# forck.live — 7–13 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-W41, 7–13 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-W41 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Replit — Shell2: 200× faster, persisted, multiplayer-native Shells - Company: Replit (replit.com) - Announced: 2024-10-11T23:55:14+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/shell2 - Record: https://forck.live/items/17882-shell2-200-faster-persisted-multiplayer-native-shells - Subject: Replit Agent Replit introduced Shell2, a new shell backend that is approximately 200 times faster than the previous implementation, with persistent and multiplayer-native shells. The performance improvement comes from technical changes such as eliminating string-byte conversions, avoiding stdio, and reducing protocol introspection. The new system also enables shells to be shared and persisted across users in a multiplayer environment. ### LG AI Research — Projection Regret: Reducing Background Bias for Novelty Detection via Diffusion Models - Company: LG AI Research (lgresearch.ai) - Announced: 2024-10-11 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=491 - Record: https://forck.live/items/4824-projection-regret-reducing-background-bias-for-novelty-detection-via-diffusion - Subject: EXAONE LG AI Research introduces Projection Regret, a method for novelty detection using diffusion models that reduces background bias by measuring semantic distance between original and diffusion-edited images. ### Vercel — How Emburse increased site performance by 4x with Vercel - Company: Vercel (vercel.com) - Announced: 2024-10-10T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-emburse-increased-site-performance-by-4x-with-vercel - Record: https://forck.live/items/20938-how-emburse-increased-site-performance-by-4x-with-vercel - Subject: v0 / AI SDK / AI Gateway Emburse, a travel and expense management company serving over 12 million users in 120 countries, modernized its legacy stack by adopting Vercel, Next.js, and Sanity with the help of consultancy Rangle. The new platform launched 1,550 pages, achieved Interaction to Next Paint (INP) below 200ms, and loads 4x faster than the previous site, outperforming 99% of global websites. The marketing team can now independently manage and update content in multiple languages without developer assistance. ### Vercel — Add 3D to your web projects with v0 and React Three Fiber - Company: Vercel (vercel.com) - Announced: 2024-10-10T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/add-3d-to-your-web-projects-with-v0-and-react-three-fiber - Record: https://forck.live/items/20937-add-3d-to-your-web-projects-with-v0-and-react-three-fiber - Subject: v0 / AI SDK / AI Gateway Vercel's guide explains how to build 3D scenes on the web using React Three Fiber (R3F) with v0, its AI-powered development assistant. It covers scene organization and provides example prompts for creating interactive 3D designs through natural-language chat with v0. ### OpenAI — MLE-bench: Evaluating Machine Learning Agents on Machine Learning Engineering - Company: OpenAI (openai.com) - Announced: 2024-10-10T10:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/mle-bench - Record: https://forck.live/items/681-mle-bench-evaluating-machine-learning-agents-on-machine-learning-engineering - Subject: GPT / ChatGPT / API Introduction of MLE-bench, a benchmark for evaluating machine learning agents on machine learning engineering tasks. ### Supabase — Live Share: Connect to in-browser PGlite with any Postgres client - Company: Supabase (supabase.com) - Announced: 2024-10-10 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://supabase.com/blog/database-build-live-share - Record: https://forck.live/items/21393-live-share-connect-to-in-browser-pglite-with-any-postgres-client - Subject: Supabase Supabase added Live Share to database.build, enabling users to connect to in-browser PGlite databases from external Postgres clients via a unique connection string. The feature uses a websocket-to-tcp proxy and pg-gateway to relay Postgres wire protocol messages. PGlite has a single-connection limit, so only one client can connect at a time. ### Replicate — FLUX is fast and it's open source - Company: Replicate (replicate.com) - Announced: 2024-10-10 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replicate.com/blog/flux-is-fast-and-open-source - Record: https://forck.live/items/2511-flux-is-fast-and-it-s-open-source - Subject: Platform Replicate has optimized FLUX.1 schnell and FLUX.1 dev models to run significantly faster, and has open-sourced the optimization code. The optimizations