# AI IN LIFE

> Daily AI news in German and English, AI-produced with editorial review. Operator: CULTURETEK PTE. LTD. (Singapur).

## About
- What: bilingual (DE/EN) news site on artificial intelligence — models, companies, chips, research, robotics, regulation.
- Format per article: headline, one-sentence answer ("Kurz gesagt" / "In brief"), fact box, body, FAQ and linked primary sources.
- Transparency: texts, images and the anchor Lina Noa are AI-generated and labelled as such; sources are linked under every report. Details: https://ai-in-life.com/ki-hinweis/
- Editorial principles: https://ai-in-life.com/ueber-uns/ · https://ai-in-life.com/en/about/
- Legal notice: https://ai-in-life.com/impressum/
- Corrections & contact: redaktion@ai-in-life.com
- Video roundups: https://ai-in-life.com/sendung/ · https://ai-in-life.com/en/show/
- Feeds: https://ai-in-life.com/feed.xml · https://ai-in-life.com/en/feed.xml
- Sitemap: https://ai-in-life.com/sitemap.xml

## Archive & Search
- Archiv (DE): https://ai-in-life.com/archiv/
- Archive (EN): https://ai-in-life.com/en/archive/
- Suche (DE): https://ai-in-life.com/suche/
- Search (EN): https://ai-in-life.com/en/search/

## Topics / Rubriken
- OpenAI / OpenAI: https://ai-in-life.com/thema/openai/ · https://ai-in-life.com/en/topic/openai/
- Anthropic / Anthropic: https://ai-in-life.com/thema/anthropic/ · https://ai-in-life.com/en/topic/anthropic/
- Google & DeepMind / Google & DeepMind: https://ai-in-life.com/thema/google/ · https://ai-in-life.com/en/topic/google/
- Meta / Meta: https://ai-in-life.com/thema/meta/ · https://ai-in-life.com/en/topic/meta/
- xAI / xAI: https://ai-in-life.com/thema/xai/ · https://ai-in-life.com/en/topic/xai/
- Alibaba / Qwen / Alibaba / Qwen: https://ai-in-life.com/thema/alibaba/ · https://ai-in-life.com/en/topic/alibaba/
- ByteDance / ByteDance: https://ai-in-life.com/thema/bytedance/ · https://ai-in-life.com/en/topic/bytedance/
- KI-Modelle / AI Models: https://ai-in-life.com/thema/modelle/ · https://ai-in-life.com/en/topic/modelle/
- Sicherheit / Safety & Security: https://ai-in-life.com/thema/sicherheit/ · https://ai-in-life.com/en/topic/sicherheit/
- Regulierung & Recht / Regulation & Policy: https://ai-in-life.com/thema/regulierung/ · https://ai-in-life.com/en/topic/regulierung/
- Robotik / Robotics: https://ai-in-life.com/thema/robotik/ · https://ai-in-life.com/en/topic/robotik/
- Chips & Infrastruktur / Chips & Compute: https://ai-in-life.com/thema/chips/ · https://ai-in-life.com/en/topic/chips/
- Wirtschaft & Deals / Business & Deals: https://ai-in-life.com/thema/wirtschaft/ · https://ai-in-life.com/en/topic/wirtschaft/
- Tools & Anwendungen / Tools & Apps: https://ai-in-life.com/thema/tools/ · https://ai-in-life.com/en/topic/tools/
- Forschung / Research: https://ai-in-life.com/thema/forschung/ · https://ai-in-life.com/en/topic/forschung/
- Gesellschaft / Society: https://ai-in-life.com/thema/gesellschaft/ · https://ai-in-life.com/en/topic/gesellschaft/
- Tests / Reviews: https://ai-in-life.com/thema/test/ · https://ai-in-life.com/en/topic/test/

## News (DE) — latest 40
- [Tokio Marine: Versicherung für KI-Roboter in Japan](https://ai-in-life.com/news/tokio-marine-first-japan-nonlife-insurer-to-cover-ai/): Der japanische Versicherer startet im November eine Police für KI-gesteuerte Roboter. Die Prämien liegen rund 20 Prozent über jenen für Industrieroboter.
