# forck.live — 2–8 March 2026

> 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 2026-W10, 2–8 March 2026, 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/2026-W10
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Google Research — WAXAL: A large-scale open resource for African language speech technology

- Company: Google Research (research.google)
- Announced: 2026-03-06T20:06:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/waxal-a-large-scale-open-resource-for-african-language-speech-technology/
- Record: https://forck.live/items/1362-waxal-a-large-scale-open-resource-for-african-language-speech-technology
- Subject: Research

Google Research introduces WAXAL, a large-scale open-access speech dataset for 27 Sub-Saharan African languages, featuring approximately 1,846 hours of ASR data and over 565 hours of TTS data, released under a CC-BY-4.0 license to support African speech technology.

### Google Research — Where wild things roam: Identifying wildlife with SpeciesNet

- Company: Google Research (research.google)
- Announced: 2026-03-06T17:59:38+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://research.google/blog/where-wild-things-roam-identifying-wildlife-with-speciesnet/
- Record: https://forck.live/items/1363-where-wild-things-roam-identifying-wildlife-with-speciesnet
- Subject: Research

SpeciesNet is an open-source AI model developed by Google Research that identifies wildlife in camera trap images. It was released open-source a year ago. It can classify nearly 2,500 animal categories and was trained on 65 million labeled images. The post highlights its performance and use by research groups worldwide.

### Sarvam AI — Open-Sourcing Sarvam 30B and 105B

- Company: Sarvam AI (sarvam.ai)
- Announced: 2026-03-06T12:00:00+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.sarvam.ai/blogs/sarvam-30b-105b/
- Record: https://forck.live/items/18388-open-sourcing-sarvam-30b-and-105b
- Subject: Sarvam / Bulbul / Saaras

Sarvam AI released two open-source reasoning models, Sarvam 30B and Sarvam 105B, trained from scratch on in-house datasets. Training was conducted entirely in India under the IndiaAI mission. Both models are already in production, powering Sarvam's conversational agent platform and AI assistant.

### OpenAI — Codex Security: now in research preview

- Company: OpenAI (openai.com)
- Announced: 2026-03-06T10:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/codex-security-now-in-research-preview
- Record: https://forck.live/items/238-codex-security-now-in-research-preview
- Subject: GPT / ChatGPT / API

OpenAI announces Codex Security, an AI application security agent now available in research preview.

### OpenAI — How Descript engineers multilingual video dubbing at scale

- Company: OpenAI (openai.com)
- Announced: 2026-03-06
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/descript
- Record: https://forck.live/items/240-how-descript-engineers-multilingual-video-dubbing-at-scale
- Subject: GPT / ChatGPT / API

Descript uses OpenAI reasoning models to automatically localize large content libraries while preserving timing and meaning.

### OpenAI — How Balyasny Asset Management built an AI research engine

- Company: OpenAI (openai.com)
- Announced: 2026-03-06
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/balyasny-asset-management
- Record: https://forck.live/items/239-how-balyasny-asset-management-built-an-ai-research-engine
- Subject: GPT / ChatGPT / API

Balyasny Asset Management built an AI research engine that uses OpenAI's full platform and agent workflows to transform investment research.

### Harvey — GPT-5.4, Now Live in Harvey

- Company: Harvey (harvey.ai)
- Announced: 2026-03-05T18:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/gpt-5-4-now-live-in-harvey
- Record: https://forck.live/items/17602-gpt-5-4-now-live-in-harvey
- Subject: Harvey

Harvey made GPT-5.4 available in its platform. The model, OpenAI's latest frontier model, scored 91.0% on Harvey's BigLaw Bench evaluation suite, up from GPT-5.2's 89.8%, with perfect scores on risk assessment and compliance tasks. It will roll out to eligible US and EU clients in the coming days, with AU availability to follow.

