Siegfried Handschuh (Q64851946)

The Swiss AI initiative (ETH Zürich, EPFL, CSCS) just released something remarkable in AI, and we benchmarked it 🇨🇭 While everyone's focused on the latest ChatGPT updates, Switzerland quietly dropped Apertus (lat: open). lnkd.in/dta-Ciu5 The first truly open model at the 70B parameter scale. Not just "open weights", like Llama, but actually open. Training data, scripts, the whole recipe. Why this matters: - They respected robots.txt retroactively - Filtered out copyrighted & toxic content - Built in safeguards against memorization - Supports 1,811 languages, including Swiss German (Schwizerdütsch) and Romansh! Our benchmark results show where Apertus stands and how it performs in our lab: Let's be honest: Apertus is roughly at GPT-3.5 level, capable but not cutting-edge. Apertus 8B It can follow instructions reasonably well (44% IFEval), handle basic math better than some competitors (5.29% vs Mistral's 2.95%), and tackle general knowledge questions (31% MMLU-Pro). But it struggles with complex multi-step reasoning (36% MuSR) and spatial tasks (only 24%). For context: you wouldn't use this for production applications yet. It's more like a very capable research prototype: good enough to build on, not good enough to deploy at scale. The reality? Open models are about 1-2 years behind the frontier. Apertus won't write your code like Claude or reason like o1. But that gap is closing, and now we have a fully transparent foundation to build on. This isn't about competing with GPT-5 today. It's about ensuring that in 5 years, AI isn't controlled by three companies. Full benchmark details & analysis on our blog: blog.nlp-lab.ai/2025/09/05/Ape… Huge gratitude to our NLP lab team members Götz-Henrik Wiegand @GH_Wiegand and Michael Gaus who dedicated extra hours to rigorously benchmark this model and ensure reproducible results. Their commitment to open science makes work like this possible. 🙏 #AI #OpenSource #ResponsibleAI #SwissInnovation #OpenScience #AIBenchmarking #LLM #LargeLanguageModels #NLP #NaturalLanguageProcessing #GenerativeAI #AIResearch #DataScience #HSG #ETHZurich #EPFL LLM Benchmark Evaluation - Apertus-8B

Chris Manning says Yann LeCun sees language as a low bandwidth communication channel compared to vision. But the gap between a chimp and a human wasn’t produced by superior eyes. What took off for humans was language. Not just for communication, but as a cognitive tool.

Our chair is currently exploring different devices for brain activity measurement and I got to test out voluntary. So interesting to see how that reacts to different things I do. Lets see what happens if we watch #brainrot 🤔 #phdlife #phd #research #brain

Got the #BestPaperAward yesterday at #KDIR #IC3K conference in #Marbella for the paper: "A Convexity-Dependent Two-Phase Training Algorithm for Deep Neural Networks" Huge thanks to the team! There will be a #arXiv version soon! Stay tuned! #paper #HSG #LLM #Transformers #ML

I thought my command line days were behind me. DOS in the 90s (yes, I’m that old), Linux in the 2000s, then Mac GUIs took over. The paradox of #genAI and #LLM: it gave me natural language #NLP, but pushed me back to #CLI. Because once the machine generates the commands, the CLI becomes the fastest interface. Feels like rediscovering an old love. #AI #CommandLine #Productivity #Coding #Automation #DevTools #GenAI #LLM #NLP #AI #ArtificialIntelligence #MachineLearning #CLI #CommandLine #Coding #DevTools #Automation #Terminal #Unix #Linux

Mental disorders are not brain disorders in isolation. Poor body health is a more pronounced manifestation of mental illness than poor brain health.

Chocolate intake is associated with the number of Nobel Prize recipients.

So @Apple just gave us the Babel fish! AirPod Pro’s 3 live translation finally makes Douglas Adams’ dream come true. Don’t panic: we can now understand everyone in the galaxy! 🐟✨ #AirPod #Apple #AI #DontPanic

Und der KI-Gewinner heisst . . . Google! Im Gespräch mit der @NZZ durfte ich zusammen mit Thilo Stadelmann @thilo_on_data unsere Einschätzung zur aktuellen KI-Landschaft teilen. Unser Fazit: Obwohl OpenAI und ChatGPT im Mittelpunkt der öffentlichen Wahrnehmung stehen, sieht es so aus, als ob Google -- noch vor zwei Jahren der vermeintliche KI-Verlierer -- der eigentliche Marktsieger wird. Den vollständigen Artikel mit allen Details gibt es in der NZZ am Sonntag (vom 7. September 2025) lnkd.in/d99hsb2v #KI #AI #Google #ChatGPT #GenerativeAI #Innovation #DataScience #LLM #LargeLanguageModel #NLP #ZHAW #ICS #DSNLP #HSG #NZZ #NZZamSonntag

I enjoy AI meetups, but they sometimes leave me uneasy. Because LLMs are non-deterministic, debates often leap to visions of non-deterministic killer robots. That makes for great drama, but not for realistic engineering. In practice: hybrids, guardrails, controls. Less theater, more engineering.

