# forck.live — 14–20 April 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-W16, 14–20 April 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-W16 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Cua — Build Your Own Operator on macOS - Part 2 - Company: Cua (cua.ai) - Announced: 2025-04-19 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cua.ai/blog/build-your-own-operator-on-macos-2 - Record: https://forck.live/items/16845-build-your-own-operator-on-macos-part-2 - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua's tutorial walks through setting up the cua-agent framework, a high-level abstraction that automates the Computer-Use Agent control loop — VM interaction, screenshot processing, model communication, and action execution — that Part 1 built manually. It covers agent loop types, local models for cost-effective workflows, and a simple UI for the operator. ### Vercel — Becoming an AI engineering company - Company: Vercel (vercel.com) - Announced: 2025-04-18T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/becoming-an-ai-engineering-company - Record: https://forck.live/items/20883-becoming-an-ai-engineering-company - Subject: v0 / AI SDK / AI Gateway Vercel's CTO describes the company's approach to integrating AI into its existing web framework and application infrastructure business, highlighting the AI SDK and v0 as examples. The post argues that companies can compete with AI giants by leveraging proprietary data, feedback cycles, and domain expertise. It advocates for an iterative development cycle and embedding AI into core application logic rather than adding it as a conversational layer. ### Luma — Introducing Camera Angle Concepts - Company: Luma (lumalabs.ai) - Announced: 2025-04-18 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lumalabs.ai/news/camera-angle-concepts - Record: https://forck.live/items/7988-introducing-camera-angle-concepts - Subject: Ray / Dream Machine Luma released nine Camera Angle Concepts for its Dream Machine product, giving creators prompt-level control over framing such as Low Angle, Over the Shoulder, and POV, composable with existing controls in Ray2. ### LG AI Research — [AAAI-25] ImagePiece: A Novel Content-Aware Re-Tokenization Technique for Semanticizing Tokens in Efficient ViT - Company: LG AI Research (lgresearch.ai) - Announced: 2025-04-18 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=554 - Record: https://forck.live/items/4794-aaai-25-imagepiece-a-novel-content-aware-re-tokenization-technique-for - Subject: EXAONE LG AI Research presented a paper at AAAI-25 introducing ImagePiece, a content-aware re-tokenization technique for Vision Transformers that re-tokenizes non-semantic visual tokens into semantic units, inspired by NLP's WordPiece, to improve efficiency and maintain expressiveness. ### Stability AI — Stable Diffusion Now Optimized for AMD Radeon™ GPUs and Ryzen™ AI APUs - Company: Stability AI (stability.ai) - Announced: 2025-04-16T13:02:52+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://stability.ai/news-updates/stable-diffusion-now-optimized-for-amd-radeon-gpus - Record: https://forck.live/items/17490-stable-diffusion-now-optimized-for-amd-radeon-gpus-and-ryzen-ai-apus - Subject: Stable Diffusion Stability AI released ONNX-optimized versions of Stable Diffusion 3.5 Large, 3.5 Large Turbo, XL 1.0, and XL Turbo for AMD Radeon GPUs and Ryzen AI APUs, achieving up to 2.6x faster inference on SD3.5 models and up to 3.8x faster on SDXL models compared to base PyTorch models. The models are available on Hugging Face with an "_amdgpu" suffix and can be used with Amuse 3.0. ### Hugging Face — Prefill and Decode for Concurrent Requests - Optimizing LLM Performance - Company: Hugging Face (huggingface.co) - Announced: 2025-04-16T10:10:58+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/tngtech/llm-performance-prefill-decode-concurrent-requests - Record: https://forck.live/items/1709-prefill-and-decode-for-concurrent-requests-optimizing-llm-performance - Subject: Platform The impact of concurrent request processing on LLM latency, throughput, and GPU utilization, focusing on the prefill and decode phases. ### OpenAI — OpenAI o3 and o4-mini System Card - Company: OpenAI (openai.com) - Announced: 2025-04-16T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/o3-o4-mini-system-card - Record: https://forck.live/items/581-openai-o3-and-o4-mini-system-card - Subject: GPT / ChatGPT / API OpenAI o3 and o4-mini combine state-of-the-art reasoning with full tool capabilities including web browsing, Python, image and file analysis, image generation, canvas, automations, file search, and memory. ### Replicate — Easel AI is now on Replicate - Company: Replicate (replicate.com) - Announced: 2025-04-16 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replicate.com/blog/easel - Record: https://forck.live/items/2498-easel-ai-is-now-on-replicate - Subject: Platform Replicate announces that two new models from Easel AI—advanced face swap and AI avatars—are now available on its platform. The advanced face swap model replaces faces and bodies while preserving likeness, skin tone, racial features, and gender, and supports