Xin Zou

Inference at scale is evolving. The future is composable systems where offline batch, online batch (near-realtime), and realtime traffic can all be served from a single, unified inference endpoint for most use-cases. This simplifies your architecture, reduces operational overhead, and increases throughput. Our new tutorial series will show you how to build it step by sep. In the first installment, Erik Saarenvirta walks you through creating a scalable image classification system on GKE that can be adapted to a variety of use-cases. Leave us feedback in comments or ask questions we are happy to answer. Kent Hua Ishmeet Mehta Erik Saarenvirta https://lnkd.in/dVNiUhwE

16

Yesterday I read this great article for the upcoming EMNLP about product review summaries from a few folks at Wayfair: End-to-End Aspect-Guided Review Summarization at Scale by Ilya Boytsov, Vinny DeGenova, Mikhail Balyasin, Joe Walt, Caitlin Eusden, Marie-Claire Rochat, Margaret Pierson Retailers: this is how review summaries should be done. Wayfair’s new paper shows an end‑to‑end, aspect‑guided approach that turns thousands of raw reviews into a short, trustworthy product summary shoppers can actually use. What’s novel for retail: - Evidence‑first UX: The model writes ~300–500‑character copy anchored to representative reviews. Shoppers can tap aspects like comfort or delivery to dive in. - Real‑time ops: Auto‑generate once a product hits 10 reviews and then auto‑refresh when new reviews grow enough. - Built for scale: Cached aspect mappings + sampling keep costs and latency in check without losing signal. - Proven impact: AB testing shows lifted ATC and CVR and cut bounce with no hit to revenue or page speed. - Open data: ~12m anonymized reviews across 92k products with extracted aspects + generated summaries available on Hugging Face. Open data is a key to comparing models across industries and use cases. Why it matters: A trustworthy, interpretable, continuously fresh review UX that boosts confidence and carts.

27

11 Comments

What are the most underrated Databricks capabilities? Watch my conversation with Shiv Trisal (Global Industrials GTM @ Databricks) on the topic below. For the full, practical, no-fluff discussion on real-world data + AI: https://lnkd.in/gMPsFHbY

89

Today we're announcing that Semantic View Autopilot is generally available in Snowflake! When we GA’ed Cortex Analyst a year ago, we learned that the barrier to AI-powered analytics was never the LLM, it was data definition. When "Monthly Recurring Revenue" means different things to product and finance, AI agents inherit that inconsistency. No amount of prompt engineering fixes that. You need a governed semantic layer - but building one manually takes weeks. So we built a system that learns from how organizations actually use their data. SVA analyzes query history, Tableau dashboards, and trusted SQL to propose governed definitions automatically. Teams review and certify instead of coding from scratch. The more you use it, the smarter it gets. Since launching in preview, we've watched customers go from "I'll spend the next two weeks building a semantic layer" to "I had something working in my first session." Proud of the whole team that shipped this. Read the full announcement: https://lnkd.in/g7Dvhn9D Get started with the docs: https://lnkd.in/guMjU92w #Snowflake #AI #DataAnalytics #Cortex #SemanticView

439

17 Comments

We have released TimesFM-2.5, the new leader in the GIFT-Eval on all accuracy metrics among zero-shot foundation models. This improved foundation model is now available on Hugging Face, with an upcoming release on GCP's BigQuery and Model Garden. This release represents a significant advancement. TimesFM-2.5 outperforms TimesFM 2.0 on leading benchmarks by up to 25%, while using half the number of parameters (200M). It also features a longer maximum context length of 16K. 𝗪𝗵𝗮𝘁 𝗶𝘀 𝗧𝗶𝗺𝗲-𝗦𝗲𝗿𝗶𝗲𝘀 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴? Time-series analysis is the process of studying data points collected over time to make predictions and identify trends. It is a critical tool for a wide range of applications, like: ➡️ Forecasting future product demand. ➡️ Tracking weather and precipitation ➡️ Optimizing supply chains and energy grids. 𝗪𝗵𝘆 𝗶𝘀 𝗶𝘁 𝗖𝗵𝗮𝗹𝗹𝗲𝗻𝗴𝗶𝗻𝗴? Time-series forecasting is difficult because data patterns are often complex, can change over time, and are influenced by numerous factors. Developing a single model that can perform well across diverse datasets without being explicitly trained on each one has been a major challenge. 𝗧𝗶𝗺𝗲𝘀𝗙𝗠-𝟮.𝟱: 𝗔 𝗡𝗲𝘄 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱 𝗳𝗼𝗿 𝗭𝗲𝗿𝗼-𝗦𝗵𝗼𝘁 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝗶𝗻𝗴 TimesFM-2.5 sets a new standard for a decoder-only foundation model trained on a large time-series corpus. 🤗 Leading Performance: TimesFM-2.5 holds the top position on the GIFT-Eval leaderboard for point forecasting accuracy (MASE) and probabilistic forecasting accuracy (CRPS). 🤗 Efficiency: The model's efficiency is a key feature, with a small parameter count that makes it practical for a wide range of production environments. 🤗 Longer Context: The increased context length allows the model to process more historical data, leading to more accurate forecasts. This work reiterates that building a single foundation model for time series forecasting is possible. Google Research keeps pushing forward the frontier of time series forecasting research. We are grateful to our community and customers who have provided feedback and deployed TimesFM in production. We are interested to hear more about how you are using TimesFM. Read more in our repository and see the leaderboard. GiFT-Eval: https://lnkd.in/dAwAcKA7 GitHub: https://lnkd.in/dWXH7BAm  Hugging Face: https://lnkd.in/dtx9iMHE To learn more about the foundational model read the Paper: https://lnkd.in/dRw3zzXT

