# forck.live — 18–24 March 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-W12, 18–24 March 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-W12 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Vercel — Protecting AI apps from bots and bad actors with Vercel and Kasada - Company: Vercel (vercel.com) - Announced: 2024-03-22T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/protecting-ai-apps-with-vercel-and-kasada - Record: https://forck.live/items/20999-protecting-ai-apps-from-bots-and-bad-actors-with-vercel-and-kasada - Subject: v0 / AI SDK / AI Gateway Vercel describes how it uses Next.js Middleware and Kasada's bot protection to secure its AI SDK Playground. The platform saw bot traffic drop from 84% of all traffic to negligible levels after integrating Kasada. ### Hugging Face — Binary and Scalar Embedding Quantization for Significantly Faster & Cheaper Retrieval - Company: Hugging Face (huggingface.co) - Announced: 2024-03-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/embedding-quantization - Record: https://forck.live/items/1920-binary-and-scalar-embedding-quantization-for-significantly-faster-cheaper - Subject: Platform Hugging Face introduces binary and scalar embedding quantization techniques to improve retrieval speed, reduce memory usage, and lower costs. The methods involve converting float32 embeddings to 1-bit (binary) or int8 (scalar) values, enabling 32x reduction in memory. A demo with 41 million Wikipedia texts demonstrates practical benefits. Experiments show up to ~96% performance retention with rescoring. ### Hugging Face — Total noob’s intro to Hugging Face Transformers - Company: Hugging Face (huggingface.co) - Announced: 2024-03-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/noob_intro_transformers - Record: https://forck.live/items/1919-total-noob-s-intro-to-hugging-face-transformers - Subject: Platform A tutorial on using Hugging Face Transformers, with a worked example using Microsoft's Phi-2 model. ### Runway — Partnering with Musixmatch to enhance the experience of music worldwide - Company: Runway (runway.com) - Announced: 2024-03-21T10:10:10.110000+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/runway-partners-with-musixmatch - Record: https://forck.live/items/7928-partnering-with-musixmatch-to-enhance-the-experience-of-music-worldwide - Subject: Gen / Aleph Runway is partnering with Musixmatch to give Musixmatch's community of over 1 million artists access to Gen-2 and other Runway models for creating synchronized lyric music videos. ### OpenAI — Embedding AI into developer software - Company: OpenAI (openai.com) - Announced: 2024-03-21T07:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/jetbrains - Record: https://forck.live/items/795-embedding-ai-into-developer-software - Subject: GPT / ChatGPT / API JetBrains is using OpenAI's API to develop its fastest-growing product. ### Upstage — Solar Mini Now Available on AWS - Company: Upstage (upstage.ai) - Announced: 2024-03-21 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.upstage.ai/news/solarmini-aws - Record: https://forck.live/items/18443-solar-mini-now-available-on-aws - Subject: Solar Pro / Solar Mini Upstage made its Solar Mini large language model available on Amazon Web Services through Amazon SageMaker JumpStart and AWS Marketplace, letting developers integrate the 10.7-billion-parameter model into applications without building models from scratch. The model handles Korean and English tasks such as summarization, translation, and content generation, with Thai and Japanese support planned. Upstage says its smaller size enables efficient inference at lower costs than larger models. ### Suno — Introducing v3 - Company: Suno (suno.com) - Announced: 2024-03-21 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://suno.com/blog/v3 - Record: https://forck.live/items/10352-introducing-v3 - Subject: Music / Studio Suno released v3, its first model capable of producing radio-quality music, enabling users to create full two-minute songs in seconds. The model is now available to all users at app.suno.ai and includes improvements in audio quality, style and genre variety, and prompt adherence. ### Hugging Face — Introducing the Chatbot Guardrails Arena - Company: Hugging Face (huggingface.co) - Announced: 2024-03-21 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/arena-lighthouz - Record: https://forck.live/items/1921-introducing-the-chatbot-guardrails-arena - Subject: Platform Hugging Face and Lighthouz AI launch the Chatbot