# forck.live — 25 September – 1 October 2023 > 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 2023-W39, 25 September – 1 October 2023, 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/2023-W39 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Hugging Face — Finetune Stable Diffusion Models with DDPO via TRL - Company: Hugging Face (huggingface.co) - Announced: 2023-09-29 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/trl-ddpo - Record: https://forck.live/items/1989-finetune-stable-diffusion-models-with-ddpo-via-trl - Subject: Platform Hugging Face releases a new DDPOTrainer in the trl library, enabling users to fine-tune Stable Diffusion models with DDPO (Denoising Diffusion Policy Optimization) reinforcement learning. ### Hugging Face — Ethics and Society Newsletter #5: Hugging Face Goes To Washington and Other Summer 2023 Musings - Company: Hugging Face (huggingface.co) - Announced: 2023-09-29 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/ethics-soc-5 - Record: https://forck.live/items/1988-ethics-and-society-newsletter-5-hugging-face-goes-to-washington-and-other - Subject: Platform Hugging Face's fifth ethics and society newsletter summarizes the team's summer 2023 activities, including advocacy with U.S., E.U., and U.K. legislators, media appearances, workshops, papers, and internal work on bias testing and transparency. ### Replit — Changes to Hosting on Replit - Company: Replit (replit.com) - Announced: 2023-09-28T19:43:00+00:00 - Category: deprecation - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/hosting-changes - Record: https://forck.live/items/18020-changes-to-hosting-on-replit - Subject: Replit Agent Replit announced it is deprecating its legacy Always On hosting product, which will be fully removed on January 1st, 2024. After that date, Deployments will be the only way to host applications on Replit. The company is providing a three-month migration period and offering a Pro trial for Cycles customers who were paying for Always On without a plan. ### Runway — Scale, Speed and Stepping Stones: The path to Gen-2 - Company: Runway (runway.com) - Announced: 2023-09-28T10:10:10.110000+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://runway.com/research/scale-speed-and-stepping-stones-the-path-to-gen-2 - Record: https://forck.live/items/7936-scale-speed-and-stepping-stones-the-path-to-gen-2 - Subject: Gen / Aleph Runway developed Gen-2 as a text-to-video model building on latent diffusion, first solving temporal consistency with Gen-1 using input video conditioning, then removing that requirement for Gen-2. The company's north star is enabling generation of a two-hour film, and it follows a staged release approach for safety and organic use case emergence. ### Hugging Face — Non-engineers guide: Train a LLaMA 2 chatbot - Company: Hugging Face (huggingface.co) - Announced: 2023-09-28 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/Llama2-for-non-engineers - Record: https://forck.live/items/1990-non-engineers-guide-train-a-llama-2-chatbot - Subject: Platform Hugging Face published a tutorial for non-engineers to fine-tune a LLaMA 2 base model and deploy a chatbot using Spaces, AutoTrain, and ChatUI without writing code. ### Replit — Replit Core: Go from idea to software, fast - Company: Replit (replit.com) - Announced: 2023-09-27T19:43:00+00:00 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/core-plan-value - Record: https://forck.live/items/18024-replit-core-go-from-idea-to-software-fast - Subject: Replit Agent Replit announced Core, a $20/month subscription that bundles a cloud development environment (8 GiB RAM, 4 vCPU), Replit AI powered by GPT-4, autoscale deployments with six million compute units (covering ~10 projects per month), 100 GiB network bandwidth, 50 GiB storage, SSH, private Repls, and a monthly PostgreSQL allotment. The company states a similar setup on GitHub Codespaces can cost up to 10x more and that high usage on Vercel can incur overages 3-4x more expensive. ### Mistral AI — Bringing open AI models to the frontier - Company: Mistral AI (mistral.ai) - Announced: 2023-09-27T08:00:00+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://mistral.ai/news/about-mistral-ai/ - Record: https://forck.live/items/7380-bringing-open-ai-models-to-the-frontier - Subject: Mistral / Le Chat Mistral AI released Mistral 7B, a 7-billion-parameter open-weight model that outperforms all currently available open models up to 13B parameters on standard English and code benchmarks, under the Apache 2.0 license. ### Mistral AI — Mistral 7B - Company: Mistral AI (mistral.ai) - Announced: 2023-09-27T08:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://mistral.ai/news/announcing-mistral-7b/ - Record: https://forck.live/items/7379-mistral-7b - Subject: Mistral / Le Chat Mistral AI released Mistral 7B, a 7.3B parameter language model that outperforms Llama 2 13B on all benchmarks and is released under the Apache 2.0 license. The model uses grouped-query attention and sliding window attention for efficient inference. They also provide a fine-tuned chat version, Mistral 7B Instruct, which outperforms Llama 2 13B chat. ### Replit — Superagent.sh on Replit: An open-source framework for creating AI-assistants - Company: Replit (replit.com) - Announced: 2023-09-26T19:36:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replit.com/blog/superagent - Record: https://forck.live/items/18033-superagent-sh-on-replit-an-open-source-framework-for-creating-ai-assistants - Subject: Replit Agent Superagent.sh, an open-source framework for creating AI assistants, is now available on Replit. The framework uses large language models and integrates with Replit's code-exec and agent-env features, allowing developers to build and deploy AI assistants without AI/ML expertise. Examples include customer support chatbots, legal AI assistants, code review bots, and content generators. ### Hugging Face — Llama 2 on Amazon SageMaker a Benchmark - Company: Hugging Face (huggingface.co) - Announced: 2023-09-26 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/llama-sagemaker-benchmark - Record: https://forck.live/items/1991-llama-2-on-amazon-sagemaker-a-benchmark - Subject: Platform Hugging Face published a benchmark of Llama 2 model sizes (7B, 13B, 70B) deployed on Amazon SageMaker using the Hugging Face LLM Inference Container, comparing latency and throughput across different instance types, quantization (GPTQ), and concurrency levels to provide recommendations on cost-effective, high-throughput, and low-latency deployments. ### LG AI Research — LG Aimers, Completing the Third Journey - Company: LG AI Research (lgresearch.ai) - Announced: 2023-09-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/news/view?seq=340 - Record: https://forck.live/items/11045-lg-aimers-completing-the-third-journey - Subject: EXAONE LG AI Research held the third LG Aimers Hackathon on September 16-17 at LG Academy, where 108 participants from 43 teams developed AI models to forecast product sales volume in online distribution channels. The winning teams received benefits including exemption from document screening in job applications and a total prize pool of 60 million won. A job fair with eight LG affiliates was also held for participants. ### LG AI Research — LG Aimers, Completing the Third Journey - Company: LG AI Research (lgresearch.ai) - Announced: 2023-09-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/blog/view?seq=340 - Record: https://forck.live/items/4892-lg-aimers-completing-the-third-journey - Subject: EXAONE LG AI Research held the third LG Aimers Hackathon, a youth AI talent development program, where participants developed a predictive AI model for product sales volume. The event included a job fair with LG affiliates and awarded prizes to winning teams.