# forck.live — 18–24 August 2025

> 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 2025-W34, 18–24 August 2025, 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/2025-W34
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Sakana AI — Competition and Attraction Improve Model Fusion

- Company: Sakana AI (sakana.ai)
- Announced: 2025-08-24T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/m2n2
- Record: https://forck.live/items/4709-competition-and-attraction-improve-model-fusion
- Subject: Sakana models

Sakana AI announces a new research paper, 'Competition and Attraction Improve Model Fusion', presented at GECCO'25, which proposes M2N2 (Model Merging of Natural Niches), a method that evolves merging boundaries, uses competition for diversity, and employs attraction for mate selection to improve model fusion.

### Vercel — AI-powered prototyping with design systems

- Company: Vercel (vercel.com)
- Announced: 2025-08-22T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/ai-powered-prototyping-with-design-systems
- Record: https://forck.live/items/20831-ai-powered-prototyping-with-design-systems
- Subject: v0 / AI SDK / AI Gateway

Vercel explains how AI-native design systems enable v0 to generate brand-aware, production-aligned prototypes, and describes how design systems built for AI provide the context models need.

### OpenAI — Accelerating life sciences research

- Company: OpenAI (openai.com)
- Announced: 2025-08-22T08:30:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/accelerating-life-sciences-research-with-retro-biosciences
- Record: https://forck.live/items/489-accelerating-life-sciences-research
- Subject: GPT / ChatGPT / API

OpenAI discusses how a specialized AI model, GPT-4b micro, assisted in engineering more effective proteins for stem cell therapy and longevity research in collaboration with Retro Bio.

### Vercel — AI Gateway: Production-ready reliability for your AI apps

- Company: Vercel (vercel.com)
- Announced: 2025-08-21T13:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/ai-gateway-is-now-generally-available
- Record: https://forck.live/items/20832-ai-gateway-production-ready-reliability-for-your-ai-apps
- Subject: v0 / AI SDK / AI Gateway

Vercel made AI Gateway generally available, a service that provides failover, rate-limit handling, and unified billing across hundreds of models through a single API, with zero markup on token prices.

### Runway — Introducing Runway Game Worlds

- Company: Runway (runway.com)
- Announced: 2025-08-21T10:10:10.110000+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/research/runway-game-worlds
- Record: https://forck.live/items/7903-introducing-runway-game-worlds
- Subject: Gen / Aleph

Runway released the beta of Runway Game Worlds, an AI-powered platform for non-linear narrative experiences that generates personalized stories, characters, and media in real time. The beta includes preset games in various genres and allows users to create custom games, with images generated to complement gameplay.

### OpenAI — Blue J’s approach for scaling fast in complex, regulated domains

- Company: OpenAI (openai.com)
- Announced: 2025-08-21T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/blue-j
- Record: https://forck.live/items/4152-blue-j-s-approach-for-scaling-fast-in-complex-regulated-domains
- Subject: GPT / ChatGPT / API

OpenAI highlights how Blue J uses GPT-4.1 with Retrieval-Augmented Generation to deliver fast, accurate, and cited tax answers for professionals in the US, Canada, and the UK.

### OpenAI — Scaling domain expertise in complex, regulated domains

- Company: OpenAI (openai.com)
- Announced: 2025-08-21T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/blue-j
- Record: https://forck.live/items/490-scaling-domain-expertise-in-complex-regulated-domains
- Subject: GPT / ChatGPT / API

Blue J uses GPT-4.1 to build AI-powered tax research tools that combine domain expertise with Retrieval-Augmented Generation to deliver fast, accurate, and fully-cited tax answers trusted by professionals in the US, Canada, and the UK.

### Genspark — Introducing Genspark AI Designer — The AI Employee That Designs Anything with One Prompt

- Company: Genspark (genspark.ai)
- Announced: 2025-08-21
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/genspark-ai-designer
- Record: https://forck.live/items/18270-introducing-genspark-ai-designer-the-ai-employee-that-designs-anything-with
- Subject: Genspark

### Amp — Through the Agent, Into the Shell

- Company: Amp (ampcode.com)
- Announced: 2025-08-21
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/through-the-agent-into-the-shell
- Record: https://forck.live/items/7813-through-the-agent-into-the-shell
- Subject: Amp

Amp CLI introduced shell mode, allowing users to execute shell commands directly in the prompt input by starting a message with $. Commands and their output are included in the context window for the agent. An incognito mode ($$) executes commands without including them in the context.