- [OpenTakeoff: Aufmaß-Engine für KI-Agenten per MCP](https://ai-in-life.com/news/kentucky-ai-opentakeoff-open-source-agpl-3-0-pdf-takeoff/): Die AGPL-Engine läuft komplett im Browser und stellt Agenten 58 MCP-Werkzeuge bereit – doch unterschreiben darf nur ein Mensch.
- [Google: KI-Tuberkulose-Screening in 40 Kliniken aktiv](https://ai-in-life.com/news/bringing-foundational-models-to-the-front-lines-of-tb/): Googles Forschungsteam bündelt offene Medizinmodelle mit einem zertifizierten Röntgen-Dienst: 40 Einrichtungen in sechs Ländern screenen damit bereits.
- [Asiens KI-Rechenzentren: Börsengänge geraten ins Stocken](https://ai-in-life.com/news/asia-s-ai-data-center-boom-faces-reality-check-from-public/): Gestoppte Börsengänge und wackelige Debüts drücken auf die Bewertungen asiatischer Rechenzentren – Nikkei Asia sieht eine Prüfung durch den Kapitalmarkt.
- [Enact Intelligence: 30 Millionen Dollar Pre-Seed](https://ai-in-life.com/news/robotics-startup-enact-intelligence-nabs-30m-funding-round/): Der Schweizer Robotik-Entwickler sammelt 30 Millionen Dollar ein – unter den Geldgebern Yann LeCun, Mistral-Chef Arthur Mensch und Thomas Wolf von Hugging Face.
- [Nvidia-Chef Huang wirbt in Berlin für Robotik-Pläne](https://ai-in-life.com/news/nvidia-so-umschmeichelt-jensen-huang-die-deutschen/): Der Nvidia-Gründer lobt Europa in Berlin als idealen Partner für seine Robotik-Offensive – und hat dabei eigene Interessen im Blick.
- [Kimi K3.1 Pro rückt auf Platz 4 nahe an Claude Opus 5.5](https://ai-in-life.com/news/chinesisches-ki-modell-kimi-k3-1-pro-schliesst-fast-zu/): Das Modell von Moonshot AI erreicht 55 Punkte im Artificial Analysis Intelligence Index und soll bis 4. November 2026 als Open Weights erscheinen.
- [Warnstreik bei privaten Banken: Verdi will KI-Regeln](https://ai-in-life.com/news/business-ticker-verdi-ruft-zu-bundesweiten-warnstreiks-bei/): Rund 150.000 Bankbeschäftigte sind betroffen: Verdi fordert sechs Prozent mehr Gehalt – und verbindliche Regeln für den KI-Einsatz in den Instituten.
- [Uber gibt allen 30.000 Mitarbeitenden ChatGPT Work](https://ai-in-life.com/news/uber-brings-chatgpt-work-to-all-employees-with-openai/): Nach einem Pilotprojekt mit 1.000 Beschäftigten erhalten alle 30.000 Uber-Mitarbeitenden Zugriff auf OpenAIs Unternehmens-Agenten ChatGPT Work.
- [GPT-6 Astra kommt in Googles Model Garden](https://ai-in-life.com/news/openai-models-are-coming-to-google-cloud-2/): Astra, Sol und Luna sollen in wenigen Wochen in Googles Modellkatalog laufen – bezahlbar über bestehende Cloud-Zusagen.
- [Google baut Agent Compute Engine für KI-Agenten](https://ai-in-life.com/news/google-stellt-agent-compute-engine-vor-eigene/): Jeder Agent läuft in einer eigenen microVM, die bei Leerlauf pausiert. Preview kommt in den nächsten Monaten, Preise nennt Google noch nicht.
