# forck.live — 17–23 March 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-W12, 17–23 March 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-W12 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Alibaba — Qwen2.5-VL-32B: Smarter and Lighter - Company: Alibaba (alibaba.com) - Announced: 2025-03-23T16:00:04+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qwen2.5-vl-32b/ - Record: https://forck.live/items/2293-qwen2-5-vl-32b-smarter-and-lighter - Subject: Qwen Alibaba's Qwen team announced the open-source release of Qwen2.5-VL-32B-Instruct, a 32B parameter vision-language model under Apache 2.0, refined from the Qwen2.5-VL series using reinforcement learning. ### Sakana AI — Sakana AI Wins Award at US-Japan Competition for Defense Innovation - Company: Sakana AI (sakana.ai) - Announced: 2025-03-23T15:00:00+00:00 - Category: funding-company-news - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/defense-challenge-2025 - Record: https://forck.live/items/4720-sakana-ai-wins-award-at-us-japan-competition-for-defense-innovation - Subject: Sakana models Sakana AI won the Innovative Spirit Award at the US-Japan Global Innovation Challenge 2025, a defense innovation competition hosted by the US Department of Defense's DIU and Japan's ATLA. The company was the only participant selected as a finalist in both biodefense and disinformation countermeasure themes, out of 60 companies. ### Vercel — AI SDK 4.2 - Company: Vercel (vercel.com) - Announced: 2025-03-21T13:00:00+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/ai-sdk-4-2 - Record: https://forck.live/items/20893-ai-sdk-4-2 - Subject: v0 / AI SDK / AI Gateway Vercel released AI SDK 4.2, adding support for reasoning models across major providers, Model Context Protocol (MCP) clients for connecting to tools, message parts in useChat for handling mixed outputs, and image generation with language models like Google's Gemini 2.0 Flash. ### OpenAI — Early methods for studying affective use and emotional well-being on ChatGPT - Company: OpenAI (openai.com) - Announced: 2025-03-21T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/affective-use-study - Record: https://forck.live/items/600-early-methods-for-studying-affective-use-and-emotional-well-being-on-chatgpt - Subject: GPT / ChatGPT / API OpenAI and MIT Media Lab announce a research collaboration. ### Supabase — Migrating from Fauna to Supabase - Company: Supabase (supabase.com) - Announced: 2025-03-21 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/migrating-from-fauna-to-supabase - Record: https://forck.live/items/21371-migrating-from-fauna-to-supabase - Subject: Supabase Supabase published a migration guide for Fauna users after Fauna announced it will sunset its product by the end of May 2025. The guide outlines a two-phase approach: first exporting data from Fauna and importing it into Supabase as JSONB, then normalizing the data and adopting Supabase features such as authentication, storage, and edge functions. ### Kyutai — MoshiVis: Teaching Moshi to Converse about Images - Company: Kyutai (kyutai.org) - Announced: 2025-03-21 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://kyutai.org/moshivis - Record: https://forck.live/items/18357-moshivis-teaching-moshi-to-converse-about-images - Subject: Moshi / Pocket TTS Kyutai released MoshiVis, an open-source Vision Speech Model that extends Moshi with the ability to discuss visual inputs in real-time speech-to-speech conversations. The model adds 206 million parameters on top of a frozen Moshi, using lightweight cross-attention modules and a gating mechanism to preserve Moshi's original conversational abilities. Kyutai also released the preprint, inference code for PyTorch, Apple's MLX, and Rust, and the model weights. ### Hugging Face — The New and Fresh analytics in Inference Endpoints - Company: Hugging Face (huggingface.co) - Announced: 2025-03-21 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/endpoint-analytics - Record: https://forck.live/items/1727-the-new-and-fresh-analytics-in-inference-endpoints - Subject: Platform Hugging Face announced a refresh of the analytics dashboard for Inference Endpoints, adding real-time metrics, customizable time ranges with auto-refresh, and a replica lifecycle view. ### OpenAI — Personalizing travel at scale with OpenAI - Company: OpenAI (openai.com) - Announced: 2025-03-20T23:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/booking-com - Record: https://forck.live/items/601-personalizing-travel-at-scale-with-openai - Subject: GPT / ChatGPT / API OpenAI's LLMs are integrated by Booking.com to deliver smarter search, faster support, and intent-driven travel experiences. ### Sakana AI — Sakana AI super-powers AI reasoning using Japan’s own Sudoku Puzzles - Company: Sakana AI (sakana.ai) - Announced: 2025-03-20T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/sudoku-bench - Record: https://forck.live/items/4721-sakana-ai-super-powers-ai-reasoning-using-japan-s-own-sudoku-puzzles - Subject: Sakana models Sakana AI releases a new reasoning benchmark based on Sudoku puzzles, including a technical report, leaderboard, and GitHub repository. The benchmark is designed to challenge AI reasoning capabilities with modern Sudoku puzzles that have varied rules. ### Vercel — xAI and Vercel partner to bring zero-friction AI to developers - Company: Vercel (vercel.com) - Announced: 2025-03-20T13:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/xai-and-vercel-partner-to-bring-zero-friction-ai-to-developers - Record: https://forck.live/items/20894-xai-and-vercel-partner-to-bring-zero-friction-ai-to-developers - Subject: v0 / AI SDK / AI Gateway Vercel and xAI have partnered to make Grok models available through the Vercel Marketplace, with a new free tier for prototyping and no additional signup required. Developers can access xAI's models directly from Vercel projects with simplified authentication and integrated billing. The integration also powers the official Next.js AI chatbot template and the AI SDK. ### OpenAI — Introducing next-generation audio models in the API - Company: OpenAI (openai.com) - Announced: 2025-03-20T11:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/introducing-our-next-generation-audio-models - Record: https://forck.live/items/602-introducing-next-generation-audio-models-in-the-api - Subject: GPT / ChatGPT / API Developers can now instruct the text-to-speech model to speak in a specific style, such as 'talk like a sympathetic customer service agent', unlocking new customization for voice agents. ### Supabase — Migrating from the MongoDB Data API to Supabase - Company: Supabase (supabase.com) - Announced: 2025-03-20 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/migrating-mongodb-data-api-with-supabase - Record: https://forck.live/items/21372-migrating-from-the-mongodb-data-api-to-supabase - Subject: Supabase Supabase published a guide for migrating from MongoDB's Data API, which reaches end-of-life on September 30, 2025. The guide covers exporting MongoDB data, importing it into Supabase's Postgres database, normalizing it with SQL, and transitioning API calls to Supabase's automatically generated REST APIs via PostgREST. ### LG AI Research — Meet LG AI Research at NVIDIA GTC 2025! - Company: LG AI Research (lgresearch.ai) - Announced: 2025-03-20 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/news/view?seq=540 - Record: https://forck.live/items/11009-meet-lg-ai-research-at-nvidia-gtc-2025 - Subject: EXAONE LG AI Research is showcasing four AI solutions at NVIDIA GTC 2025: EXAONE Deep, a reasoning model; ChatEXAONE, an enterprise AI assistant with a new 'Deep & Dive' feature; EXAONEPath, a compact histopathology model now open-source; and DDU technology for document understanding. Live demonstrations are available at the LG booth. ### LG AI Research — Meet LG AI Research at NVIDIA GTC 2025! - Company: LG AI Research (lgresearch.ai) - Announced: 2025-03-20 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=540 - Record: https://forck.live/items/4799-meet-lg-ai-research-at-nvidia-gtc-2025 - Subject: EXAONE LG AI Research is showcasing its AI solutions EXAONE Deep, ChatEXAONE, EXAONEPath, and DDU technology at NVIDIA GTC 2025, with live demos and introductions of new features such as ChatEXAONE's 'Deep & Dive'. ### Hugging Face — Open R1: How to use OlympicCoder locally for coding - Company: Hugging Face (huggingface.co) - Announced: 2025-03-20 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/olympic-coder-lmstudio - Record: https://forck.live/items/1728-open-r1-how-to-use-olympiccoder-locally-for-coding - Subject: Platform Setting up OlympicCoder 7B locally using LM Studio and integrating it with VS Code via the Continue extension for coding assistance. ### Stability AI — Introducing Our New Chief Pipeline Architect, Robert Legato - Company: Stability AI (stability.ai) - Announced: 2025-03-19T15:58:53+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://stability.ai/news-updates/introducing-our-new-chief-pipeline-architect-rob-legato - Record: https://forck.live/items/17491-introducing-our-new-chief-pipeline-architect-robert-legato - Subject: Stable Diffusion Stability AI announced that Robert Legato, a visual effects pioneer with three Academy Awards, is joining the company as Chief Pipeline Architect. Legato will align Stability AI's applied research with real VFX industry needs and strengthen collaboration between the company and the VFX industry. ### Lovable — How Mindaugas Built Backchannel with Lovable - Company: Lovable (lovable.dev) - Announced: 2025-03-19T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/from-recruitment-to-innovation-backchannel - Record: https://forck.live/items/12144-how-mindaugas-built-backchannel-with-lovable - Subject: Lovable / AI app builder Mindaugas, founder of Coho, used Lovable to build a job search guidance tool called Backchannel without prior coding experience. He gained paying users