Pascal Biese

One to Watch: Black Forest Labs - 𝐄𝐮𝐫𝐨𝐩𝐞'𝐬 𝐋𝐞𝐚𝐝𝐢𝐧𝐠 𝐕𝐢𝐬𝐮𝐚𝐥 𝐀𝐈 𝐒𝐭𝐚𝐫𝐭𝐮𝐩 Black Forest Labs is pushing the boundaries of generative AI with their Flux model suite-delivering top-tier image generation, in-context editing, and industry-leading benchmark results. Based between Freiburg and Delaware, they’re the minds behind Stable Diffusion and VQGAN, now building some of the strongest text-to-image tools on the market. 🚀 Big Moves: 🖼️ Launched Flux.1 and Flux Pro-outperforming Midjourney v6 and DALL·E 3 📲 Integrated into xAI’s Grok and Mistral’s Le Chat 💸 Raised $31M seed with backing from a16z, General Catalyst, and more ✨Latest Release:  Flux.1 Kontext [dev] Released June 26, 2025, this open-weight 12B model offers top-tier image editing. You can perform iterative edits, preserve characters, and run locally with Hugging Face Diffusers, ComfyUI, TensorRT. In a noisy AI landscape, Andreas Blattmann and Black Forest Labs are quietly leading from the front. #generativeai #visualai #startups

42

New on the Anthropic Engineering blog: writing effective tools for LLM agents. AI agents are only as powerful as the tools we give them. So how do we make those tools maximally effective? In this post, we share methods for improving the performance of tools, including how to: -Build and test prototypes of your tools; -Create and run comprehensive tool evaluations; -Use agents like Claude Code to improve the performance of your tools. Read the post: https://lnkd.in/dWm2VVuY

2.509

118 Kommentare

A new study by researchers Markus Mueller, Kathrin Gruber, and Dennis Fok examines the use of AI for generating tabular data, such as spreadsheets or databases. Their research focuses on diffusion-probabilistic models—the same technology behind popular AI image generators like Stable Diffusion and DALL-E. Instead of creating images, however, these models generate entirely new data points that reflect the patterns of existing datasets. The findings will be presented at the prestigious International Conference on Learning Representations (ICLR) in April, 2025, and published in the proceedings of the conference. Interested? Read more: https://lnkd.in/e6WTM7gu #artificialintelligence #diffusionprobabilisticmodels #science

101

Very nice - #Apertus, the Swiss open #AI model, does very well on this massive, participatory commonsense reasoning benchmark for over 100 languages, constructed by hand by 335 researchers from 65 countries around the world. Note the color scheme: open-weight in purple vs. fully open in pink, including open data 😍 Not bad for a V1! Paper: https://lnkd.in/e9U7djC4

389

4 Kommentare

A very interesting research by Johannes Wachs and his colleagues on the second order impacts of ChatGPT on the society. The posts on stack-overflow have decreased, thus the data that was in public domain and were used by openai to train their models is shrinking while the data that is in private hands and is used for model training is presumably increasing. Interesting finding on what are the programming languages that were the most affected by chatgpt - typescript, python and R. https://lnkd.in/dxEtkyyZ

33

1 Kommentar

💭 “What if we could not only see language drift, but actually predict it?” After countless late nights, that question led us to a patent, a paper, and a milestone I’ll never forget. 🌍 Our work “Koopman Spectral Dynamics for Interpretable and Predictive Semantic Drift in Language Models” has just been accepted at IEEE ICDM 2025 (A* tier). This isn’t just a paper, it comes alongside our first published patent: the world’s first application of Koopman operator theory in NLP. No better venue than IEEE ICDM to share this. Its value will be proven with time. 💡 The problem we tackled: Language models are powerful, but words don’t stay still, their meanings drift. Models capture this drift, but we rarely understand or predict why and how it happens. 🔑 Our solution: We used Koopman operators (a tool from dynamical systems) to model semantic drift as if meanings were “trajectories” evolving in time. This lets us both forecast drift and interpret the hidden forces shaping it. 👦 Satyam’s Explanation: Imagine words are little boats floating on a big river of meaning. Over time, the current shifts: “cloud” once meant sky fluff, then computers. Our method builds a “map of the river” so we can see where each boat is heading, and why it drifts that way. 🙏 Gratitude: Huge thanks to my co-authors Vivian Phung, Vũ Phúc, SHIVAM MISHRA, Sundaram Mishra, and Prof. Dr. Vishwanath Bijalwan. This journey was long, but your teamwork and creativity made it possible. Excited to present this at ICDM 2025 and see how the community builds on it. ✨ This one’s for every sleepless night that finally paid off. #ICDM2025 #AI #MachineLearning #KoopmanOperators #NLP #satmis