- [Claude Duet: Anthropics Agenten-App startet als Preview](https://ai-in-life.com/news/claude-duet-app-is-anthropic-s-bet-on-agents/): Die neue App läuft im Research Preview auf iOS, Android und Web, ist Pro- und Max-Kunden per Warteliste vorbehalten und erledigt Aufgaben im Hintergrund.
- [Canva Craft: KI-Modell plant das Design vor dem Bild](https://ai-in-life.com/news/canva-s-first-ai-image-model-plans-designs-before/): Canva bringt sein erstes eigenes Bildmodell: Craft entwirft zuerst einen Layoutplan, rendert dann bis 2K in 30 Seitenverhältnissen.
- [Claude Duet: Anthropics Dauer-Agent im Research Preview](https://ai-in-life.com/news/anthropic-schickt-proaktiven-dauer-agenten-claude-duet/): Die Research Preview läuft auf Claude Opus 5.5, prüft verbundene Konten alle paar Stunden und fragt vor Zahlungen und Signaturen nach.
- [1Password: Login für KI-Agenten ohne Passwortfreigabe](https://ai-in-life.com/news/1password-launches-a-way-for-ai-agents-to-log-in-on-your/): Der Passwortmanager prüft Agenten per eigenem Schlüssel, fragt den Nutzer auf dessen Gerät und tippt die Zugangsdaten selbst ein — der Agent sieht sie nie.
- [OpenAI-Modelle kommen in die Google Cloud](https://ai-in-life.com/news/openai-models-are-coming-to-google-cloud/): GPT-6 Astra, Sol und Luna sollen in wenigen Wochen über Googles Model Garden laufen – samt Antigravity und Gemini Agent.
- [Inception Mercury 3: Benchmarks ohne externe Prüfung](https://ai-in-life.com/news/inception-s-mercury-3-diffusion-model-claims-to-beat-gpt-6/): Inception nennt 1.068 Token pro Sekunde und acht Siege in zwölf Tests. Unabhängige Messungen dazu gibt es bisher nicht.
- [Blacksmith: Sequoia-Runde macht CI-Startup zum Unicorn](https://ai-in-life.com/news/blacksmith-hits-unicorn-valuation-after-sequoia-leads/): Zwei Monate nach der Series B bei 550 Millionen Dollar steigt die Bewertung auf 1 Milliarde Dollar – der ARR wuchs laut Firma um das 17-Fache.
- [Anthropic plant Gemeinwohl-Fonds mit fünf Prozent Aktien](https://ai-in-life.com/news/anthropic-will-fuenf-prozent-seiner-aktien-in-gemeinwohl/): Der Fonds soll Mittel an NGOs und bestimmte Bevölkerungsgruppen geben, um Folgen der KI am Arbeitsmarkt abzufedern – politische Spenden sind ausgeschlossen.
- [Claude Duet: Anthropics proaktiver KI-Agent im Preview](https://ai-in-life.com/news/anthropic-takes-on-muse-with-claude-duet-its-new-ai-agent/): Anthropic oeffnet eine Warteliste fuer einen Agenten, der morgens briefen, Entwuerfe vorlegen und in ueber 6.000 Diensten selbst arbeiten soll.
- [DeepSeek-Verbot für zivile US-Bundesbehörden](https://ai-in-life.com/news/us-orders-federal-agencies-to-remove-china-s-deepseek-ai/): Das Heimatschutzministerium zwingt zivile US-Behörden, DeepSeek-Produkte und darauf aufbauende Dienste zu entfernen – ohne genannte Frist.
- [Otter.ai: 320 Millionen Dollar Finanzierung ohne Anteile](https://ai-in-life.com/news/otter-ai-raises-320m-from-general-catalyst-s-growth-fund/): General Catalyst finanziert Otter.ai über seinen Customer Value Fund: 320 Millionen Dollar für Vertrieb und Marketing, ohne Firmenanteile zu übernehmen.
