# forck.live — 25–31 October 2021 > 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 2021-W43, 25–31 October 2021, 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/2021-W43 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### OpenAI — Solving math word problems - Company: OpenAI (openai.com) - Announced: 2021-10-29T07:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://openai.com/index/solving-math-word-problems - Record: https://forck.live/items/925-solving-math-word-problems - Subject: GPT / ChatGPT / API OpenAI trained a system that solves grade school math problems with nearly twice the accuracy of a fine-tuned GPT-3 model, achieving 55% accuracy on a test where 9-12 year olds scored 60%. ### Replit — Replit Art Gallery: An introduction to Replit's Illustrator - Joe Baker - Company: Replit (replit.com) - Announced: 2021-10-27T08:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/artist-introduction - Record: https://forck.live/items/17902-replit-art-gallery-an-introduction-to-replit-s-illustrator-joe-baker - Subject: Replit Agent Who makes Replit art? Hi! Thats me! My name is Joe and I’m Replit’s illustrator. I have been making art/graphics/multi media for the last 10 years. I studied Visual Media and excelled in experimental artwork. I draw for fun almost daily and I can’t stop thinking about aesthetics and concepts. It started with colouring in books when I was a kid I went from there basically. I have a huge love for making and appreciating art. I’m specifically drawn to art that rocks you to your core! Anything with wild colours, strange or abstract concepts, stuff that makes you feel something. My major influences come from surrealist art, psychedelic comics from the 70’s, 90’s cartoons, pop art, large scale public installations (sculpture and murals) and any festival artwork! My work is a fusion of these influences and I’m so lucky that I now get to create art for Replit on a full time basis. How did you hear about Replit? This all started in 2018 when I was freelancing. Amjad hit me up after seeing an album cover I made and asked me to create some promo for a “bot building competition”. I was instantly interested in the project. … ### Replit — Design Systems @ Replit: Better Tokens - Company: Replit (replit.com) - Announced: 2021-10-26 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/rui-tokens - Record: https://forck.live/items/18220-design-systems-replit-better-tokens - Subject: Replit Agent Part 1 of a series about our evolving design system, RUI (Replit User Interface). Replit is growing fast, as an application and a team. New features are being added, new people are joining the platform, and new designers and engineers are building it all. Unfortunately, this means that different parts of the product start to look and behave differently, because no single designer or engineer can keep all the interface states in their head. With dozens of people working on Replit, what happens when you want to update your color or text scheme across the whole site? What if you want users to be familiar with how components work anywhere they see them? It doesn't happen by accident — it requires strong infrastructural basics that you can rely on. So, we spent a few months this year building a stronger foundation for our design system. This is how our system is structured now: Each layer is a set of patterns that are used to compose the layer above it. We started with the most primitive layer: tokens. Tokens are just core visual attributes that we can apply in design tools (Figma) and on in our codebase (CSS). They describe appearance, but don’t describe behaviour. … ### Hugging Face — Course Launch Community Event - Company: Hugging Face (huggingface.co) - Announced: 2021-10-26 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/course-launch-event - Record: https://forck.live/items/2243-course-launch-community-event - Subject: Platform Hugging Face announces the release of part 2 of the Hugging Face Course on November 15th, 2021, and organizes a community event with talks and team projects. ### Replit — Betting on Nix: donating $25K to the NixOS Foundation - Company: Replit (replit.com) - Announced: 2021-10-25T00:06:40+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/betting-on-nix - Record: https://forck.live/items/18127-betting-on-nix-donating-25k-to-the-nixos-foundation - Subject: Replit Agent As building software grows more like snapping Legos together, how people find and use those Legos becomes more important. That's why we are donating $25,000 to the NixOS Foundation and betting on Nix as the future of software distribution. In software, the Lego pieces are called packages. A package may be some code your program needs to call (a library) or another program your code needs to run. Historically, people have used package managers to find and install packages into their projects. Each language ecosystem has its own package manager. Replit built and open-sourced the Universal Package Manager to unify this fragmented landscape. Package managers are a holdover from the old model of programming, when development environments lived on personal computers. Now that development environments (ahem, repls) can run entirely in the cloud, fetching packages from a central server, unzipping them, and installing them into a filesystem seems shockingly archaic. The future is instant. When you press the run button in a repl, your code should run immediately, not pause to fetch packages. The Nix project unlocks instant repl runs. … ### Hugging Face — Train a Sentence Embedding Model with 1B Training Pairs - Company: Hugging Face (huggingface.co) - Announced: 2021-10-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/1b-sentence-embeddings - Record: https://forck.live/items/2245-train-a-sentence-embedding-model-with-1b-training-pairs - Subject: Platform To train a sentence embedding model using 1 billion training pairs, covering the training methodology including Multiple Negative Ranking Loss, batch composition strategies, and similarity functions.