### Lovable — How Lovable built an industry intelligence tool

- Company: Lovable (lovable.dev)
- Announced: 2026-03-05T14:50:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/how-lovable-built-an-industry-intelligence-tool
- Record: https://forck.live/items/12100-how-lovable-built-an-industry-intelligence-tool
- Subject: Lovable / AI app builder

Lovable's post describes how its team built a competitive intelligence hub in less than a day, using the platform to automatically capture competitive signals from Slack and call transcripts and distribute them to the GTM team. The hub includes a shared feed, emerging threat tracking, a competitor cheatsheet, and win/loss case analysis, aiming to replace manual, periodic competitive intelligence processes. The post also provides a starter template for others to use.

### Hugging Face — Bringing Robotics AI to Embedded Platforms: Dataset Recording, VLA Fine‑Tuning, and On‑Device Optimizations

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-05T14:16:49+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/nxp/bringing-robotics-ai-to-embedded-platforms
- Record: https://forck.live/items/1539-bringing-robotics-ai-to-embedded-platforms-dataset-recording-vla-fine-tuning
- Subject: Platform

NXP's hands-on best practices for recording reliable robotic datasets, fine-tuning VLA policies (ACT and SmolVLA), and highlights the real-time performance that NXP i.MX 95 SoC achieves after optimization. It covers dataset recording tips, use of gripper camera, prehension improvements, and diversity in data splits.

### Vercel — From idea to secure checkout in minutes with Stripe

- Company: Vercel (vercel.com)
- Announced: 2026-03-05T13:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/from-idea-to-secure-checkout-in-minutes-with-stripe
- Record: https://forck.live/items/20736-from-idea-to-secure-checkout-in-minutes-with-stripe
- Subject: v0 / AI SDK / AI Gateway

Vercel announced that Stripe is now generally available on the Vercel Marketplace and in v0, with full support for connecting production accounts. The integration allows developers to connect an existing live Stripe account directly to a Vercel project or import one into their environment, with environment variables provisioned automatically. The release includes a new set of key management APIs built in partnership with Stripe, enabling cryptographic key exchange rather than manual credential handling.

### OpenAI — Reasoning models struggle to control their chains of thought, and that’s good

- Company: OpenAI (openai.com)
- Announced: 2026-03-05T10:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/reasoning-models-chain-of-thought-controllability
- Record: https://forck.live/items/243-reasoning-models-struggle-to-control-their-chains-of-thought-and-that-s-good
- Subject: GPT / ChatGPT / API

OpenAI introduces CoT-Control and finds reasoning models struggle to control their chains of thought, reinforcing monitorability as an AI safety safeguard.

### OpenAI — Introducing GPT-5.4

- Company: OpenAI (openai.com)
- Announced: 2026-03-05T10:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/introducing-gpt-5-4
- Record: https://forck.live/items/241-introducing-gpt-5-4
- Subject: GPT / ChatGPT / API

OpenAI announces GPT-5.4, a frontier model with state-of-the-art coding, computer use, tool search, and a 1M-token context window.

### OpenAI — Ensuring AI use in education leads to opportunity

- Company: OpenAI (openai.com)
- Announced: 2026-03-05T09:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/ai-education-opportunity
- Record: https://forck.live/items/244-ensuring-ai-use-in-education-leads-to-opportunity
- Subject: GPT / ChatGPT / API

OpenAI announces new tools, certifications, and measurement resources for schools and universities to close AI capability gaps and expand opportunity.

### Supabase — Supabase Storage: major performance, security, and reliability updates

- Company: Supabase (supabase.com)
- Announced: 2026-03-05
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-storage-performance-security-reliability-updates
- Record: https://forck.live/items/21300-supabase-storage-major-performance-security-and-reliability-updates
- Subject: Supabase

Supabase shipped a large set of updates to its Storage service, including a full rewrite of object listing using a hybrid skip-scan algorithm and cursor-based pagination, which on a table with 60 million rows makes deep pagination up to 14.8 times faster. The update also closes a path traversal vulnerability in the file backend, blocks direct SQL deletes that could orphan objects, adds query cancellation and statement timeouts, and makes all migrations idempotent. These changes are live in all regions and affect every project using Storage.