- [OpenAI kauft Orchestra Bio für Rosalind-Einheit](https://ai-in-life.com/news/openai-acquires-orchestra-bio/): Die Agenten-Firma Orchestra Bio geht in OpenAIs Life-Sciences-Einheit Rosalind auf – Kaufpreis nicht genannt, Team und Technik ziehen um.
- [Merz kündigt KI-Deal mit Nvidia-Chef Huang an](https://ai-in-life.com/news/ki-investitionen-wenigstens-jensen-huang-glaubt-an/): Beim Gipfel mit ausländischen Investoren verkündet der Kanzler eine KI-Partnerschaft mit Nvidia – offen bleibt, was darin mehr ist als eine Ankündigung.
- [Anthropic: 5 Prozent der Aktien für Gemeinwohlfonds](https://ai-in-life.com/news/ki-anthropic-will-5-prozent-seiner-aktien-dem-gemeinwohl/): Der KI-Anbieter legt fünf Prozent seiner Anteile in einen Fonds für NGOs und Arbeitsmarktfolgen – der Anteil kann mit dem Firmenwert wachsen.
- [DeepSeek macht Inferenz-Framework DeepInfer open source](https://ai-in-life.com/news/deepinfer-high-performance-inference-framework-for-deepseek/): Das Framework hinter DeepSeeks Produktionsdiensten liegt unter Apache 2.0 offen – lauffaehig auf Nvidia SM100 und Huawei Ascend 950, ohne KV-Cache-System.
- [Pi-Harness: Desktop-Aufsicht für den Pi Coding Agent](https://ai-in-life.com/news/wangmiaozero-pi-harness-the-operational-superset-of-pi/): Das Open-Source-Projekt Pi-Harness legt eine Electron-Oberfläche mit Checkpoints, Budgets und Agent-Teams über den Pi Coding Agent — unter AGPL-3.0.
- [Timestral: Mistrals Zeitreihen-Modelle laufen auf CPUs](https://ai-in-life.com/news/timestral-mistral-s-time-series-foundation-models/): Mistral veröffentlicht vier Zeitreihen-Modelle von 5 bis 209 Millionen Parametern unter modifizierter MIT-Lizenz — und zwar für reine CPU-Hardware.
- [EBU und NVIDIA setzen Standards für KI-Agenten im Rundfunk](https://ai-in-life.com/news/nvidia-advances-agentic-security-for-media-from-european/): Die EBU öffnet ihren Souveränitätsrahmen für Cloud-Anbieter, offene Modelle und Integratoren – mit Microsoft, Nebius, Qvest und NVIDIA als erste Partner.
- [Merz und NVIDIA: Absichtserklärung für Industrie-KI](https://ai-in-life.com/news/germany-wants-to-lead-the-world-in-industrial-ai-how-nvidia/): Im Kraftwerk Berlin unterzeichneten Kanzleramt, zwei Bundesministerien und NVIDIA eine unverbindliche Absichtserklärung zu physischer KI.
- [DeepSeek legt bei seinem Billig-KI-Modell nach](https://ai-in-life.com/news/china-s-deepseek-debuts-improved-low-cost-ai-as-tech-doubts/): Nikkei Asia meldet ein verbessertes Modell, das laut DeepSeek US-Modelle bei geringeren Kosten übertrifft — Zahlen nennt der Bericht nicht.
- [Nikkei-Umfrage: Japans Englischlernende wollen Menschen](https://ai-in-life.com/news/ai-savvy-english-learners-in-japan-want-human-instructors/): Laut einer am 20. Oktober veröffentlichten Nikkei-Erhebung halten erwachsene Lernende menschliche Lehrkräfte für unverzichtbar – gerade dann, wenn sie KI nutzen.
- [RTX Spark Laptops: Marktstart ab 2.399 Dollar](https://ai-in-life.com/news/rtx-spark-ignites-global-celebration-as-new-windows-pcs/): NVIDIA bringt RTX Spark in schlanke Windows-Notebooks von sechs Herstellern: Blackwell-GPU, Grace-CPU, bis zu 128 GB Unified Memory, 1 Petaflop.
