Open Innovation Models

I’m thrilled to share that our paper, "Opening Up Military Innovation: Causal Effects of Reforms to US Defense Research," co-authored with the brilliant Sabrina Howell, John Van Reenen, and Jun Wong, has been officially published in the Journal of Political Economy. For decades, the conventional wisdom in government R&D procurement has been for the government to tightly specify the products it wants. Think of the Henry Ford quote: "If I had asked people what they wanted, they would have said faster horses". Our research explores what happens when the government stops asking for "faster horses" and instead opens the door for industry to propose the "automobile." We studied the US Air Force Small Business Innovative Research (SBIR) program (e.g., R&D for small businesses) where two approaches were run simultaneously: a conventional model with highly specific topics and a new open model where firms could propose any technology they thought the Air Force might need. Using a sharp regression discontinuity design, we uncovered striking causal effects for companies that won an open topic award. These companies saw significant, tangible economic benefits: * 📈 A 12 point increase in the probability of receiving subsequent Venture Capital investment * 🚀 An 11.4 point increase in winning larger, non-SBIR Department of Defense (DoD) contracts, a key measure of military adoption and scale. * 💡 An 8.9 point increase in securing a patent and a 7 point increase in securing a high-originality patent, signaling novel innovation. In stark contrast, winning a conventional award had no positive effects on commercial innovation or military adoption. In fact, its main effect was increasing the chances of winning another small SBIR award, a form of program "lock-in”. Our research demonstrates that the open approach doesn't just work by attracting different firms; the open incentive structure itself drives greater innovation. It provides an avenue for firms to identify technological opportunities the government isn't yet aware of, creating an entry point to much larger public sector contracts and private investment. This work has powerful implications for how we procure innovation across the public sector. I'm incredibly proud of what our team accomplished and hope it contributes to building a more dynamic and innovative industrial base. You can read the full paper here: [https://lnkd.in/eV7uEqeH] #Innovation #Economics #NationalSecurity #DefenseTech #SBIR #VentureCapital #DualUse #AirForce #JPE

Google’s Gemma uses an open model, not an open source strategy. The difference is critical because it has massive implications for open source LLMs. The open model paradigm means the weights are made public, but significant constraints on model usage limit what developers can build. Developers get access to the weights, but nothing else, and the new license gives Google more control over how Gemma is used. The Gemma open model strategy breaks a cardinal rule for AI platforms. If developers can’t build what they need on your platform, they’ll go elsewhere. Companies that release AI platforms have some use cases in mind. The developer community brings its ideas and innovations to the platform. They see uses that AI platform companies like Google don’t. The applications AI platform providers haven’t imagined are often the most lucrative. Open source models crowdsource innovation that’s monetized at the platform vs. model level. Will open models create the same virtuous cycle? If terms of use can change anytime, building an innovative product on Gemma is risky. Google could charge for a novel application or, if it directly competes with Google, restrict it altogether. Other Big Tech companies may follow Google, which would be a huge loss for the open source LLM community and transparency in general. Google implemented the new strategy for safety reasons, but it leaves room for Google to monetize open models better than they could with open source. It’s an intelligent business strategy, but will the developer community embrace it or choose to build the most innovative products on a different AI platform? Winning on the model monetization front could cause Google to lose the AI platform race. #GenerativeAI #AIStrategy #DataScience

