# forck.live — 12–18 August 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-W33, 12–18 August 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-W33
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Supabase — vec2pg: Migrate to pgvector from Pinecone and Qdrant

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/vec2pg
- Record: https://forck.live/items/21410-vec2pg-migrate-to-pgvector-from-pinecone-and-qdrant
- Subject: Supabase

Supabase released vec2pg, a CLI utility for migrating data from Pinecone and Qdrant vector databases to any Postgres instance with pgvector. The tool copies entire indexes or collections with associated ids and metadata into a new schema, and supports cloud and locally hosted Qdrant instances. Throughput ranges from 700-1100 records per second for Pinecone and 900-2500 records per second for Qdrant under specified conditions.

### Supabase — The Supabase Book by David Lorenz

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/supabase-book-by-david-lorenz
- Record: https://forck.live/items/21409-the-supabase-book-by-david-lorenz
- Subject: Supabase

David Lorenz, a long-time Supabase user, published a 500+ page book teaching Supabase through building a multi-tenant ticket management system. The book covers database design, Row Level Security, authentication, realtime features, file management, Edge functions, and AI-based search. It is available for purchase and includes community support via a Discord channel.

### Supabase — Supabase Python

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/python-support
- Record: https://forck.live/items/21408-supabase-python
- Subject: Supabase

Supabase announced official support for its Python client libraries, which had previously been community-maintained. The libraries include supabase-py and related packages, and recent updates added HTTP2, keep-alive, and redirect handling by default, along with new features for Realtime V2, anonymous logins, edge function regions, and a tool for migrating vector data from Pinecone and Qdrant.

### Supabase — PostgREST 12.2: Prometheus metrics

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/postgrest-12-2
- Record: https://forck.live/items/21407-postgrest-12-2-prometheus-metrics
- Subject: Supabase

Supabase announced the release of PostgREST 12.2, which adds Prometheus-compatible metrics for the schema cache and connection pool, hoisted function settings for per-function timeouts, and a max-affected preference to limit affected rows in mutations.

### Supabase — Postgres Foreign Data Wrappers with Wasm

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/postgres-foreign-data-wrappers-with-wasm
- Record: https://forck.live/items/21406-postgres-foreign-data-wrappers-with-wasm
- Subject: Supabase

Supabase released support for WebAssembly (Wasm) Foreign Data Wrappers (FDWs) in public alpha for all new projects. Wasm FDWs allow developers to create and run FDWs from URL-accessible storage like GitHub or S3, with improved security, simplified development, and near-native performance. The feature currently supports data sources with HTTP-based JSON APIs and includes two built-in Wasm FDWs for Snowflake and Paddle.

### Supabase — Introducing New Platform Access Control

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/platform-access-control
- Record: https://forck.live/items/21405-introducing-new-platform-access-control
- Subject: Supabase

Supabase introduced Platform Access Control, a Role-Based Access Control (RBAC) feature that lets administrators assign permissions at the organization or individual project level. The feature uses the existing roles (Owner, Administrator, Developer, Read-Only) but now allows scoping them to specific projects rather than the entire organization.

### Supabase — Top 10 Launches of Launch Week 12

- Company: Supabase (supabase.com)
- Announced: 2024-08-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/launch-week-12-top-10
- Record: https://forck.live/items/21404-top-10-launches-of-launch-week-12
- Subject: Supabase

Supabase published a recap of its Launch Week 12, highlighting ten updates including the open-sourcing of Snaplet under the MIT license, the release of pg_replicate for data copying, vec2pg for migrating vector databases, an official VS Code and GitHub Copilot extension, official Python library support, Log Drains for exporting logs, Realtime Broadcast and Presence Authorization, support for third-party Auth providers, Wasm-based Foreign Data Wrappers, and the in-browser Postgres sandbox postgres.new with an AI interface.

### Replicate — Replicate Intelligence #11

- Company: Replicate (replicate.com)
- Announced: 2024-08-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replicate.com/blog/replicate-intelligence-2024-08-16
- Record: https://forck.live/items/2517-replicate-intelligence-11
- Subject: Platform

Replicate announced that users can now fine-tune the FLUX.1 image generation model on their platform, allowing custom model training with 12-20 images and a trigger word.