include using torch.compile, fast CuDNN attention kernels, and a new synchronous HTTP API. Users can disable optimizations via the go_fast input. ### Hugging Face — A Security Review of Gradio 5 - Company: Hugging Face (huggingface.co) - Announced: 2024-10-10 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/gradio-5-security - Record: https://forck.live/items/1816-a-security-review-of-gradio-5 - Subject: Platform Hugging Face announces a security audit of Gradio by Trail of Bits, detailing vulnerabilities fixed in Gradio 5.0. ### Hugging Face — Introducing the AMD 5th Gen EPYC™ CPU - Company: Hugging Face (huggingface.co) - Announced: 2024-10-10 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/huggingface-amd-turin - Record: https://forck.live/items/1815-introducing-the-amd-5th-gen-epyc-cpu - Subject: Platform Hugging Face announces support for AMD's 5th Gen EPYC CPU (Turin) and presents benchmark results showing ~2x throughput improvement over the previous Genoa CPU. ### Vercel — How Chatbase scaled rapidly with Vercel's developer experience and AI SDK - Company: Vercel (vercel.com) - Announced: 2024-10-09T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-chatbase-scaled-rapidly-with-vercels-developer-experience-and-ai-sdk - Record: https://forck.live/items/20940-how-chatbase-scaled-rapidly-with-vercel-s-developer-experience-and-ai-sdk - Subject: v0 / AI SDK / AI Gateway Chatbase, a platform for building chat-based AI agents, scaled to 500K monthly visitors and $4M ARR in 1.5 years using Vercel's AI SDK, Next.js, and Vercel's infrastructure. The AI SDK enabled rapid implementation of custom chats and model optimizations, including an LLM Compare Mode for users to test different configurations. ### Vercel — Leveraging Vercel and the AI SDK to deliver a seamless, AI-powered experience as a solo founder - Company: Vercel (vercel.com) - Announced: 2024-10-09T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/leveraging-vercel-and-the-ai-sdk-to-deliver-a-seamless-ai-powered-experience - Record: https://forck.live/items/20939-leveraging-vercel-and-the-ai-sdk-to-deliver-a-seamless-ai-powered-experience - Subject: v0 / AI SDK / AI Gateway Vercel published a case study on ChatPRD, an AI co-pilot for product managers built by solo founder Claire Vo. ChatPRD uses GPT-4 Omni and the Vercel AI SDK to implement streaming responses and streaming UI components, which reduced perceived latency and improved user satisfaction. The platform reached 20,000 active users within nine months of launch. ### Supabase — MongoDB Realm & Device Sync Alternatives - Supabase - Company: Supabase (supabase.com) - Announced: 2024-10-09 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/mongodb-realm-and-device-sync-alternatives - Record: https://forck.live/items/21394-mongodb-realm-device-sync-alternatives-supabase - Subject: Supabase Supabase published a guide listing five alternatives to MongoDB Realm for offline-first and real-time data sync: Legend-State, WatermelonDB, PowerSync, Replicache, and ElectricSQL. Each tool integrates with Supabase to handle offline data and synchronization. ### Hugging Face — Scaling AI-based Data Processing with Hugging Face + Dask - Company: Hugging Face (huggingface.co) - Announced: 2024-10-09 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/dask-scaling - Record: https://forck.live/items/1818-scaling-ai-based-data-processing-with-hugging-face-dask - Subject: Platform Scaling AI-based data processing using Hugging Face and Dask, showing how to process 100 rows locally with pandas and then scale to 211 million rows with Dask across multiple GPUs on the cloud, using the FineWeb dataset and FineWeb-Edu classifier. ### Hugging Face — Welcome, Gradio 5 - Company: Hugging Face (huggingface.co) - Announced: 2024-10-09 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/gradio-5 - Record: https://forck.live/items/1817-welcome-gradio-5 - Subject: Platform Hugging Face announces the stable release of Gradio 5, a major update to the open-source Python library for building machine learning web applications. The release focuses on performance improvements, modernized design, low-latency streaming, and an experimental AI Playground for generating apps. It is intended to make Gradio production-ready while maintaining a simple API. ### Replit — From localhost to live - Company: Replit (replit.com) - Announced: 2024-10-08T17:38:49+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/from-localhost-to-live - Record: https://forck.live/items/17974-from-localhost-to-live - Subject: Replit Agent Replit introduced the Replit Agent, an LLM interface that creates full