Grassroots Innovation: Kaizen in Indian Street Engineering Workshops Street engineering workshops in India, found in market areas and narrow lanes, excel in grassroots innovation through kaizen, meaning continuous improvement. These small, family-run establishments understand customer needs and deliver simple, effective home-related solutions using basic mechanics. Here are some examples: 1. Improvised Spare Parts : When specific home appliance spare parts are unavailable or too expensive, street engineers fabricate parts using basic metalworking tools and local materials. This keeps appliances functional without costly imports or long waits. 2. Affordable Automation Solutions : For home-based businesses, street engineers develop simple automation solutions. These include motorized devices for sewing machines, automated irrigation systems for gardens using recycled materials, and mechanized tools for small-scale production. These solutions enhance productivity and reduce manual labor. 3. Cooling Solutions for Appliances : In regions with extreme heat, home appliances like fans and coolers often overheat. Street workshops devise simple cooling solutions, such as installing small fans powered by the appliance’s own power supply or creating custom vents for better air circulation. These modifications maintain performance and extend appliance life. 4. Noise Reduction in Home Equipment : Noise pollution from home equipment can be a nuisance. Street workshops offer noise-reducing solutions, such as adding custom mufflers, using rubber mounts to dampen vibrations, or retrofitting soundproofing materials around noisy components. These solutions significantly improve the home environment. 5. Water Pump Innovations : Efficient water pumps are critical for home gardens and small-scale farming. Street engineers innovate by modifying hand pumps to work with electric motors or creating hybrid systems that can switch between manual and motorized operation, ensuring reliable water access. 6. Enhanced Ergonomics for Tools : Home tools often need ergonomic adjustments to reduce user fatigue and improve efficiency. Street workshops modify handles, grips, and control systems to better suit individual needs, typically done on-site. The street engineering workshops of India embody kaizen through their continuous pursuit of better, simpler home-related solutions. Their deep connection with the community and understanding of customer problems enable effective innovation with limited resources, proving that impactful solutions often come from simple ideas #india #engineering #innovation #motivation #inspiration #design #education

Over the coming weeks and months you are going to hear a lot more about open-source models. Not only because of June 12 and the US export-control directive that pulled Fable 5 for every customer worldwide. Whatever you think of the decision, there was a direct lesson for enterprises: a hosted model is revocable by forces neither you nor your vendor control. Open weights you already run are the one version nobody can take back. But there is more: 1 - The data stays put. You run the model where the data already lives (your VPC, your on-prem, your regulated environment), so legal and compliance stop being the thing that quietly might kill your project. 2 - You own the cost and the customization. Open weights mean you can fine-tune on domain data, distill, quantize, and run cheap inference at scale. For a lot of narrow enterprise tasks, a tuned open model is good enough and a fraction of the price. 3 - No commercial lock-in. You are not hostage to one company's pricing, rate limits, or deprecation schedule. The permissive licenses on half this list (MIT, Apache 2.0) are what make that commercially safe, not just technically possible. None of this means open wins everywhere. For the hardest frontier reasoning I would still reach for a closed lab, and I would not pretend the operational burden of self-hosting is free. But the applied layer is where most enterprise value actually gets created, and that is exactly where the open-weight option is getting stronger (DeepSeek shipping under MIT, GLM topping SWE-Bench Pro, OLMo releasing its full training pipeline). ByteByteGo put together a clean map below, with some alternatives to watch. — 𝗘𝘃𝗲𝗿𝘆 𝘄𝗲𝗲𝗸, 𝗜 𝘀𝗵𝗮𝗿𝗲 𝘁𝗵𝗲 𝗯𝗲𝘀𝘁 𝗻𝗲𝘄 𝗔𝗜 𝗰𝗼𝘂𝗿𝘀𝗲𝘀, 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀, 𝗮𝗻𝗱 𝗶𝗻𝘀𝗶𝗴𝗵𝘁𝘀 𝘁𝗼 𝗵𝗲𝗹𝗽 𝘆𝗼𝘂 𝗹𝗲𝗮𝗿𝗻, 𝘂𝗽𝘀𝗸𝗶𝗹𝗹, 𝗮𝗻𝗱 𝘀𝘁𝗮𝘆 𝗮𝗵𝗲𝗮𝗱: https://lnkd.in/dbf74Y9E