23

4 Kommentare

What happens when hardcore theoreticians, practitioners and industry leaders all tackle the same database challenges from completely different angles at ACM SIGMOD PODS 2025? As RelationalAI's Leonid Libkin and Wim Martens described in their real-time conversation with Alec Coughlin from the floor of the conference in Berlin: "Different takes, but on the same essential problem, right? And this is quite exciting from a purely scientific perspective." An incredibly thought provoking environment where "you meet somebody random here, you can just grab them, talk to them, and they can come from the total opposite spectrum of the one where you're working from, and you're going to discover something interesting." The conference buzz around good language design - from addressing SQL's complexity to the value of intuitive solutions like REL - shows how thoughtful design principles are reshaping the future of data systems. With 15+ accepted papers from RelationalAI and invited talks including Leonid's GQL keynote and Wim's Gems of PODS presentation, this research is actively shaping enterprise AI systems within Snowflake today. Watch the quick fireside chat below to learn more about the experience of Leonid and Wim so far at this year's conference and their upcoming presentation.

28

1 Kommentar

About how deep tech makes its way into industry; while spending time around the research ecosystem in Germany, I noticed something interesting. I saw moments where research, applied institutes like Deutsches Forschungszentrum für Künstliche Intelligenz (DFKI), and industry were at least in the same room. The handoff wasn’t seamless or guaranteed, but there was an observable effort to bridge research and real-world use. I felt the same dynamic during SSDA ’23 at Université de Fribourg - Universität Freiburg, where researchers and industry practitioners were actively exchanging constraints. Not everyone gets access to that kind of proximity. And even when they do, adoption is still hard especially in slow-moving industries. That’s where many deep-tech startups struggle today. What the deep-tech startups that collaborate with us during the build & production phase tend to get right is this: they don’t wait for industry validation. They manufacture it by - running short pilots. - bench-marking against existing industrial systems. - measuring things industry actually cares about: latency, power, reliability, integration. Those comparisons change the conversation. That’s often the difference between pure research and real adoption. If you’re building in robotics, neuromorphic systems, or embedded AI, what’s the hardest thing for you to prove before industry will even engage?

56

1 Kommentar

🚨🎉 New paper accepted at ICLR 2026! 🎉🚨 🕵️ Can you spot the AI peer reviewer? ❌ Spoiler: probably not. 🤖 Everyone worries about LLMs writing papers and we’ve already seen NeurIPS 2025 papers slip through with hallucinated citations and AI generated content. 🔍 We asked the next uncomfortable question: what if LLMs are writing the peer reviews too? 📊 Built the largest benchmark to date 📚 ~789K AI + human peer reviews 🏛️ 8 years of ICLR & NeurIPS 🧪 Tested 18 detection methods + a new context-aware approach (Anchor) ⚠️ Result: even SOTA detectors fail at the individual review level 🔥 AI writing is improving 😴 Negligent reviewers are getting bolder 🧱 The integrity of peer review is at risk 📦 Dataset + benchmark will be publicly released 👀 Paper dropping soon Work done with Sungduk Yu, Man Luo, Vasudev Lal, Phillip Howard #ICLR #NeurIPS #AI #PeerReview #AcademicIntegrity #LLMs

303

8 Kommentare

von Mises stress - we see it in every analysis tool we open. But what is it…exactly, and why is it so useful? We just published our latest deep-dive tutorial on EngineeringSkills.com. This is one for coders AND non-coders and is available to all… Understanding Tresca and von Mises Elastic Failure Theories - linked below 👇 If you’re one of the many engineers with only a vague understanding of what the terms Tresca and von Mises refer to or what a failure theory actually is, this tutorial was written for you! In it, we’ll go from the fundamentals of elastic failure theory to implementing section analysis in Python using sectionproperties - the fantastic open-source section analysis library. Guest contributor, Julian Haudek, does an excellent job of presenting the core concepts of Tresca and von Mises failure theories before demonstrating their application with some manual calculations. Then, Julian supercharges proceedings by introducing the sectionproperties Python library to speed up the analysis process dramatically. The full tutorial, complete with downloadable code, is now available in the EngineeringSkills tutorial feed. #structuralengineering #civilengineering #engineeringskills #python #sectionproperties #stressanalysis #vonmises #fea