Guardrails Arena, a platform to stress-test LLMs and privacy guardrails by having users try to trick chatbots into revealing sensitive financial information. The arena features 12 guardrailed LLMs, including closed-source and open-source models, and will compile votes into a public leaderboard based on the Elo rating system. ### Sakana AI — Evolving New Foundation Models: Unleashing the Power of Automating Model Development - Company: Sakana AI (sakana.ai) - Announced: 2024-03-20T15:00:00+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://sakana.ai/evolutionary-model-merge - Record: https://forck.live/items/4712-evolving-new-foundation-models-unleashing-the-power-of-automating-model - Subject: Sakana models Sakana AI announces Evolutionary Model Merge, a method using evolutionary techniques to combine open-source models, and releases three Japanese foundation models: EvoLLM-JP (LLM), EvoVLM-JP (VLM), and EvoSDXL-JP (image generation). The models achieve state-of-the-art results on Japanese benchmarks. The associated paper was accepted to Nature Machine Intelligence. ### Vercel — Revolutionizing video editing on the web with Next.js and Vercel - Company: Vercel (vercel.com) - Announced: 2024-03-20T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/revolutionizing-video-editing-on-the-web-with-next-js-and-vercel - Record: https://forck.live/items/21000-revolutionizing-video-editing-on-the-web-with-next-js-and-vercel - Subject: v0 / AI SDK / AI Gateway Ozone, a web-based video editing platform, migrated from Heroku to Vercel, resulting in an 85% decrease in build times and a 5x acceleration in development. The platform also integrated AI models from Stability AI, Meta, and ElevenLabs using Vercel's AI Integrations and AI SDK. ### Runway — Partnering with Media.Monks to expand creative horizons - Company: Runway (runway.com) - Announced: 2024-03-20T10:10:10.110000+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/partnering-with-media-monks - Record: https://forck.live/items/7929-partnering-with-media-monks-to-expand-creative-horizons - Subject: Gen / Aleph Runway announced a partnership with Media.Monks to leverage its generative AI technology to enhance content generation and creative workflows for Media.Monks and its clients. ### Hugging Face — GaLore: Advancing Large Model Training on Consumer-grade Hardware - Company: Hugging Face (huggingface.co) - Announced: 2024-03-20 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/galore - Record: https://forck.live/items/1924-galore-advancing-large-model-training-on-consumer-grade-hardware - Subject: Platform GaLore is a method that reduces memory footprint for training large language models by projecting gradients into low-rank subspaces, enabling training of up to 7 billion parameter models on consumer GPUs like the NVIDIA RTX 4090. It achieves over 82.5% reduction in memory for optimizer states and can be combined with 8-bit optimizers for further savings. ### Hugging Face — Cosmopedia: how to create large-scale synthetic data for pre-training Large Language Models - Company: Hugging Face (huggingface.co) - Announced: 2024-03-20 - Category: research-paper - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/cosmopedia - Record: https://forck.live/items/1923-cosmopedia-how-to-create-large-scale-synthetic-data-for-pre-training-large - Subject: Platform The creation of Cosmopedia, a large-scale synthetic dataset for pre-training LLMs, aiming to replicate the training data used for Phi-1.5. It details the challenges of scaling synthetic data generation, prompt curation using curated sources and web data, and the use of Mixtral-8x7B-Instruct-v0.1. The dataset, code, and a 1B model called cosmo-1b are released openly. ### Hugging Face — A Chatbot on your Laptop: Phi-2 on Intel Meteor Lake - Company: Hugging Face (huggingface.co) - Announced: 2024-03-20 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/phi2-intel-meteor-lake - Record: https://forck.live/items/1922-a-chatbot-on-your-laptop-phi-2-on-intel-meteor-lake - Subject: Platform To run the Phi-2 small language model locally on a laptop with an Intel Meteor Lake CPU by applying 4-bit quantization using Intel OpenVINO and Optimum Intel, enabling local inference with benefits like privacy, lower latency, offline work, and cost savings. ### Harvey — Harvey Launches AI-Powered Professional Services Platform on Microsoft Azure - Company: Harvey (harvey.ai) - Announced: 2024-03-19T21:33:00+00:00 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/harvey-launches-ai-powered-professional-services-platform-on-microsoft-azure - Record: https://forck.live/items/17735-harvey-launches-ai-powered-professional-services-platform-on-microsoft-azure - Subject: Harvey Harvey announced the deployment of its generative AI platform on Microsoft Azure and its availability in the Azure Marketplace. The collaboration aims to market and sell the Harvey on Azure solution to law firms, in-house teams, professional service providers, and private equity firms globally. Customers can now purchase Harvey's solutions directly from the Azure Marketplace. ### Replit — More Reliable Connections to Your Repls - Company: Replit (replit.com) - Announced: 2024-03-18T16:00:00+00:00 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/eval - Record: https://forck.live/items/18002-more-reliable-connections-to-your-repls - Subject: Replit Agent Replit introduced Eval, a dedicated reverse WebSocket proxy that separates proxying from container management. This change reduces client disconnections during Conman updates, simplifies autoscaling, and improves observability for diagnosing session failures. ### Vercel — WordPress monolith to Vercel: How Personio elevated site performance and efficiency - Company: Vercel (vercel.com) - Announced: 2024-03-18T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/from-wordpress-monolith-to-vercel-personio-elevates-site-performance - Record: https://forck.live/items/21002-wordpress-monolith-to-vercel-how-personio-elevated-site-performance-and - Subject: v0 / AI SDK / AI Gateway Personio migrated its website from a monolithic WordPress implementation to a composable stack of Vercel, Next.js, and Contentful. The company reported a more than 90% improvement in Core Web Vitals scores, a 29% improvement in mobile Largest Contentful Paint, and roughly 100% uptime. The headless architecture also improved developer experience and site security. ### Vercel — Leonardo generates 4.5M images daily with Next.js and Vercel - Company: Vercel (vercel.com) - Announced: 2024-03-18T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/leonardo-ai-performantly-generates-4-5-million-images-daily-with-next-js-and-vercel - Record: https://forck.live/items/21001-leonardo-generates-4-5m-images-daily-with-next-js-and-vercel - Subject: v0 / AI SDK / AI Gateway Leonardo.Ai, which generates over 4.5 million images daily, migrated its web application to Vercel and reduced build times from over 10 minutes to as little as 2 minutes, and cut page load times by 95% through improved caching and CDN performance. The move also enabled the team to iterate and launch new products in 1–4 weeks instead of months. ### OpenAI — Building a data-driven, efficient culture with AI - Company: OpenAI (openai.com) - Announced: 2024-03-18T07:00:00+00:00 - Category: availability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/holiday-extras - Record: https://forck.live/items/798-building-a-data-driven-efficient-culture-with-ai - Subject: GPT / ChatGPT / API Holiday Extras deploys ChatGPT Enterprise organization-wide and reports a productivity gain of 500 hours per week. ### OpenAI — Enterprise-ready trust and safety - Company: OpenAI (openai.com) - Announced: 2024-03-18T07:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/salesforce - Record: https://forck.live/items/796-enterprise-ready-trust-and-safety - Subject: GPT / ChatGPT / API Salesforce is integrating OpenAI's enterprise-ready large language models to transform customer applications. ### Hugging Face — Quanto: a PyTorch quantization backend for Optimum - Company: Hugging Face (huggingface.co) - Announced: 2024-03-18 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/quanto-introduction - Record: https://forck.live/items/1926-quanto-a-pytorch-quantization-backend-for-optimum - Subject: Platform Hugging Face introduces Quanto, a PyTorch quantization backend for Optimum, designed for versatility and simplicity, supporting various quantization schemes and devices. ### Hugging Face — Easily Train Models with H100 GPUs on NVIDIA DGX Cloud - Company: Hugging Face (huggingface.co) - Announced: 2024-03-18 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/train-dgx-cloud - Record: https://forck.live/items/1925-easily-train-models-with-h100-gpus-on-nvidia-dgx-cloud - Subject: Platform Hugging Face launches Train on DGX Cloud, a service on the Hugging Face Hub that enables Enterprise Hub organizations to fine-tune open models using NVIDIA H100 GPUs with pay-as-you-go pricing and no-code training via AutoTrain.