# forck.live — 13–19 October 2025 > AI lab announcements, news reports and podcasts. forck.live follows what companies announce about their own products. Press may corroborate an official story; it never leads and never appears as an item of its own. A missing date stays missing. A title is the publisher's words — where a company does not publish in English, an English rendering is listed separately as `Title in English` and named as this site's translation, never as theirs. This file is one issue of the weekly archive: every confirmed first-party announcement of ISO week 2025-W42, 13–19 October 2025, newest first. The window never moves, so this URL always names the same seven days — but a post published inside them and detected later is added when it arrives, so a closed week can still gain entries. Every entry is one first-party announcement, and carries the same eight fields in this order before any others: Company · Announced · Category · Coverage · Announcement · Group · Source · Record `Source` is the company's own page the announcement was traced to and `Record` is this site's page for it. Both are safe to cite. `Coverage` is a number of outlets, or `not counted` — which means this record reached no count of its own, never that nobody wrote about it. `Category` is this tracker's own analysis, and is authoritative wherever it is stated. `not stated` means either that nothing has analysed the post yet or that the analysis placed it outside this taxonomy; neither is a verdict about the announcement. `Announcement` is whether the post is news of something the company did, rather than a how-to, a customer story, a changelog listing or a staffing note. `Group` is which of this record's groups the entry falls in, carried on the entry itself so it holds where the groups are headings and where they are not. One of: `models`, `covered`, `announcements`, `routine`. Anything after `Record` is held detail that only some announcements carry. - This week's page: https://forck.live/week/2025-W42 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Google Research — Solving virtual machine puzzles: How AI is optimizing cloud computing - Company: Google Research (research.google) - Announced: 2025-10-17T17:56:35+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://research.google/blog/solving-virtual-machine-puzzles-how-ai-is-optimizing-cloud-computing/ - Record: https://forck.live/items/1405-solving-virtual-machine-puzzles-how-ai-is-optimizing-cloud-computing - Subject: Research Google Research introduces LAVA, a trio of algorithms (NILAS, LAVA, LARS) that use continuous reprediction of VM lifetimes to optimize resource allocation in cloud data centers. ### Hugging Face — AI for Food Allergies - Company: Hugging Face (huggingface.co) - Announced: 2025-10-16T22:38:11+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/hugging-science/ai-for-food-allergies - Record: https://forck.live/items/1603-ai-for-food-allergies - Subject: Platform Hugging Face announces the AI for Food Allergies project, a community-driven research lab exploring how AI can advance food allergy research. The article reviews current AI applications in food allergy research, including molecular-level deep learning models, clinical diagnostics, and consumer/regulatory NLP systems, and highlights the need for high-quality data. ### Supabase — Snap, Inc. Launches Snap Cloud, Powered by Supabase - Company: Supabase (supabase.com) - Announced: 2025-10-16T17:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://supabase.com/blog/snap-launches-snap-cloud - Record: https://forck.live/items/21323-snap-inc-launches-snap-cloud-powered-by-supabase - Subject: Supabase Snap Inc. launched Snap Cloud, a managed backend platform for developers building on Spectacles, powered by Supabase. The platform provides Spectacles developers with a Postgres database, real-time APIs, and globally distributed Edge Functions directly inside Lens Studio without infrastructure setup. Snap Cloud is currently in alpha as part of Snap's upcoming Spectacles program. ### Google Research — Using AI to identify genetic variants in tumors with DeepSomatic - Company: Google Research (research.google) - Announced: 2025-10-16T16:33:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: no - Group: routine - Source: https://research.google/blog/using-ai-to-identify-genetic-variants-in-tumors-with-deepsomatic/ - Record: https://forck.live/items/1406-using-ai-to-identify-genetic-variants-in-tumors-with-deepsomatic - Subject: Research Google Research announces DeepSomatic, an AI-powered tool published in Nature Biotechnology that uses convolutional neural networks to identify cancer-related somatic variants in tumor cells from sequencing data. The tool works across major sequencing platforms and is made openly available to the research community. ### Cognition — Introducing SWE-grep and SWE-grep-mini: RL for Multi-Turn, Fast Context Retrieval - Company: Cognition (cognition.com) - Announced: 2025-10-16 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cognition.com/blog/swe-grep - Record: https://forck.live/items/7689-introducing-swe-grep-and-swe-grep-mini-rl-for-multi-turn-fast-context-retrieval - Subject: Devin / SWE Cognition released SWE-grep and SWE-grep-mini, fast agentic models specialized in multi-turn, highly parallel context retrieval. These match the retrieval capabilities of frontier coding models while taking an order of magnitude less time. They power the new Fast Context subagent in Windsurf. ### World Labs — RTFM: A Real-Time Frame Model - Company: World Labs (worldlabs.ai) - Announced: 2025-10-16 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.worldlabs.ai/blog/rtfm - Record: https://forck.live/items/7345-rtfm-a-real-time-frame-model - Subject: Marble / Atlas World Labs announced RTFM, a real-time generative world model that generates video in real-time as users interact with it. RTFM is available as a research preview and can be used to explore generated 3D worlds and real-world locations. It runs inference on a single H100 GPU at interactive framerates and models persistent 3D worlds. ### Baidu — PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model - Company: Baidu (baidu.com) - Announced: 2025-10-16 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://ernie.baidu.com/blog/posts/paddleocr-vl/ - Record: https://forck.live/items/4652-paddleocr-vl-boosting-multilingual-document-parsing-via-a-0-9b-ultra-compact - Subject: ERNIE Baidu releases PaddleOCR-VL, a compact vision-language model for document parsing with SOTA performance, multilingual support for 109 languages, and Apache 2.0 license. ### Replicate — How to prompt Veo 3.1 - Company: Replicate (replicate.com) - Announced: 2025-10-16 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/veo-3-1 - Record: https://forck.live/items/2475-how-to-prompt-veo-3-1 - Subject: Platform Replicate published a guide to prompting Google's Veo 3.1, highlighting new capabilities: reference-to-video generation using up to three images, first and last frame interpolation, enhanced image-to-video with better prompt following, and fast generation options with reduced cost and slightly lower quality. ### Hugging Face — Google Cloud C4 Brings a 70% TCO improvement on GPT OSS with Intel and Hugging Face - Company: Hugging Face (huggingface.co) - Announced: 2025-10-16 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/gpt-oss-on-intel-xeon - Record: https://forck.live/items/1604-google-cloud-c4-brings-a-70-tco-improvement-on-gpt-oss-with-intel-and-hugging - Subject: Platform Hugging Face, Intel, and Google Cloud collaborated to benchmark Google Cloud C4 VMs running Intel Xeon 6 processors against previous-generation C3 VMs for inferencing the open-source GPT OSS model, reporting up to 1.7x TCO improvement and up to 1.7x TPOT throughput per vCPU per dollar. ### Sakana AI — ShinkaEvolve in Action: How a Human-AI Partnership Conquered a Coding Challenge - Company: Sakana AI (sakana.ai) - Announced: 2025-10-15T15:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://sakana.ai/icfp-2025 - Record: https://forck.live/items/4706-shinkaevolve-in-action-how-a-human-ai-partnership-conquered-a-coding-challenge - Subject: Sakana models Sakana AI's ShinkaEvolve framework, an evolutionary code optimization tool powered by LLMs, was applied to optimize the SAT encoding in a solution for the ICFP Programming Contest. It achieved up to a 10x speedup in solver execution time, enabling the