91

3 Kommentare

Bridging the Gap: Why a Hybrid Approach to Code Generation is the Future I had a fantastic time presenting our Natural AI Code Assistant at the recent Software AG IUG conference and engaging with so many bright minds! A recurring theme in my conversations with customers was a healthy skepticism surrounding the readiness of today's LLMs for generating enterprise-grade solutions. It’s a fair concern. While generative AI, low-code, and no-code platforms all promise rapid code creation, each comes with its strengths and limitations. On one hand, low-code/no-code tools empower a wider audience to build applications through intuitive visual interfaces and pre-built templates. This is achieved by abstracting the underlying code, which is a massive leap in efficiency for many use cases. On the other hand, generative AI for code generation operates without the guardrails of templates and component libraries. It takes a developer's natural language prompts and generates code snippets from the ground up. While the accuracy of these models is constantly improving, the generated code can still have flaws and requires rigorous human oversight, editing, and refinement to be production-ready. So, what’s the solution? ✅ A Hybrid Approach: Generative AI + Low-Code Principles A hybrid approach that merges these worlds for precise code generation across common and domain-specific programming languages. In this context, LLMs serve as LLMs as smart orchestrators, while the templates encode standards, patterns, and compliance rules. LLMs interpret requirements (from user input or specs). Then, they select, compose, and fill in code templates based on context. Moreover, they help glue templates together, resolve conflicts, and generate custom code when templates fall short. The future of enterprise development isn't just about writing code faster; it's about writing the right code, faster and more reliably. #GenAI #LLMs #LowCode #NoCode #FutureOfCoding #SoftwareDevelopment #EnterpriseSoftware #AI #Innovation #SoftwareAG #IUG #CodeGeneration

44

1 Kommentar

The Secret to Effortless Data Extraction from Documents (No Code Headaches!) Ever felt like pulling structured data from documents is like herding cats? PDFs, contracts, reports—you name it. Extracting insights manually is slow. Using LLMs? Often more complex than the problem itself. Until now. ⚡ Meet ContextGem – The Open-Source Shortcut to LLM-Powered Extraction No more wrestling with boilerplate code or API quirks. ContextGem cuts setup time by 3-5X with smart abstractions that just work. ✅ Works with ANY LLM – OpenAI, Anthropic, or local models ✅ Minimal code – Focus on what you need, not how to get it ✅ Open-source & free – No paywalls, no nonsense 👉 Try it yourself! GitHub: https://lnkd.in/dZg2MbAj Upvote us on Product hunt : https://lnkd.in/dJBEzEfV 🔥 Pro Tip: Pair it with semi-structured docs (invoices, forms, etc.) for max time savings. 🤔 Question for you:** What’s the most frustrating part of extracting data with LLMs right now?** (Let’s brainstorm in the comments!) P.S. If you’ve ever wasted hours tweaking prompts or parsing JSON outputs… you’ll love this. [Subtle note] This gem of a tool is built by Sergii Shcherbak – give the repo a star ⭐ if you dig open-source that actually saves time!

270

10 Kommentare

> Sharing Resource < Very interesting: "Double descent: When do neural quantum states generalize?" by M. Schuyler Moss, Alev Orfi, Christopher Roth, Anirvan M. Sengupta, Antoine Georges, Dries Sels, Anna Dawid, and Agnes Valenti Abstract: Neural quantum states (NQS) provide flexible wavefunction parameterizations for numerical studies of quantum many-body physics. While inspired by deep learning, it remains unclear to what extent NQS share characteristics with neural networks used for standard machine learning tasks. We demonstrate that NQS exhibit the double descent phenomenon, a key feature of modern deep learning, where generalization worsens as network size increases before improving again in an overparameterized regime. Notably, we find the second descent to occur only for network sizes much larger than the Hilbert space dimension, indicating that NQS typically operate in an underparameterized regime, where increasing network size can degrade generalization. Our analysis reveals that the optimal network size in this regime depends on the number of unique training samples, highlighting the importance of sampling strategies. These findings suggest the need for symmetry-aware, physics-informed architecture design, rather than directly adopting machine learning heuristics. Link: https://lnkd.in/eqhKbZ-s #quantumcomputing #doubledescent #quantummachinelearning #research #paper

32

💥 Community Member of the Month: 💻🎓 Eldar Rakhmatullaev, founder and CEO of Qemie, an AI automation agency and workshop organizer. Eldar is also the driving force behind Data Science Group Frankfurt, Hessen’s largest community for data scientists and AI enthusiasts. At Qemie, Eldar and his team are currently rolling out a new series of workshops on AI and automation, designed to help businesses integrate #artificialintelligence into their daily processes. In parallel, they’re also preparing a video series that reviews popular AI tools and explores how powerful they can be for organizations of all sizes. 👓 Read the blog post: https://lnkd.in/dTn6Zctu

64

3 Kommentare