### Amp — Multi-File Tab

- Company: Amp (ampcode.com)
- Announced: 2025-08-21
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/multi-file-tab
- Record: https://forck.live/items/7812-multi-file-tab
- Subject: Amp

Amp Tab now suggests edits across multiple recently-viewed files when an accepted edit introduces diagnostic errors.

### Amp — JetBrains Diagnostics for Java, Kotlin, and Scala

- Company: Amp (ampcode.com)
- Announced: 2025-08-21
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/jetbrains-diagnostics
- Record: https://forck.live/items/7811-jetbrains-diagnostics-for-java-kotlin-and-scala
- Subject: Amp

Amp CLI has added the ability to read diagnostics from JetBrains IDEs for JVM languages, including Java, Kotlin, and Scala.

### DeepSeek — DeepSeek-V3.1 Release

- Company: DeepSeek (deepseek.com)
- Announced: 2025-08-21
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.deepseek.com/en/news/deepseek-v3-1
- Record: https://forck.live/items/4323-deepseek-v3-1-release
- Subject: DeepSeek models

DeepSeek announces DeepSeek-V3.1, a model update with hybrid inference (Think & Non-Think), faster thinking, stronger agent skills, API updates with 128K context, open-source weights, and pricing changes.

### Together AI — How Together AI Uses AI Agents to Automate Complex Engineering Tasks: Lessons from Developing Efficient LLM Inference Systems

- Company: Together AI (together.ai)
- Announced: 2025-08-21
- Category: research-paper
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/ai-agents-to-automate-complex-engineering-tasks
- Record: https://forck.live/items/2413-how-together-ai-uses-ai-agents-to-automate-complex-engineering-tasks-lessons
- Subject: Inference platform

Together AI discusses patterns for building AI agents to automate complex engineering tasks, using a case study of accelerating LLM inference via speculative decoding.

### NVIDIA — NVIDIA Releases 6 Million Multi-Lingual Reasoning Dataset

- Company: NVIDIA (nvidia.com)
- Announced: 2025-08-20T22:13:18+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/nvidia/multilingual-reasoning-v1
- Record: https://forck.live/items/1631-nvidia-releases-6-million-multi-lingual-reasoning-dataset

NVIDIA released a 6 million multilingual reasoning dataset in five languages (French, Spanish, German, Italian, Japanese), translated from English reasoning data using Qwen2.5 models, and made it available under an open license.

### OpenAI — Mixi reimagines communication with ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2025-08-20T17:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/mixi
- Record: https://forck.live/items/491-mixi-reimagines-communication-with-chatgpt
- Subject: GPT / ChatGPT / API

MIXI, a Japanese digital entertainment and lifestyle services company, uses ChatGPT Enterprise to improve productivity, increase AI adoption across teams, and create a secure innovation environment.

### Supabase — Supabase Launch Week 15 Hackathon Winner Announcement

- Company: Supabase (supabase.com)
- Announced: 2025-08-20T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/lw15-hackathon-winners
- Record: https://forck.live/items/21335-supabase-launch-week-15-hackathon-winner-announcement
- Subject: Supabase

Supabase announced the winners of its Launch Week 15 Hackathon, with Figma AI Tickets winning Best Overall Project. The hackathon featured categories including Best Use of AI, Most Fun, Most Technically Impressive, and Most Visually Pleasing.

### Vercel — <script type="text/llms.txt">

- Company: Vercel (vercel.com)
- Announced: 2025-08-20T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/a-proposal-for-inline-llm-instructions-in-html
- Record: https://forck.live/items/20834-script-type-text-llms-txt
- Subject: v0 / AI SDK / AI Gateway

A proposal for inline LLM instructions in HTML based on llms.txt How do you tell an AI agent what it needs to do when it hits a protected page? Most systems rely on external documentation or pre-configured knowledge, but there's a simpler approach.

### Vercel — Rethinking prototyping, requirements, and project delivery at Code and Theory

- Company: Vercel (vercel.com)
- Announced: 2025-08-20T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/rethinking-prototyping-requirements-and-project-delivery-at-code-and-theory
- Record: https://forck.live/items/20833-rethinking-prototyping-requirements-and-project-delivery-at-code-and-theory
- Subject: v0 / AI SDK / AI Gateway

Code and Theory, a creative technology agency, reports using Vercel's v0 to reduce time-to-prototype by 75% and cut deployment horizons by 50-75%. The tool replaces traditional requirement documents and wireframes with prompt-driven workflows, enabling teams to go from idea to working software in minutes rather than weeks.

