# forck.live — 15–21 July 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-W29, 15–21 July 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-W29
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Replit — Introducing Replit Projects

- Company: Replit (replit.com)
- Announced: 2024-07-19T21:28:20+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/introducing-replit-projects
- Record: https://forck.live/items/17980-introducing-replit-projects
- Subject: Replit Agent

Replit introduced Projects, a new collaboration feature in beta for Replit Teams customers. It allows team members to fork code, make changes in isolated environments, and merge changes back into the main version. The feature is built on git but does not require users to learn git.

### Mistral AI — Mistral NeMo

- Company: Mistral AI (mistral.ai)
- Announced: 2024-07-18T08:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://mistral.ai/news/mistral-nemo/
- Record: https://forck.live/items/7367-mistral-nemo
- Subject: Mistral / Le Chat

Mistral AI released Mistral NeMo, a 12B model developed with NVIDIA, featuring a 128k token context window and Apache 2.0 license. The model includes a new tokenizer, Tekken, and is available as pre-trained base and instruction-tuned checkpoints.

### Hugging Face — TGI Multi-LoRA: Deploy Once, Serve 30 Models

- Company: Hugging Face (huggingface.co)
- Announced: 2024-07-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/multi-lora-serving
- Record: https://forck.live/items/1852-tgi-multi-lora-deploy-once-serve-30-models
- Subject: Platform

Hugging Face announces Multi-LoRA serving for Text Generation Inference (TGI), allowing deployment of a single base model with multiple LoRA adapters to serve many fine-tuned models efficiently.

### Hugging Face — Docmatix - a huge dataset for Document Visual Question Answering

- Company: Hugging Face (huggingface.co)
- Announced: 2024-07-18
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/docmatix
- Record: https://forck.live/items/1851-docmatix-a-huge-dataset-for-document-visual-question-answering
- Subject: Platform

Hugging Face releases Docmatix, a large-scale dataset for Document Visual Question Answering (DocVQA), 240 times larger than previous datasets. It is generated from PDFA OCR dataset using Phi-3-small model for Q/A pair creation, with filtering for hallucinations. The dataset contains 2.4 million images and 9.5 million Q/A pairs from 1.3 million PDFs. Ablation studies using Florence-2 model show a 20% improvement in DocVQA performance.

### OpenAI — New compliance and administrative tools for ChatGPT Enterprise

- Company: OpenAI (openai.com)
- Announced: 2024-07-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/new-tools-for-chatgpt-enterprise
- Record: https://forck.live/items/733-new-compliance-and-administrative-tools-for-chatgpt-enterprise
- Subject: GPT / ChatGPT / API

OpenAI announced new compliance and administrative tools for ChatGPT Enterprise, including Compliance API integrations, SCIM, and GPT controls to support compliance programs, data security, and user access at scale.

### OpenAI — Prover-Verifier Games improve legibility of language model outputs

- Company: OpenAI (openai.com)
- Announced: 2024-07-17T10:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/prover-verifier-games-improve-legibility
- Record: https://forck.live/items/734-prover-verifier-games-improve-legibility-of-language-model-outputs
- Subject: GPT / ChatGPT / API

Prover-verifier games improve the legibility of language model outputs, making AI solutions clearer, easier to verify, and more trustworthy.

### Replit — Introducing Replit Teams

- Company: Replit (replit.com)
- Announced: 2024-07-16T18:06:32+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/teams
- Record: https://forck.live/items/17984-introducing-replit-teams
- Subject: Replit Agent

Replit announced the general availability of Replit Teams, a collaborative software development platform designed for modern, cross-functional teams. The platform includes AI assistance for planning, code generation, and debugging, and offers features like SAML SSO, role-based access control, centralized billing, and SOC-2 compliance for enterprise use.

### Mistral AI — MathΣtral

- Company: Mistral AI (mistral.ai)
- Announced: 2024-07-16T08:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://mistral.ai/news/mathstral/
- Record: https://forck.live/items/7368-math-tral
- Subject: Mistral / Le Chat

Mistral AI released Mathstral, a model specialized in STEM reasoning, built on Mistral 7B. It achieves 56.6% on MATH and 63.47% on MMLU benchmarks. The weights are hosted on HuggingFace for use and fine-tuning.

### Supabase — Announcing Supabase on JSR

- Company: Supabase (supabase.com)
- Announced: 2024-07-16
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-js-on-jsr
- Record: https://forck.live/items/21418-announcing-supabase-on-jsr
- Subject: Supabase

Supabase announced that its JavaScript client library, supabase-js, is now available on JSR, a modern open source JavaScript registry. The library supports TypeScript, auto-generated documentation, and can be used across browsers, servers, and edge environments.

### Hugging Face — How we leveraged distilabel to create an Argilla 2.0 Chatbot

- Company: Hugging Face (huggingface.co)
- Announced: 2024-07-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/argilla-chatbot
- Record: https://forck.live/items/1854-how-we-leveraged-distilabel-to-create-an-argilla-2-0-chatbot
- Subject: Platform

Building a chatbot for Argilla 2.0 using distilabel, including synthetic data generation, fine-tuning an embedding model, creating a vector database, and deploying the chatbot on Hugging Face Spaces.

### Hugging Face — SmolLM - blazingly fast and remarkably powerful

- Company: Hugging Face (huggingface.co)
- Announced: 2024-07-16
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/smollm
- Record: https://forck.live/items/1853-smollm-blazingly-fast-and-remarkably-powerful
- Subject: Platform

Hugging Face introduces SmolLM, a family of state-of-the-art small language models with 135M, 360M, and 1.7B parameters, trained on a new high-quality dataset called SmolLM-Corpus. The models outperform other models in their size categories across reasoning and knowledge benchmarks.

### Suno — The Summer of Suno, Vol. 2

- Company: Suno (suno.com)
- Announced: 2024-07-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://suno.com/blog/sos-vol-2
- Record: https://forck.live/items/10347-the-summer-of-suno-vol-2
- Subject: Music / Studio

Suno announced The Summer of Suno, Vol. 2, a two-week contest starting July 15, 2024, with a $100,000 prize pool split among 200 creators who submit reaction videos to songs made with Suno. The contest is part of a previously announced $1 million payout to top creators in 2024. Eligibility was expanded to include creators in India, Israel, and Japan.

### LG AI Research — [CVPR 2024] Introduction to Papers Selected for the Best Paper Award and Low-Level Vision Research

- Company: LG AI Research (lgresearch.ai)
- Announced: 2024-07-15
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=455
- Record: https://forck.live/items/4841-cvpr-2024-introduction-to-papers-selected-for-the-best-paper-award-and-low
- Subject: EXAONE

LG AI Research introduces two papers presented at CVPR 2024: 'Generative Image Dynamics', which won the Best Paper Award and focuses on generating naturally swaying videos from a single image using a diffusion model in the frequency domain, and 'Learning to Control Camera Exposure via Reinforcement Learning', a low-level vision study aimed at improving image quality during acquisition.