But what if a caller says something unusual, such as, “I left my wallet behind at my last appointment.” Yes, it’s possible that the system could mistakenly lead the caller to a new appointment scheduler. But typically an IVR is set up to confirm that it has understood a caller’s request prior to routing the call. If the IVR can’t interpret the request, it could trigger an agent intervention or route the caller to a basic touch-tone menu. As you can see from this example, a natural language IVR is going to be overkill for a small business with relatively few options for callers to navigate. Simple scheduling can be handled by traditional IVR

XOcean has secured $115mn to expand its fleet of uncrewed surface vessels. Founded in 2017 by James Ives, XOcean builds autonomous boats that zip around the ocean, using sensors to gather large amounts of data on everything from the subsurface structures to the temperature and clarity of the seawater.  The bots — which are about the size of a small car — then relay this information in real time to a ground team via satellite link. They then turn the numbers and measurements into surveys, maps, or reports. This data is especially helpful for ocean  research. It’s also of great value to companies in offshore wind, oil and gas, and carbon capture. XOcean’s funding round was led by Big Oil-backed Climate Investment and SGS, an American VC focused on clean energy. The other key investors were Morgan Stanley’s 1GT fund and an affiliate of the Crown Family’s CC Industries.  How Startup Amsterdam Boosts Innovation and Growth at TNW Conference Discover how the City of Amsterdam partnered with TNW to amplify its startup ecosystem, attract global talent, and foster innovation that drives economic impact. XOcean also raised $30mn back in June, bringing its total funding to date to $180mn. Although it has not publicly disclosed its valuation, Dealroom puts it north of $500mn. Whatever the precise figure, it’s certainly a lot of cash, especially for a company building tiny little boats that gather data that we’ve been able to collect through other methods for decades.  However, the real added value is that XOcean can gather this data in a way that’s “safer, cost-effective, and ultra-low-impact,” says Ives.  “Traditionally, mapping the ocean floor and collecting marine data requires a ship with a large crew working for extended periods at sea, Ives told TNW. “When a customer needs data, it can be months before a traditional supplier has availability and costs are uncontrolled.” Now, people who need this data can rely on a small, low-energy autonomous vessel to do the hard work.  XOcean also promises that its vessels are greener. The company estimates that a fleet of its drones emit just 0.1% of the CO2 of the equivalent surveying boats with a crew on board.  And it’s a pitch that seems to be paying off, not just in VC cash, but in real-world deployments. XOCEAN already works with offshore giants including SSE Renewables, Ørsted, BP, and Shell. The scaleup has delivered data solutions to commercial and government clients in over 23 territories, it said. As we build more and more infrastructures out at sea, it’s no surprise that we’re seeing new technology hopping on board. A second example comes from Zelim, based out of Edinburgh, which is developing AI-powered person-overboard detection that aims to increase survival rates. Meanwhile, another British company, Beam, has built an underwater robot for inspecting offshore wind farms.     source

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More As enterprises continue to adopt large language models (LLMs) in various applications, one of the key challenges they face is improving the factual knowledge of models and reducing hallucinations. In a new paper, researchers at Meta AI propose “scalable memory layers,” which could be one of several possible solutions to this problem. Scalable memory layers add more parameters to LLMs to increase their learning capacity without requiring additional compute resources. The architecture is useful for applications where you can spare extra memory for factual knowledge but also want the inference speed of nimbler models. Dense and memory layers Traditional language models use “dense layers” to encode vast amounts of information in their parameters. In dense layers, all parameters are used at their full capacity and are mostly activated at the same time during inference. Dense layers can learn more complex functions as they grow larger, but increasing their size requires additional computational and energy resources.  In contrast, for simple factual knowledge, much simpler layers with associative memory architectures resembling lookup tables would be more efficient and interpretable. This is what memory layers do. They use simple sparse activations and key-value lookup mechanisms to encode and retrieve knowledge. Sparse layers take up more memory than dense layers but only use a small portion of the parameters at once, which makes them much more compute-efficient. Memory layers have existed for several years but are rarely used in modern deep learning architectures. They are not optimized for current hardware accelerators.  Current frontier LLMs usually use some form of “mixture of experts” (MoE) architecture, which uses a mechanism vaguely similar to memory layers. MoE models are composed of many smaller expert components that specialize in specific tasks. At inference time, a routing mechanism determines which expert becomes activated based on the input sequence. PEER, an architecture recently developed by Google DeepMind, extends MoE to millions of experts, providing more granular control over the parameters that become activated during inference. Upgrading memory layers Memory layers are light on compute but heavy on memory, which presents specific challenges for current hardware and software frameworks. In their paper, the Meta researchers propose several modifications that solve these challenges and make it possible to use them at scale. Memory layers can store knowledge in parallel across several GPUs without slowing down the model (source: arXiv) First, the researchers configured the memory layers for parallelization, distributing them across several GPUs to store millions of key-value pairs without changing other layers in the model. They also implemented a special CUDA kernel for handling high-memory bandwidth operations. And, they developed a parameter-sharing mechanism that supports a single set of memory parameters across multiple memory layers within a model. This means that the keys and values used for lookups are shared across layers. These modifications make it possible to implement memory layers within LLMs without slowing down the model. “Memory layers with their sparse activations nicely complement dense networks, providing increased capacity for knowledge acquisition while being light on compute,” the researchers write. “They can be efficiently scaled, and provide practitioners with an attractive new direction to trade-off memory with compute.” To test memory layers, the researchers modified Llama models by replacing one or more dense layers with a shared memory layer. They compared the memory-enhanced models against the dense LLMs as well as MoE and PEER models on several tasks, including factual question answering, scientific and common-sense world knowledge and coding. A 1.3B memory model (solid line) trained on 1 trillion tokens approaches the performance of a 7B model (dashed line) on factual question-answering tasks as it is given more memory parameters (source: arxiv) Their findings show that memory models improve significantly over dense baselines and compete with models that use 2X to 4X more compute. They also match the performance of MoE models that have the same compute budget and parameter count. The model’s performance is especially notable on tasks that require factual knowledge. For example, on factual question-answering, a memory model with 1.3 billion parameters approaches the performance of Llama-2-7B, which has been trained on twice as many tokens and 10X more compute.  Moreover, the researchers found that the benefits of memory models remain consistent with model size as they scaled their experiments from 134 million to 8 billion parameters. “Given these findings, we strongly advocate that memory layers should be integrated into all next generation AI architectures,” the researchers write, while adding that there is still a lot more room for improvement. “In particular, we hope that new learning methods can be developed to push the effectiveness of these layers even further, enabling less forgetting, fewer hallucinations and continual learning.” source