### Upstage — Upstage Completes $45M Series B Bridge to Accelerate Enterprise-Grade GenAI and Global Expansion

- Company: Upstage (upstage.ai)
- Announced: 2025-08-20
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/news/upstage-series-b-bridge
- Record: https://forck.live/items/18427-upstage-completes-45m-series-b-bridge-to-accelerate-enterprise-grade-genai-and
- Subject: Solar Pro / Solar Mini

Upstage raised $45 million in a Series B bridge round from KDB, Amazon, and AMD, bringing total funding to over $150 million. The company will use the funds to accelerate development of its Solar language model, advance document AI products, and expand in the U.S. and APAC markets.

### Amp — A New Model for Amp Tab

- Company: Amp (ampcode.com)
- Announced: 2025-08-20
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://ampcode.com/news/new-model-for-amp-tab
- Record: https://forck.live/items/7815-a-new-model-for-amp-tab
- Subject: Amp

Amp announced that its Amp Tab feature now uses an updated custom model to provide better completions. The model was trained on a rebuilt dataset with synthetic examples and uses DPO post-training to reduce poor suggestions, particularly for follow-up edits.

### Amp — From AGENT.md to AGENTS.md

- Company: Amp (ampcode.com)
- Announced: 2025-08-20
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/AGENTS.md
- Record: https://forck.live/items/7814-from-agent-md-to-agents-md
- Subject: Amp

Amp now supports AGENTS.md files (plural) in addition to the original AGENT.md files, maintaining backwards compatibility, after OpenAI adopted the plural form and the condition (OpenAI acquiring the agents.md domain) was met.

### Harvey — How Harvey’s Building a Culture of Privacy

- Company: Harvey (harvey.ai)
- Announced: 2025-08-19T14:00:00+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/how-harveys-building-a-culture-of-privacy
- Record: https://forck.live/items/17685-how-harvey-s-building-a-culture-of-privacy
- Subject: Harvey

Harvey, an AI platform for law firms and enterprises, describes its privacy and security practices, including commitments to not train on customer data, zero-day data retention, and no human review of customer data. The company has built infrastructure for EU- and Swiss-only data processing and allows customers to set data retention periods as short as three hours.

### Upstage — Upstage and AWS Join Forces to Drive AI Innovation in APAC and the United States

- Company: Upstage (upstage.ai)
- Announced: 2025-08-19
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/news/upstage-aws-sca-global
- Record: https://forck.live/items/18428-upstage-and-aws-join-forces-to-drive-ai-innovation-in-apac-and-the-united-states
- Subject: Solar Pro / Solar Mini

Upstage announced a strategic collaboration with AWS covering co-sales and go-to-market efforts, with Amazon making a minority investment in Upstage. Upstage will use AWS as its preferred cloud provider for building, training, and deploying future foundation models, leveraging SageMaker, Trainium, and Inferentia chips. The companies aim to empower governments and regulated industries with generative AI solutions across APAC and the US.

### Factory AI — Factory joins AGENTS.md collaboration with OpenAI

- Company: Factory AI (factory.com)
- Announced: 2025-08-19
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://factory.com/news/agents-md
- Record: https://forck.live/items/17814-factory-joins-agents-md-collaboration-with-openai
- Subject: Droid

Factory joined the industry working group convened by OpenAI to define AGENTS.md, an open standard for coding agents. The standard is a Markdown file placed at the repository root that captures operational details such as build commands, test instructions, and security notes. Factory is contributing to the draft specification and will support AGENTS.md across its products.

### Cartesia — Introducing Line: The Modern Voice Agent Development Platform

- Company: Cartesia (cartesia.ai)
- Announced: 2025-08-19
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.cartesia.ai/blog/introducing-line-for-voice-agents
- Record: https://forck.live/items/8015-introducing-line-the-modern-voice-agent-development-platform
- Subject: Sonic / Ink

Cartesia introduced Line, a code-first voice agent development platform. Line is available to all developers, with subscription tiers receiving prepaid credits. It integrates Cartesia's Sonic text-to-speech and Ink speech-to-text models.

### Amp — New Threads without Leaving the CLI

- Company: Amp (ampcode.com)
- Announced: 2025-08-19
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/new-threads-without-leaving-the-cli
- Record: https://forck.live/items/7816-new-threads-without-leaving-the-cli
- Subject: Amp

Amp has added the ability to start new threads and continue old threads directly from the CLI using the /new and /continue commands.