283

9 Comments

New from the Wayfair Tech Blog! My team built and shipped a multimodal GenAI system to validate product dimensions at scale. The problem sounds simple - make sure the sofa dimensions on the website match reality. But, across 30M+ products from 20K+ suppliers, it gets complicated fast. Conflicting sources, inconsistent conventions, and edge cases galore (how do you measure an inflatable palm tree?). We use Gemini to cross-reference text, images, and supplier data. Results: 85%+ precision and 70%+ recall, up from <50% precision and 7% recall with earlier classical ML approaches. ➡️ Full write-up: https://lnkd.in/e7MxEjxW #MachineLearning #GenAI #Ecommerce #DataQuality #MultimodalAI

66

5 Comments

🚀 Just caught the newest a16z Podcast on the state of consumer AI, and the team lays out a crisp framework for where the next breakout products could land: 1️⃣ Information – AI that helps us find, organize, and reason about the world’s knowledge (picture a Google-for-the-AI-era). 2️⃣ Utility – AI that quietly does the work for us: filing, summarizing, scheduling, automating. Think the Box/Dropbox moment, but with copilots instead of clouds. 3️⃣ Creativity & Expression – AI as a personal studio that lets anyone design, compose, code, or direct—fast. From text-to-video to voice cloning, this bucket is exploding. (Mobile ex. Instagram, VSCO, Wattpad, CapCut, etc.) 4️⃣ Connection – AI that forges new social graphs and shared experiences. This white space is still wide-open—there’s no Facebook or Snap of the AI age yet, only early experiments. 🎧 I’ve attached a 30 sec clip from the episode that sums up the taxonomy and a link to the episode below. Highly recommend. 💡 Which bucket do you think is ripest for a category-defining leader in 2025-26? Drop your thoughts below!👇 #AI #Startups #ProductStrategy #a16z #ConsumerTech https://lnkd.in/gAq6Xbag

8

Here's a snippet from our latest Analytics@Meta blog post: "LLMs are revolutionizing the analysis of customer feedback. For example, large online retailers are leveraging LLMs to highlight key aspects of product reviews for their consumers, eliminating the need for manual sifting and summarizing. For Meta, the value of customer feedback lies in refining our products and services to meet the evolving needs of our users. By tuning into their voices, we can ensure we’re delivering products and services that truly matter to them." https://lnkd.in/gDDCPWzH #DataScience #Meta #AI #LLM

39

Predictive ML x Agent = Trust and Adoption Traditional Machine Learning is the workhorse of the modern enterprise. The world is moving faster than ever. Frontline employees make 6x more decisions than 5 years ago - and faster. Forecasting, classification, and prediction give them the gift of time. Many enterprises are great at building accurate ML models. The problem is the last mile. Too often, we display the inference from ML models on a dashboard. That doesn't work. I learned this the hard way years ago from users of our revenue forecasting tool. The model was accurate across 1000s of forecasts with a low MAPE. But users didn’t not care about backtests. They wanted to know whether the forecast reflected: ➡️ A deal that closed yesterday ➡️ Their good feelings about the high win probability of this other deal ➡️ Pandemic macro slowdown Our beautiful, powerful dashboard simply could not give that context. Accuracy was high, but trust was low. So people fell back to their own back-of-the-envelope estimates. The success metric for an ML model is not accuracy. It’s whether the user takes the right action. A lightweight agent wrapped around a traditional ML model can bridge that trust gap and compound the ROI on the models you already have. If people understand the number, they will trust it. If they trust it, they will act on it. Check out my Substack for a practical guide on how to build this, using Customer Success Score as an example: https://lnkd.in/ePydSDAa

14

🧠 While LLMs continue to grab the spotlight, Small Language Models (SLMs) are quietly reshaping how we think about efficiency in AI. I recently explored how SLMs can run entirely on CPUs, unlocking new possibilities for low-cost, privacy-friendly, and scalable inference, ideal for teams experimenting with AI or deploying it at the edge. In my latest piece for Towards AI publication, I cover: 💡 Why CPUs are back in the AI conversation: smarter models, quantization, and optimized runtimes like llama.cpp ⚙️ SLMs vs LLMs: how a hybrid strategy can balance capability and cost 🧩 CPU Inference Tech Stack: GGUF, Ollama, and other tools making CPU inference practical 🧠 Hands-on section: building a translation app using llama.cpp, Streamlit, and AWS EC2, fully CPU-based The best part? You can try it locally - no GPU required. It’s been exciting to see the response so far, the article has already crossed 200+ likes on Medium! 📖 Read my full blog post here: 👉 https://lnkd.in/gby-EBmJ #SLM #llamacpp #cpu #BeginnerFriendly #AWS

7

Given Turing's role as a strategic data partner to all the leading AI labs, lots of folks have reached out following Meta's ~$15B investment in Scale AI Turing's goals and stance remain unchanged: we aim to deliver world-class data, talent, and technology, enabling all our clients to advance their AGI roadmaps And we're fully neutral, remaining "Switzerland" in the space, working to advance all the models, proprietary or open source We have extensive experience across Coding, Reasoning, and STEM, as well as Audio, Video, RL Gyms, and Industry-specific solutions. We're also now investing heavily in our Robotics/EmbodiedAI and 3D capabilities And we're bringing this expertise to not only advancing AI via the leading labs, but also to deploying AI at Fortune 500 companies through Turing Intelligence Another exciting week in AI world.

337

7 Comments