Michael Jordan: The Next Evolution of AI: Markets, Uncertainty, and Engineering Intelligence at Scale🔭
I'm building a GenAI course and would love your input 🤗
Welcome back to Vanishing Gradients! This newsletter is where I dig into the latest in data science, ML, and AI, sharing what’s happening in the field and what I’ve been working on. It’s always evolving, so let me know what you’d like to see more (or less!) of as we go.
This week, I’m thrilled to share the launch of High Signal, a podcast designed to bring you real-world conversations with leaders in AI. I’ll also be diving into strategies for making AI accessible beyond big tech, practical frameworks for scaling AI platforms, and how data science teams can drive measurable value in organizations. I'm also excited to get your input on a new course I'm developing on GenAI applications (there's a quick survey below). Plus, there are some key upcoming workshops and events that offer hands-on experience in generative AI and multimodal applications. Let's get into it! 🚀
📖 Reading time: 8 minutes
🚀 Launching High Signal: Conversations with Leaders in AI, Data Science, and Machine Learning 🎧
I’m stoked to have launched a new podcast High Signal last week with my friends at Delphina. It features conversations with the experts driving AI, data science, and machine learning forward. Our first three episodes are available, bringing real-world applications and research insights on today’s biggest challenges. Check out the clips below:
Michael Jordan (UC Berkeley) 🌐 – AI, economics, and intelligent infrastructure. Mike explains his approach to designing AI systems that manage uncertainty and make decisions at scale, combining technology and economic strategies for meaningful impact.
Andrew Gelman (Columbia University) 🔄 – Simulation and statistical thinking. Andy discusses why simulating data before collection is essential, likening it to "writing SimCity before you play it." He also emphasizes that statistics is always about making comparisons, not just finding single estimates.
Chiara Farronato (Harvard Business School) 🔍 – Collaborative decision-making in data science. Based on her Harvard case study with Uber, Chiara explores how data scientists, engineers, and product managers align on goals and tackle challenges like “treatment and control interference.” She shares Uber’s “trip party” analysis as a powerful example of cross-functional teamwork.
And this is just the start. Upcoming episodes feature Guido Imbens (Nobel Laureate), Hilary Mason, Emily Sands, Ramesh Johari, and more.
Produced by Delphina, with Duncan Gilchrist and Jeremy Hermann, High Signal is here to help you advance your career and make an impact in AI, data science, and machine learning. Head to our landing page for full episodes and more. Or just watch my full conversation with Michael Jordan on YouTube!
💫 Help Shape Our New Course: Building Gen-AI Applications Using First Principles 🤖
I’m excited to announce a new course, Building Gen-AI Applications Using First Principles, co-developed with Stefan Krawczyk (CEO of Dagworks, ex-Stitchfix). This course is crafted for software engineers, data scientists, and ML practitioners looking to create production-ready AI systems—not just models.
Why Take This Course?
Build Complete Systems 👨💻: Go beyond just models and create end-to-end, scalable AI applications.
Master the Development Lifecycle ⚙️: From rapid prototyping to iterative refinement, you’ll learn how to manage the entire AI software lifecycle.
Handle Non-Determinism 📊: Confront the unique challenges of data-driven applications, ensuring reliability even with non-deterministic systems.
Develop Real Applications 📄: Gain practical experience by building functional applications, like a PDF-querying system powered by multimodal models.
Help Us Shape It!
We’re gathering feedback to ensure this Maven course is as effective and impactful as possible. Take our quick 5-10 minute survey to share your thoughts, join the waitlist, and be among the first to know when enrollment opens.
⭐ Reasonable Scale AI: How Can the “99%” Put AI into Production? 🌍
In a recent fireside chat for Outerbounds, I spoke with Jacopo Tagliabue, co-founder of Bauplan, about making AI practical for companies outside big tech. Jacopo’s experience spans NLP, information retrieval, and serverless data solutions, making him a key voice on bringing scalable data transformation to the “99%.”
In our conversation, Jacopo shared:
What “reasonable scale” AI really means – Jacopo explains how companies without the infrastructure of Netflix or Uber can still get value from data and ML by focusing on scalable, cost-efficient solutions.
Data over modeling – He emphasizes that high-quality data is the cornerstone of impactful AI, often more important than advanced models.
Practical tips for companies – Jacopo offers advice on adopting AI with a lean approach, covering monitoring, cost efficiency, and the essentials of data transformation without massive infrastructure.
Jacopo’s approach to AI is tailored for businesses looking to make data work without an elaborate tech stack, and his insights are a valuable resource for teams aiming to make AI real at a reasonable scale.
Catch the full conversation here and/or a short clip below!
💡 Data Dialogs: Building AI Systems, Not Just Models with Savin Goyal (Metaflow, Outerbounds, ex-Netflix)
Our latest Data Dialogs session brought together AI/ML leaders from Netflix, LinkedIn, Uber, Peloton, CVS, and more, with Savin Goyal from Metaflow and Outerbounds leading an in-depth discussion on scaling AI platforms, common failure patterns, and the integration of generative AI into existing systems.
Key highlights from the session with Savin:
AI/ML Patterns & Failure Modes:
Mismatched Expectations: AI development isn’t as predictable as software—continuous iteration is essential.
