Rob J. Wang

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

282

9 Comments

A quick update on #OSI (Open Semantic Interchange), a spec focused on making semantic definitions portable across data, AI and BI tools: we’ve published the first public Apache 2.0 version of the spec, an important step toward more open, shared semantics! We’ve also added a number of new partners to the OSI ecosystem: AtScale, Coalesce.io, Collate, Credible, Databricks, JetBrains, Lightdash, and Qlik. It's great to see continued momentum and collaboration. In our latest working group session we identified several key working groups that will focus on the continued evolution of the specification, including things like: Metrics Expression Language, Composability, Catalog Integration, Ontology Representation, Model Converters, and Developer Tools More context in the blog below, and a link to the repo is in the comments 👇 https://lnkd.in/gEaQXSME

458

22 Comments

If you are getting started in AI engineering or trying to level up, eval engineering is a concept you need to understand early. Traditional AI evals break down because most teams rely on LLM-as-a-judge. Accuracy plateaus around 60–70%, costs scale fast, and teams end up evaluating only a small slice of production traffic. That means real failures slip through. Eval engineering treats evaluation as a system, not a metric. ✦ Start with LLM-as-a-judge to establish a baseline ✦ Involve subject matter experts to surface real failure modes ✦ Refine prompts and criteria using production data ✦ Train a small, fast judge model ✦ Deploy it for low-cost, low-latency evals across 100% of traffic ✦ Continuously feed failures back into the loop The goal is simple: move from sampled guesses to full observability, and from “it looks fine” to “we trust this in production.” Image inspiration: Galileo ✦✦✦ ♻️ Share this with your network to help them learn about latest data/AI concepts Follow me (Aishwarya Srinivasan) for more data science/ML/AI insights and subscribe to my Substack for more educational resources: https://lnkd.in/dpBNr6Jg

195

18 Comments

Anthropic just announced they’re launching Claude-powered AI tools for finance, using data from providers like PitchBook and S&P Global to automate research and market analysis. This is a glimpse into where AI is really headed. We’re moving beyond “general intelligence” into vertical intelligence - AI that doesn’t just know everything in a shallow sense, but knows your industry deeply. Finance is one of the first big use cases where the ROI is obvious. Healthcare, law, logistics, retail - any industry with structured data and repeatable decisions is next. Anthropic's CEO Dario Amodei even said this shift could eliminate half of all entry-level white-collar jobs. When I look at how the GenAI landscape is evolving, this feels like a second wave. The first wave was: “Let’s build general-purpose LLMs and see what they can do.” The second wave is: “Let’s plug those models into real industry data, workflows, and decision-making pipelines, and see what they can replace or reimagine.” This is exactly what I’m seeing from startups in places like Y Combinator too - Vertical AI is the next frontier. And if you’re building in this space, you’ll want to start thinking this way. For anyone entering AI right now, here’s how I’d approach it: 💡 Stop thinking in terms of standalone tools. Think in terms of workflows. How can AI handle 80% of a task pipeline - research, synthesis, drafts, action steps - while the human stays in the loop for the final 20%? 💡 Pick a domain. Get obsessed with it. If you’re in finance, healthcare, law - learn the actual workflows people use. The real bottlenecks. That’s where the best AI applications will come from. 💡 Learn how to chain things together. RAG pipelines, APIs, function calling, multi-agent systems - this is the architecture of AI-powered workflows. It’s more than just prompt engineering now. 💡 Focus on augmentation, not just replacement. Yes, some jobs will get automated. But the real winners will be the people who learn to work with AI and design better ways of doing things. This isn’t just about building smarter models. It’s about building AI that knows your business better than your new analyst does. That’s where things are going, And it’s happening faster than most people realize. Happy learning ❤️

531

57 Comments

A lot of times we get asked about knowledge graphs at Ragie but my hot take is that the overwhelming majority of customer use cases don't require them. It's also no surprise to me that knowledge graph hype is evangelized by graph database vendors wherein vendors often compare KG search against pure semantic search but I think it's a bad comparison. They should be comparing against Agentic Retrieval which I think is more likely to be the future of retrieval in the long term. We recently launched Agentic Retrieval at Ragie and Matt Kauffman did a really nice write up explaining how it works. Check it out: https://lnkd.in/gMeyqD7X

