Apurva Mody

Putting together the opening keynote for UC’s Data Science Symposium I delivered this morning was a super fun challenge. It is seemingly a technical conference, of practitioners and academics. But I knew they would pull in executives and other curious attendees as well (SOLD OUT btw). So I wanted to be thoughtful about how "deep" I went while still capturing the attention of the technical folks. I started with a historical and philosophical journey about AI, inspired by the book "God, Human, Animal, Machine" (thanks Natalie Monbiot for that recommendation - may or may not have spiraled for days after reading this). I think in order to talk about how to use AI systems well, we must understand what we are "trying" to do with them in the first place. I then had the opportunity to reflect on several of the technical shifts I have witnessed in the last decade of my career. I covered 4 core areas: 1️⃣ Data: We’re now dealing with terabytes of unstructured, multimodal data—text, images, voice, and more. Data labeling, redaction, and synthetic data generation has become more common place to ensure we’re capturing the right context with the highest quality. 2️⃣ Architecture: We have gone from simple implementations of modeling techniques like decision trees to complex architectures to support LLMs. We now need to support things like transfer learning, mixture of experts (MOE), reinforcement learning from human feedback (RLHF), and retrieval-augmented generation (RAG). 3️⃣ Testing & Evaluation: Today, measuring a model’s performance goes beyond traditional metrics. We now emphasize helpfulness, harmfulness, and robustness, with rigorous red-teaming, behavioral monitoring, and specialized benchmarks. 4️⃣ Risk Management: As AI capabilities grow, so do the risks. We’re facing challenges like excessive agency, prompt injection, and data poisoning, each with the potential to compromise sensitive information or lead to unintended model behaviors. Let's not forget the ever evolving regulatory environment. I ended with the ultimate of question on everyone's mind, where is this going? Focusing on the near terms trends - the ones that are ready to be useful. Concepts like AI Agentic Systems, Virtual Humans, and the ever evolving Human Computer Interaction paradigms. Thanks Tim Cholvat and Marilyn Kump for inviting me to kick off the Symposium today! AlignAI Nicholas Woo Kristy Wedel #DataScience #AI #UCincy #Agents

120

11 Comments

The new Reduce function will be amazing if you process arrays or json with lateral flatten or udfs. I am going to have to test this out soon. https://lnkd.in/gcQVtZmt

7

Excited to try this one out! I'll be speaking at Minnesota Technology Association Tech Connect in a few weeks on Open Source Large Language Models and will discuss this and many more. Take read into why Snowflake's Arctic should be a model you explore with Streamlit. I'll give you a hint. "Arctic is on-par or better than both Llama 3 8B and Llama 2 70B on all metrics, while using less than ½ of the training compute budget." Look for covering in the upcoming Applied Artificial Intelligence Weekly News issue! https://lnkd.in/gM-wECeu

16

https://lnkd.in/gK7AJq27 great article on the need for enterprise semantic layer by #cube. There are other upcoming players too in this space like #atscale and others. so far the process of building, owning and administering is an art and not well defined. By far these are taxonomy based , as a first step and not #ontology based. One may want to consider #graph to represent #ontologies and #palantir is a great candidate , though is is not as dominant or wide spread as others. Thanks to Artyom Keydunov for describing the need , sadly we saw the raise of these in early 2000s from SAP, Microstrategy and IBM and saw their stagnation too due to non-openness.

1

Ease of use and the ability to bring your own AI models are two reasons why Bedrock should be a part of your Gen AI strategy. With Bedrock, you can avoid the risk of model lock-in which carries the burden of higher cost and complexity. Check out Amazon's new innovations in generative AI applications with Bedrock. Learn more: https://lnkd.in/ekd9dFZW #AmazonBedrock #GenAI

5

I wrapped up a conversation this morning with Jason Krantz, Laurence Kirk, and Dan Lovell about external data. This brings up an interesting point: do you see organizations touting they have lots and lots and lots of data for you? Does the amount of data matter? Well, yes, to a degree, the more data you have can be powerful IF you know what to do with it. External data to an organization can bring new insight and power to your data, analytics, and AI work, but just because you have this external data doesn't mean that you have the answers. You need to be able to use it for the right solutions at the right time for the right objectives. How can you tell what data you should be using? Harken back to your goals and objectives. How are you bringing in data to your organization to help hit your goals and objectives? How are you setting up that data? Do you have good governance? Good connections? Are you using it to build a powerful tool? Overall, yes, having a good amount of data can bring power BUT you need to know how to bring that data to life. Just remember: more isn't always better. Stay nerdy, my friends #dataliteracy #AILiteracy #data #AI Meme Source: Brosix

65

1 Comment

I’m excited to speak at Snowflake's #DataCloudSummit ❄️ in June with Matt Luizzi. Stop by our session, "How WHOOP Migrated from Redshift and Created an LLM Chatbot with Streamlit", at June 03, 2024 | 03:30 PM PDT, to learn more about streamlit, LLM's, chatbots and a lot more in snowflake. https://lnkd.in/egEGMCFf

33

💡 NEW BLOG: Mastering Spark - RDDs vs. DataFrames 💡 In distributed computing with #ApacheSpark, DataFrames often steal the spotlight, but understanding RDDs (Resilient Distributed Datasets) can unlock powerful use cases! In my latest blog post, I dive deep into RDDs vs DataFrames and show you when to use one vs. the other and how they fundamentally differ. A bit of a preamble to my upcoming post on parallelizing API calls and other non-distributed tasks. Link: https://lnkd.in/gyr4eWXT #Spark #DataEngineering #MicrosoftFabric