### Supabase — Log Drains: Now available on Pro

- Company: Supabase (supabase.com)
- Announced: 2026-03-05
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/log-drains-now-available-on-pro
- Record: https://forck.live/items/21299-log-drains-now-available-on-pro
- Subject: Supabase

Supabase made Log Drains available on the Pro tier, allowing Pro users to send logs from Postgres, API Gateway, Auth, Storage, Edge Functions, and Realtime to external destinations such as Sentry, Grafana Loki, Datadog, and AWS S3. The feature batches up to 250 logs per batch or flushes every second, with gzip compression supported.

### Krea — Turn Any Image Into a Prompt

- Company: Krea (krea.ai)
- Announced: 2026-03-05
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/image-to-prompt
- Record: https://forck.live/items/15087-turn-any-image-into-a-prompt
- Subject: Krea 2 / AI creative platform

Krea introduced a feature that generates a detailed prompt from any image dropped into its image tool. The prompt, typically 30 to 100 words, is tuned for models like Flux and K1 and streams into the prompt box in batches of five words every 60ms. The tool analyzes medium, style, subject, lighting, composition, color, environment, and visible text.

### Amp — GPT-5.4, The New Oracle

- Company: Amp (ampcode.com)
- Announced: 2026-03-05
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://ampcode.com/news/gpt-5.4-the-new-oracle
- Record: https://forck.live/items/7754-gpt-5-4-the-new-oracle
- Subject: Amp

Amp introduced GPT-5.4 as an oracle in its smart mode, complementing Opus 4.6, and also added it to deep mode alongside GPT-5.3-Codex. Internal evaluations show improvement in response quality and reduced latency compared to GPT-5.2.

### LG AI Research — [AAAI 2026] RL-Studio: A System for Multi-Phase Reinforcement Learning Experimentation

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-03-05
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=631
- Record: https://forck.live/items/4757-aaai-2026-rl-studio-a-system-for-multi-phase-reinforcement-learning
- Subject: EXAONE

LG AI Research presented RL-Studio, a platform for multi-phase reinforcement learning experimentation, at the AAAI 2026 Demonstration Program.

### Cursor — Automations

- Company: Cursor (cursor.com)
- Announced: 2026-03-05
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/03-05-26
- Record: https://forck.live/items/2627-automations
- Subject: Cursor

Cursor introduces automations, allowing users to create always-on agents that run on schedules or trigger from events (Slack, Linear, GitHub, PagerDuty, webhooks). Agents use cloud sandboxes and configured MCPs and models, with a memory tool for learning from past runs.

### Together AI — FlashAttention-4: Algorithm and Kernel Pipelining Co-Design for Asymmetric Hardware Scaling

- Company: Together AI (together.ai)
- Announced: 2026-03-05
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/flashattention-4
- Record: https://forck.live/items/2374-flashattention-4-algorithm-and-kernel-pipelining-co-design-for-asymmetric
- Subject: Inference platform

Together AI presents FlashAttention-4, an algorithm and kernel co-design for Blackwell GPUs that maximizes overlap between matmul and resource bottlenecks like exponential units and shared memory traffic, achieving up to 1605 TFLOPs/s on B200 with BF16.

### Together AI — Key research and product announcements at the AI Native Conf

- Company: Together AI (together.ai)
- Announced: 2026-03-05
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/ai-native-conf-research-and-product-announcements
- Record: https://forck.live/items/2373-key-research-and-product-announcements-at-the-ai-native-conf
- Subject: Inference platform

Together AI announces seven research and product releases at the AI Native Conf, including FlashAttention-4, the Reinforcement Learning API, ThunderAgent, together.compile, and Megakernel, covering kernels, reinforcement learning, and algorithmic inference optimization.

### Hugging Face — Introducing Modular Diffusers - Composable Building Blocks for Diffusion Pipelines

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-05
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/modular-diffusers
- Record: https://forck.live/items/1540-introducing-modular-diffusers-composable-building-blocks-for-diffusion-pipelines
- Subject: Platform

Hugging Face introduces Modular Diffusers, a new composable block-based system for building diffusion pipelines, allowing users to mix and match reusable blocks instead of writing entire pipelines from scratch. The post includes a quickstart example using FLUX.2 Klein 4B, instructions for creating custom blocks (e.g., a DepthProcessorBlock using Depth Anything V2), and integration with a node-based visual interface called Mellon.