### Vercel — Using the AI SDK to fix edge-case errors in our code

- Company: Vercel (vercel.com)
- Announced: 2024-08-15T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/using-the-ai-sdk-to-fix-edge-case-errors-in-our-code
- Record: https://forck.live/items/20957-using-the-ai-sdk-to-fix-edge-case-errors-in-our-code
- Subject: v0 / AI SDK / AI Gateway

Vercel used the AI SDK with GPT-4o to determine the correct language code for Punycode-encoded international domain names, fixing a bug that caused domain purchases with non-English characters to fail. The solution achieved a 100% success rate on previously failed purchases and has been deployed to production.

### OpenAI — Delivering contextual job matching for millions with OpenAI

- Company: OpenAI (openai.com)
- Announced: 2024-08-15T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/indeed
- Record: https://forck.live/items/723-delivering-contextual-job-matching-for-millions-with-openai
- Subject: GPT / ChatGPT / API

Indeed's mission and key statistics, but does not contain any announcement about an AI product or model.

### Nous Research — Freedom at the Frontier: Hermes 3

- Company: Nous Research (nousresearch.com)
- Announced: 2024-08-15T03:15:22+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://nousresearch.com/freedom-at-the-frontier-hermes-3
- Record: https://forck.live/items/24037-freedom-at-the-frontier-hermes-3
- Subject: Hermes models

Closed-source, “frontier” models today lack flexibility and adaptability. Many refuse to answer simple questions, hallucinate an authority’s form of morality, or require convoluted prompts in order to trigger a coherent answer.

### Supabase — pg_graphql 1.5.7: pagination and multi-tenancy support

- Company: Supabase (supabase.com)
- Announced: 2024-08-15
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/pg-graphql-1-5-7
- Record: https://forck.live/items/21412-pg-graphql-1-5-7-pagination-and-multi-tenancy-support
- Subject: Supabase

pg_graphql 1.5.7 introduces offset-based pagination (First/Offset), schema-based multi-tenancy that caches only permitted schemas per role, and filtering on array columns with contains, containedBy, and overlaps operators. The multi-tenancy improvement reduces cache rebuild overhead in environments with many tenants, demonstrated with a project of over 2,200 tenants.

### Supabase — Introducing Log Drains

- Company: Supabase (supabase.com)
- Announced: 2024-08-15
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/log-drains
- Record: https://forck.live/items/21411-introducing-log-drains
- Subject: Supabase

Supabase released Log Drains for all Team and Enterprise users, allowing developers to export logs from Database, Storage, Realtime, and Auth to external destinations like Datadog or custom HTTP endpoints. The feature supports extended retention for compliance and can be used for custom alerting and observability pipelines. Log Drains are available as a project add-on at $60 per month per project plus processing and egress fees.

### Replicate — Fine-tune FLUX.1 with your own images

- Company: Replicate (replicate.com)
- Announced: 2024-08-15
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replicate.com/blog/fine-tune-flux
- Record: https://forck.live/items/2518-fine-tune-flux-1-with-your-own-images
- Subject: Platform

Replicate announces that users can fine-tune FLUX.1 models using Ostris's AI Toolkit, which employs LoRA for fast and low-cost training (under 2 minutes, under $2). The post provides instructions for preparing training data and running fine-tunes via the web or API.

### Replit — How Replit makes sense of code at scale

- Company: Replit (replit.com)
- Announced: 2024-08-14T22:52:33+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replit.com/blog/how-replit-makes-sense-of-code-at-scale-ai-data
- Record: https://forck.live/items/17958-how-replit-makes-sense-of-code-at-scale
- Subject: Replit Agent

Replit describes its data infrastructure for analyzing user code at scale, covering over 300 million repositories and using Operational Transformation logs, execution data, and debugging trajectories. The company states it only uses public Repls for analytics and AI training, with all user code anonymized and PII removed. The post outlines challenges in reading, writing, and making sense of petabytes of code data while maintaining user privacy and accessibility for non-technical teams.

### Sarvam AI — Sarvam product launch

- Company: Sarvam AI (sarvam.ai)
- Announced: 2024-08-14T12:00:00+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sarvam.ai/blogs/sarvam-launch/
- Record: https://forck.live/items/18407-sarvam-product-launch
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI launched a full-stack generative AI platform for India, including voice-enabled multilingual agents (Sarvam Agents) starting at Rs. 1/min, the open‑source 2B‑parameter Indic LLM Sarvam 2B, the open‑source AudioLM Shuka 1.0, and an API platform for Indic models. The platform supports 10 Indian languages and targets sectors such as financial services, legal services, and consumer goods.

