Ankur Jain

There is a powerful and transformative change that is happening right now in the world of business insights creation and activation that can deliver AI readiness that is accurate, secure, fast, and scalable. I’d recommend starting with this video if you’re on point to make AI real for your organization.

10

Composable CDPs have some clear advantages over out-of-the-box solutions. Think about the example of reverse ETL tools, which take your first party data and push it to other destinations. If a company has had to build everything from scratch, they simply won't have the resources to invest in reverse ETL like a dedicated specialist would. They have so many other components to focus on that connecting with dozens of important destinations won't be possible. That's where the beauty of composable CDPs comes in. Imagine going to a company that specializes solely in reverse ETL and having them tailor their solution to fit with your existing tech stack. The result? A customized solution that meets your unique needs and surpasses anything a single CDP provider could offer. Instead of settling for a one-size-fits-all solution, brands can pick and choose the best-of-breed components, such as reverse ETL, from specialized providers. By connecting these components together, brands can create a purpose-built solution that outshines any one off-the-shelf option.

7

Interesting point of view on Apple's approach to AI. I believe that Apple approach is pretty thoughtful for the consumer space (that is generally grossly overhyped). I strongly believe that the real next frontier for LLM value creation is in developing LLMs for enterprise use cases that offer concrete practical value. This is very much an (yet) unsolved area! #GenerativeAI #largelanguagemodels #artificialintelligence #airevolution

53

5 Comments

🚀 Transforming Data Strategy for the Future: The Rise of Data Mesh and Composable Architecture 🚀 🔗 Dive into our latest feature at The CDO TIMES: "The Evolution of Organizational Data Ecosystems" – a comprehensive look at the seismic shifts in data management paradigms from traditional, rigid frameworks to dynamic, integrated digital ecosystems. 🌐 In this era of rapid digital transformation, understanding the strategic implementation of Data Mesh and Composable Data Architecture is not just advantageous—it's essential for staying ahead. These architectures aren't just about managing data; they're about empowering businesses with flexibility, agility, and the capability to innovate faster than ever. 💡 Key Insights: Data Mesh decentralizes data management, turning data into a product that drives business efficiency and innovation. Composable Data Architecture offers the flexibility needed in today’s fast-paced market, allowing businesses to adapt swiftly and efficiently to changing data needs. 🔍 We're at a pivotal moment where data isn't just part of the business; it is the business. As we move towards more advanced integrations of AI and potentially quantum computing, the stakes will only get higher. 📊 Whether you're a C-level executive strategizing for the next big shift or a data professional looking to stay on the cutting edge, this article provides essential insights and strategic frameworks that will help chart the course of your organization’s data future. 👉 Read the full article and join the conversation about how these revolutionary data frameworks are shaping the future of business. https://lnkd.in/eAkacPfp #DataStrategy #DigitalTransformation #DataMesh #ComposableArchitecture #BusinessIntelligence #CDO #Innovation #AI #QuantumComputing #TheCDOTimes

3

I was working with a big #CPG brand who were shopping for a new vendor. Their main concerns: 1. It takes too long, 3-4 months after data delivery 2. Its not granular enough. (These were PPG models which are granular according to most MMMs) 3. The optimization takes too long and we are unable to use in time These are all valid points to look around. If your 5-10 year MMM partnership hasn't solved for this problem till now, likely, they will not be able to solve it in the 11th year. The decision to move on, makes sense. However, the vendor that they were about to choose also couldn't solve for #1 and #2. Their timeline was similar. Also, they didn't specialise in CPGs and didn't even know what a PPG was. I said to them, this won't work because this vendor doesn't solve for #1 and #2. We moved on. But most brands don't know the insides of an MMM and a relationship until much later after spending months, years and hundreds of thousands of dollars in a problem that cannot be fixed. Because, its not the right FIT. If you are considering a new MMM, or reconsidering your #MMM, reach out. Naavics #media #attribution #vendor

8

6 Comments

Chatting and Generative AI - Not Ideal In Every Case What if the conversational UI understood the boundaries and strengths of the RAG corpus and prevented you from wasting time while putting you on top of the answers you seek in milliseconds? Ideally, we want generative AI solutions to be smart enough to guide us, but this often requires more than guardrails that prevent jailbreaking and hallucinations. We need generative AI solutions — the interface itself — to possess information that dynamically tests the legitimacy of new prompts before you consume CPU and GPU cycles on your company’s services. The UX should give you a clear sense that what you’re about to ask will probably result in a successful or poor outcome. I believe we waste a massive amount of resources blindly chatting into the LLM ether. I used CustomGPT.ai for this RAG infrastructure. https://lnkd.in/gqk4i4H9