### OpenAI — The five AI value models driving business reinvention

- Company: OpenAI (openai.com)
- Announced: 2026-03-05
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/the-five-ai-value-models-driving-business-reinvention
- Record: https://forck.live/items/248-the-five-ai-value-models-driving-business-reinvention
- Subject: GPT / ChatGPT / API

OpenAI outlines five AI value models that guide leaders in sequencing AI adoption from workforce fluency to process reinvention for durable business advantage.

### OpenAI — Introducing ChatGPT for Excel and new financial data integrations

- Company: OpenAI (openai.com)
- Announced: 2026-03-05
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/chatgpt-for-excel
- Record: https://forck.live/items/247-introducing-chatgpt-for-excel-and-new-financial-data-integrations
- Subject: GPT / ChatGPT / API

OpenAI launches ChatGPT for Excel alongside new financial data integrations, leveraging GPT-5.4 to speed up modeling, research, and analysis in regulated environments.

### OpenAI — Introducing the Adoption news channel

- Company: OpenAI (openai.com)
- Announced: 2026-03-05
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/introducing-the-adoption-news-channel
- Record: https://forck.live/items/246-introducing-the-adoption-news-channel
- Subject: GPT / ChatGPT / API

OpenAI introduces a news channel that offers practical insights and frameworks to help businesses turn AI progress into business advantage.

### Lovable — Routing Billions of Tokens per Minute

- Company: Lovable (lovable.dev)
- Announced: 2026-03-04T23:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/routing-billions-of-tokens-per-minute
- Record: https://forck.live/items/12101-routing-billions-of-tokens-per-minute
- Subject: Lovable / AI app builder

Lovable's infrastructure team describes how it routes traffic across multiple LLM providers at over one billion tokens per minute, using multiple fallback chains with project-level affinity to preserve prompt caching and automatically adjusting provider weights via a PID controller to handle outages and rate limits.

### Google Research — Teaching LLMs to reason like Bayesians

- Company: Google Research (research.google)
- Announced: 2026-03-04T20:29:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/teaching-llms-to-reason-like-bayesians/
- Record: https://forck.live/items/1364-teaching-llms-to-reason-like-bayesians
- Subject: Research

Google Research teaches LLMs to reason in a Bayesian manner by training them to mimic the predictions of an optimal Bayesian model, improving performance on a recommendation task and enabling generalization to other tasks.

### Harvey — Harvey Accelerates Enterprise AI With Agent‑Powered Platform and Microsoft 365 Copilot

- Company: Harvey (harvey.ai)
- Announced: 2026-03-04T13:30:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/agent-platform-and-microsoft-365-copilot
- Record: https://forck.live/items/17603-harvey-accelerates-enterprise-ai-with-agent-powered-platform-and-microsoft-365
- Subject: Harvey

Harvey announced an integration with Microsoft 365 Copilot, enabling legal teams to access its legal intelligence directly in Copilot and use agentic document capabilities in Word. The integration aims to reduce time spent switching between tools and support legal work from questions to execution.

### OpenAI — Extending single-minus amplitudes to gravitons

- Company: OpenAI (openai.com)
- Announced: 2026-03-04T10:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/extending-single-minus-amplitudes-to-gravitons
- Record: https://forck.live/items/249-extending-single-minus-amplitudes-to-gravitons
- Subject: GPT / ChatGPT / API

OpenAI announces a new preprint that extends single-minus amplitudes to gravitons, with GPT-5.2 Pro assisting in deriving and verifying nonzero graviton tree amplitudes in quantum gravity.

