# forck.live — 21–27 October 2024 > 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 2024-W43, 21–27 October 2024, 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/2024-W43 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Vercel — Recap: Next.js Conf 2024 - Company: Vercel (vercel.com) - Announced: 2024-10-25T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/recap-next-js-conf-2024 - Record: https://forck.live/items/20930-recap-next-js-conf-2024 - Subject: v0 / AI SDK / AI Gateway Vercel recapped Next.js Conf 2024, announcing that Turbopack is now stable for development with up to 90% faster code updates, and that Next.js 15 is officially stable and ready for production. The conference also highlighted a new Next.js community GitHub organization with ten deployment templates and a donation of ticket proceeds to open source projects. ### LG AI Research — How AI Researchers Can Create a Better World - Language Lab Hyewon Choi - Company: LG AI Research (lgresearch.ai) - Announced: 2024-10-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=493 - Record: https://forck.live/items/4823-how-ai-researchers-can-create-a-better-world-language-lab-hyewon-choi - Subject: EXAONE An interview with Hyewon Choi, a researcher at LG AI Research's Language Lab, discussing her work on natural language processing and information retrieval, her personal journey into AI research, and her goal of using AI to improve people's lives. ### Sarvam AI — Sarvam 1: The first Indian language LLM - Company: Sarvam AI (sarvam.ai) - Announced: 2024-10-24 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.sarvam.ai/blogs/sarvam-1/ - Record: https://forck.live/items/18404-sarvam-1-the-first-indian-language-llm - Subject: Sarvam / Bulbul / Saaras Sarvam AI introduced Sarvam-1, a 2-billion parameter language model built from the ground up for 10 Indian languages (Bengali, Gujarati, Hindi, Kannada, Malayalam, Marathi, Oriya, Punjabi, Tamil, and Telugu) alongside English. The model uses a custom tokenizer achieving token fertility rates of 1.4–2.1 across supported languages, compared to 4–8 tokens per word for existing multilingual models. Sarvam-1 outperforms Gemma-2-2B and Llama-3.2-3B on benchmarks including MMLU, Arc-Challenge, and IndicGenBench, and achieves similar numbers to Llama 3.1 8B while enabling 4–6x faster inference. The company also released Sarvam-2T, a ~2-trillion-token Indic pretraining corpus with improved document quality and content distribution compared to existing datasets like Sangraha. ### Hugging Face — A Deepdive into Aya Expanse: Advancing the Frontier of Multilinguality - Company: Hugging Face (huggingface.co) - Announced: 2024-10-24 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/aya-expanse - Record: https://forck.live/items/1804-a-deepdive-into-aya-expanse-advancing-the-frontier-of-multilinguality - Subject: Platform Cohere For AI releases the Aya Expanse family of multilingual models (8B and 32B parameters) as open weights, claiming state-of-the-art multilingual performance that outperforms larger models like Llama 3.1 70B and rivals in its class. The blog details technical innovations including data arbitrage, multilingual preference training, safety tuning, and model merging. ### Vercel — Maximizing outputs with v0: From UI generation to code creation - Company: Vercel (vercel.com) - Announced: 2024-10-23T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/maximizing-outputs-with-v0-from-ui-generation-to-code-creation - Record: https://forck.live/items/20932-maximizing-outputs-with-v0-from-ui-generation-to-code-creation - Subject: v0 / AI SDK / AI Gateway Vercel's guide covers prompt engineering best practices for v0, a tool for generating UIs and code with deep integrations with libraries and frameworks like Next.js and React. It explains how to craft detailed prompts to improve UI generation, code generation, and developer support, including tips for using third-party libraries, fetching data, and creating 3D graphics. ### Vercel — What's new in Svelte 5 - Company: Vercel (vercel.com) - Announced: 2024-10-23T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/whats-new-in-svelte-5 - Record: https://forck.live/items/20931-what-s-new-in-svelte-5 - Subject: v0 / AI SDK / AI Gateway Svelte 5 introduces runes for universal reactivity, snippets for reusable markup, and compiler improvements. Runes like $state, $derived, and $effect provide explicit control over application state. Components are now plain JavaScript functions, and native TypeScript support is included. ### OpenAI — Simplifying, stabilizing, and scaling continuous-time consistency models - Company: OpenAI (openai.com) - Announced: 2024-10-23T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/simplifying-stabilizing-and-scaling-continuous-time-consistency-models - Record: https://forck.live/items/678-simplifying-stabilizing-and-scaling-continuous-time-consistency-models - Subject: GPT / ChatGPT / API OpenAI simplified, stabilized, and scaled continuous-time consistency models, achieving sample quality comparable to leading diffusion models with only two sampling steps. ### Hugging Face — CinePile 2.0 - making stronger datasets with adversarial refinement - Company: Hugging Face (huggingface.co) - Announced: 2024-10-23 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/cinepile2 - Record: https://forck.live/items/1807-cinepile-2-0-making-stronger-datasets-with-adversarial-refinement - Subject: Platform Hugging Face and collaborators announce CinePile 2.0, an improved long video QA dataset, and describe their adversarial refinement method for dataset creation. ### Hugging Face — Introducing HUGS - Scale your AI with Open Models - Company: Hugging Face (huggingface.co) - Announced: 2024-10-23 - Category: product-launch - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/hugs - Record: https://forck.live/items/1806-introducing-hugs-scale-your-ai-with-open-models - Subject: Platform Hugging Face announced HUGS, optimized zero-configuration inference microservices for deploying open models in your own infrastructure, with pricing starting at $1 per hour per container on AWS and GCP. ### Hugging Face — Introducing SynthID Text - Company: Hugging Face (huggingface.co) - Announced: 2024-10-23 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/synthid-text - Record: https://forck.live/items/1805-introducing-synthid-text - Subject: Platform Hugging Face and Google DeepMind launch SynthID Text in Transformers v4.46.0, a watermarking technology for AI-generated text that allows applying and detecting watermarks without affecting LLM quality. ### Runway — Introducing Act-One - Company: Runway (runway.com) - Announced: 2024-10-22T16:22:35.195000+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/research/introducing-act-one - Record: https://forck.live/items/7920-introducing-act-one - Subject: Gen / Aleph Runway released Act-One, a tool for generating expressive character performances using video and voice inputs, available inside Gen-3 Alpha. ### Vercel — How Vercel adopted microfrontends - Company: Vercel (vercel.com) - Announced: 2024-10-22T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-vercel-adopted-microfrontends - Record: https://forck.live/items/20934-how-vercel-adopted-microfrontends - Subject: v0 / AI SDK / AI Gateway Vercel adopted vertical microfrontends for its main website, splitting the monolithic Next.js application into three separate apps for marketing, documentation, and the logged-in dashboard. This change resulted in over a 40% improvement in preview build times and local development compilation, along with reduced page weight and improved Core Web Vitals. The migration was done incrementally using Next.js Multi-Zones and feature flags, with shared components maintained in a monorepo. ### Vercel — BNP Paribas Open: Serving up scores and experiences in real time with Work & Co and Vercel - Company: Vercel (vercel.com) - Announced: 2024-10-22T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/bnp-paribas-open-serving-up-scores-and-experiences-in-real-time-with-work - Record: https://forck.live/items/20933-bnp-paribas-open-serving-up-scores-and-experiences-in-real-time-with-work-co - Subject: v0 / AI SDK / AI Gateway Vercel published a case study documenting how Work & Co rebuilt the BNP Paribas Open website and mobile apps on Vercel, replacing separate WordPress and white-labeled systems. The new platform created 83,000 user accounts between launch and tournament end and handled millions of users during the event. ### OpenAI — OpenAI and the Lenfest Institute AI Collaborative and Fellowship program - Company: OpenAI (openai.com) - Announced: 2024-10-22T06:05:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/lenfest-institute - Record: https://forck.live/items/679-openai-and-the-lenfest-institute-ai-collaborative-and-fellowship-program - Subject: GPT / ChatGPT / API OpenAI announces the Lenfest Institute AI Collaborative and Fellowship program. ### Replicate — Stable Diffusion 3.5 is here - Company: Replicate (replicate.com) - Announced: 2024-10-22 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://replicate.com/blog/stable-diffusion-3-5-is-here - Record: https://forck.live/items/2510-stable-diffusion-3-5-is-here - Subject: Platform Stability AI's Stable Diffusion 3.5 text-to-image model is now available on Replicate, offering three variants with improved image quality and prompt understanding. ### Replicate — Ideogram v2 is an outstanding new inpainting