You’re going to be hearing a lot about Open-source v Open-weights. Open-source AI models and "Open-weights" are distinct approaches to transparency and accessibility in AI. So what’s the difference? Open-source AI models refer to machine learning models whose entire codebase, including architecture, training pipeline, and often training data, is publicly available. This approach offers full transparency of the model's structure and training process, allows for reproduction and improvement of the model, often includes the training data, and typically comes with an open-source licence. Examples include BERT by Google and GPT-2 by OpenAI. "Open-weights", on the other hand, is the practice of releasing only the trained parameters (primarily weights) of an AI model, without necessarily disclosing the full architecture, training code, or data. This approach provides access to the model's learned parameters, allows for fine-tuning and adaptation of the model, does not necessarily reveal the full model architecture or training process, and may come with specific usage restrictions. Open-source and Open-weights differ in several key aspects. Open-source models offer comprehensive transparency, revealing the entire process from architecture to training, while openweights provide partial transparency, focusing on the end result rather than the process. Open-source models can be fully reproduced, whereas open-weights models can only be used or fine-tuned, as the training process isn't disclosed. Open-source models typically grant more extensive rights to users, while open-weights may come with more restrictions on use and modification. Open-weights can be easier to implement quickly, as users don't need to understand or reproduce the training process. Open-source models allow for more fundamental innovations in architecture and training, while open-weights focus innovation on applications and fine-tuning of existing architectures. Reproducing open-source models often requires significant computational resources for training, whereas using openweights is generally less resource-intensive, as the costly training process is already complete. From a commercial perspective, open-source models may pose challenges for companies wanting to maintain a competitive edge, while open-weights can strike a balance between openness and protecting proprietary aspects of the training process. The choice between these approaches often depends on balancing innovation fostering, intellectual property protection, and democratising access to advanced AI capabilities. By sharing learned parameters, open weights enable users to leverage and build upon sophisticated AI models without extensive computational resources or access to large training datasets. This approach has gained traction, especially with large language models, as it allows for wider use and adaptation of powerful AI systems while maintaining some level of proprietary advantage for the original developers.

Everything is framed as open models vs. closed models in a zero sum fashion. That’s the wrong framing. It’s an ecosystem of AI that gets used together and advances the industry and pushes the applied use-cases forward. The amount of creativity that exists when you have competing approaches for what the future looks like drives the cycle of innovation that we’ve seen for all participants. You get to have layers of the stack that emerge to post train models for highly specific purposes, which makes AI more useful in real world scenarios. Instead of waiting for just a few labs to go deep in a domain, you get dozens or hundreds of attempts at that vertical, like in finance, life sciences, legal, healthcare, and more. You get to see variance in how to handle safety and cyber risks. Instead of just one approach, you get a peek into what happens advanced capabilities can be used to build better systems to are used to defend systems. You get alternative approaches to training and building AI models. In more compute constrained environments, you develop more novel and efficient approaches to model training, which every other lab can learn from. And you get different cost structures for different workloads. High end and orchestration tasks can go to the closed frontier models and specific workhorse tasks can be done more cheaply. The reason why you want strong open weights models is because it pushes the entire AI industry forward.

In many nonprofits, innovation often mirrors privilege. Who gets to dream up solutions? Whose ideas are embraced as “bold” or “innovative”? Too often, decision-making is concentrated in leadership or external consultants, leaving grassroots, community-driven insights underutilized. This perpetuates inequity and stifles transformative potential within our own organizations. Here’s the truth: Privilege shapes perceptions of innovation: Ideas from leadership or external experts are often prioritized, while community-driven ideas are dismissed as “too risky” or “impractical.” Communities with lived experience are sidelined: Those who deeply understand systemic challenges are excluded from shaping the solutions meant to address them. The result? Nonprofits risk replicating the same inequities they aim to dismantle by ignoring the imaginative potential of those closest to the issues. When imagination is confined to decision-makers in positions of power, we limit our ability to create truly transformative solutions. As nonprofit practitioners, we can start shifting this dynamic by fostering equity within our organizations: * Redistribute decision-making power: Engage community members and frontline staff in brainstorming and strategic discussions. Elevate their voices in decision-making processes. * Value lived experience as expertise: Treat the insights of those who experience systemic challenges as central to innovation, not secondary. * Create space for experimentation: Advocate for internal processes that allow for piloting bold, community-driven ideas, even if they challenge traditional approaches. * Focus on capacity-mobilisation: Invest in staff and community partners through training, mentorship, and resources that empower them to lead imaginative projects. * Rethink impact metrics: Develop evaluation systems that prioritize community-defined success over traditional donor-centric metrics. What practices has your organization used to centre community-driven ideas? Share your insights—I’d love to learn from you! Want to hear more: https://lnkd.in/gXp76ssF