### Swiss AI Initiative — GPAI Training Transparency

- Company: Swiss AI Initiative (swiss-ai.org)
- Announced: 2026-03-04T09:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://apertus-ai.org/articles/2026-02-gpai/
- Record: https://forck.live/items/18378-gpai-training-transparency
- Subject: Apertus

Researchers at the AI Accountability Lab in Trinity College Dublin developed a documentation quality assessment framework for evaluating transparency and usefulness of AI model documentation under the EU AI Act. They found that many AI companies are not complying with the requirement to publish summaries on model training data, and that the existing EU template for summaries led to inconsistent submissions. The researchers recommend a centralized portal to host these summaries and have shared their findings, including evaluations of models such as Switzerland's Apertus, through a website and report accepted at the FAccT conference.

### Ant Ling — Ming-Omni-TTS: Simple and Efficient Unified Generation of Speech, Music, and Sound with Precise Control

- Company: Ant Ling (ant-ling.com)
- Announced: 2026-03-04
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://developer.ant-ling.com/en/blogs/ming-omni-tts
- Record: https://forck.live/items/22333-ming-omni-tts-simple-and-efficient-unified-generation-of-speech-music-and
- Subject: Ling / Ring / Ming

Ant Ling introduced Ming-Omni-TTS, a unified audio generation model that can synthesize speech, environmental sounds, and music in a single channel. The model features fine-grained vocal control over speech rate, pitch, volume, emotion, and dialects, achieving 93% accuracy for Cantonese and 46.7% for emotional control. It also includes over 100 built-in voices, supports zero-shot voice design via natural language descriptions, and uses a Patch-by-Patch compression strategy to reduce inference frame rate to 3.1Hz.

### LG AI Research — [LG Accountability Report on AI Ethics EP.2] Proving Safety, Ensuring Trust: Inside the 2025 Report

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-03-04
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/news/view?seq=630
- Record: https://forck.live/items/10994-lg-accountability-report-on-ai-ethics-ep-2-proving-safety-ensuring-trust
- Subject: EXAONE

LG AI Research published its 2025 Accountability Report on AI Ethics, detailing its commitment to safety and trustworthiness. The company established K-AUT (Korea-Augmented Universal Taxonomy), an AI risk taxonomy with 226 detailed items reflecting universal human values and Korea's legal and cultural characteristics. LG AI Research also conducted research on hallucination mitigation in multimodal AI and deepfake detection, fostered an internal AI ethics culture through seminars, and became the first Korean company to serve as an Authorized Assessor for IEEE CertifAIEd, certifying LG Electronics' AI home hub as the first private-sector product to receive the certification.

### LG AI Research — [LG Accountability Report on AI Ethics EP.2] Proving Safety, Ensuring Trust: Inside the 2025 Report

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-03-04
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=630
- Record: https://forck.live/items/4758-lg-accountability-report-on-ai-ethics-ep-2-proving-safety-ensuring-trust
- Subject: EXAONE

LG AI Research published a report detailing its 2025 progress in AI ethics and trustworthiness, including the K-AUT risk taxonomy, hallucination mitigation research, deepfake detection, internal ethics culture programs, the EXAONE and K-EXAONE open-source models, AI education initiatives, and being the first Korean company to achieve IEEE CertifAIEd product certification for LG ThinQ On.

### Cursor — Cursor in JetBrains IDEs

- Company: Cursor (cursor.com)
- Announced: 2026-03-04
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/03-04-26
- Record: https://forck.live/items/2628-cursor-in-jetbrains-ides
- Subject: Cursor

Cursor is now available in JetBrains IDEs (IntelliJ IDEA, PyCharm, WebStorm, etc.) through the Agent Client Protocol (ACP), enabling developers to use frontier models from OpenAI, Anthropic, Google, and Cursor for agent-driven development.

### Together AI — Cache-aware prefill–decode disaggregation (CPD) for up to 40% faster long-context LLM serving

- Company: Together AI (together.ai)
- Announced: 2026-03-04
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/cache-aware-disaggregated-inference
- Record: https://forck.live/items/2375-cache-aware-prefill-decode-disaggregation-cpd-for-up-to-40-faster-long-context
- Subject: Inference platform

Together AI announces cache-aware prefill–decode disaggregation (CPD), a serving architecture that separates cold and warm workloads by cache hit rate to improve throughput and reduce time-to-first-token for long-context inference.