development environments, installs languages and packages, adds services, and deploys apps. The Agent leverages Replit's deployment parity, where the development environment is identical to what gets deployed. ### OpenAI — OpenAI and Hearst Content Partnership - Company: OpenAI (openai.com) - Announced: 2024-10-08T10:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/hearst - Record: https://forck.live/items/682-openai-and-hearst-content-partnership - Subject: GPT / ChatGPT / API OpenAI announces a partnership with Hearst, bringing curated lifestyle and local news content from Hearst's brands to OpenAI's products. ### Supabase — Building offline-first mobile apps with Supabase, Flutter and Brick - Company: Supabase (supabase.com) - Announced: 2024-10-08 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/offline-first-flutter-apps - Record: https://forck.live/items/21395-building-offline-first-mobile-apps-with-supabase-flutter-and-brick - Subject: Supabase Supabase's guide explains how to build offline-first mobile apps using Brick, a Flutter data manager that syncs data between Supabase and local caches like SQLite. Brick handles querying, uploading, and offline-first data parity, allowing apps to function with or without connectivity. The guide covers setup, code generation, repository usage, queries, reactivity, upserting, and testing. ### Cartesia — Introducing our next 8 languages on Sonic Multilingual - Company: Cartesia (cartesia.ai) - Announced: 2024-10-08 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.cartesia.ai/blog/sonic-multilingual-next-8-languages - Record: https://forck.live/items/8029-introducing-our-next-8-languages-on-sonic-multilingual - Subject: Sonic / Ink Cartesia announced the alpha release of 8 new languages for Sonic Multilingual, expanding total language support to 15. The model, built on state space model architecture, offers low-latency, lifelike speech generation across languages. ### Hugging Face — Faster Assisted Generation with Dynamic Speculation - Company: Hugging Face (huggingface.co) - Announced: 2024-10-08 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/dynamic_speculation_lookahead - Record: https://forck.live/items/1819-faster-assisted-generation-with-dynamic-speculation - Subject: Platform Hugging Face and Intel labs announce dynamic speculative decoding, a new method for accelerating text generation up to 2.7x, which becomes the default in Transformers version 4.45.0. The method adjusts the number of draft tokens based on confidence, outperforming the heuristic approach. ### Harvey — BigLaw Bench: Hallucinations - Company: Harvey (harvey.ai) - Announced: 2024-10-07T17:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/biglaw-bench-hallucinations - Record: https://forck.live/items/17723-biglaw-bench-hallucinations - Subject: Harvey Harvey published a benchmark evaluating hallucinations on legal tasks. Its Assistant model hallucinated at a rate of 0.2% (1 in 500 claims), lower than Claude (0.7%), ChatGPT (1.3%), and Gemini (1.9%), despite providing longer and more detailed answers. The company defines hallucinations as factual claims demonstrably disproven by a source of truth and uses a system of models to detect and measure them. ### Runway — Foundations for Safe Generative Media - Company: Runway (runway.com) - Announced: 2024-10-07T16:22:35.195000+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/research/foundations-for-safe-generative-media - Record: https://forck.live/items/7921-foundations-for-safe-generative-media - Subject: Gen / Aleph Runway shared its guardrails for safety, fairness, and integrity in generative models, including an in-house visual moderation system, reporting of child sexual abuse material to the National Center for Missing and Exploited Children, a fairness fine-tuned model to reduce biased representations of professions, and adoption of C2PA provenance standards with invisible watermarks. ### Vercel — How Supabase increased signups through the Vercel Marketplace - Company: Vercel (vercel.com) - Announced: 2024-10-07T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-supabase-increased-signups-through-the-vercel-marketplace - Record: https://forck.live/items/20941-how-supabase-increased-signups-through-the-vercel-marketplace - Subject: v0 / AI SDK / AI Gateway Supabase has seen a notable increase in new signups since launching on the Vercel Marketplace, which has become their largest partner channel. The integration provides frictionless onboarding, allowing developers to set up and manage Supabase databases directly from the Vercel CLI or dashboard with integrated billing.