# forck.live — 24–30 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-W13, 24–30 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-W13 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Supabase — Postgres Language Server: Initial Release - Company: Supabase (supabase.com) - Announced: 2025-03-29 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://supabase.com/blog/postgres-language-server - Record: https://forck.live/items/21370-postgres-language-server-initial-release - Subject: Supabase Supabase released the initial version of Postgres Language Server, an LSP implementation for Postgres that provides autocompletion, syntax error highlighting, type-checking via EXPLAIN error insights, and a linter. The language server is available for Neovim and as a CLI via GitHub Releases and npm. ### Cartesia — Cartesia Python SDK v2.0.0 - Company: Cartesia (cartesia.ai) - Announced: 2025-03-28 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.cartesia.ai/blog/python-sdk - Record: https://forck.live/items/8020-cartesia-python-sdk-v2-0-0 - Subject: Sonic / Ink Cartesia released v2.0.0 of its Python SDK, improving the developer experience for using Cartesia's AI voice capabilities with Python. The SDK includes async client support, streaming, WebSocket integration, automatic retries, configurable timeouts, and custom httpx client support. ### Replicate — Creative roundup: avatars, lightsabers, and LoRA tricks - Company: Replicate (replicate.com) - Announced: 2025-03-28 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/creative-roundup - Record: https://forck.live/items/2500-creative-roundup-avatars-lightsabers-and-lora-tricks - Subject: Platform This blog post by Replicate highlights recent AI model releases and updates available on the platform, including ShieldGemma 2 for NSFW detection, Hunyuan3D 2Mini for 3D generation, CSM-1B and Orpheus-3B speech models with human-like quirks, faster Luma video generation, Kling v1.6 Pro with end frame support, and LoRA fine-tuning experiments on Wan2.1 and Flux. ### Hugging Face — 🚀 Accelerating LLM Inference with TGI on Intel Gaudi - Company: Hugging Face (huggingface.co) - Announced: 2025-03-28 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/intel-gaudi-backend-for-tgi - Record: https://forck.live/items/1723-accelerating-llm-inference-with-tgi-on-intel-gaudi - Subject: Platform Hugging Face announces native integration of Intel Gaudi hardware support into Text Generation Inference (TGI), enabling deployment of LLMs on Gaudi accelerators with features like multi-card inference, vision-language models, and FP8 precision, and supporting models such as Llama 3.1, Mixtral, Mistral, and more. ### Alibaba — QVQ-Max: Think with Evidence - Company: Alibaba (alibaba.com) - Announced: 2025-03-27T16:00:04+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qvq-max-preview/ - Record: https://forck.live/items/2291-qvq-max-think-with-evidence - Subject: Qwen Alibaba's Qwen team officially released the first version of QVQ-Max, a visual reasoning model that can understand and reason with content from images and videos. ### OpenAI — Moving from intent-based bots to proactive AI agents - Company: OpenAI (openai.com) - Announced: 2025-03-27T09:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/zendesk - Record: https://forck.live/items/594-moving-from-intent-based-bots-to-proactive-ai-agents - Subject: GPT / ChatGPT / API OpenAI describes moving from intent-based bots to proactive AI agents. ### Replit — Secure Vibe Coding Made Simple - Company: Replit (replit.com) - Announced: 2025-03-26T21:21:54+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replit.com/blog/secure-vibe-coding-made-simple - Record: https://forck.live/items/17967-secure-vibe-coding-made-simple - Subject: Replit Agent Replit published a guide covering five security fundamentals for building on its platform: version control, encrypted storage, database security, Google Cloud infrastructure, and enterprise features. The post also provides a checklist of 16 best practices for secure vibe coding. ### Replit — What is Vibe Coding? - Company: Replit (replit.com) - Announced: 2025-03-26T19:11:36+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/what-is-vibe-coding - Record: https://forck.live/items/17961-what-is-vibe-coding - Subject: Replit Agent From idea to app, fast—that's always been our mission at Replit. But what if you could turn your ideas into working apps without knowing how to code at all? ### Hugging Face — Open R1: Update #4 - Company: Hugging Face (huggingface.co) - Announced: 2025-03-26T18:47:29+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/open-r1/update-4 - Record: https://forck.live/items/1724-open-r1-update-4 - Subject: Platform An updated version of DeepSeek-V3 (DeepSeek-V3-0324) has been released, featuring improvements in instruction following, code and math capabilities, and now using an MIT license. The model shows strong benchmark gains. ### Alibaba — Qwen2.5 Omni: See, Hear, Talk, Write, Do It All! - Company: Alibaba (alibaba.com) - Announced: 2025-03-26T16:00:45+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qwen2.5-omni/ - Record: https://forck.live/items/2292-qwen2-5-omni-see-hear-talk-write-do-it-all - Subject: Qwen Alibaba releases Qwen2.5-Omni, a new end-to-end multimodal model that processes text, images, audio, and video and generates real-time streaming responses via text and speech. ### OpenAI — Security on the path to AGI - Company: OpenAI (openai.com) - Announced: 2025-03-26T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/security-on-the-path-to-agi - Record: https://forck.live/items/595-security-on-the-path-to-agi - Subject: GPT / ChatGPT / API OpenAI describes its proactive approach