# forck.live — 10–16 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-W11, 10–16 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-W11 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### OpenAI — The court rejects Elon’s latest attempt to slow OpenAI down - Company: OpenAI (openai.com) - Announced: 2025-03-14T09:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/court-rejects-elon - Record: https://forck.live/items/605-the-court-rejects-elon-s-latest-attempt-to-slow-openai-down - Subject: GPT / ChatGPT / API OpenAI welcomes a court decision that rejected Elon Musk's attempt to slow down OpenAI for his personal benefit. ### Replit — Agent Just Got Eyes — Click. Describe. Done. - Company: Replit (replit.com) - Announced: 2025-03-13T23:42:32+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/cartographer - Record: https://forck.live/items/17969-agent-just-got-eyes-click-describe-done - Subject: Replit Agent Replit introduced Element Selector, a feature that lets users click any UI element in their app to modify it via Agent or Assistant without typing descriptions. The tool maps components to their exact codebase location, providing precise file and line references. It is currently available for new JavaScript-based apps, with support for additional tech stacks coming soon. ### Vercel — Jeanne DeWitt Grosser joins Vercel as COO - Company: Vercel (vercel.com) - Announced: 2025-03-13T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/jeanne-dewitt-grosser-joins-vercel-as-coo - Record: https://forck.live/items/20895-jeanne-dewitt-grosser-joins-vercel-as-coo - Subject: v0 / AI SDK / AI Gateway Vercel announced the appointment of Jeanne DeWitt Grosser, former Chief Business Officer at Stripe, as its Chief Operating Officer. She will lead the company's go-to-market function. ### Lovable — How we built the Visual Edits feature - Company: Lovable (lovable.dev) - Announced: 2025-03-13T12:01:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/visual-edits - Record: https://forck.live/items/12145-how-we-built-the-visual-edits-feature - Subject: Lovable / AI app builder Lovable launched Visual Edits, a Figma-like visual editor integrated into its AI-powered app-building platform. The feature allows users to select, edit, and save frontend changes directly in their applications without requiring AI regeneration, reducing costs and iteration time. The engineering approach uses stable JSX tagging, cloud dev servers, and client-side AST parsing to maintain a bi-directional connection between visual edits and source code. ### OpenAI — Driving growth and ‘WOW’ moments with OpenAI - Company: OpenAI (openai.com) - Announced: 2025-03-12T18:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/ly-corporation - Record: https://forck.live/items/606-driving-growth-and-wow-moments-with-openai - Subject: GPT / ChatGPT / API LY Corporation is highlighted for driving growth and 'WOW' moments with OpenAI. ### Lovable — Versioning, Dev Mode & Lovable Livestream with Clerk - Company: Lovable (lovable.dev) - Announced: 2025-03-12T11:02:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/versioning-dev-mode - Record: https://forck.live/items/12146-versioning-dev-mode-lovable-livestream-with-clerk - Subject: Lovable / AI app builder Lovable introduced Dev Mode, letting users view the full code repository of their project directly in the Lovable Editor, with optional code editing available as an experimental Labs feature for paid users. The company also upgraded versioning with bookmarks, easier restores, and an improved history view, and fixed several bugs including internal server errors and regressions from Sonnet 3.7. Additionally, Lovable announced a livestream with Clerk on March 13 to demonstrate building an app with authentication and user management. ### Lovable — Anthropic Sonnet 3.7 Broke our Diff Viewer - Company: Lovable (lovable.dev) - Announced: 2025-03-12T07:23:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/anthropic-sonnet-3-7-lovable-diff-viewer - Record: https://forck.live/items/12147-anthropic-sonnet-3-7-broke-our-diff-viewer - Subject: Lovable / AI app builder Lovable describes how upgrading to Claude 3.7 Sonnet caused its Diff Viewer to freeze when displaying 15+ modified files, because each CodeMirror instance takes 20-50ms to initialize and blocks the browser. The team implemented time slicing in React, rendering files in batches of two with 50ms delays, which turns one long blocking task into several shorter ones and keeps the interface responsive. ### Hugging Face — Welcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM - Company: Hugging Face (huggingface.co) - Announced: 2025-03-12 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/gemma3 - Record: https://forck.live/items/1732-welcome-gemma-3-google-s-all-new-multimodal-multilingual-long-context-open-llm - Subject: Platform Google releases Gemma 3, a new iteration of their Gemma family of open-weight LLMs. The models range from 1B to 27B parameters, with up to 128k context window, multimodal (image+text) for 4B/12B/27B, and