### OpenAI — Awakening Sleeping Beauties at The Met

- Company: OpenAI (openai.com)
- Announced: 2024-08-14T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/the-met-museum
- Record: https://forck.live/items/724-awakening-sleeping-beauties-at-the-met
- Subject: GPT / ChatGPT / API

OpenAI announces a collaborative exhibit with The Met's Costume Institute called 'Sleeping Beauties: Reawakening Fashion', highlighting AI's artistic potential.

### Supabase — Supabase Auth: Bring-your-own Auth0, Cognito, or Firebase

- Company: Supabase (supabase.com)
- Announced: 2024-08-14
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/third-party-auth-mfa-phone-send-hooks
- Record: https://forck.live/items/21414-supabase-auth-bring-your-own-auth0-cognito-or-firebase
- Subject: Supabase

Supabase announced three new features for its Auth product: support for third-party authentication providers (AWS Cognito, Auth0, and Firebase Auth in private alpha), phone-based multi-factor authentication via SMS and WhatsApp, and new Auth Hooks for customizing email and SMS sending via HTTP endpoints.

### Supabase — Snaplet is now open source

- Company: Supabase (supabase.com)
- Announced: 2024-08-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/snaplet-is-now-open-source
- Record: https://forck.live/items/21413-snaplet-is-now-open-source
- Subject: Supabase

Snaplet, a startup that is shutting down, has open-sourced three data tools under the MIT license: Copycat (deterministic fake data generation), Seed (synthetic data generation based on database schema), and Snapshot (database snapshot capture, transformation, and restoration with subsetting). The Snaplet team members who joined Supabase are migrating these projects to the Supabase GitHub organization and will maintain them going forward.

### LG AI Research — LG AI Research Unveils EXAONE 3.0 at ACL 2024

- Company: LG AI Research (lgresearch.ai)
- Announced: 2024-08-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/news/view?seq=464
- Record: https://forck.live/items/11021-lg-ai-research-unveils-exaone-3-0-at-acl-2024
- Subject: EXAONE

LG AI Research presented EXAONE 3.0 and ChatEXAONE at ACL 2024. The model is designed for high performance in real-world industrial use cases, excelling in multi-turn conversations, instruction execution, reasoning, and mathematics. The company also introduced ChatEXAONE, a generative AI service based on EXAONE 3.0, and discussed its future as an Enterprise AI Agent.

### LG AI Research — LG AI Research Unveils EXAONE 3.0 at ACL 2024

- Company: LG AI Research (lgresearch.ai)
- Announced: 2024-08-14
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.lgresearch.ai/blog/view?seq=464
- Record: https://forck.live/items/4837-lg-ai-research-unveils-exaone-3-0-at-acl-2024
- Subject: EXAONE

LG AI Research introduces EXAONE 3.0 and ChatEXAONE at ACL 2024, highlighting the model's performance in multi-turn conversations, instruction execution, reasoning, and mathematics.

### Hugging Face — A failed experiment: Infini-Attention, and why we should keep trying?

- Company: Hugging Face (huggingface.co)
- Announced: 2024-08-14
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/infini-attention
- Record: https://forck.live/items/1838-a-failed-experiment-infini-attention-and-why-we-should-keep-trying
- Subject: Platform

Infini-Attention, a method for extending context length, fails to perform well; its performance degrades with increased memory compression, and existing methods like ring attention, YaRN, and rope scaling remain superior.

### OpenAI — Introducing SWE-bench Verified

- Company: OpenAI (openai.com)
- Announced: 2024-08-13T10:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-swe-bench-verified
- Record: https://forck.live/items/725-introducing-swe-bench-verified
- Subject: GPT / ChatGPT / API

OpenAI is releasing a human-validated subset of SWE-bench to better evaluate AI models on real-world software issues.

### Supabase — Supabase Realtime: Broadcast and Presence Authorization

- Company: Supabase (supabase.com)
- Announced: 2024-08-13
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-realtime-broadcast-and-presence-authorization
- Record: https://forck.live/items/21415-supabase-realtime-broadcast-and-presence-authorization
- Subject: Supabase

Supabase released authorization for Realtime's Broadcast and Presence features using Row Level Security policies. Developers can create RLS policies on a new `realtime.messages` table to control which authenticated clients can send or receive messages on specific channels. The feature is available in Public Beta.