15

13 Comments

Gen AI: Coming to an On-Prem Theatre Near You? In the march to embrace generative AI, most IT leaders have generally assumed the SaaS model is equally applicable to gen AI as it is to business solutions. My assessment suggests otherwise. I sense there's something unmistakable about the economic factors surrounding decisions involving generative AI adoption for workers. Ya know that smell of fresh salt air as the tide turns? That's what I'm sensing for on-prem AI - there's a certain freshness and happiness I sense when speaking with those who have experienced localized generative AI. - Most organizations (SMEs especially) have become shadows of their former IT selves. - Most have also become vastly dependent on the SaaS model. - Many believe the high technical bar of DevOps, on-prem deployments, and sustained localized application management is out of reach financially and managerially. - Many want to insulate their teams from privacy and security issues, especially regarding generative AI. They rationalize a trade-off, which is a decision point that falls in favor of SaaS solutions. - Expectedly, they lean into SaaS with generative AI services and fail to realize that workers are infinitely more likely to expose sensitive information when a conversational UI provides the gateway to better search, improved Q&A, and hypothetical queries. SaaS is a foregone conclusion because (i) we are all pretty satisfied with the cost matrix, performance, and security, and (ii) generative AI vendors have [thus far] offered only off-prem solutions. What's your take? Are we about to experience a sea change as we understand the risks of trusting SaaS AI solutions with our deepest and most sensitive thoughts and questions?

10

10 Comments

You definitely want to attend this talk! Desireé Abbott's book is coming out AND this talk features interactivity in visualization. #DataViz #DataFam #Tableau

14

Wake up! LLMs have jolted companies from their data slumber. But mountains of disjointed info, scattered silos, ignored guidelines, and skyrocketing costs are getting in the way. Sanjay Deshmukh tells us how Snowflake is stepping up as the new data sheriff and bring order to this new chaotic data frontier. #data #snowflake #LLM #generativeai #AI #datamanagement

14

*Three silos for the marketing-kings under the sky, Seven for the data-lords in their halls of ETLs, Nine for mortal analysts, doomed to write SQL, One for the Data Architect on his dark throne In the Land of Data where Silos lie. One Lakehouse to rule them all, One Share to find them, One Share to bring them all and in the governance bind them. In the Land of Data where the Silos lie.* Data sharing in a lakehouse is a massive leap forward in the customer data space. I have been building software for customer data for over a decade and over that time there have been so many attempts at breaking down the silos an enterprise company collects over time. The key problem has always been figuring out how to unify those silos without massive burdens on data eng teams. Each "silo" has always had its own storage and interface to interact with it. The history is full of iterations. There were various evolutions of the data warehouse, which was great for analytics work. There was the data lake, which provided cheap storage at an incredible scale. The cloud data warehouses separated compute and storage like data lakes while also providing great analyst interfaces and finally toolkits for AI/ML. The lakehouse era and the innovation of data sharing is the missing piece I have been craving for all of this time. With lakehouse data sharing and open-format storage, each application an enterprise company needs can optimize its own compute and features while the data can be accessed from a single interface. This provides truly unified data governance and unlimited flexibility with all of the benefits we have been chasing over time with big data storage technologies. Data engineering teams won’t have to focus on building and maintaining an endless array of integrations. Instead they can spend their time refining and expanding on the data in their lakehouse and all of the tools that plug into that data can get the benefit. One layer to rule them all. I wrote a blog about it for Amperity 👇 https://lnkd.in/gQBxqz5b

61

7 Comments

Excellent quotes by David Mathison for a Financial Times article below about the rise of the #ChiefAIOfficer. A google search for a "Chief AI Officer" doesn't show many - but lots of companies have "Chief Data Officer" roles that actually include AI and Data foundations. It may not sound as cool, but the reality is that good AI needs good Data and the two can't be separate. Separating the function into AI and Data will cause issues in execution - a good Executive leading AI needs to have Data, AI, Governance, Infrastructure and Analytics experience as a practitioner. Critical for the person to understand both stakeholder and engineering/technical. The person also needs to know how to balance the strategic needs of the Board with the tactical needs of the Business. So, what skills do you think a Chief Data and AI Analytics Officer needs to have? - AI/ML - Data/Analytics - ML Ops, Cloud Ops - Governance, Risk and Responsible AI - Data Infrastructure, Platforms - Business Stakeholders, Strategy What else?