to integrating security measures into its infrastructure and models. ### Krea — Announcing Krea Ship Week - Company: Krea (krea.ai) - Announced: 2025-03-26 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.krea.ai/blog/ship-week - Record: https://forck.live/items/15097-announcing-krea-ship-week - Subject: Krea 2 / AI creative platform At Krea, we are proud of being the fastest to bring the most powerful AI research into delightful creative tools. We embrace that AI changes every week — so do we. ### Hugging Face — Training and Finetuning Reranker Models with Sentence Transformers - Company: Hugging Face (huggingface.co) - Announced: 2025-03-26 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/train-reranker - Record: https://forck.live/items/1725-training-and-finetuning-reranker-models-with-sentence-transformers - Subject: Platform To use the Sentence Transformers library to finetune reranker (cross-encoder) models, including dataset loading, loss functions, training arguments, evaluators, and the trainer class. It also presents two finetuned reranker models that outperform existing public reranker models on the author's evaluation dataset. ### Vercel — Postmortem on Next.js Middleware bypass - Company: Vercel (vercel.com) - Announced: 2025-03-25T13:00:00+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/postmortem-on-next-js-middleware-bypass - Record: https://forck.live/items/20892-postmortem-on-next-js-middleware-bypass - Subject: v0 / AI SDK / AI Gateway Vercel published a postmortem on CVE-2025-29927, a critical severity vulnerability in Next.js that allowed bypassing Middleware via an internal header. The vulnerability was disclosed on February 27, 2025, and patches were released for Next.js 14.2.25 and 15.2.3 on March 17, 2025, with backports for 13.5.9 and 12.3.5 later. Vercel confirmed that static exports, Vercel's platform, Netlify, and Cloudflare Workers were not affected due to decoupled routing. ### OpenAI — Introducing 4o Image Generation - Company: OpenAI (openai.com) - Announced: 2025-03-25T11:05:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/introducing-4o-image-generation - Record: https://forck.live/items/596-introducing-4o-image-generation - Subject: GPT / ChatGPT / API OpenAI has integrated a new, advanced image generation capability into GPT-4o, making it the most advanced image generator they have built yet. ### OpenAI — Addendum to GPT-4o System Card: 4o image generation - Company: OpenAI (openai.com) - Announced: 2025-03-25T11:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/gpt-4o-image-generation-system-card-addendum - Record: https://forck.live/items/597-addendum-to-gpt-4o-system-card-4o-image-generation - Subject: GPT / ChatGPT / API OpenAI announces 4o image generation, a new image generation approach that is more capable than DALL·E 3, can create photorealistic output, and can take images as inputs and transform them. ### OpenAI — Automating 90% of finance and legal work with agents - Company: OpenAI (openai.com) - Announced: 2025-03-25T10:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/hebbia - Record: https://forck.live/items/598-automating-90-of-finance-and-legal-work-with-agents - Subject: GPT / ChatGPT / API Hebbia's deep research, powered by OpenAI, automates 90% of finance and legal work. ### Upstage — シン・大規模言語モデル | The Next-Generation LLM for Japanese Enterprise - Company: Upstage (upstage.ai) - Announced: 2025-03-25 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.upstage.ai/news/introducing-upstage-japan - Record: https://forck.live/items/18436-the-next-generation-llm-for-japanese-enterprise - Title in English (translated by forck.live from ja): Syn: The Next-Generation LLM for Japanese Enterprise - Subject: Solar Pro / Solar Mini Upstage and Karakuri Inc. jointly released Syn, a Japanese language model with fewer than 14 billion parameters, which achieved top-tier benchmark scores on the Nejumi Leaderboard. The model is designed for enterprise use, balancing cost optimization and high accuracy, and is available on AWS Marketplace from March 25. Upstage also established a Japanese subsidiary, Upstage AI Co., Ltd., on March 14, 2025. ### Cartesia — Cartesia Named to 7th Annual Enterprise Tech 30 List Presented by Wing Venture Capital - Company: Cartesia (cartesia.ai) - Announced: 2025-03-25 - Category: funding-company-news - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.cartesia.ai/blog/enterprise-tech-30 - Record: https://forck.live/items/8021-cartesia-named-to-7th-annual-enterprise-tech-30-list-presented-by-wing-venture - Subject: Sonic / Ink Cartesia was named to the seventh annual Enterprise Tech 30 list by Wing Venture Capital, which recognizes the most promising private enterprise tech companies across all stages of maturity. ### OpenAI — Leadership updates - Company: OpenAI (openai.com) - Announced: 2025-03-24T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/leadership-updates-march-2025 - Record: https://forck.live/items/599-leadership-updates - Subject: GPT / ChatGPT / API OpenAI announces that it has grown and now delivers products used by hundreds of millions of people. ### Hugging Face — Introducing Gradio's new Dataframe! - Company: Hugging Face (huggingface.co) - Announced: 2025-03-24 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/gradio-dataframe-upgrade - Record: https://forck.live/items/1726-introducing-gradio-s-new-dataframe - Subject: Platform Hugging Face announced a major update to Gradio's Dataframe component, adding ten new features such as multi-cell selection, row numbers, column pinning, copy and full-screen buttons, scroll-to-top, accessibility improvements, row/column selection, static columns, search, and filter functionality.