- [ClawPanel: Multi-Engine-Panel für fünf Agent-Engines](https://ai-in-life.com/news/qingchencloud-clawpanel-openclaw-hermes-agent-ai-ai-tauri/): Das Open-Source-Panel verwaltet OpenClaw, Hermes Agent, DeepSeek Harness, OpenCode und das experimentelle Pi aus einer Tauri-v2-App.
- [Llama.cpp auf gejailbreakter PS5: GGUF-Modelle lokal](https://ai-in-life.com/news/llama-cpp-on-a-jailbroken-ps5-run-any-gguf-model-natively/): Das Open-Source-Projekt PS5LM startet llama.cpp über Vulkan auf der RDNA-2-GPU der Konsole und stellt OpenAI- und Anthropic-kompatible Endpunkte bereit.
- [LangChain Core 1.6.12 behebt Streaming-Metadaten](https://ai-in-life.com/news/langchain-ai-langchain-langchain-core-1-6-12-released/): Der Patch gibt Extras bei content-block-start-Events wieder weiter – ein kleiner Fix, der Streaming-Metadaten in Agenten-Pipelines rettet.
- [Driving-Bench: Sprachmodelle fahren einen echten Toyota](https://ai-in-life.com/news/chatbots-am-steuer-entwickler-lassen-gpt-claude-und-grok/): Drei Entwickler des Startups Axiom ließen vier Chatbots ohne Fahrtraining ans Lenkrad – nur GPT-6 Astra brachte den Parcours zu Ende.
- [Muse Spark Open: Metas Rückkehr mit 816 Milliarden](https://ai-in-life.com/news/muse-spark-open-meta-meldet-sich-mit-open-weight-modell/): Das offene Modell bringt 42 von 816 Milliarden Parametern pro Token in Bewegung, fasst eine Million Token Kontext – und läuft laut Meta auf zwei Mac Studio.
- [Bundesregierung vereinbart KI-Kooperation mit Nvidia](https://ai-in-life.com/news/kuenstliche-intelligenz-bundesregierung-vereinbart-ki/): Nvidia soll Ingenieure stellen, der Bund das Geld: Die Kooperation zielt auf KI in der Industrie. Summe, Ressort und Zeitplan bleiben offen.
- [Google-Agenten steuern Laborroboter: VERID im Nasslabor](https://ai-in-life.com/news/grounding-ai-agents-in-biological-reality-building-blocks/): Drei Systeme aus Google Research koppeln digitalen Zwilling und Lernschleife — die Aptamer-Trefferquote stieg in drei Runden von 10,2 auf 33,5 Prozent.

## News (EN) — latest 40
- [Tokio Marine launches AI robot insurance in Japan](https://ai-in-life.com/en/news/tokio-marine-first-japan-nonlife-insurer-to-cover-ai/): The insurer starts selling cover for AI-driven robots in November, with premiums priced about 20 percent above those for conventional industrial robots.
- [OpenTakeoff Gives AI Agents 58 MCP Takeoff Tools](https://ai-in-life.com/en/news/kentucky-ai-opentakeoff-open-source-agpl-3-0-pdf-takeoff/): The AGPL engine runs entirely in the browser and hands agents 58 MCP tools, yet only a human click turns a proposal into signed ink.
- [Google AI TB Screening Now Live in 40 Clinics](https://ai-in-life.com/en/news/bringing-foundational-models-to-the-front-lines-of-tb/): Google Research pairs its open medical models with a CE-certified X-ray service: roughly 40 facilities in six countries already screen with it.
- [Asia's AI Data Center IPOs Hit a Market Reality Check](https://ai-in-life.com/en/news/asia-s-ai-data-center-boom-faces-reality-check-from-public/): Stalled listings and shaky debuts are pressuring valuations across Asia's data center sector, Nikkei Asia reports, as investors demand proof of returns.