team to solve larger problem instances, and also provided insights that improved human development. ### Google Research — Coral NPU: A full-stack platform for Edge AI - Company: Google Research (research.google) - Announced: 2025-10-15T13:07:00+00:00 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://research.google/blog/coral-npu-a-full-stack-platform-for-edge-ai/ - Record: https://forck.live/items/1407-coral-npu-a-full-stack-platform-for-edge-ai - Subject: Research Google Research introduces Coral NPU, a full-stack, open-source platform for edge AI, including a RISC-V-based NPU architecture, software toolchain, and documentation. ### Vercel — Just another (Black) Friday - Company: Vercel (vercel.com) - Announced: 2025-10-15T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/just-another-black-friday - Record: https://forck.live/items/20810-just-another-black-friday - Subject: v0 / AI SDK / AI Gateway Vercel customers can treat Black Friday like just another day, ready to scale to billions of requests. For teams on Vercel, Black Friday is just another Friday. ### Vercel — Talha Tariq joins Vercel as CTO of Security - Company: Vercel (vercel.com) - Announced: 2025-10-15T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/talha-tariq-joins-vercel-as-cto-security - Record: https://forck.live/items/20809-talha-tariq-joins-vercel-as-cto-of-security - Subject: v0 / AI SDK / AI Gateway Talha Tariq joined Vercel as CTO of Security, bringing expertise from HashiCorp and IBM. He will lead security research, product R&D, engineering, and trust & safety. ### Vercel — Running Next.js inside ChatGPT: A deep dive into native app integration - Company: Vercel (vercel.com) - Announced: 2025-10-15T13:00:00+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/running-next-js-inside-chatgpt-a-deep-dive-into-native-app-integration - Record: https://forck.live/items/20808-running-next-js-inside-chatgpt-a-deep-dive-into-native-app-integration - Subject: v0 / AI SDK / AI Gateway Vercel published a technical guide explaining how to run a Next.js application inside ChatGPT's triple-iframe architecture using the ChatGPT Apps SDK with MCP support. The post details patches for asset loading, relative URL resolution, client-side navigation, and React Server Components, and provides a starter template deployable to Vercel. ### Vercel — Agents at work, a partnership with Salesforce and Slack - Company: Vercel (vercel.com) - Announced: 2025-10-15T13:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/agents-at-work-a-partnership-with-salesforce-and-slack - Record: https://forck.live/items/20807-agents-at-work-a-partnership-with-salesforce-and-slack - Subject: v0 / AI SDK / AI Gateway Vercel and Salesforce are partnering to enable teams to build, ship, and scale AI agents within the Salesforce ecosystem, starting with Slack. The Vercel AI Cloud serves as the agent runtime, allowing agents to operate directly in Slack threads. Vercel also introduced a v0 Slack Agent built on its platform. ### Harvey — Work Without Boundaries: The AI Platform for Legal Work - Company: Harvey (harvey.ai) - Announced: 2025-10-15T07:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/work-without-boundaries-ai-platform-for-legal-work - Record: https://forck.live/items/17663-work-without-boundaries-the-ai-platform-for-legal-work - Subject: Harvey Harvey announced integrations with Microsoft Word, Outlook, and SharePoint, a mobile app, and iManage, along with a new Matter OS feature that creates a single workspace for legal matters. The platform also introduced external sharing capabilities for playbooks, Vault projects, and workflows. These additions aim to reduce context-switching and make institutional knowledge more accessible across devices and tools. ### Genspark — Genspark Custom Super Agent: Build Your Own AI Agent with Just One Prompt - Company: Genspark (genspark.ai) - Announced: 2025-10-15 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/genspark-custom-agent - Record: https://forck.live/items/18266-genspark-custom-super-agent-build-your-own-ai-agent-with-just-one-prompt - Subject: Genspark ### Krea — Krea Realtime 14B: Real-Time, Long-Form AI Video Generation - Company: Krea (krea.ai) - Announced: 2025-10-15 