model - Company: Replicate (replicate.com) - Announced: 2024-10-22 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replicate.com/blog/ideogram-v2-inpainting - Record: https://forck.live/items/2509-ideogram-v2-is-an-outstanding-new-inpainting-model - Subject: Platform Replicate announces that Ideogram v2 and v2-turbo are now available on its API, with a focus on inpainting capabilities, and provides code examples and usage tips. ### Hugging Face — Deploying Speech-to-Speech on Hugging Face - Company: Hugging Face (huggingface.co) - Announced: 2024-10-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/s2s_endpoint - Record: https://forck.live/items/1812-deploying-speech-to-speech-on-hugging-face - Subject: Platform Deploying the Speech-to-Speech (S2S) pipeline on Hugging Face Inference Endpoints using a custom Docker image. S2S combines Voice Activity Detection, Speech-to-Text, a Language Model, and Text-to-Speech to create a seamless speech interaction experience, with support for multiple languages. ### Hugging Face — Releasing Outlines-core 0.1.0: structured generation in Rust and Python - Company: Hugging Face (huggingface.co) - Announced: 2024-10-22 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/outlines-core - Record: https://forck.live/items/1811-releasing-outlines-core-0-1-0-structured-generation-in-rust-and-python - Subject: Platform Hugging Face and dottxt announce outlines-core 0.1.0, a Rust port of outlines' core algorithms for structured generation, offering 2x speed improvement in index compilation, separation of concerns, and portability. The Python bindings are available. ### Hugging Face — Diffusers welcomes Stable Diffusion 3.5 Large - Company: Hugging Face (huggingface.co) - Announced: 2024-10-22 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/sd3-5 - Record: https://forck.live/items/1810-diffusers-welcomes-stable-diffusion-3-5-large - Subject: Platform Hugging Face announces that Stable Diffusion 3.5 Large is now available in the Diffusers library. The release includes two checkpoints: an 8B parameter model and a timestep-distilled version for few-step inference. The post covers architectural changes (QK normalization, dual attention layers), how to use the models with Diffusers for inference and training LoRAs with quantization, and notes that the model is gated and requires acceptance. ### Hugging Face — Transformers.js v3: WebGPU Support, New Models & Tasks, and More… - Company: Hugging Face (huggingface.co) - Announced: 2024-10-22 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/transformersjs-v3 - Record: https://forck.live/items/1809-transformers-js-v3-webgpu-support-new-models-tasks-and-more - Subject: Platform Hugging Face announces the release of Transformers.js v3, featuring WebGPU support, new quantization formats (dtypes), 120 supported architectures, 25 new example projects, over 1200 pre-converted models on the Hub, and compatibility with Node.js (ESM + CJS), Deno, and Bun. ### Hugging Face — Hugging Face Teams Up with Protect AI: Enhancing Model Security for the ML Community - Company: Hugging Face (huggingface.co) - Announced: 2024-10-22 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/protectai - Record: https://forck.live/items/1808-hugging-face-teams-up-with-protect-ai-enhancing-model-security-for-the-ml - Subject: Platform Hugging Face announces a partnership with Protect AI to integrate Guardian, a security scanner, into the Hugging Face Hub to automatically scan all public model repositories for vulnerabilities. ### Harvey — Connecting Cuatrecasas’ Knowledge Base to Harvey - Company: Harvey (harvey.ai) - Announced: 2024-10-21T17:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.harvey.ai/blog/connecting-cuatrecasas-knowledge-base-to-harvey - Record: https://forck.live/items/17722-connecting-cuatrecasas-knowledge-base-to-harvey - Subject: Harvey Harvey detailed how it enabled Cuatrecasas to securely and privately leverage their firm's internal data within Harvey via a model exclusive to Cuatrecasas. The integration includes a logically separated datastore accessible only by Cuatrecasas users and a model trained to keep data private. Since deployment, daily usage of Harvey at Cuatrecasas has increased significantly. ### Hugging Face — “Llama 3.2 in Keras” - Company: Hugging Face (huggingface.co) - Announced: 2024-10-21 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/keras-llama-32 - Record: https://forck.live/items/1813-llama-3-2-in-keras - Subject: Platform Llama 3.2 models are already supported in Keras, as Keras Llama3 can load from any standard Hugging Face checkpoint, including Llama 3.2 checkpoints. It provides a code example and discusses Keras features like multi-backend, tokenizer, and fine-tuning.