# forck.live — 24–30 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-W26, 24–30 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-W26 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Replicate — Replicate Intelligence #6 - Company: Replicate (replicate.com) - Announced: 2024-06-28 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/replicate-intelligence-2024-06-28 - Record: https://forck.live/items/2526-replicate-intelligence-6 - Subject: Platform Google released new Gemma2 models in 9b and 27b sizes, described as overtrained on tokens and distilled from larger Gemini models, with alternating global/local attention layers. ### OpenAI — Finding GPT-4’s mistakes with GPT-4 - Company: OpenAI (openai.com) - Announced: 2024-06-27T10:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://openai.com/index/finding-gpt4s-mistakes-with-gpt-4 - Record: https://forck.live/items/736-finding-gpt-4-s-mistakes-with-gpt-4 - Subject: GPT / ChatGPT / API OpenAI announces CriticGPT, a model based on GPT-4 that writes critiques of ChatGPT responses to help human trainers spot mistakes during RLHF. ### OpenAI — Strategic Content Partnership with TIME - Company: OpenAI (openai.com) - Announced: 2024-06-27T06:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/strategic-content-partnership-with-time - Record: https://forck.live/items/737-strategic-content-partnership-with-time - Subject: GPT / ChatGPT / API OpenAI announces a partnership with TIME to use its archival content to improve responses and link to stories on Time.com. ### Upstage — Upstage Signs Multi-Year Strategic Collaboration Agreement with AWS to Accelerate Generative AI Innovation - Company: Upstage (upstage.ai) - Announced: 2024-06-27 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.upstage.ai/news/upstage-aws-sca - Record: https://forck.live/items/18440-upstage-signs-multi-year-strategic-collaboration-agreement-with-aws-to - Subject: Solar Pro / Solar Mini Upstage announced a multi-year Strategic Collaboration Agreement with AWS to extend their partnership, and joined the AWS ISV Accelerate Program to enable co-selling and expand global reach of its SOLAR large language model. SOLAR MINI, a small language model supporting Korean, Japanese, and English (Thai coming in 2024), is available on Amazon SageMaker JumpStart and AWS Marketplace. Over 400 customers across industries are using Upstage to build custom LLMs, chatbots, and knowledge bases. ### Hugging Face — Welcome Gemma 2 - Google’s new open LLM - Company: Hugging Face (huggingface.co) - Announced: 2024-06-27 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/gemma2 - Record: https://forck.live/items/1864-welcome-gemma-2-google-s-new-open-llm - Subject: Platform Google released Gemma 2, an open LLM in 9B and 27B parameter sizes, with base and instruction-tuned versions, available on Hugging Face Hub. ### Supabase — Generate Vector Tiles with PostGIS - Company: Supabase (supabase.com) - Announced: 2024-06-26 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/postgis-generate-vector-tiles - Record: https://forck.live/items/21422-generate-vector-tiles-with-postgis - Subject: Supabase Supabase published a tutorial on generating vector tiles from PostGIS data. The guide covers downloading open map data via Overture Maps, transforming GeoJSON into SQL, importing into Supabase Postgres, and using PostGIS' ST_AsMVT to generate vector tiles served to a MapLibre GL client with supabase-js. ### LG AI Research — [AAMAS 2024] Multi-Agent Reinforcement Learning for Real-World Application (Part 1) - Company: LG AI Research (lgresearch.ai) - Announced: 2024-06-26 - Category: research-paper - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=444 - Record: https://forck.live/items/4847-aamas-2024-multi-agent-reinforcement-learning-for-real-world-application-part-1 - Subject: EXAONE LG AI Research's Data Intelligence Lab presented two papers at AAMAS 2024 on multi-agent reinforcement learning for real-world applications: one on naphtha cracking center scheduling optimization and one on agent-oriented centralized critic for asynchronous MARL. ### LG AI Research — [AAMAS 2024] Multi-Agent Reinforcement Learning for Real-World Application (Part 2) - Company: LG AI Research (lgresearch.ai) - Announced: 2024-06-26 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=446 - Record: https://forck.live/items/4846-aamas-2024-multi-agent-reinforcement-learning-for-real-world-application-part-2 - Subject: EXAONE Two papers from AAMAS 2024 on improvements in asynchronous multi-agent reinforcement learning and offline multi-agent reinforcement learning. ### Hugging Face — XLSCOUT Unveils ParaEmbed 2.0: a Powerful Embedding Model Tailored for Patents and IP with Expert Support from Hugging Face - Company: Hugging Face (huggingface.co) - Announced: 2024-06-25 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/xlscout-case-study - Record: https://forck.live/items/1865-xlscout-unveils-paraembed-2-0-a-powerful-embedding-model-tailored-for-patents - Subject: Platform XLSCOUT released ParaEmbed 2.0, a proprietary embedding model for patent analysis, developed with Hugging Face's Expert Support Program, achieving a 23% accuracy improvement over ParaEmbed 1.0. ### Hugging Face — Fine-tuning Florence-2 - Microsoft's Cutting-edge Vision Language Models - Company: Hugging Face (huggingface.co) - Announced: 2024-06-24 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/finetune-florence2 - Record: https://forck.live/items/1866-fine-tuning-florence-2-microsoft-s-cutting-edge-vision-language-models - Subject: Platform This blog post by Hugging Face demonstrates how to fine-tune Microsoft's Florence-2 vision-language model on the DocVQA dataset, showing that after fine-tuning the Levenshtein similarity score improved from 0 to 57.0. The post includes code walkthroughs and links to a demo space and GitHub repository.