@keypousttchi @PKoss82 Ja, mein Nachname kann vielfältige Assoziationen auslösen. Manchmal freue ich mich auch darüber, bspw. wenn es um den GloVe Algorithmus aus Stanford geht. arxiv.org/abs/2507.18103

@PKoss82 Ich glaube, das fände der Kollege @sha_hsg nicht so witzig 😅

Das ficht Herrn Merz aber nicht an. Er hält es wie Frau Merkel, hat sich seit Amtsantritt zu keiner der Taten geäußert und redet alles schön. Und dann wundern sie sich bei der CDU, wenn ihnen keiner mehr glaubt, daß sie das Problem lösen wollen. Und die Leute alle AfD wählen.

@EPFL , @ETH_en and #CSCS today released Apertus, Switzerland's first large-scale, multilingual language model (LLM). As a fully open LLM, it serves as a building block for developers and organizations to create their own applications: cscs.ch/science/comput… #Apertus #AI

Ab heute ist o3 öffentlich verfügbar. Jetzt müssen sich Programmierer besonders anstrengen.

Europa ist raus, China hält mit

OpenAI o1 is now out of preview in ChatGPT. What’s changed since the preview? A faster, more powerful reasoning model that’s better at coding, math & writing. o1 now also supports image uploads, allowing it to apply reasoning to visuals for more detailed & useful responses.

Picture of the Year 2023 on Wikimedia Commons commons.wikimedia.org/wiki/File:Ince… All finalists: commons.wikimedia.org/wiki/Commons:P… Previous winners: commons.wikimedia.org/wiki/Commons:P…

I don’t know if we live in a Matrix, but I know for sure that robots will spend most of their lives in simulation. Let machines train machines. I’m excited to introduce DexMimicGen, a massive-scale synthetic data generator that enables a humanoid robot to learn complex skills from only a handful of human demonstrations. Yes, as few as 5! DexMimicGen addresses the biggest pain point in robotics: where do we get data? Unlike with LLMs, where vast amounts of texts are readily available, you cannot simply download motor control signals from the internet. So researchers teleoperate the robots to collect motion data via XR headsets. They have to repeat the same skill over and over and over again, because neural nets are data hungry. This is a very slow and uncomfortable process. At NVIDIA, we believe the majority of high-quality tokens for robot foundation models will come from simulation. What DexMimicGen does is to trade GPU compute time for human time. It takes one motion trajectory from human, and multiplies into 1000s of new trajectories. A robot brain trained on this augmented dataset will generalize far better in the real world. Think of DexMimicGen as a learning signal amplifier. It maps a small dataset to a large (de facto infinite) dataset, using physics simulation in the loop. In this way, we free humans from babysitting the bots all day. The future of robot data is generative. The future of the entire robot learning pipeline will also be generative. 🧵

Meine Einschätzung zu #Strawberry #o1: Spannende Tech-Vorschau. Brückenschlag vom reaktiven Antworten zu systematischem Problemlösen. Stark in Mathe, Logik, Programmierung. Wichtiger #AI-Meilenstein. Prognose: GPT-4/o1-Hybrid im November. #AI #NLP youtube.com/watch?v=lgtgg4…

arXiv -> alphaXiv Students at Stanford have built alphaXiv, an open discussion forum for arXiv papers. @askalphaxiv You can post questions and comments directly on top of any arXiv paper by changing arXiv to alphaXiv in any URL!

Please consider signing this letter as I did. SB1047 attempts to regulate AI research and development, creating obstacles to the dissemination open research in AI and open source AI platforms. Regulating the deployment of AI applications is fine. But regulating R&D would have apocalyptic consequences on the AI ecosystem. The sad thing is that the regulation of AI R&D is predicated on the illusion of "existential risks" pushed by a handful of delusional think-tanks, and dismissed as nonsense (or at least widely premature) by the vast majority of researchers and engineers in academia, startups, larger companies, and investment firms.