Optimizing Experimentation: Fast, scalable testing helps avoid stagnation and fosters innovation.
Simplifying Infrastructure: Keeping systems streamlined can prevent productivity from being blocked by excessive complexity.
Building Scalable ML Infrastructure:
Empowering Data Scientists: Platforms like Metaflow allow data scientists to focus on business outcomes without getting mired in infrastructure.
Maintaining Velocity: Savin shared how Netflix scaled from running a handful of tests to thousands, ensuring continuous innovation.
Modular Approach: A strategic build-and-buy model can combine standard tools with tailored solutions for optimal scalability.
Integrating Generative AI:
Compute Efficiency: With increased demand for GPUs, flexibility in compute is critical.
Open Source Flexibility: Starting with commercial models like GPT-4 and gradually transitioning to open-source options like LLaMA ensures both adaptability and cost control.
For those interested in exploring the value side of data science, join us for the next session on October 29 with Ali Rauh, Director of Applied Science and Data Science at Uber. Ali will cover how to measure and drive value in data science through frameworks like OKRs, long-term value modeling, and impactful experimentation processes. You can apply to join here!
Level Up Your AI Skills: Free Tickets to the NYC CALM Summit on Generative AI, Conversational Tech, and Building Reliable AI Assistants 🗽
I have a few free tickets available for the upcoming AI CALM Summit in NYC, plus some highly discounted ones! The summit will bring together leaders and experts from top companies and universities, covering the latest in generative AI, conversational tech, and best practices for building robust, reliable AI assistants.
Here’s a quick look at what’s on the agenda:
Industry Insights 🌐: Hear from leaders at Verizon, Mastercard, and Capital One on using generative AI and AI assistants at scale.
Hands-On Masterclasses 🛠️: Dive into practical sessions on building with LLMs, fine-tuning, and voice technology—ideal for developers and conversational AI practitioners.
Academic Expertise 🎓: Learn from experts at NYU and Cornell about scaling language models and designing effective conversational AI systems.
If I were attending, here are some of the sessions I’d look forward to:
🚀 “From 1M to 1B Customers: Scaling with Conversational AI” with Hans van Dam, CEO of Conversation Design Institute.
🎙️ “The Future of Voice Technology” with Alan Nichol (Rasa Co-Founder & CTO) and Ilan Avner (Director of Product Management, AudioCodes).
🧠 “Fine-Tuning Large Language Models” led by Daksh Varshneya, Senior Product Manager at Rasa.
When: October 30, 2024
Where: New York City, NY
If you’re interested in attending, reach out to me for ticket details! This is a great opportunity to learn, connect, and engage with some of the brightest minds in AI.
🎙️ The Art of Freelance AI Consulting and Products: Data, Dollars, and Deliverables
In a recent live podcast recording, I had an in-depth conversation with Jason Liu on the world of AI consulting, freelancing, and building scalable AI systems. Building on our previous episode, How to Build Terrible AI Systems, this discussion dives into the strategies and mindset needed to grow a successful consulting business while creating impactful AI products.
Jason, an independent consultant with experience at Meta and Stitch Fix, brings a wealth of knowledge in designing and deploying production-level AI systems. His work includes developing a $50 million revenue-generating vector-based product similarity search system at Stitch Fix and architecting the widely used Flight recommendation framework.
Highlights from the Episode:
Jason’s Consulting Playbook 📘: Key steps for building production-ready AI applications.
From Freelancing to High-Value Contracts 💼: How Jason moved from hourly consulting to securing substantial, long-term engagements, including his journey from $170/hour to $60k upfront contracts.
The Consultant’s Mindset 🧠: Transitioning from hourly rates to value-based pricing and aligning incentives with clients to ensure lasting impact.
Navigating Non-Deterministic AI 🔄: Jason’s experience in helping teams adapt to the shift from deterministic software to probabilistic AI systems.
Live Role-Playing Session 🎭: Jason coached me through common pitfalls in client engagement and pricing, providing practical advice for freelancers looking to scale their businesses.
This episode also explores Jason’s journey from software engineering to AI consulting, where he helps companies not only solve AI problems but guide their teams towards a scientific, experimental approach to AI development. You can watch the full livestream below or wait until we drop the podcast episode later this week:
Upcoming Workshops and Conferences 🚀
I’m excited to announce some upcoming conferences where I’ll be leading hands-on workshops focused on Generative AI and multimodal applications!
🗽 PyData NYC (Nov 6-8, New York): I’ll be leading a workshop on Building Your First Multimodal Generative AI App. This session will guide you through the practical steps of creating AI applications that can handle multiple forms of input and output. Sign up here to join me and explore the exciting world of multimodal AI!
🤖 MLOps World and Generative AI World Conference (Nov 7-8, Austin): In Austin, I’ll be teaching a workshop on Generative AI for Software Engineers. This is perfect for developers looking to integrate AI into their software workflows. I’m also honored to be on the conference steering committee 🤗. Register here with my 15% discount code!
I’ll be announcing more livestreams, events, and podcasts soon, so subscribe to the Vanishing Gradients lu.ma calendar to stay up to date. Also subscribe to our YouTube channel, where we livestream, if that’s your thing!
That’s it for now. Please let me know what you’d like to hear more of, what you’d like to hear less of, and any other ways I can make this newsletter more relevant for you,
Hugo