# forck.live — 28 April – 4 May 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-W18, 28 April – 4 May 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-W18 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Vercel — How v0 is building SEO-optimized sites by default - Company: Vercel (vercel.com) - Announced: 2025-05-02T13:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-v0-is-building-seo-optimized-sites-by-default - Record: https://forck.live/items/20880-how-v0-is-building-seo-optimized-sites-by-default - Subject: v0 / AI SDK / AI Gateway Vercel's v0 now generates SEO-optimized sites by default, integrating with Next.js to provide structured metadata, performance tuning, and server-side rendering. The tool automatically applies optimizations for Core Web Vitals, metadata, mobile responsiveness, and accessibility, and includes built-in analytics via Vercel Analytics and Speed Insights. ### OpenAI — Expanding on what we missed with sycophancy - Company: OpenAI (openai.com) - Announced: 2025-05-02T08:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/expanding-on-sycophancy - Record: https://forck.live/items/574-expanding-on-what-we-missed-with-sycophancy - Subject: GPT / ChatGPT / API OpenAI provides a deeper dive on their findings regarding sycophancy, discusses what went wrong, and outlines future changes they are making. ### Vercel — iOS developers can now offer commission-free payments on web - Company: Vercel (vercel.com) - Announced: 2025-05-01T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/ios-developers-can-now-offer-commission-free-payments-on-web - Record: https://forck.live/items/20881-ios-developers-can-now-offer-commission-free-payments-on-web - Subject: v0 / AI SDK / AI Gateway A federal court ruled that Apple violated a 2021 injunction by restricting developers from linking to external payment methods and by imposing a 27% fee on those transactions. The ruling confirms developers can guide users to external payment pages without Apple's tax or obstruction, ending the so-called 'Apple Tax' on off-platform payments. Vercel released a starter kit for using its platform to build web-based checkout flows in native apps. ### Cua — Training Computer-Use Models: Creating Human Trajectories with Cua - Company: Cua (cua.ai) - Announced: 2025-05-01 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cua.ai/blog/training-computer-use-models-trajectories-1 - Record: https://forck.live/items/16844-training-computer-use-models-creating-human-trajectories-with-cua - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua published a tutorial on using its Computer-Use Interface (CUI) with Gradio UI to record, tag, and share human-generated computer interaction trajectories for training computer-use AI models. ### Suno — Introducing v4.5 - Company: Suno (suno.com) - Announced: 2025-05-01 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://suno.com/blog/introducing-v4-5 - Record: https://forck.live/items/10335-introducing-v4-5 - Subject: Music / Studio Suno released v4.5, its newest and most expressive music model, featuring expanded genres and genre mashups, enhanced vocal range and emotional depth, more complex sounds, smarter prompt interpretations, a prompt enhancement helper, upgraded Covers and Personas, faster generation, extended song length up to 8 minutes, and improved audio quality. ### Kyutai — Helium 1: a modular and multilingual LLM - Company: Kyutai (kyutai.org) - Announced: 2025-04-30 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://kyutai.org/blog/2025-04-30-helium - Record: https://forck.live/items/18356-helium-1-a-modular-and-multilingual-llm - Subject: Moshi / Pocket TTS Kyutai announced Helium 1, a 2B parameter modular and multilingual language model, which is open-sourced for reproducibility. ### Hugging Face — The 4 Things Qwen-3’s Chat Template Teaches Us - Company: Hugging Face (huggingface.co) - Announced: 2025-04-30 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/qwen-3-chat-template-deep-dive - Record: https://forck.live/items/1703-the-4-things-qwen-3-s-chat-template-teaches-us - Subject: Platform The blog post analyzes the new Qwen-3 chat template, highlighting four key improvements: optional reasoning via an enable_thinking flag, dynamic context management with a rolling checkpoint system, better tool argument serialization to avoid