41

3 Comments

'We’re All Going to Be CEOs Companies that invested in big data over the past two decades have a big advantage because the analytical tools are finally catching up. “That data is probably worth more than anyone thought it would be,” Lange said. “AI allows you to distill that information.” This means decision-makers can look to AI for help answering strategic business questions about customers, deals, products, pricing, operations, and more.' #AI #DecisionMaking

17

An interesting and relevant repost by Ian Coyle Data Ideology, LLC. The competition between Databricks and Snowflake isn't a zero-sum game. Do you agree? #HealthcareData #HealthcareAnalytics #AIinHealthcare #HealthcareInnovation #DigitalHealth #HealthTech #HealthcareTransformation #HealthDataGovernance #ValueBasedCare #DataDrivenHealthcare #PopulationHealth #HealthIT #HealthcareAI #DataScienceInHealthcare #HealthcareQuality #Snowflake #Databricks #Teradata

2

Over the last few months, I've been speaking to many customers around LLM Governance. The reality is long term, organizations will need implement custom guardrails specific to their use cases and industries in addition to standard off the shelf guardrails(e.g. hate speech detection, violence, etc). NVIDIA's NeMo Guardrails is an excellent library to achieve this. And you can pretty easily deploy this as middleware in Snowflake Snowpark Container Services wrapping our Cortex LLM functions which provide easy, instant access to a large library of both closed source and open source LLMs. Checkout how to build and deploy such an architecture here 👇🏻: https://lnkd.in/eXWt6MXe

130

"Spanner now supports vector search! This is a huge win for data scientists working on generative AI. 🤖 #Spanner #AI #DataScience"

5

At ASU-GSV, Google Cloud announced the launch of new certificates & courses in generative AI, cybersecurity, and data analytics - a game-changer for public sector skilling. Imagine the possibilities: 🏛️ Govts leveraging AI to improve citizen services 🚔 Law enforcement using analytics to enhance public safety 🏥 Public health orgs tackling complex challenges with AI 🌐 State & local govts bolstering cybersecurity These no-cost offerings, available through Google Cloud Career Launchpad, will help agencies bridge skills gaps & drive innovation. Partners like U.S. Department of the Treasury and Purdue Global are already on board. With 50% of employees needing reskilling by 2025, these pathways are critical for building a future-ready public sector workforce. The potential impact on government efficiency, security & citizen services is tremendous. Check out the post below for details. ⬇️

68

June has officially become Cloud ADA month (AI Data Analytics) - this week is Databricks Data+AI Summit, let’s see what new exciting announcements are. NVIDIA is in the fore-front of all vendors with GenAI at the front of the stage. Here at NorthBay Solutions, we are also helping many enterprises around the world on GenAI and LLM adoption, wanna hear our stories? Let’s chat with Dennis Kelly , Robert Murphy or myself ☕️☕️👨🏻‍💻 #Databricks #GenAI #LLM #Partner #DPP #Cloud #Data #Analytics #Lakehouse #DeltaTable #DataCatalog #NorthBay #AWS #APN #DAIS #DataEngineering #AI #MachineLearning

70

4 Comments

📅 *Today* - Tuesday, October 22nd, 2024 11:00 AM PT [▶][▶] 👨‍💻👨‍💻Join Abhishek Narain and Wilson Lee for "Unlocking AI at Scale with Data Factory in Microsoft Fabric"! 📊 Efficiently managing high-volume, complex data is critical for building scalable AI solutions. Data Factory in Microsoft Fabric streamlines the process of ingesting, transforming, and orchestrating data, making it easier to prepare and integrate diverse data sources for AI applications 🔬Join this session to explore how Data Factory can help automate data workflows, enabling seamless data preparation for AI models. Learn how to design scalable data pipelines, automate data transformation, and optimize data integration, all within the Microsoft Fabric ecosystem. This is your chance to unlock the full potential of AI using Data Factory!. 🚀🔥Don’t miss this opportunity to learn and do more with AI in Microsoft Fabric. Register now for the Microsoft Fabric & AI Learning Hackathon and livestream series! Check out the "Microsoft Fabric & AI Learning Hackathon" on Devpost: https://lnkd.in/gwSAyWhE See the full "Microsoft Fabric & AI Learning Hackathon" livestream schedule on the MS Reactor: https://aka.ms/fabrichack Watch today's livestream event on-demand on YouTube: https://lnkd.in/gbN9qFTr

15

Microsoft researchers Sebastien Bubeck and Ronen Eldan have developed #PHI-3, a groundbreaking family of small yet powerful language models. Designed for efficiency and high performance, PHI-3 models can be deployed on everyday devices, making advanced AI more accessible than ever. #PHI-3 models offer significant improvements in speed and efficiency, enabling AI integration into more consumer devices and enhancing user experiences. Compared to larger models, PHI-3 demonstrates competitive capabilities while being more resource-efficient. The future looks bright for PHI-3, with ongoing research promising even more powerful capabilities for a broader range of devices. This innovation is set to reshape the AI landscape! More in the Turing Post - PHI-3 Article: https://lnkd.in/dX4XXFMC #AI #MachineLearning #LanguageModels #Microsoft #Innovation #Tech #ArtificialIntelligence #PHI3 #SebastienBubeck #RonenEldan

5

1 Comment