31

1 Comment

Just-in-Time UI: At Aligned Labs we frequently share reports about the data collection, fine-tuning and eval work that we do with clients.  They often need different views or refined analysis, which is why we built Whiz-bang Boom! Our new tool takes a CSV or JSON file and generates a shareable, interactive report. You can easily modify the report or focus the analysis with a simple prompt, and it creates a new revision. You can organize reports into spaces and customize those spaces with a simple prompt. Try it out. https://wizbangboom.com #businessIntelligence #AI

25

2 Comments

Whether you believe Databricks is worth $100B or not, and whether you like what it offers or not, here’s the reality: Databricks has pushed all the major players to deliver more value out of the box. The days of stitching together a thousand different cloud tools are fading, except for when you need a very specialized tool. Databricks’ strategy of bundling many affordable tools in one platform has been a major driver of this shift. Again, whether you like them or not, every other platform is now fighting for market share by trying to look more like Databricks. This is good for customers.

380

27 Comments

A major investment bank who just deployed Pulse showed us their previous document ingestion pipeline. They'd spent years building automated document processing that promised leadership fully automated analysis with beautiful dashboards fed by real-time data extraction. Then they showed us what their system could actually handle. It worked fine on simple tables - basic financial statements with clean formatting. But anything complex broke immediately. Multi-page tables lost their structure. Charts and graphs were completely ignored. Documents with nested data hierarchies got scrambled. Their data engineering team was spending most of their time manually reconstructing what the "automated" system couldn't parse properly. The pipeline was technically automated for maybe 35-40% of their document types. Everything else required manual intervention to fix what the AI got wrong. We’re at an inflection point with document processing. The competitive advantages are going to be massive.

38

3 Comments

The classic forecasting dilemma: Do you want an interpretable model (like ARIMA) that's hard to scale, or a complex "black box" deep learning model that is? A new paper from Google, ARIMA_PLUS, suggests we can now have the best of both worlds, while leveraging it all inside BigQuery. The paper introduces a system for large scale, in-database time series forecasting and anomaly detection. It's designed to be automatic, accurate, and, most importantly, interpretable. The performance claims are massive: -Accuracy: On the standard Monash benchmarks, ARIMA_PLUS outperformed popular DL models like N-BEATS, N-HiTS, and PatchTST. -Scalability: It forecasted 100 million time series in just 1.5 hours with a single SQL query. -Interpretability: This is the best part. It’s not a black box. ARIMA_PLUS is a structural model that automatically decomposes every forecast into its core components. This means you can finally see why your forecast looks the way it does. The model can explicitly show you the impact of: -The long term trend -Multiple Seasonalities (like daily, weekly, or yearly cycles) -Holidays and custom Events -External covariates (like ad spend or price changes) This is a game changer for any data team running forecasting on BigQuery. It makes powerful, understandable results accessible to anyone who can write SQL. Paper: https://lnkd.in/gju5p38K

134

16 Comments

We are launching Pulse’s Formula Recognition Submodule. This builds on Pulse’s state-of-the-art hybrid extraction model pipeline, already deployed in leading financial institutions, research labs, and Fortune 50 enterprises. Formulas have always broken traditional OCR. Variables like α, β, fractions, and matrices require spatial reasoning, not just text recognition. Our dedicated mini-model, trained on 10M+ formula/LaTeX pairs from academic papers, handwritten notes, and scientific documents, treats formulas as structured mathematical objects. The result is a precise LaTeX transcription that preserves meaning at production scale. Pulse processes tens of millions of pages every week. Now, the same pipeline can handle complex mathematical content for AI teams, academic institutions, and enterprises that rely on accuracy in scientific data extraction. DM for a demo or early access to formula recognition. Full technical blog below.