### Together AI — Transform OpenAI gpt-oss Models into Domain Experts with Together AI Fine-Tuning

- Company: Together AI (together.ai)
- Announced: 2025-08-19
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/fine-tune-gpt-oss-models-into-domain-experts-together-ai
- Record: https://forck.live/items/2414-transform-openai-gpt-oss-models-into-domain-experts-with-together-ai-fine-tuning
- Subject: Inference platform

Together AI announces that OpenAI's open-weight gpt-oss-120B and gpt-oss-20B models are now available on its platform for fine-tuning, enabling organizations to customize these models for domain-specific tasks using Together AI's fine-tuning infrastructure.

### Hugging Face — Generate Images with Claude and Hugging Face

- Company: Hugging Face (huggingface.co)
- Announced: 2025-08-19
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/claude-and-mcp
- Record: https://forck.live/items/1632-generate-images-with-claude-and-hugging-face
- Subject: Platform

To connect Claude to Hugging Face Spaces to generate images using the FLUX.1 Krea Dev and Qwen-Image models, enabling Claude to help build prompts and iterate on results.

### Alibaba — Qwen-Image-Edit: Image Editing with Higher Quality and Efficiency

- Company: Alibaba (alibaba.com)
- Announced: 2025-08-18T17:30:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwenlm.github.io/blog/qwen-image-edit/
- Record: https://forck.live/items/2282-qwen-image-edit-image-editing-with-higher-quality-and-efficiency
- Subject: Qwen

Alibaba introduces Qwen-Image-Edit, an image editing model built upon the 20B Qwen-Image model, extending text rendering capabilities to image editing for precise text editing, and using Qwen2.5-VL for visual semantic control and VAE Encoder for visual appearance control to enable both semantic and appearance editing.

### Lovable — Lovable is where vibe coding meets security

- Company: Lovable (lovable.dev)
- Announced: 2025-08-18T14:06:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/lovable-security
- Record: https://forck.live/items/12134-lovable-is-where-vibe-coding-meets-security
- Subject: Lovable / AI app builder

At Lovable, we know that trust is earned, not given. When you build, publish, and run apps on Lovable, you trust us with your ideas, your data, and your users.

### Vercel — If agents are building your app, who gets the W-2?

- Company: Vercel (vercel.com)
- Announced: 2025-08-18T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/if-agents-are-building-your-app-who-gets-the-w-2
- Record: https://forck.live/items/20835-if-agents-are-building-your-app-who-gets-the-w-2
- Subject: v0 / AI SDK / AI Gateway

Vercel argues that under U.S. GAAP (ASC 350-40), costs from AI agents performing capitalizable development work—such as coding, integration, and testing—can be capitalized like human developer salaries, rather than treated as overhead. The post explains that detailed logging now allows precise allocation of AI agent usage to the application development stage, and includes a quantitative example showing the impact on financial statements.

### Amp — Queue Messages in the CLI

- Company: Amp (ampcode.com)
- Announced: 2025-08-18
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/queue-messages-in-the-cli
- Record: https://forck.live/items/7817-queue-messages-in-the-cli
- Subject: Amp

Amp has added the ability to queue messages in its CLI. Users can use /queue <message> to enqueue a message and /dequeue to dequeue them. Queued messages are sent to the agent once it finishes its current turn.

### Hugging Face — MCP for Research: How to Connect AI to Research Tools

- Company: Hugging Face (huggingface.co)
- Announced: 2025-08-18
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/mcp-for-research
- Record: https://forck.live/items/1634-mcp-for-research-how-to-connect-ai-to-research-tools
- Subject: Platform

To use the Model Context Protocol (MCP) to connect AI agents to research discovery tools, covering three layers of abstraction: manual research, scripted tools, and MCP integration. It provides a setup guide for the Research Tracker MCP via Hugging Face MCP Settings.

### Hugging Face — From Zero to GPU: A Guide to Building and Scaling Production-Ready CUDA Kernels

- Company: Hugging Face (huggingface.co)
- Announced: 2025-08-18
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/kernel-builder
- Record: https://forck.live/items/1633-from-zero-to-gpu-a-guide-to-building-and-scaling-production-ready-cuda-kernels
- Subject: Platform

This is a tutorial guide on building and scaling production-ready CUDA kernels, introducing the kernel-builder library for developing, building, and deploying custom kernels via the Hugging Face Hub.

### OpenAI — Q&A with DoorDash’s CPO, Mariana Garavaglia

- Company: OpenAI (openai.com)
- Announced: 2025-08-18
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/doordash-mariana-garavaglia
- Record: https://forck.live/items/492-q-a-with-doordash-s-cpo-mariana-garavaglia
- Subject: GPT / ChatGPT / API

OpenAI published a Q&A with DoorDash's Chief People Officer about how DoorDash is scaling AI adoption to empower employees.