- [Enact Intelligence raises $30M pre-seed for robots](https://ai-in-life.com/en/news/robotics-startup-enact-intelligence-nabs-30m-funding-round/): The Swiss startup builds its own force-sensing actuators and pre-trains robot models from scratch, with Yann LeCun and Arthur Mensch as angel backers.
- [Nvidia's Huang Pitches Robotics Plans to Europe in Berlin](https://ai-in-life.com/en/news/nvidia-so-umschmeichelt-jensen-huang-die-deutschen/): In Berlin, Nvidia's founder courted Europe as the ideal partner for his robotics push – a compliment that also serves his own order book.
- [Kimi K3.1 Pro Hits No. 4, Just Behind Claude Opus 5.5](https://ai-in-life.com/en/news/chinesisches-ki-modell-kimi-k3-1-pro-schliesst-fast-zu/): Moonshot AI's model scores 55 points on the Artificial Analysis Intelligence Index, and its weights are slated for release by November 4, 2026.
- [Verdi Bank Warning Strike Puts AI Rules on the Table](https://ai-in-life.com/en/news/business-ticker-verdi-ruft-zu-bundesweiten-warnstreiks-bei/): About 150,000 bank employees are covered: Verdi wants six percent more pay plus binding rules on how banks deploy artificial intelligence.
- [Uber ChatGPT Work Rollout Reaches 30,000 Employees](https://ai-in-life.com/en/news/uber-brings-chatgpt-work-to-all-employees-with-openai/): A 1,000-person pilot in finance, operations and product convinced Uber to hand its entire 30,000-strong workforce OpenAI's enterprise agent.
- [OpenAI's GPT-6 models land in Google Model Garden](https://ai-in-life.com/en/news/openai-models-are-coming-to-google-cloud-2/): Astra, Sol and Luna are due in Google's model catalog within weeks, payable from committed cloud spend rather than a new AI budget.
- [Google Cloud's Agent Compute Engine runs AI agents](https://ai-in-life.com/en/news/google-stellt-agent-compute-engine-vor-eigene/): Every agent gets its own microVM that pauses when idle. Preview arrives in the coming months, and Google has not published any prices yet.
- [Claude Duet: Anthropic's Agent App Opens as Preview](https://ai-in-life.com/en/news/claude-duet-app-is-anthropic-s-bet-on-agents/): The research preview runs on web, iOS and Android, is limited to Pro and Max subscribers via waitlist, and works through tasks on a cloud machine.
- [Canva Craft: first image model plans before drawing](https://ai-in-life.com/en/news/canva-s-first-ai-image-model-plans-designs-before/): Canva's first in-house image generator drafts a layout plan first, then renders up to 2K across 30 aspect ratios, trained on templates.
- [Claude Duet: Anthropic's Always-On Agent Preview](https://ai-in-life.com/en/news/anthropic-schickt-proaktiven-dauer-agenten-claude-duet/): The research preview runs on Claude Opus 5.5, checks connected accounts every few hours, and asks for approval before payments or signatures.
- [1Password for Agents: AI login without the password](https://ai-in-life.com/en/news/1password-launches-a-way-for-ai-agents-to-log-in-on-your/): The password manager verifies each agent with its own key, asks the account owner to approve on their device, then types the credentials into the page itself.
- [OpenAI Models Coming to Google Cloud Model Garden](https://ai-in-life.com/en/news/openai-models-are-coming-to-google-cloud/): Google Cloud's Model Garden will host GPT-6 Astra alongside the cheaper Sol and Luna tiers, with general availability a few weeks out.
- [Inception's Mercury 3 benchmarks lack outside checks](https://ai-in-life.com/en/news/inception-s-mercury-3-diffusion-model-claims-to-beat-gpt-6/): Inception claims 1,068 tokens per second and wins in eight of 12 tests. So far no outside lab has published its own measurements.