278

27 Comments

Among other things, my team has been focusing on AI research for agents. We just blogged about a recent project: the Test-Time Diffusion Deep Researcher (TTD-DR). This framework uses a Deep Research agent to draft and revise its own drafts using high-quality retrieved information. This approach achieves new state-of-the-art results in writing long-form research reports and completing complex reasoning tasks. The work achieves better results than any other deep research agent we compared to. It also shows that self-evaluation and self-improvement are must-have components of a successful agent. https://lnkd.in/gBZpVMQy

297

8 Comments

Just getting started with LLMs? Don’t let the jargon slow you down. This 5-part LLM Glossary is your shortcut to mastering the fundamentals👇 1️⃣ Model Types • Foundation Model: Pretrained on large-scale, diverse datasets to learn general capabilities. • Instruction-Tuned Model: Further trained to follow user instructions precisely. • Multi-modal Model: Handles multiple input/output formats like text, images, and audio. • Reasoning Model: Optimized for logic, problem-solving, and step-by-step thinking. • Small Language Model (SLM): Lightweight models for fast, efficient, on-device tasks. 2️⃣ Training & Fine-Tuning • Pretraining: Initial learning phase using massive, diverse data sources. • RLHF (Reinforcement Learning with Human Feedback): Aligns model behavior with human preferences. • DPO (Direct Preference Optimization): Uses ranked preferences instead of rewards for training. • Synthetic Data: AI-generated data used to supplement real datasets. • Fine-Tuning: Targeted retraining on specific domains or tasks. • LoRA / QLoRA: Techniques for efficient low-resource fine-tuning. • Guardrails: Rules or filters applied to enforce safety, ethics, and compliance. 3️⃣ Prompt Engineering • System/User Prompts: Define model behavior and user context in a session. • Chain-of-Thought (CoT): Prompts that guide the model to reason step-by-step. • Few-Shot / Zero-Shot Learning: Demonstrating tasks with few or no examples. • Prompt Tuning: Optimizing prompts via training for specific outcomes. • Context Window: The total number of tokens the model can "remember" per interaction. 4️⃣ Inference & Generation • Temperature: Adjusts output randomness; lower values are more deterministic. • Max Tokens: Limits the number of tokens in a generated response. • Seed: Controls reproducibility of outputs. • Latency: Time taken by the model to respond. • Hallucination: When the model generates factually incorrect but plausible-sounding information. 5️⃣ Retrieval-Augmented Generation (RAG) • Retrieval: Fetching relevant data from external sources before generation. • Semantic Search: Search based on meaning, not just keywords. • Embeddings: Vector representations of text used for similarity matching. • Chunks: Dividing documents into smaller segments for better retrieval. • Vector Databases (VectorDBs): Stores embeddings for fast and accurate search. • Reranking: Rescoring retrieved documents to prioritize relevance. • Indexing: Structuring data for efficient retrieval during generation. 🌟 Follow the AIKaDoctor (Free AI & Data Science Resources) channel on WhatsApp: https://lnkd.in/dCTCEKKc 📌Follow Habib Shaikh For more such content.

232

36 Comments

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

432

17 Comments

I met one AI development agency founder, and they said our 50% RAG apps failed in production last year. Because teams don’t know how to handle: chunking + embeddings + vector DB + reranking + hallucinations + evaluation + monitoring. So I'm sharing a RAG Playbook that shows exactly how to build production-ready RAG systems, with practical examples. ✅ What you’ll learn from this playbook This playbook helps you go from: “RAG demo is working” → “RAG system is reliable in production” You’ll learn how to: - reduce hallucinations using prompting techniques - use advanced chunking methods (not random splits) - select the right embedding model for your use case - choose the best vector database (based on real needs) - add reranking to improve relevance - follow clear steps to build an enterprise-grade RAG system - run evaluation before production using key scenarios - monitor and optimize performance after deployment - track RAG quality using 4 powerful metrics If you’re building a RAG-based feature for your product, this playbook will save you weeks of trial and error. ✅ Save this post (you’ll need it later) 🔁 Repost for your team/network ➕ Follow Shubham Vora to learn more on AI agents.

139

77 Comments