single or dual swaps. The AI avatars model generates avatars from a single prompt without requiring LoRAs or training, supporting single and multiplayer modes. Both models are designed for production use in messaging, social, creative, and marketing applications. ### Hugging Face — Introducing HELMET: Holistically Evaluating Long-context Language Models - Company: Hugging Face (huggingface.co) - Announced: 2025-04-16 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/helmet - Record: https://forck.live/items/1712-introducing-helmet-holistically-evaluating-long-context-language-models - Subject: Platform HELMET, a comprehensive benchmark for evaluating long-context language models (LCLMs) that addresses limitations of existing evaluations by providing diverse, controllable, and reliable metrics. The benchmark is presented at ICLR 2025. ### Hugging Face — Cohere on Hugging Face Inference Providers 🔥 - Company: Hugging Face (huggingface.co) - Announced: 2025-04-16 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/inference-providers-cohere - Record: https://forck.live/items/1711-cohere-on-hugging-face-inference-providers - Subject: Platform Cohere is now a supported Inference Provider on Hugging Face Hub, allowing serverless inference for a range of Cohere and Cohere Labs models including Command variants, Aya Expanse, and Aya Vision. ### Hugging Face — 17 Reasons Why Gradio Isn't Just Another UI Library - Company: Hugging Face (huggingface.co) - Announced: 2025-04-16 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/why-gradio-stands-out - Record: https://forck.live/items/1710-17-reasons-why-gradio-isn-t-just-another-ui-library - Subject: Platform 17 features that make Gradio distinct, positioning it as a framework for building interactive AI/ML applications with built-in APIs, server-side rendering, queue management, streaming, and multi-page support. ### Lovable — How Lovable Empowered Grammy Nominated Music Producer and Filmmaker Mike Burns to Build an AI Filmmaking Studio Platform - Company: Lovable (lovable.dev) - Announced: 2025-04-15T16:03:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/mike-burns-ai-filmmaking-studio-platform - Record: https://forck.live/items/12142-how-lovable-empowered-grammy-nominated-music-producer-and-filmmaker-mike-burns - Subject: Lovable / AI app builder Lovable's case study describes how Hollywood filmmaker Mike Burns, who has no coding experience, used Lovable to build MVP (Movipods Virtual Producer), an AI-powered filmmaking platform that integrates AI-generated video, audio, and storytelling tools. The platform is currently in closed beta with production companies and top AI creatives, and a broader launch is planned. ### Vercel — Update on Spain and LALIGA blocks of the internet - Company: Vercel (vercel.com) - Announced: 2025-04-15T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/update-on-spain-and-laliga-blocks-of-the-internet - Record: https://forck.live/items/20885-update-on-spain-and-laliga-blocks-of-the-internet - Subject: v0 / AI SDK / AI Gateway Vercel reports that a Spanish court order allows LALIGA to block IP addresses associated with unauthorized football streaming, without distinguishing between infringing and non-infringing services. Spanish ISPs are now blocking entire IP ranges, affecting legitimate websites hosted on Vercel, Cloudflare, GitHub Pages, and BunnyCDN. Vercel is working with LALIGA to remove illegal content and exploring mitigation strategies to restore access for Spanish users. ### Vercel — Life of a Vercel request: Application-aware routing - Company: Vercel (vercel.com) - Announced: 2025-04-15T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/life-of-a-request-application-aware-routing - Record: https://forck.live/items/20884-life-of-a-vercel-request-application-aware-routing - Subject: v0 / AI SDK / AI Gateway Vercel describes how its platform uses application-aware routing, where the gateway understands the codebase's routes, layouts, middleware, and functions to make routing decisions. This approach reduces configuration overhead, minimizes latency, and enables instant global propagation for rollouts and rollbacks without DNS propagation or warmup time. ### OpenAI — OpenAI announces nonprofit commission advisors - Company: OpenAI (openai.com) - Announced: 2025-04-15T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/nonprofit-commission-advisors - Record: https://forck.live/items/583-openai-announces-nonprofit-commission-advisors - Subject: GPT / ChatGPT / API OpenAI appoints four new advisors to support its philanthropic efforts. ### Upstage — Document Parse got stronger: Better at forms, rotation, and complex tables - Company: Upstage (upstage.ai) - Announced: 2025-04-15 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.upstage.ai/blog/en/upstage-document-parse-now-supports-rotated-docs-multi-page-tables-and-long-image-processing - Record: https://forck.live/items/18472-document-parse-got-stronger-better-at-forms-rotation-and-complex-tables - Subject: Solar Pro / Solar Mini Upstage updated Document Parse (version document-parse-250404) with better table recognition in form-based documents, multi-page