- [Blacksmith Becomes a Unicorn on Sequoia-Led Series C](https://ai-in-life.com/en/news/blacksmith-hits-unicorn-valuation-after-sequoia-leads/): Two months after a $550 million Series B, the CI startup is valued at $1 billion; it says annual recurring revenue has grown 17-fold.
- [Anthropic public-benefit fund to hold five percent stake](https://ai-in-life.com/en/news/anthropic-will-fuenf-prozent-seiner-aktien-in-gemeinwohl/): The money is earmarked for NGOs and selected population groups to cushion labor-market effects of AI, and the company rules out political donations.
- [Claude Duet: Anthropic's Proactive AI Agent Preview](https://ai-in-life.com/en/news/anthropic-takes-on-muse-with-claude-duet-its-new-ai-agent/): Anthropic has opened a waitlist for an assistant that briefs you each morning, drafts replies for approval and works across 6,000-plus tools.
- [DeepSeek Ban Hits US Federal Civilian Agencies](https://ai-in-life.com/en/news/us-orders-federal-agencies-to-remove-china-s-deepseek-ai/): Homeland Security told civilian agencies to strip DeepSeek products and anything built on its models out of federal IT, with no deadline set.
- [Otter.ai Raises $320M in Non-Dilutive Funding](https://ai-in-life.com/en/news/otter-ai-raises-320m-from-general-catalyst-s-growth-fund/): General Catalyst's Customer Value Fund is backing the meeting assistant with $320 million for sales and marketing, taking revenue share instead of equity.
- [OpenAI Acquires Orchestra Bio for Its Rosalind Unit](https://ai-in-life.com/en/news/openai-acquires-orchestra-bio/): Orchestra Bio's operations agents move into OpenAI's Rosalind life sciences unit; the purchase price stays undisclosed, the team joins.
- [Merz Announces AI Deal With Nvidia's Jensen Huang](https://ai-in-life.com/en/news/ki-investitionen-wenigstens-jensen-huang-glaubt-an/): Germany's chancellor used an investor summit to unveil an AI partnership with Nvidia, leaving open how much of it is signed and how much is intent.
- [Anthropic Pledges 5% of Shares to Public Benefit Fund](https://ai-in-life.com/en/news/ki-anthropic-will-5-prozent-seiner-aktien-dem-gemeinwohl/): The AI company is routing five percent of its equity into a fund for NGOs and labor-market support, with the stake able to grow alongside its valuation.
- [DeepSeek open-sources DeepInfer inference framework](https://ai-in-life.com/en/news/deepinfer-high-performance-inference-framework-for-deepseek/): The engine behind DeepSeek's production services is now Apache 2.0, running on Nvidia SM100 and Huawei Ascend 950 — but the KV cache layer stays in-house.
- [Pi-Harness Adds Run Control to Pi Coding Agent](https://ai-in-life.com/en/news/wangmiaozero-pi-harness-the-operational-superset-of-pi/): Pi-Harness wraps the Pi Coding Agent in an Electron desktop app with checkpoints, per-run budgets and agent teams, released under AGPL-3.0.
- [Timestral: Mistral's Time-Series Models Run on CPUs](https://ai-in-life.com/en/news/timestral-mistral-s-time-series-foundation-models/): Mistral released four forecasting models from 5 to 209 million parameters under a modified MIT license, built to run on server CPUs.
- [EBU and NVIDIA Set Agentic Security Rules for Broadcasters](https://ai-in-life.com/en/news/nvidia-advances-agentic-security-for-media-from-european/): The EBU is widening its sovereignty framework to cloud vendors, open models and integrators, naming Microsoft, Nebius, Qvest and NVIDIA as first partners.
- [Merz and NVIDIA Sign Industrial AI Letter of Intent](https://ai-in-life.com/en/news/germany-wants-to-lead-the-world-in-industrial-ai-how-nvidia/): At Berlin's Kraftwerk venue, the chancellery, two federal ministries and NVIDIA signed a non-binding letter of intent on physical AI.
