James Corcoran

Pushing SHAP explainers to production often feels impossible. Everything works fine in a notebook, but once you try to use SHAP explainers on new data, you hit serialization errors, environment mismatches, and broken explainers. In my latest article, I walk through how to make SHAP work in production with MLflow, CatBoost, and SageMaker. The piece covers why SHAP explainers break, how MLflow's 𝖾𝗏𝖺𝗅𝗎𝖺𝗍𝖾() API solves it, and how to package explainers for production use 🚀. If you've struggled with explainability pipelines in real-world ML systems, this guide might save you a lot of time. 👉 Read here: https://lnkd.in/dN-6JD4N #MLflow #CatBoost #MLOps #DataScience #ExplainableAI

180

6 Comments

The fight for AI dominance continues to move at a relentless pace. Events this week have underscored compute themes I have been writing about recently. Compute infrastructure has become AI's existential bottleneck. It is the new moat. Here's OpenAI President Greg Brockman on CNBC talking about the $100 billion NVIDIA investment: "You want every person to be able to have their own dedicated GPU, right? So you're talking on order of 10 billion GPUs. We're going to need this deal we're talking about, it's for millions of GPUs. We're still three orders of magnitude off of where we need to be. We're heading to this world where the economy is powered by compute, and it's going to be a compute-scarce one." CEO Sam Altman published a short but sweeping manifesto called “Abundant Intelligence": “To be able to deliver what the world needs—for inference compute to run these models, and for training compute to keep making them better and better—we are putting the groundwork in place to be able to significantly expand our ambitions for building out AI infrastructure." Meta CEO Mark Zuckerberg discussed his infrastructure spending plans a few days ago on a podcast: “…if we end up misspending a couple of hundred billion dollars, I think that that is going to be very unfortunate, obviously. But...the risk is higher on the other side. If you build too slowly and then super intelligence is possible in 3 years, but you built it out assuming it would be there in 5 years, then you’re just out of position on what I think is going to be the most important technology that enables the most new products and innovation and value creation in history.” None of these players can afford to be cautious. Waiting means falling irrecoverably behind. However, the risks of this "Damn the torpedoes, full speed ahead" can not be understated. Bain & Company Capital this week released its sixth annual Global Technology Report, which calculates that these companies will need $2 trillion in new annual revenue by 2030 to pay for this compute, but currently appear to be on pace to fall about $800 billion short. At the same time, a new Financial Times analysis of SEC filings and transcripts for S&P 500 companies raised fresh questions about AI’s short-term utility in the enterprise, concluding: “The biggest US-listed companies keep talking about artificial intelligence. But other than the ‘fear of missing out’, few appear to be able to describe how the technology is changing their businesses for the better." So, is this a bubble? This remains the wrong question. By using the lens of discontinuity, we can avoid focusing solely on the short-term bubble narrative, which misses the larger truth: we're witnessing the early stages of a general-purpose technology deployment on par with electricity or the internet. The question isn't whether AI will transform the economy. It's which companies will capture the value. And when.

62

5 Comments

RAG is evolving—and it’s changing how we build intelligent AI systems Most people think of Retrieval-Augmented Generation (RAG) as a simple retrieve + generate setup. But in reality, RAG has grown into a sophisticated ecosystem of architectures—each solving real-world problems like context continuity, multi-source retrieval, and dynamic reasoning. In my latest article on DZone, I break down the evolution of RAG architectures step by step: 🔹 Simple RAG → Lookup from a static knowledge base 🔹 RAG with Memory → Maintain context across conversations 🔹 Branched RAG → Route queries to the right data source 🔹 HyDe → Use hypothetical embeddings for better retrieval 🔹 Adaptive & Corrective RAG → Optimize efficiency and ensure accuracy 🔹 Self-RAG → Fetch missing information mid-answer 🔹 Agentic RAG → Multi-agent systems that plan, reason, and act This guide isn’t just theory—I’ve included visual diagrams and practical workflows to make it easy to understand and apply. 📖 Read the full article here: https://lnkd.in/gxyNYv6b If you’re working on LLM-powered apps, copilots, or autonomous agents, this is a must-read. #RAG #LLM #AIagents #GenerativeAI #AgenticAI #KnowledgeGraphs #DZone