She couldn't finish her homework before dark. In the Wayúu communities of northern Colombia, sunset means the day ends. No grid. No generators. Just night. A Colombian startup, E-Dina, built a different kind of lamp. You pour in saltwater. Magnesium and copper plates do the rest — an electrochemical reaction that produces light. No fuel. No noise. No toxic batteries. The numbers: ↳ 500 ml of saltwater powers it for up to 45 days ↳ One unit delivers about 5,600 hours over its lifetime ↳ It charges small devices via USB ↳ Built for reuse, fully recyclable The chemistry is clever. But what stopped me is how they built it. E-Dina didn't design this in a city lab and ship it out. They worked with Wayúu families — around the problems they actually live with. A child studying after dark. A phone that stays charged during an outage. A backup light when storms hit. Not a gadget. A tool people can depend on when everything else goes dark. The Multiplication Effect: 1 home with steady light = kids can study after dark 10 coastal villages = emergency power without diesel 100 off-grid communities = electricity without waiting for infrastructure At scale = we stop asking who deserves power and start building where it's needed The ocean has always fed these communities. Now it lights their homes, too. The best innovations don't start where the resources are. They start where the problem is. Where have you seen proximity to a problem matter more than access to funding? Video Credit: Harald Friedl

While national agencies play a vital role, the true strength of disaster management often lies at the grassroots. Community-Based Disaster Management (CBDM) empowers local populations to become their own first responders, transforming vulnerability into collective resilience. When a disaster hits, local communities are the first on the scene, often before external aid can arrive. By equipping them with knowledge, skills, and resources, we foster self-reliance and accelerate effective response. This approach focuses on local risk assessment, tailored preparedness plans, and empowering community leaders who can coordinate efforts and disseminate information effectively. According to a systematic review of disaster management approaches, communities with CBDM plans experience up to 50% fewer casualties in disasters. This is a testament to the power of local knowledge and collective action. From remote villages in Pakistan organizing local flood watch groups, to Indigenous communities in Australia revitalizing traditional fire management techniques, CBDM leverages intimate local knowledge for powerful results. It's about collective ownership and shared safety that builds strength from the ground up. Is your community empowered to respond? Support community-based disaster management for a stronger, more resilient future. #CommunityResilience #CBDM #LocalAction #UNICEF

We’ve gone from a “DeepSeek” moment to a “DeepSeek Month”. In rapid succession, four open models were released that have over one trillion training parameters, the magic number that open model labs had been chasing. That parameter threshold is believed to have played a critical role in the superior performance of models from the two leading proprietary AI labs. The release of three Chinese and one American model, all with open weights, crosses a Rubicon with open models, finally hitting performance thresholds very close to those of the most cutting-edge proprietary models. The story is not just one of competition but also of coopetition and distillation. Large model builders are adapting techniques introduced by their competitors and actively using distillation — effectively learning from other competing models — to improve their own results. The net effect is a rapid acceleration of AI progress. Not surprisingly, firms that act as AI routing services have seen a steady increase in usage of these large models, a shift likely driven by the golden trio of improved performance, lower price, and the promise of AI sovereignty.  Which leads us to the elephant in the room that may prove to be the most decisive factor in determining ultimate adoption patterns and winners in AI. After restrictions were placed on foreign usage of cutting-edge proprietary AI models, businesses and governments all over the world were forced to consider whether their choice of AI models could generate an unacceptable business continuity risk. In many countries, the restrictions spurred efforts to shift away from proprietary models to open weight AI models under permissive licenses equivalent to open source software like Linux and Kubernetes, which now dominate enterprise computing. Bill Gurley a leading voice in technology and long an advocate for open source business models, loudly proclaimed this week that open weight models have already won in AI. The rapid shift in market share does seem to bear out Gurley’s point. Unfortunately, many business leaders now face the untenable choice of deploying a model that might be blocked or “nerfed” by government regulations, or deploying a model that is open but developed in a different country with its own brand of security risks. The AI company Hugging Face was forced this week to use a foreign country’s open frontier models to remediate a cyberattack after model restrictions placed on a proprietary frontier model blocked efforts to probe code for vulnerabilities. According to Hugging Face, the fastest growing component of AI model usage on its site is actually for smaller open models that can run on local servers — effectively a vote for complete and total local control. Ultimately, the crowd will vote with its tokens and business leaders would do well to follow the token flows to understand which way the AI wind blows. https://lnkd.in/e5w7qMaw