### OpenAI — Understanding AI and learning outcomes

- Company: OpenAI (openai.com)
- Announced: 2026-03-04
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/understanding-ai-and-learning-outcomes
- Record: https://forck.live/items/251-understanding-ai-and-learning-outcomes
- Subject: GPT / ChatGPT / API

OpenAI introduces the Learning Outcomes Measurement Suite, a tool to assess AI's impact on student learning over time.

### OpenAI — How Axios uses AI to help deliver high-impact local journalism

- Company: OpenAI (openai.com)
- Announced: 2026-03-04
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/axios-allison-murphy
- Record: https://forck.live/items/250-how-axios-uses-ai-to-help-deliver-high-impact-local-journalism
- Subject: GPT / ChatGPT / API

Axios COO Allison Murphy explains how the company uses AI to support local reporters, streamline newsroom workflows, and deliver high-impact local journalism at scale.

### Hugging Face — PRX Part 3 — Training a Text-to-Image Model in 24h!

- Company: Hugging Face (huggingface.co)
- Announced: 2026-03-03T16:50:49+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/Photoroom/prx-part3
- Record: https://forck.live/items/1541-prx-part-3-training-a-text-to-image-model-in-24h
- Subject: Platform

Photoroom describes a 24-hour training speedrun for a text-to-image model using a combination of techniques such as x-prediction in pixel space, perceptual losses, token routing with TREAD, and representation alignment with REPA, and open-sources the code.

### Google DeepMind — Gemini 3.1 Flash-Lite: Built for intelligence at scale

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-03-03T16:35:55+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://deepmind.google/blog/gemini-3-1-flash-lite-built-for-intelligence-at-scale/
- Record: https://forck.live/items/1249-gemini-3-1-flash-lite-built-for-intelligence-at-scale
- Subject: Gemini

Google DeepMind announces Gemini 3.1 Flash-Lite as the fastest and most cost-efficient model in the Gemini 3 series.

### Harvey — Work Securely From Anywhere With Harvey Mobile

- Company: Harvey (harvey.ai)
- Announced: 2026-03-03T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/work-from-anywhere-with-harvey
- Record: https://forck.live/items/17604-work-securely-from-anywhere-with-harvey-mobile
- Subject: Harvey

Legal teams can get up to speed, capture new information in real time, and keep work moving from anywhere with Harvey Mobile. Legal work doesn’t always happen at a desk.

### Vercel — Scaling redirects to infinity on Vercel

- Company: Vercel (vercel.com)
- Announced: 2026-03-03T13:00:00+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/scaling-redirects-to-infinity-on-vercel
- Record: https://forck.live/items/20738-scaling-redirects-to-infinity-on-vercel
- Subject: v0 / AI SDK / AI Gateway

Vercel developed a new redirect system that uses sharded Bloom filters and binary search to support millions of static redirects per project with low latency and memory usage. The system avoids parsing the entire redirect set on each request by first checking a Bloom filter and then only loading a small shard when a match is possible. JSON parsing was identified as a bottleneck on positive lookups, leading to a binary search approach over sorted keys.

### Vercel — Building Slack agents can be easy

- Company: Vercel (vercel.com)
- Announced: 2026-03-03T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/building-slack-agents-can-be-easy
- Record: https://forck.live/items/20737-building-slack-agents-can-be-easy
- Subject: v0 / AI SDK / AI Gateway

eve , our open-source framework for building agents, is now available. Build your Slack agent with the Slack channel now or follow the guide .

### Swiss AI Initiative — SME Circle #3

- Company: Swiss AI Initiative (swiss-ai.org)
- Announced: 2026-03-03T09:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://apertus-ai.org/articles/2026-03-sme-circle/
- Record: https://forck.live/items/18379-sme-circle-3
- Subject: Apertus

The Swiss AI Initiative held the third edition of its SME Circle at EPFL's Industry Day, gathering nearly 40 participants from Swiss SMEs. The roundtable identified a strong demand for sovereign AI alternatives to hyperscalers, with Apertus seen as a key solution. Key themes included data sovereignty, practical integration for SMEs, governance and ethics, and community collaboration.