double-escaping, and removal of the default system prompt. These changes provide greater flexibility and reliability, especially for agentic workflows. ### Hugging Face — How to Build an MCP Server with Gradio - Company: Hugging Face (huggingface.co) - Announced: 2025-04-30 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/gradio-mcp - Record: https://forck.live/items/1702-how-to-build-an-mcp-server-with-gradio - Subject: Platform To use Gradio to build an MCP (Model Context Protocol) server for LLMs, allowing any Python function to be exposed as a tool that LLMs can call. It includes examples, setup instructions, and advanced features like resources and prompts. ### OpenAI — Sycophancy in GPT-4o: what happened and what we’re doing about it - Company: OpenAI (openai.com) - Announced: 2025-04-29T18:00:00+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://openai.com/index/sycophancy-in-gpt-4o - Record: https://forck.live/items/575-sycophancy-in-gpt-4o-what-happened-and-what-we-re-doing-about-it - Subject: GPT / ChatGPT / API OpenAI rolled back a GPT-4o update in ChatGPT because it was overly flattering and agreeable (sycophantic), reverting users to an earlier version with more balanced behavior. ### Hugging Face — Introducing AutoRound: Intel’s Advanced Quantization for LLMs and VLMs - Company: Hugging Face (huggingface.co) - Announced: 2025-04-29 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/autoround - Record: https://forck.live/items/1705-introducing-autoround-intel-s-advanced-quantization-for-llms-and-vlms - Subject: Platform Intel introduces AutoRound, a new quantization tool for LLMs and VLMs that uses signed gradient descent to optimize weight rounding and clipping, supporting low-bit quantization with minimal accuracy loss. ### Hugging Face — Welcoming Llama Guard 4 on Hugging Face Hub - Company: Hugging Face (huggingface.co) - Announced: 2025-04-29 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/llama-guard-4 - Record: https://forck.live/items/1704-welcoming-llama-guard-4-on-hugging-face-hub - Subject: Platform Meta releases Llama Guard 4, a dense 12B multimodal safety model, along with two new Llama Prompt Guard 2 models. The models are available on Hugging Face Hub with open checkpoints and an interactive notebook. ### Alibaba — Qwen3: Think Deeper, Act Faster - Company: Alibaba (alibaba.com) - Announced: 2025-04-28T20:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qwen3/ - Record: https://forck.live/items/2290-qwen3-think-deeper-act-faster - Subject: Qwen Alibaba announces the release of the Qwen3 family of large language models, including Qwen3-235B-A22B, Qwen3-30B-A3B, and Qwen3-4B, with competitive benchmark results. ### Upstage — Experience stronger Korean fluency in the new Solar Pro & Mini - Company: Upstage (upstage.ai) - Announced: 2025-04-28 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.upstage.ai/blog/en/experience-stronger-korean-fluency-in-the-new-solar-llms - Record: https://forck.live/items/18470-experience-stronger-korean-fluency-in-the-new-solar-pro-mini - Subject: Solar Pro / Solar Mini Upstage updated Solar Pro and Solar Mini (released Apr 22) with improved Korean fluency. Solar Pro's ko-fluency score rose from 74.4 to 82.0, and Solar Mini's from 81.2 to 82.1. Solar Pro also improved on Ko-MT-Bench from 71.8 to 81.0. ### Genspark — Super Agent Becomes Personalized - Your AI, Now Tailored for You! - Company: Genspark (genspark.ai) - Announced: 2025-04-28 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/genspark-super-agent-personalized - Record: https://forck.live/items/18288-super-agent-becomes-personalized-your-ai-now-tailored-for-you - Subject: Genspark ### LG AI Research — [AAAI-25] Enhanced Anomaly/Out-of-Distribution Detection with Foundation Models - Company: LG AI Research (lgresearch.ai) - Announced: 2025-04-28 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=556 - Record: https://forck.live/items/4792-aaai-25-enhanced-anomaly-out-of-distribution-detection-with-foundation-models - Subject: EXAONE LG AI Research presented a paper at AAAI 2025 on a new outlier generation framework called SONA (Diffusion-based Semantic Outlier Generation via Nuisance Awareness) for out-of-distribution detection, which uses diffusion models to generate semantic outliers and achieves state-of-the-art performance on both near-OOD and far-OOD datasets.