Michael J. Franklin

At Ray Summit 2025, Elizabeth Hu, Goku Mohandas Mohandes and Akshay Malik from Anyscale will announce Lineage Tracking on Anyscale, a new observability tool for end-to-end ML lineage powered by OpenLineage. It maps datasets and models across Workspaces, Jobs, and Services and renders an interactive lineage graph for fast reproduction, audit, and debugging. It also integrates with your existing catalogs and registries like MLflow, Weights & Biases, and Unity Catalog. What you will learn: - How OpenLineage automatically captures datasets, models, code versions, and parameters across development to production on Anyscale - How to navigate the lineage graph to reproduce runs, trace failures upstream, and support governance and audits - How to connect lineage to MLflow, W&B, and Unity Catalog to keep one source of truth for artifacts and metadata Nov 3–5 at the San Francisco Marriott Marquis. Register now: https://lnkd.in/ghAANMa6

53

1 Comment

AI development is advancing rapidly. Compute is cheaper. Models are larger. Infrastructure is improving every month. Yet most AI teams face the same constraint. Not compute. Not GPUs. Not frameworks. Human intelligence. High quality datasets require domain experts who understand nuance, context, and accuracy. Without expert generated and evaluated data, even the most powerful models plateau. The future of AI will not be defined only by algorithms. It will be defined by the quality of the human intelligence behind them. At Adzzat, we help AI teams scale expert driven datasets, evaluation workflows, and model training pipelines. Because better AI begins with better human expertise.

3

“I think we're finally at the cusp of understanding and learning the science, and bringing AI to bear,” said Carnegie Mellon President Farnam Jahanian. “I predict that within the next 10 years, we will see a fundamental shift in how we teach [and] how we learn … for students of all ages, from K-12 [to] higher ed.” Jahanian spoke Monday on a panel Monday at the University of Pittsburgh during Forging the Future: The Intersection of Health, AI & Tech. GFCC | The Global Federation of Competitiveness Councils | Council on Competitiveness

113

3 Comments

Anaxi Labs partners with Carnegie Mellon University on AI economics research Read more on our website: https://lnkd.in/gDmtwRuv Anaxi Labs has teamed up with Carnegie Mellon University to study how value is created and distributed in AI ecosystems. “Kate Shen said AI systems need new economic models to fairly reward contributors.” The research focuses on agent interactions, data valuation, and sustainable monetization models for generative AI. #AI #GenerativeAI #DataEconomy #DigitalTransformation #ITTechPulse #BestPlaceToWork

7

Now on Substack: "Machine-First FAIR: Realigning Academic Data for the AI Research Revolution" - by our VP of Open Research, Mark Hahnel. Why should we prioritize machines over people when sharing data? 🔗 Read Mark's post: https://ow.ly/2hBo50Xu7tb #FAIRdata #OpenData #OpenResearch #AI #ArtificialIntelligence

5

In our latest Data Insights episode, recorded at AI Infra Connect's 2025 summit, Allyson Klein and Jeniece Wnorowski of Solidigm sit down with DANIEL WU, a prominent AI strategist, educator, and author. Daniel shares lessons on building trustworthy frameworks, securing AI systems, and bridging academia with enterprise. 🎧 See the preview directly below. Link to full conversation in comments. #Solidigm #DataInsights #AIInfra

2

1 Comment

Our former intern, Selena, spoke with Carter Abdallah to share her advice for students. ✨ How to take advantage of the NVIDIA DGX Spark? Experiment. With preloaded tools, unified memory, and no GPU queues, the only limit is what you can imagine. 💭 See the demo Selena worked on during her internship here: https://lnkd.in/gc83sG4j #SparkSomethingBig 💫

127

6 Comments

During last month’s NeurIPS 2025 conference, YC’s Diana Hu sat down with Thomas Wolf, co-founder and CSO of Hugging Face, to discuss his unconventional journey from physics and law to building one of the most influential open-source AI platforms. They discussed how open research accelerates innovation, the real challenges of turning AI demos into products, and how great open models and the application layer unlock the biggest opportunities for founders. https://lnkd.in/eDtzAja6

110

12 Comments

A new report from CMIST's Harry Krejsa and CMU Software Engineering Institute's Thomas Şerban von Davier offers a vision forward for AI, arguing that openness should be central in an industry where things cannot and will not stay the same. Read more: https://lnkd.in/erS8hN9U

1

Co-founders Sid Pardeshi and Brian Elliott explore Anthropic's "circuit tracing" - a technique that reveals the specific computational pathways and decision-making processes happening inside AI models, giving us unprecedented visibility into the "why" behind AI responses. Full conversation on YT @Blitzyai #circuittracing #anthropic

7

1 Comment

In early 2026 DataSeer will join Proofig AI as a partner on their PubShield program. The collaboration will enable researchers at dozens of leading institutions to automatically check their manuscripts for compliance with institutional, funder, or publisher Open Science policies. https://lnkd.in/gJkb5zkq #openscience #opendata #researchintegrity #openresearch #scholcomm #scholcomms

12

Presented at the ML+X community meeting at UW-Madison today on deploying a retrieval-augmented generation (RAG) system on AWS Bedrock and evaluating LLM performance at scale. Our team took KohakuRAG, the #1 solution from the 2025 WSDM WattBot Challenge, and deployed it as a fully serverless pipeline on AWS Bedrock. I built the Bedrock integration, evaluation framework, and cost tracking infrastructure, then benchmarked 9 LLMs and 3 ensemble strategies across 282 questions to answer a practical question: which model should you actually deploy in production? A few results that stood out... - Ensemble majority voting (0.840) outperformed every individual model. - Llama 4 Maverick delivered 98% of the top score at a fraction of the cost and latency. - Model behavior matters as much as raw accuracy; our highest-citation model finished last overall due to aggressive refusal behavior. - Text-only embeddings create a ceiling on figure-based questions that no LLM can overcome. The full recording will be available on ML+X Nexus: (https://lnkd.in/gSTe_EdW) soon. I would definitely recommend watching the whole thing if you are interested in deploying or learning more about RAG applications! Update: Now live at https://lnkd.in/gv7_jHkv Thank you to Christopher Endemann for the mentorship and for bringing me onto this project; I learned more about applied ML infrastructure than I have in any course in just a few months. Thank you to Blaise Manga Enuh, Ph.D. for the collaboration on the local deployment pipeline.