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.krea.ai/blog/krea-realtime-14b - Record: https://forck.live/items/15093-krea-realtime-14b-real-time-long-form-ai-video-generation - Subject: Krea 2 / AI creative platform Krea released Krea Realtime 14B, a 14-billion parameter autoregressive video model distilled from Wan 2.1 14B using Self-Forcing. The model generates video at 11 fps on a single NVIDIA B200 GPU and allows users to modify prompts mid-generation, restyle videos on-the-fly, and see first frames in 1 second. It is over 10x larger than existing realtime video models and introduces techniques for mitigating error accumulation. ### Luma — Hartbeat and Luma AI Partner on Prompt Side Story — A First-of-Its-Kind Live AI Film Event at LA Tech Week - Company: Luma (lumalabs.ai) - Announced: 2025-10-15 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lumalabs.ai/news/hartbeat-and-luma-ai-partner-on-prompt-side-story - Record: https://forck.live/items/7978-hartbeat-and-luma-ai-partner-on-prompt-side-story-a-first-of-its-kind-live-ai - Subject: Ray / Dream Machine Hartbeat and Luma AI announced a creative collaboration to power Prompt Side Story, a live AI film battle during LA Tech Week on October 18, 2025, where teams use Luma's Dream Machine and Ray3 model to create cinematic shorts in real time. ### Amp — Amp Free - Company: Amp (ampcode.com) - Announced: 2025-10-15 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://ampcode.com/news/amp-free - Record: https://forck.live/items/7798-amp-free - Subject: Amp Amp introduced a free mode that is supported by advertisements, funded by various dev and infra companies. The free mode offers agentic programming with a mix of top OSS models, frontier models with limited context windows, and pre-release frontier models. Amp Free no longer requires users to share data for training, as ads alone cover costs. ### Together AI — Announcing the Together AI Startup Accelerator, purpose-built for AI Native Apps - Company: Together AI (together.ai) - Announced: 2025-10-15 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.together.ai/blog/announcing-together-ai-startup-accelerator - Record: https://forck.live/items/2406-announcing-the-together-ai-startup-accelerator-purpose-built-for-ai-native-apps - Subject: Inference platform Together AI launches a Startup Accelerator for AI-native app startups, offering compute credits, engineering support, and GTM help. ### Hugging Face — Get your VLM running in 3 simple steps on Intel CPUs - Company: Hugging Face (huggingface.co) - Announced: 2025-10-15 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/openvino-vlm - Record: https://forck.live/items/1605-get-your-vlm-running-in-3-simple-steps-on-intel-cpus - Subject: Platform Hugging Face published a tutorial showing how to run a Vision Language Model (VLM) on Intel CPUs using Optimum Intel and OpenVINO, with three steps: model conversion, quantization, and inference. The tutorial uses SmolVLM2-256M-Video-Instruct as the example model. ### OpenAI — Plex Coffee delivers fast, personal service with ChatGPT - Company: OpenAI (openai.com) - Announced: 2025-10-15 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/plex-coffee - Record: https://forck.live/items/421-plex-coffee-delivers-fast-personal-service-with-chatgpt - Subject: GPT / ChatGPT / API Plex Coffee uses ChatGPT Business to centralize knowledge, train staff faster, and preserve personal connections while expanding. ### Harvey — Harvey In Practice: How to Turn Precedent into Tailored Drafts in Minutes - Company: Harvey (harvey.ai) - Announced: 2025-10-14T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/harvey-in-practice-turn-precedent-into-tailored-drafts - Record: https://forck.live/items/17664-harvey-in-practice-how-to-turn-precedent-into-tailored-drafts-in-minutes - Subject: Harvey Harvey's guide demonstrates how lawyers can upload prior agreements and generate revised contract clauses with side-by-side comparisons and citations. The workflow is designed to speed drafting and maintain continuity across deals without switching tools. ### OpenAI — Expert Council on Well-Being and AI - Company: OpenAI (openai.com) - Announced: 