# forck.live — 3–9 June 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-W23, 3–9 June 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-W23 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### OpenAI — Expanding on how Voice Engine works and our safety research - Company: OpenAI (openai.com) - Announced: 2024-06-07T17:45:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/expanding-on-how-voice-engine-works-and-our-safety-research - Record: https://forck.live/items/749-expanding-on-how-voice-engine-works-and-our-safety-research - Subject: GPT / ChatGPT / API OpenAI discusses the technology behind its text-to-speech model and its safety research. ### Upstage — Upstage Optimizes Large Language Model for Intel Core Ultra Processors - Company: Upstage (upstage.ai) - Announced: 2024-06-07 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.upstage.ai/news/upstage-intel - Record: https://forck.live/items/18441-upstage-optimizes-large-language-model-for-intel-core-ultra-processors - Subject: Solar Pro / Solar Mini Upstage announced it will optimize its Solar Mini LLM for Intel Core Ultra processors and integrate the WriteUp writing assistant app to run offline on Windows PCs based on Intel processors, enabling on-device generative AI capabilities. ### LG AI Research — Unsure about AGI, LG AI Research’s Junior Members Just Want to Succeed - Company: LG AI Research (lgresearch.ai) - Announced: 2024-06-07 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=434 - Record: https://forck.live/items/4852-unsure-about-agi-lg-ai-research-s-junior-members-just-want-to-succeed - Subject: EXAONE This blog post profiles junior researchers at LG AI Research, discussing their work on anomaly detection, image editing, instruction tuning, and data generation, as well as their experiences and growth within the company. It does not announce any AI product or model. ### Replicate — Replicate Intelligence #3 - Company: Replicate (replicate.com) - Announced: 2024-06-07 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://replicate.com/blog/replicate-intelligence-2024-06-07 - Record: https://forck.live/items/2535-replicate-intelligence-3 - Subject: Platform Garden State Llama is a language model fine-tuned to focus on New Jersey. ### Hugging Face — Introducing the Hugging Face Embedding Container for Amazon SageMaker - Company: Hugging Face (huggingface.co) - Announced: 2024-06-07 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/sagemaker-huggingface-embedding - Record: https://forck.live/items/1875-introducing-the-hugging-face-embedding-container-for-amazon-sagemaker - Subject: Platform Hugging Face announced the general availability of the Hugging Face Embedding Container for Amazon SageMaker, a purpose-built inference container powered by Text Embedding Inference (TEI) for deploying embedding models on SageMaker. ### Hugging Face — Making sense of this mess - Company: Hugging Face (huggingface.co) - Announced: 2024-06-07 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/transformers-docs-redesign - Record: https://forck.live/items/1874-making-sense-of-this-mess - Subject: Platform Hugging Face announces a redesign of the Transformers documentation to make it more code-first, solution-oriented, and flexible. ### Replit — Announcing Replit Guides - Company: Replit (replit.com) - Announced: 2024-06-06T17:14:52+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replit.com/blog/replit-guides - Record: https://forck.live/items/17970-announcing-replit-guides - Subject: Replit Agent Replit launched Replit Guides, a set of educational resources that include Quickstarts (lessons under 15 minutes) and longer guides for building applications such as AI-powered chatbots. The initial guides are created in partnership with companies including Google, Groq, OpenAI, and Anthropic. ### Alibaba — Hello Qwen2 - Company: Alibaba (alibaba.com) - Announced: 2024-06-06T16:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://qwenlm.github.io/blog/qwen2/ - Record: https://forck.live/items/2312-hello-qwen2 - Subject: Qwen Alibaba announces the Qwen2 family of models, evolved from Qwen1.5, featuring five sizes, multilingual training, SOTA benchmarks, improved coding and math, and extended context length. ### Alibaba — Generalizing an LLM from 8k to 1M Context using Qwen-Agent - Company: Alibaba (alibaba.com) - Announced: 2024-06-06T03:59:59+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://qwenlm.github.io/blog/qwen-agent-2405/ - Record: https://forck.live/items/2313-generalizing-an-llm-from-8k-to-1m-context-using-qwen-agent - Subject: Qwen An agent built on Qwen2 models with an 8k context size can handle documents with 1M tokens, surpassing RAG and native long-context models. ### Hugging Face — Launching the Artificial Analysis Text to Image Leaderboard & Arena - Company: Hugging Face (huggingface.co) - Announced: 2024-06-06 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/leaderboard-artificial-analysis2 - Record: https://forck.live/items/1876-launching-the-artificial-analysis-text-to-image-leaderboard-arena - Subject: Platform Hugging Face announces the Artificial Analysis Text to Image Leaderboard & Arena, a crowdsourced human preference ranking system for comparing text-to-image models. ### OpenAI — Extracting Concepts from GPT-4 - Company: OpenAI (openai.com) - Announced: 2024-06-06 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/extracting-concepts-from-gpt-4 - Record: https://forck.live/items/751-extracting-concepts-from-gpt-4 - Subject: GPT / ChatGPT / API OpenAI announces a new technique for scaling sparse autoencoders to identify 16 million patterns in GPT-4's computations. ### Mistral AI — My Tailor is Mistral - Company: Mistral AI (mistral.ai) - Announced: 2024-06-05T09:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://mistral.ai/news/customization/ - Record: https://forck.live/items/7370-my-tailor-is-mistral - Subject: Mistral / Le Chat Mistral AI introduced model customization on la Plateforme, offering three entry points for fine-tuning: an open-source fine-tuning SDK (mistral-finetune), serverless fine-tuning services, and custom training services. The fine-tuning services are initially compatible with Mistral 7B and Mistral Small, with more models to be added. ### Mistral AI — Mistral AI Fine-tuning Hackathon - Company: Mistral AI (mistral.ai) - Announced: 2024-06-05T08:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://mistral.ai/news/2024-ft-hackathon/ - Record: https://forck.live/items/7371-mistral-ai-fine-tuning-hackathon - Subject: Mistral / Le Chat Mistral AI announced a virtual fine-tuning hackathon running from June 5 to June 30, 2024, with prizes of €2,500 in API credits for the top three projects. Participants must use Mistral's fine-tuning API and submit original work with code, a description, and a video presentation. ### Hugging Face — Introducing NPC-Playground, a 3D playground to interact with LLM-powered NPCs - Company: Hugging Face (huggingface.co) - Announced: 2024-06-05 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/npc-gigax-cubzh - Record: https://forck.live/items/1877-introducing-npc-playground-a-3d-playground-to-interact-with-llm-powered-npcs - Subject: Platform Hugging Face, Cubzh, and Gigax announce NPC-Playground, a 3D browser demo where users interact with LLM-powered NPCs. The demo uses Cubzh as the game engine and Gigax for NPC intelligence. Gigax has open-sourced its inference stack and released fine-tuned models (Phi-3 and Mistral-7B) on the Hugging Face Hub. The demo is hosted on Hugging Face Spaces and can be cloned for customization. ### Hugging Face — Faster assisted generation support for Intel Gaudi - Company: Hugging Face (huggingface.co) - Announced: 2024-06-04 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/assisted-generation-support-gaudi - Record: https://forck.live/items/1878-faster-assisted-generation-support-for-intel-gaudi - Subject: Platform Hugging Face and Intel have adapted and optimized assisted decoding (speculative sampling) for Intel Gaudi processors, now integrated into Optimum Habana to accelerate text generation. ### Vercel — Introducing bytecode caching for Vercel Functions - Company: Vercel (vercel.com) - Announced: 2024-06-03T13:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/introducing-bytecode-caching-for-vercel-functions - Record: https://forck.live/items/20976-introducing-bytecode-caching-for-vercel-functions - Subject: v0 / AI SDK / AI Gateway Vercel announced an experimental bytecode caching feature for Vercel Functions that reduces cold start latency by up to 27% for large applications. The feature caches compiled bytecode from the first execution and reuses it on subsequent cold starts, eliminating the compilation step. It is available for Node.js 20 and frameworks that compile to CommonJS, with an opt-in environment variable. ### Vercel — Mintlify: Scaling a powerful documentation platform with Vercel - Company: Vercel (vercel.com) - Announced: 2024-06-03T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/mintlify-scaling-a-powerful-documentation-platform-with-vercel - Record: https://forck.live/items/20975-mintlify-scaling-a-powerful-documentation-platform-with-vercel - Subject: v0 / AI SDK / AI Gateway Mintlify, a documentation platform, uses Vercel and Next.js to serve multiple customers across thousands of custom domains from a single codebase. The platform now has 2,500 custom domains connected, and Vercel's API automates custom domain setup and SSL certificate issuance. Mintlify plans to further leverage Vercel's infrastructure, including migrating their dashboard for improved performance.