### World Labs — 3D as code

- Company: World Labs (worldlabs.ai)
- Announced: 2026-03-03
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.worldlabs.ai/blog/3d-as-code
- Record: https://forck.live/items/7342-3d-as-code
- Subject: Marble / Atlas

World Labs argues that 3D representations serve as a spatial analogue to code, enabling communication, collaboration, and creation between people and machines. The company draws parallels between how large language models use text as a universal interface and how world models can use structured 3D representations to integrate with existing tools and pipelines.

### fal — HeyGen models are now available on fal

- Company: fal (fal.ai)
- Announced: 2026-03-02T21:04:48+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://blog.fal.ai/heygen-models-are-now-available-on-fal/
- Record: https://forck.live/items/8002-heygen-models-are-now-available-on-fal
- Subject: fal platform

HeyGen's models are now available on the fal platform, providing capabilities such as a video agent API, avatar IV API, video translation, and avatar video generation.

### Vercel — Advancing Python typing

- Company: Vercel (vercel.com)
- Announced: 2026-03-02T13:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/advancing-python-typing
- Record: https://forck.live/items/20739-advancing-python-typing
- Subject: v0 / AI SDK / AI Gateway

Vercel shared a year-long research effort resulting in PEP 827, which proposes standard type-level building blocks for introspecting and constructing types in Python. The goal is to make Python's static typing more expressive and composable, similar to TypeScript's utility types but tailored for Python's runtime model.

### Harvey — Harvey FORUM: Redefining the Future of Legal

- Company: Harvey (harvey.ai)
- Announced: 2026-03-02T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-forum-redefining-the-future-of-legal
- Record: https://forck.live/items/17605-harvey-forum-redefining-the-future-of-legal
- Subject: Harvey

Harvey hosted FORUM, a two-day event for law firm and in-house legal leaders, featuring keynotes, panels, and case studies on AI adoption. A central takeaway was that AI has moved from experimentation to foundational infrastructure for leading organizations. The event also highlighted Harvey's partnership with Intapp and its investment in features like Memory, agentic workflows, and Shared Spaces.

### Factory AI — Factory Expands to London

- Company: Factory AI (factory.com)
- Announced: 2026-03-02
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/factory-london
- Record: https://forck.live/items/17841-factory-expands-to-london
- Subject: Droid

Factory announced it will open a London office on March 9, 2026, to serve its European customers and is hiring for sales, customer success, marketing, operations, deployed engineering, and solutions engineering roles.

### Krea — Introducing Voice Mode

- Company: Krea (krea.ai)
- Announced: 2026-03-02
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/voice-mode
- Record: https://forck.live/items/15089-introducing-voice-mode
- Subject: Krea 2 / AI creative platform

Krea released Voice Mode for its iPad app, allowing users to give spoken commands like "make the background warmer" while drawing, with the canvas updating in real time. The feature is available now in Krea for iPad.

### Krea — Krea Redesign

- Company: Krea (krea.ai)
- Announced: 2026-03-02
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/redesign
- Record: https://forck.live/items/15088-krea-redesign
- Subject: Krea 2 / AI creative platform

Krea redesigned every screen with a unified sidebar, clearer model selection, drag-and-drop between tools, customizable workspace, and a rebuilt mobile experience. The homepage is now a third its previous size, animation delays were removed, and sessions persist across tools.

### Together AI — Introducing Together AI’s new look

- Company: Together AI (together.ai)
- Announced: 2026-03-02
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/introducing-together-ai-new-look
- Record: https://forck.live/items/2376-introducing-together-ai-s-new-look
- Subject: Inference platform

Together AI announced a refreshed visual brand identity, including a new logo and design language created with Pentagram, to better reflect its mission of open innovation and its platform connecting research, infrastructure, and builders.