2025-10-14T10:00:00+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/expert-council-on-well-being-and-ai - Record: https://forck.live/items/422-expert-council-on-well-being-and-ai - Subject: GPT / ChatGPT / API OpenAI has formed an Expert Council on Well-Being and AI to guide ChatGPT's support for emotional health, particularly for teens, aiming to create safer AI experiences. ### OpenAI — Argentina’s AI opportunity - Company: OpenAI (openai.com) - Announced: 2025-10-14T06:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/global-affairs/argentinas-ai-opportunity - Record: https://forck.live/items/423-argentina-s-ai-opportunity - Subject: GPT / ChatGPT / API OpenAI and Sur Energy are exploring Argentina's first Stargate project, an AI and clean energy collaboration that could make Argentina a Latin American leader in artificial intelligence, sustainable infrastructure, and digital innovation. ### Luma — Ray3 Evaluation Report – State-of-the-Art Performance for Pro Video Generation - Company: Luma (lumalabs.ai) - Announced: 2025-10-14 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://lumalabs.ai/news/ray3-eval - Record: https://forck.live/items/7979-ray3-evaluation-report-state-of-the-art-performance-for-pro-video-generation - Subject: Ray / Dream Machine Luma released Ray3, a new video generation model. Independent evaluations show it achieves state-of-the-art performance alongside Veo3 and leads over other models in realism and control. The model is designed for filmmakers and creators. ### NVIDIA — Nemotron-Personas-India: Synthesized Data for Sovereign AI - Company: NVIDIA (nvidia.com) - Announced: 2025-10-13T23:00:42+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/nvidia/nemotron-personas-india - Record: https://forck.live/items/1606-nemotron-personas-india-synthesized-data-for-sovereign-ai NVIDIA released Nemotron-Personas-India, an open synthetic dataset of 21 million Indic personas grounded in India's real-world demographic distributions, licensed under CC BY 4.0, to support development of AI systems reflecting Indian society. ### OpenAI — OpenAI and Broadcom announce strategic collaboration to deploy 10 gigawatts of OpenAI-designed AI accelerators - Company: OpenAI (openai.com) - Announced: 2025-10-13T06:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/openai-and-broadcom-announce-strategic-collaboration - Record: https://forck.live/items/424-openai-and-broadcom-announce-strategic-collaboration-to-deploy-10-gigawatts-of - Subject: GPT / ChatGPT / API OpenAI and Broadcom announce a multi-year partnership to deploy 10 gigawatts of OpenAI-designed AI accelerators, co-developing next-generation systems and Ethernet solutions for scalable, energy-efficient AI infrastructure by 2029. ### Microsoft AI — Introducing MAI-Image-1, debuting in the top 10 on LMArena | Microsoft AI - Company: Microsoft AI (microsoft.ai) - Announced: 2025-10-13 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://microsoft.ai/news/introducing-mai-image-1-debuting-in-the-top-10-on-lmarena/ - Record: https://forck.live/items/10971-introducing-mai-image-1-debuting-in-the-top-10-on-lmarena-microsoft-ai - Subject: MAI Microsoft AI announced MAI-Image-1, its first in-house image generation model, which debuted in the top 10 text-to-image models on LMArena. The model is now available in Bing Image Creator and Copilot Audio Expressions, and is designed to deliver photorealistic imagery with speed and visual diversity. ### Microsoft AI — Introducing MAI-Image-1, debuting in the top 10 on LMArena - Company: Microsoft AI (microsoft.ai) - Announced: 2025-10-13 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://microsoft.ai/news/introducing-mai-image-1-debuting-in-the-top-10-on-lmarena - Record: https://forck.live/items/8437-introducing-mai-image-1-debuting-in-the-top-10-on-lmarena - Subject: MAI Microsoft AI announced MAI-Image-1, its first in-house image generation model, which debuted in the top 10 text-to-image models on LMArena. The model is designed for photorealistic imagery and visual diversity, prioritizing rigorous data selection and evaluation based on real-world creative use cases. It began rolling out into select Microsoft products, including Bing Image Creator and Copilot Audio Expressions, starting November 4, 2025.