21

4 Comments

LLMs are impressive but they are costly, slow, and cloud-dependent. That’s why more teams are choosing Small Language Models (SLMs). SLMs: 📉 Require lower compute 🔒 Have stronger privacy 🚀 Have faster inference 🎯 Have better accuracy for specific domains In this carousel, we break down: • What small language models really are • Why they outperform LLMs in many real-world systems • How fine-tuning turns them into domain experts • Where they already deliver impact across industries If you care about practical, production-ready AI, read the full guide on implementing SLMs here 👉 https://shorturl.at/xgrpt #SmallLanguageModels #ArtificialIntelligence #MachineLearning #NLP #AIEngineering #EnterpriseAI #EdgeAI #OnDeviceAI #ResponsibleAI #HumanCenteredAI #AIForGood

12

2 Comments

What will define the next frontier of artificial intelligence? At the AIX Summit, leaders like Yann LeCun, Dr. Sanjeev Arora, Jiayuan Mao, and Saining Xie will explore how AI systems are evolving and what that means for research, industry, and society. Moderated by Jia Deng, this session brings together foundational thinkers and builders working at the edge of what AI can do next. This is where the future direction of AI takes shape. Join us on April 17: https://lnkd.in/e93CdUeN #ArtificialIntelligence #MachineLearning #Innovation #AIXSummit 

6

💾 Are embeddings too big to store? Here’s why recomputing them might be smarter. In Episode 28 of LLMs on the Run, UC Berkeley Professor Joseph Gonzalez (Joey) explains why storing embeddings at scale is more expensive than you think — and how recomputation can flip the tradeoff. Joey breaks down: ✅ Why embedding vectors often exceed the size of the original data ✅ How index structures add even more storage overhead ✅ Why cheaper embedding models make recomputation viable — even preferable The takeaway? In RAG pipelines, storage isn’t the bottleneck anymore — compute is cheap, and dynamic is back. Brought to you by RunLLM: https://lnkd.in/gA8Ckcnc #LLMsOnTheRun #RAG #AIInfrastructure #Embeddings #Retrieval #ContextualSearch #RunLLM #MachineLearning #JoeyGonzalez #PostTraining

9

LLMs are powerful, but getting consistent accuracy remains a major challenge. Lamini CEO Sharon Zhou, PhD shares how to build compound AI systems using SLMs and high-accuracy mini-agents within agentic workflows. She introduces techniques like memory RAG, which reduces hallucinations using embed-time compute for contextual embeddings, and memory tuning, which applies a Mixture of Memory Experts (MoME) to specialize models with proprietary data. The takeaway: with the right building blocks, high-accuracy mini-agents can be composed into larger, more reliable AI systems: https://lnkd.in/gMyXFs45

162

4 Comments

Every few years, the definition of technical leverage shifts (think #opensource, #APIs, #VMs). October announcements from OpenAI, Anthropic, Google Cloud mark another one of these inflection points. Claude’s new Agent SDK, Gemini’s Computer Use mode, and OpenAI’s custom chip and infrastructure partnerships give the next generation of startups leverage to scale with smaller teams, faster time-to-market, and more capital efficiency from the get-go. The new techops stack means companies can scale output with less headcount. We see this in our recent AI-first investments: Apricot, Lyric, Tavily, Promptfoo, Delve

32

2 Comments

Today, we are announcing Traverse. (https://traverse.so) We are building reinforcement learning environments for long-horizon agent journeys, our goal is to automate all economically valuable work. AI systems have made rapid progress on tasks that are easy to describe and easy to evaluate, yet they continue to struggle with the work that defines most professions. We believe this gap exists because agents are trained in environments that optimize for convenience rather than realism, and because the industry has largely converged on data collection methods that fail to capture how work is actually done. The bottleneck to agent capability is no longer architecture or compute but data quality and environment fidelity. Much of today's training data is contrived and produced by people who do not deeply understand the domains being modeled, which optimizes for volume and surface correctness but strips out the reasoning and judgment and context that define real expertise. High-quality training signal comes from observation, not fabrication. The most valuable data is generated when experts perform real work in real environments, where decisions are shaped by experience and tradeoffs and failure. We are starting with software engineering as it is the most verifiable and deterministic domain that exists today. Tests pass or fail and builds succeed or break. Software also creates a natural flywheel where agents that write code can eventually improve the infrastructure used to train them, which is a feedback loop we want to establish as early as possible. Our approach is to assemble small teams of world-class domain experts and world-class ML engineers, and to hold that bar fixed as we expand into law, medicine, finance, and beyond. The path to general intelligence is to traverse through every domain where work happens. https://traverse.so

46

6 Comments

Tomorrow at 1pm EDT our CEO is interviewing Jonathan Stray on Taout.tv, Senior Scientist at UC Berkeley’s Center for Human-Compatible AI (CHAI), about the algorithms that keep cranking up conflict online. He explores the recommender systems, polarization, and whether digital platforms can be built to reward better thinking instead of constant outrage. Less rage bait, more real insight, and a smarter look at the machine behind the madness. You're going to want to watch this! Link will be in the comments. 👇🏻