- [DeepSeek upgrades its low-cost AI model](https://ai-in-life.com/en/news/china-s-deepseek-debuts-improved-low-cost-ai-as-tech-doubts/): Nikkei Asia reports an upgraded DeepSeek model that the company says beats OpenAI and Anthropic tools at lower cost, with no figures given.
- [Nikkei Poll: Japan's English Learners Still Want Humans](https://ai-in-life.com/en/news/ai-savvy-english-learners-in-japan-want-human-instructors/): Nikkei's October 20 survey finds adult learners treat live teachers as indispensable, with the strongest conviction among those already studying with AI.
- [RTX Spark Laptops Launch at $2,399 and Up](https://ai-in-life.com/en/news/rtx-spark-ignites-global-celebration-as-new-windows-pcs/): NVIDIA's RTX Spark superchip reaches slim Windows laptops from six vendors, pairing a Blackwell RTX GPU and Grace CPU with up to 128GB of unified memory.
- [ClawPanel: One Panel for Five AI Agent Engines](https://ai-in-life.com/en/news/qingchencloud-clawpanel-openclaw-hermes-agent-ai-ai-tauri/): The AGPL-licensed desktop app bundles OpenClaw, Hermes Agent, DeepSeek Harness, OpenCode and an experimental Pi engine behind one Rust shell.
- [Jailbroken PS5 Runs Any GGUF Model on Its Own GPU](https://ai-in-life.com/en/news/llama-cpp-on-a-jailbroken-ps5-run-any-gguf-model-natively/): PS5LM pushes llama.cpp onto the console's RDNA 2 graphics core, serving OpenAI- and Anthropic-style endpoints from 11.4 GB of shared memory.
- [LangChain Core 1.6.12 Fixes Streaming Block Extras](https://ai-in-life.com/en/news/langchain-ai-langchain-langchain-core-1-6-12-released/): One bug fix, one version bump: extras now ride along on content-block-start events again, so streaming metadata stops vanishing mid-chunk.
- [Driving-Bench: Language Models Steer a Real Toyota](https://ai-in-life.com/en/news/chatbots-am-steuer-entwickler-lassen-gpt-claude-und-grok/): Four chatbots with no driving training took the wheel of a Toyota Corolla in an Axiom experiment, and only one finished the course.
- [Muse Spark Open: Meta's 816B Open-Weight Release](https://ai-in-life.com/en/news/muse-spark-open-meta-meldet-sich-mit-open-weight-modell/): Meta's open model activates 42 of 816 billion parameters per token, holds a million-token context – and, Meta says, runs on two Mac Studio desktops.
- [German Government Agrees AI Cooperation With Nvidia](https://ai-in-life.com/en/news/kuenstliche-intelligenz-bundesregierung-vereinbart-ki/): Nvidia is to supply engineers and Berlin the funding to push AI onto factory floors. The sum, the lead ministry and the timing stay open.
- [Google's VERID lets AI agents run wet-lab robots](https://ai-in-life.com/en/news/grounding-ai-agents-in-biological-reality-building-blocks/): Three Google Research systems pair a semantic digital twin with closed-loop learning, lifting aptamer hit rates from 10.2 to 33.5 percent in three rounds.

## Optional
- Full list of all reports: https://ai-in-life.com/llms-full.txt

## Legal
- agb: https://ai-in-life.com/agb/ · https://ai-in-life.com/en/terms/
- datenschutz: https://ai-in-life.com/datenschutz/ · https://ai-in-life.com/en/privacy/
- impressum: https://ai-in-life.com/impressum/ · https://ai-in-life.com/en/legal-notice/
- ki-atlas: https://ai-in-life.com/ki-atlas-sendung/ · https://ai-in-life.com/en/ai-atlas-show/
- ki-hinweis: https://ai-in-life.com/ki-hinweis/ · https://ai-in-life.com/en/ai-disclosure/
- ueber-uns: https://ai-in-life.com/ueber-uns/ · https://ai-in-life.com/en/about/