46

10 Comments

Typically most #llm work we do is to reduce the responses down to JSON, but LLms even #llama3 do well with structured data as SQL. I’ve built entire schemas fairly rapidly in the early days of #genai

4

From Princeton & Microsoft There is a lot of interest in estimating LLMs' uncertainty, but should LLMs express uncertainty to end users? If so, when and how? v/ Sunnie S. Y. Kim In our #FAccT2024 paper, we explore how users perceive and act upon LLMs’ natural language uncertainty expressions. "I'm Not Sure, But...": Examining the Impact of Large Language Models' Uncertainty Expression on User Reliance and Trust https://lnkd.in/emqtjqPQ Through a pre-registered experiment (N=404), we examine how the presence and perspective (first-person/general) of LLMs' uncertainty expression impact user reliance, trust, and task performance, when users seek medical information with a fictional LLM-infused search engine. Through a pre-registered experiment (N=404), we examine how the presence and perspective (first-person/general) of LLMs' uncertainty expression impact user reliance, trust, and task performance, when users seek medical information with a fictional LLM-infused search engine. We also find perspective matters: LLMs’ uncertainty expressions in the first person show stronger effects than those in the general perspective. We view this as evidence that "language choices" matter in how users perceive and act on the outputs of LLMs. Our results suggest teams building and deploying LLMs and LLM-infused applications should evaluate approaches to mitigate overreliance, including language choices, carefully with users before release. We hope our work serves as an example and guide for such evaluations.

45

Maximizing Profits with Precision: How Balancing Coverage and Accuracy in Data Science Fuels Revenue Growth The ability to balance "coverage" and "accuracy" in data science models significantly impacts revenue growth. Achieving the right balance can transform a theoretically perfect model into a practical, revenue-enhancing tool. Why Balance Matters? Coverage ensures that a model or dataset captures a range of scenarios, making insights comprehensive and actionable. This broad applicability directly correlates with market reach and potential revenue streams. Accuracy denotes how closely predictions match outcomes. High accuracy reduces risks and builds trust in data-driven decisions, crucial for strategic implementations that drive revenue. Agile Model Production: The Path to Quick Wins Incorporating agile practices in model development enhances adaptability and speed, key components for staying ahead in competitive markets: 1. Iterative Refinement: Agile sprints allow for continuous model adjustments based on real-world feedback, fine-tuning both coverage and accuracy. 2. Cross-Functional Collaboration: Involving diverse teams in the model development ensures that multiple perspectives drive the models' evolution, enhancing relevance and applicability to different market segments. 3. Feedback Integration: Agile methods facilitate the swift incorporation of user and stakeholder feedback, aligning model outputs with current market demands and maximizing revenue opportunities. Revenue-Driving Strategies - Enhanced Targeting Through Better Coverage: Improved model coverage allows for targeting broader customer segments, potentially capturing untapped markets which directly increase potential revenue. - Increased Conversion Via Accuracy: Precise models significantly boost conversion rates by reducing false positives and enhancing user experience, directly impacting the bottom line. - Rapid Prototyping and Market Testing: Agile workflows enable quick launches of prototype models, allowing businesses to test markets and refine strategies without significant upfront investment. The strategic balancing of coverage and accuracy, coupled with agile production techniques, not only speeds the development process but also significantly boosts profitability. These models allow businesses to expand market reach while ensuring that strategies are targeted effectively. As we refine these practices, the essential question is: How effectively are you using agile methodologies to enhance your data science projects for increased revenue? #DataScience #RevenueGrowth #MachineLearning #Agile #BigData #Analytics #AI

2

The Information has a great write-up about the intense battle for foundational data storage companies namely Snowflake and Databricks and who really has the pole position. It is a great read for a part of the technology stack that frequently does not get the headlines they deserve. What do you think? #mangement #venturecapital #privateequity #cloudcomputing #capitalmarkets

1

Join me and my colleague Abhishek Mundra for an upcoming webinar that offers buyers and service providers valuable insights into real-world gen AI use cases in retail and CPG that are moving into production, key challenges & hurdles for scaling AI, and the suggested enterprise playbook around AI governance. https://okt.to/n8eJ95

18

Have you tried Snowflake's Artic? Artic is the latest open sourced LLM launched by Snowflake. It's amazing to observe the innovation driving training and operating efficiency for LLMs. Some Specs about Artic: - Dense-MoE Hybrid transformer architecture - 480B parameters spread across 128 fine-grained experts - 17B Active Parameters - Apache 2.0 licensed The main use cases of Arctic are SQL, coding, and instruction following for enterprise applications. Generating SQL code presents an important challenge, demanding a deep understanding of both natural language and programming syntax. It involves crafting precise outputs that faithfully reflect the user's intentions. Artic is available on Hugging Face, Together AI, Microsoft Azure, NVIDIA API Catalog and Replicate. It's also coming soon on Amazon Web Services (AWS), Lamini and Perplexity.

25

6 Comments