36

👶 This week, Two OpenAI Agents Got Married - And Had a Baby 🍼 OpenAI launched something significant: the ChatGPT Agent - a bold fusion of its previous experiments, combining the Operator agent (functional but very flawed, aka the father) with the very impressive agent (near perfect, aka the mother) Deep Research. My verdict? A unified agentic system that acts like a junior analyst or personal assistant. It brings together: - Operator’s ability to interact with websites - Deep Research’s skill in synthesis and reasoning - ChatGPT’s conversational fluency ✅ What I like The execution. The interface is clean, transparent, and polished. You can see each step the agent takes - building confidence. And releasing it early for user feedback?  Smart move. It's important to build with the community, not in a silo. 🤔 What I don’t like The agent still struggles with a few significant issues: - It sometimes hangs for no reason - Visual outputs like PowerPoints and infographics often look underwhelming - It frequently produces inaccurate results 🚨 What are my bigger concerns? Beyond bugs and polish, there are strategic challenges that AI agents are facing: 1. Cybersecurity: Giving an AI agent access to your inbox, calendar, or documents is a massive trust decision. OpenAI itself warns of prompt injection - hidden commands in web pages that could hijack the agent’s actions or leak sensitive data. Even with “explicit confirmation,” one slip could be costly. And that raises a deeper question: How does the agent decide what's a consequential action vs. a trivial one? 2. Website blocking limits its power: Many websites already block AI agents (e.g. Amazon and LinkedIn). Without local execution or formal partnerships, agents will hit dead ends. If they can’t access or interact with the open internet reliably, their usefulness drops fast. This will become a bigger and bigger problem as AI agents grow more popular. 3. The “Last Mile” problem: 98% accuracy sounds great - until that final 2% undermines everything. Silent errors in finance, legal, or engineering tasks can be fatal. You can spend as much time reviewing as you would doing the task manually. That’s not productivity - that's an illusion. 🧪 The bottom line? Try it - but tread carefully. For personal use: Explore, but don’t share sensitive info. For organisations: Test, learn — but don’t scale or spend too quickly. 🤔 What are the use cases you have in mind to explore? Have a great week ahead, everyone. ----- 👉 Follow Will Liang for more and hit the 🔔 to stay ahead of the curve. 👉 DM me to chat through how AI apply to your team or your business.

53

8 Comments

Thrilled to see our portfolio company Groq announce its latest $750M fundraise at a $6.9B valuation, further cementing its position as a disruptive force in the inference compute market — and a credible challenger to the industry incumbent, NVIDIA. When we invested in August 2024, our conviction was that Groq’s specialized LPU architecture sat at the perfect intersection of soaring AI inference demand, semiconductor supply constraints, and the need for more energy-efficient compute. Over the past year, that conviction has been validated with: - A $1.5B commitment from Saudi Arabia to build the largest inference cluster in EMEA - A new inference cluster in Finland - Orders from Bell Canada - 2M+ developers on GroqCloud What makes Groq stand out: - Unmatched speed: 500–700+ tokens/second, delivering 5–10x GPU throughput - Energy efficiency: ~90% lower power usage vs. traditional GPUs - Predictable latency: Critical for scaled inference workloads - Vertical integration: From GroqSilicon → GroqRack → GroqCloud Excited to continue partnering with Jonathan Ross, Mohsen Moazami, Sunny Madra and the entire Groq team on this journey. Wishing them the very best as they scale to even greater heights ahead. Dishita Prajapati | Hridaya Zaveri | Tejas Kamat

164

8 Comments