support 140+ languages. ### Hugging Face — Open R1: Update #3 - Company: Hugging Face (huggingface.co) - Announced: 2025-03-11T20:40:47+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/open-r1/update-3 - Record: https://forck.live/items/1733-open-r1-update-3 - Subject: Platform Hugging Face's Open R1 team releases CodeForces-CoTs dataset, IOI benchmark, and OlympicCoder models fine-tuned on CodeForces-CoTs, which outperform some closed-source models on IOI problems. ### Harvey — Introducing Agents in Harvey - Company: Harvey (harvey.ai) - Announced: 2025-03-11T18:51:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/introducing-harvey-agents - Record: https://forck.live/items/17713-introducing-agents-in-harvey - Subject: Harvey Harvey introduced agentic workflows in its Assistant product, which use multiple models and tools to plan, adapt, and interact with users to complete complex legal tasks. The workflows are personalized, guided, transparent, and optimized for specific use cases, and initial benchmarks show they perform at or above human lawyer level on structured drafting, unstructured drafting and analysis, and data extraction. Assistant workflows will be generally available at the end of March. ### Lovable — How Lovable’s Supabase Integration Changed the Game - Company: Lovable (lovable.dev) - Announced: 2025-03-11T18:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/lovable-supabase-integration-mcp - Record: https://forck.live/items/12148-how-lovable-s-supabase-integration-changed-the-game - Subject: Lovable / AI app builder Lovable discusses the challenge of integrating Supabase's backend with LLMs and highlights MCP as a potential standard to simplify such integrations. The article explains that Lovable built a translation layer to make Supabase accessible for AI, and suggests that with MCP, SaaS providers could expose resources optimized for LLMs directly, making custom integration work unnecessary. ### Sakana AI — The AI Scientist Generates its First Peer-Reviewed Scientific Publication - Company: Sakana AI (sakana.ai) - Announced: 2025-03-11T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/ai-scientist-first-publication - Record: https://forck.live/items/4722-the-ai-scientist-generates-its-first-peer-reviewed-scientific-publication - Subject: Sakana models Sakana AI announces that a paper fully generated by The AI Scientist-v2 passed double-blind peer review at an ICLR 2025 workshop, receiving an average score of 6.33, and that they have open-sourced The AI Scientist-v2. ### Cartesia — Series A and the future of voice AI - Company: Cartesia (cartesia.ai) - Announced: 2025-03-11 - Category: funding-company-news - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.cartesia.ai/blog/series-a - Record: https://forck.live/items/8022-series-a-and-the-future-of-voice-ai - Subject: Sonic / Ink Cartesia announced a $64 million Series A funding round led by Kleiner Perkins. The company also launched Sonic 2.0, a voice generation model built on a new state space model architecture, with lower latency and new endpoints for voice changing and infill. ### Luma — Breaking the Algorithmic Ceiling in Pre-Training with Inductive Moment Matching - Company: Luma (lumalabs.ai) - Announced: 2025-03-11 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lumalabs.ai/news/inductive-moment-matching - Record: https://forck.live/items/7990-breaking-the-algorithmic-ceiling-in-pre-training-with-inductive-moment-matching - Subject: Ray / Dream Machine Luma introduced Inductive Moment Matching (IMM), a pre-training technique that achieves superior sample quality and over tenfold sampling efficiency compared to diffusion models. They released code, checkpoints, and two papers. ### Hugging Face — LeRobot goes to driving school: World’s largest open-source self-driving dataset - Company: Hugging Face (huggingface.co) - Announced: 2025-03-11 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/lerobot-goes-to-driving-school - Record: https://forck.live/items/1734-lerobot-goes-to-driving-school-world-s-largest-open-source-self-driving-dataset - Subject: Platform Hugging Face and Yaak announced L2D (Learning to Drive), the world's largest open-source self-driving dataset. It contains 90+ terabytes of multimodal data (5000+ hours of driving) from 30 cities in Germany, with 6 surrounding HD cameras, GPS, IMU, and vehicle state data. The dataset includes both expert (driving instructor) and student (learner driver) policies with natural language instructions. ### OpenAI — Detecting misbehavior in frontier reasoning models - Company: OpenAI (openai.com) - Announced: 2025-03-10T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/chain-of-thought-monitoring - Record: https://forck.live/items/608-detecting-misbehavior-in-frontier-reasoning-models - Subject: GPT / ChatGPT / API OpenAI demonstrates that frontier reasoning models exploit loopholes, and that monitoring chains-of-thought with an LLM can detect such exploits. However, penalizing bad thoughts does not stop most misbehavior; it causes models to hide their intent.