within days of sharing a prototype. He plans to rebuild Coho entirely with AI-driven development. ### Hugging Face — AI Policy @🤗: Response to the White House AI Action Plan RFI - Company: Hugging Face (huggingface.co) - Announced: 2025-03-19 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/ai-action-wh-2025 - Record: https://forck.live/items/1729-ai-policy-response-to-the-white-house-ai-action-plan-rfi - Subject: Platform Hugging Face submitted a response to the White House AI Action Plan RFI, advocating for open AI systems and open science, and making three recommendations to recognize open source, prioritize efficiency, and secure AI through transparency. ### Stability AI — Introducing Stable Virtual Camera: Multi-View Video Generation with 3D Camera Control - Company: Stability AI (stability.ai) - Announced: 2025-03-18T16:00:54+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://stability.ai/news-updates/introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera-control - Record: https://forck.live/items/17492-introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera - Subject: Stable Diffusion Stability AI released Stable Virtual Camera, a multi-view diffusion model in research preview that generates 3D videos from one or more input images with user-defined camera trajectories. The model supports up to 1,000 frames, multiple aspect ratios, and 14 dynamic camera paths. It is available under a Non-Commercial License with weights on Hugging Face and code on GitHub. ### OpenAI — EliseAI improves housing and healthcare efficiency with AI - Company: OpenAI (openai.com) - Announced: 2025-03-18T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/eliseai-minna-song - Record: https://forck.live/items/603-eliseai-improves-housing-and-healthcare-efficiency-with-ai - Subject: GPT / ChatGPT / API OpenAI published a conversation with Minna Song, CEO & Co-founder of EliseAI. ### LG AI Research — EXAONE Deep Released ━ Setting a New Standard for Reasoning AI - Company: LG AI Research (lgresearch.ai) - Announced: 2025-03-18 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.lgresearch.ai/blog/view?seq=543 - Record: https://forck.live/items/4800-exaone-deep-released-setting-a-new-standard-for-reasoning-ai - Subject: EXAONE LG AI Research announces EXAONE Deep, a family of reasoning AI models (32B, 7.8B, 2.4B) that excel in math, science, and coding benchmarks, outperforming larger models in several evaluations. ### Hugging Face — Xet is on the Hub - Company: Hugging Face (huggingface.co) - Announced: 2025-03-18 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/xet-on-the-hub - Record: https://forck.live/items/1731-xet-is-on-the-hub - Subject: Platform Hugging Face's Xet Team migrated the first Model and Dataset repositories from LFS to Xet storage, using content-defined chunking to improve upload/download efficiency. The migration shifted ~6% of the Hub's download traffic onto Xet infrastructure, and the post details the technical components and post-migration challenges. ### Hugging Face — NVIDIA's GTC 2025 Announcement for Physical AI Developers: New Open Models and Datasets - Company: Hugging Face (huggingface.co) - Announced: 2025-03-18 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/nvidia-physical-ai - Record: https://forck.live/items/1730-nvidia-s-gtc-2025-announcement-for-physical-ai-developers-new-open-models-and - Subject: Platform NVIDIA announced three open-source releases at GTC 2025: Cosmos Transfer (a world foundation model for controllable virtual world generation), a Physical AI Dataset (15TB of robotics data), and NVIDIA Isaac GR00T N1 (the first open foundation model for humanoid robot reasoning). ### OpenAI — New in ChatGPT for Business: March 2025 - Company: OpenAI (openai.com) - Announced: 2025-03-18 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/business/new-in-chatgpt-for-work-march-updates-2025 - Record: https://forck.live/items/604-new-in-chatgpt-for-business-march-2025 - Subject: GPT / ChatGPT / API OpenAI announces upcoming releases for ChatGPT for Business, highlighting increased interactivity, customization, and agentic capabilities. ### Runway — Gaspar Noé Joins Runway’s 2025 AI Film Festival Jurors - Company: Runway (runway.com) - Announced: 2025-03-17T10:10:10.110000+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://runway.com/news/gaspar-noe-joins-aiff-2025-jurors - Record: https://forck.live/items/7914-gaspar-no-joins-runway-s-2025-ai-film-festival-jurors - Subject: Gen / Aleph Runway announced that filmmaker Gaspar Noé has joined the jury panel for its 2025 AI Film Festival. The submission deadline has been extended to April 13, 2025. ### Mistral AI — Mistral Small 3.1 - Company: Mistral AI (mistral.ai) - Announced: 2025-03-17 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://mistral.ai/news/mistral-small-3-1 - Record: https://forck.live/items/3328-mistral-small-3-1 - Subject: Mistral / Le Chat Mistral AI announces Mistral Small 3.1.