### xAI — Grok-2 Beta Release

- Company: xAI (x.ai)
- Announced: 2024-08-13
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://x.ai/news/grok-2
- Record: https://forck.live/items/4353-grok-2-beta-release
- Subject: Grok

xAI announces the beta release of Grok-2 and Grok-2 mini on the X platform, with an enterprise API coming later in the month. The model demonstrates competitive performance, outperforming Claude 3.5 Sonnet and GPT-4-Turbo on the LMSYS leaderboard.

### Hugging Face — Introduction to ggml

- Company: Hugging Face (huggingface.co)
- Announced: 2024-08-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/introduction-to-ggml
- Record: https://forck.live/items/1839-introduction-to-ggml
- Subject: Platform

Ggml, an open-source machine learning library written in C and C++ focused on Transformer inference. It highlights ggml's minimalism, easy compilation, lightweight nature, good hardware compatibility, support for quantized tensors, and memory efficiency, as well as current disadvantages such as limited backend support and breaking changes.

### Sakana AI — The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery

- Company: Sakana AI (sakana.ai)
- Announced: 2024-08-12T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/ai-scientist
- Record: https://forck.live/items/4729-the-ai-scientist-towards-fully-automated-open-ended-scientific-discovery
- Subject: Sakana models

The AI Scientist is a system for fully automatic scientific discovery that automates the entire research lifecycle, includes automated peer review, and costs approximately $15 per paper.

### Vercel — How to build scalable AI applications

- Company: Vercel (vercel.com)
- Announced: 2024-08-12T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-to-build-scalable-ai-applications
- Record: https://forck.live/items/20958-how-to-build-scalable-ai-applications
- Subject: v0 / AI SDK / AI Gateway

Vercel's guide covers best practices for building scalable AI applications, including model selection, data management, and infrastructure optimization using the Vercel AI SDK.

### TII — Welcome Falcon Mamba: The first strong attention-free 7B model

- Company: TII (tii.ae)
- Announced: 2024-08-12T12:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://falcon-lm.github.io/blog/falcon-mamba/
- Record: https://forck.live/items/18418-welcome-falcon-mamba-the-first-strong-attention-free-7b-model
- Subject: Falcon LLM

TII released Falcon Mamba, a 7B-parameter pure Mamba architecture model trained on ~5500GT of data. The model can process sequences of arbitrary length without increased memory storage and fits on a single A10 24GB GPU. It is open access under the TII Falcon Mamba 7B License 1.0 and available via Hugging Face.

### Supabase — postgres.new: In-browser Postgres with an AI interface

- Company: Supabase (supabase.com)
- Announced: 2024-08-12
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/postgres-new
- Record: https://forck.live/items/21417-postgres-new-in-browser-postgres-with-an-ai-interface
- Subject: Supabase

Supabase launched database.build (formerly postgres.new), an in-browser Postgres sandbox with an AI interface. It uses PGlite, a WASM version of Postgres, to run databases locally in the browser, paired with GPT-4o for AI assistance. Features include drag-and-drop CSV import, chart generation, ER diagrams, and SQL migrations, all without a remote server.

### Supabase — Official Supabase extension for VS Code and GitHub Copilot

- Company: Supabase (supabase.com)
- Announced: 2024-08-12
- Category: developer-tool-release
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/github-copilot-extension-for-vs-code
- Record: https://forck.live/items/21416-official-supabase-extension-for-vs-code-and-github-copilot
- Subject: Supabase

Supabase launched a GitHub Copilot extension for VS Code that includes a chat participant (@supabase) and a guided database migration feature. The extension also allows users to inspect tables, views, functions, storage buckets, and migration history directly from the editor. It requires the Supabase CLI and a local project, with future plans to integrate the Supabase Management API for hosted projects.

### Hugging Face — Tool Use, Unified

- Company: Hugging Face (huggingface.co)
- Announced: 2024-08-12
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/unified-tool-use
- Record: https://forck.live/items/1841-tool-use-unified
- Subject: Platform

Hugging Face announces a unified tool use API in the Transformers library that works across Mistral, Cohere, NousResearch, and Llama models, with helper functionality and complete documentation.

### Hugging Face — Welcome Falcon Mamba: The first strong attention-free 7B model

- Company: Hugging Face (huggingface.co)
- Announced: 2024-08-12
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/falconmamba
- Record: https://forck.live/items/1840-welcome-falcon-mamba-the-first-strong-attention-free-7b-model
- Subject: Platform

Falcon Mamba is a new 7B parameter model by TII, based on the Mamba architecture without attention, released under a custom license. It is open access and available on Hugging Face. It achieves competitive performance on various benchmarks and offers efficient processing of long sequences.