table merging, support for rotated inputs (90°, -90°, 180°), 2× faster processing for long images, HWP file conversion, and improved layout analysis for the ocr=auto mode. ### Upstage — Struggling to process loooooooong document images with Generative AI? - Company: Upstage (upstage.ai) - Announced: 2025-04-15 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.upstage.ai/blog/en/struggling-to-process-loooooooong-document-images-with-generative-ai - Record: https://forck.live/items/18471-struggling-to-process-loooooooong-document-images-with-generative-ai - Subject: Solar Pro / Solar Mini Do you have long document images that you want to process with generative AI but can't find the right solution? You're not alone. ### Amp — How to Build an Agent - Company: Amp (ampcode.com) - Announced: 2025-04-15 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://ampcode.com/notes/how-to-build-an-agent - Record: https://forck.live/items/7850-how-to-build-an-agent - Subject: Amp Amp published a guide on building a code-editing agent using Claude and Go, explaining the process in under 400 lines of code. ### LG AI Research — [AAAI-25] Study on Efficient Time Series Forecasting Model Learning Strategies That Alleviate Variable Redundancy - Company: LG AI Research (lgresearch.ai) - Announced: 2025-04-15 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=552 - Record: https://forck.live/items/4795-aaai-25-study-on-efficient-time-series-forecasting-model-learning-strategies - Subject: EXAONE LG AI Research presented a paper titled 'VarDrop: Enhancing Training Efficiency by Reducing Variate Redundancy in Periodic Time Series Forecasting' at the AAAI-25 main track, introducing a method to reduce computational cost in multivariate time series forecasting by adaptively grouping and sampling redundant variables. ### OpenAI — Our updated Preparedness Framework - Company: OpenAI (openai.com) - Announced: 2025-04-15 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/updating-our-preparedness-framework - Record: https://forck.live/items/584-our-updated-preparedness-framework - Subject: GPT / ChatGPT / API OpenAI announced an updated Preparedness Framework for measuring and protecting against severe harm from frontier AI capabilities. ### Vercel — Migrating Grep from Create React App to Next.js - Company: Vercel (vercel.com) - Announced: 2025-04-14T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/migrating-grep-from-create-react-app-to-next-js - Record: https://forck.live/items/20886-migrating-grep-from-create-react-app-to-next-js - Subject: v0 / AI SDK / AI Gateway Vercel migrated the code search tool Grep from Create React App to Next.js, using React Server Components to improve initial page load speed while preserving the quick navigation feel of a single-page app. The migration involved conditional rendering of the search bar in the header versus the homepage to avoid layout shift, URL-based state management to keep search input in sync across pages, and server-side data fetching with TanStack Query for instant search results. ### Runway — Runway and Fabula Unveil Production Partnership - Company: Runway (runway.com) - Announced: 2025-04-14T10:10:10.110000+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/runway-fabula-partnership - Record: https://forck.live/items/7909-runway-and-fabula-unveil-production-partnership - Subject: Gen / Aleph Runway and Fabula are partnering to integrate Runway's filmmaking tools into Fabula's production pipeline, spanning concept ideation, storyboarding, and VFX workflows. The partnership is already in use by director Cutter Hodierne for his upcoming feature film THE SHEPHERD. ### OpenAI — Introducing GPT-4.1 in the API - Company: OpenAI (openai.com) - Announced: 2025-04-14T10:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://openai.com/index/gpt-4-1 - Record: https://forck.live/items/585-introducing-gpt-4-1-in-the-api - Subject: GPT / ChatGPT / API OpenAI introduced GPT-4.1, a new family of models with improvements in coding, instruction following, and long-context understanding, along with their first nano model. The models are available to developers worldwide starting today. ### Hugging Face — 4M Models Scanned: Protect AI + Hugging Face 6 Months In - Company: Hugging Face (huggingface.co) - Announced: 2025-04-14 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/pai-6-month - Record: https://forck.live/items/1714-4m-models-scanned-protect-ai-hugging-face-6-months-in - Subject: Platform Hugging Face and Protect AI have been partnering for six months to improve model security using Guardian's scanning technology, resulting in over 4 million models scanned and new threat detection modules. ### Hugging Face — Hugging Face to sell open-source robots thanks to Pollen Robotics acquisition 🤖 - Company: Hugging Face (huggingface.co) - Announced: 2025-04-14 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/hugging-face-pollen-robotics-acquisition - Record: https://forck.live/items/1713-hugging-face-to-sell-open-source-robots-thanks-to-pollen-robotics-acquisition - Subject: Platform Hugging Face acquires Pollen Robotics to sell open-source robots, starting with Reachy 2.