# forck.live — 3–9 November 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-W45, 3–9 November 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-W45
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Replit — Build AI Apps Instantly with Replit AI Integrations

- Company: Replit (replit.com)
- Announced: 2025-11-08T01:11:24+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replit.com/blog/ai-integrations
- Record: https://forck.live/items/17935-build-ai-apps-instantly-with-replit-ai-integrations
- Subject: Replit Agent

Replit introduced AI Integrations, which give users instant access to 300+ AI models from providers including GPT, Gemini, Claude, Llama, Grok, and Mistral directly inside Replit, with no setup, API keys, or extra accounts required. The feature lets users add AI to projects by asking the Replit Agent, such as building a chatbot or adding GPT-5 to a project.

### Google Research — Introducing Nested Learning: A new ML paradigm for continual learning

- Company: Google Research (research.google)
- Announced: 2025-11-07T17:37:22+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/introducing-nested-learning-a-new-ml-paradigm-for-continual-learning/
- Record: https://forck.live/items/1393-introducing-nested-learning-a-new-ml-paradigm-for-continual-learning
- Subject: Research

Google Research introduces Nested Learning, a new ML paradigm for continual learning presented at NeurIPS 2025.

### Harvey — Introducing BigLaw Bench: Arena

- Company: Harvey (harvey.ai)
- Announced: 2025-11-07T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/introducing-biglaw-bench-arena
- Record: https://forck.live/items/17652-introducing-biglaw-bench-arena
- Subject: Harvey

Harvey introduced BigLaw Bench: Arena, an internal system for head-to-head evaluation of AI systems on legal tasks through pairwise comparison by lawyers. The system produces Elo scores and qualitative feedback to determine which models experts prefer. Harvey also published recent results comparing its Harvey Assistant against prior implementations and leading foundation models.

### Vercel — How Nous Research used BotID to block automated abuse at scale

- Company: Vercel (vercel.com)
- Announced: 2025-11-07T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-nous-research-used-botid-to-block-automated-abuse-at-scale
- Record: https://forck.live/items/20791-how-nous-research-used-botid-to-block-automated-abuse-at-scale
- Subject: v0 / AI SDK / AI Gateway

Vercel's case study documents how Nous Research used Vercel BotID Deep Analysis to block automated abuse during a free tier promotion for its Hermes model, preventing a coordinated attack that spiked traffic by 3,000% without impacting legitimate users.

### OpenAI — Understanding prompt injections: a frontier security challenge

- Company: OpenAI (openai.com)
- Announced: 2025-11-07T11:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/prompt-injections
- Record: https://forck.live/items/392-understanding-prompt-injections-a-frontier-security-challenge
- Subject: GPT / ChatGPT / API

OpenAI explains prompt injection attacks as a frontier security challenge and describes its efforts in research, model training, and building safeguards.

### Thinking Machines Lab — Tinker: Call for Community Projects

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2025-11-07
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://thinkingmachines.ai/news/call-for-community-projects
- Record: https://forck.live/items/18562-tinker-call-for-community-projects
- Subject: Thinking Machines / Inkling

Thinking Machines Lab is inviting the community to submit projects built with its Tinker training platform for potential feature on its blog, and has suggested several research directions including replicating Constitutional AI and RLVR with noisy student.

### Inception — The Next Step for dLLMs: Scaling up Mercury

- Company: Inception (inceptionlabs.ai)
- Announced: 2025-11-07
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.inceptionlabs.ai/blog/mercury-refreshed
- Record: https://forck.live/items/18527-the-next-step-for-dllms-scaling-up-mercury
- Subject: Mercury

Inception released an upgraded version of its Mercury diffusion-based LLM, featuring larger models, architectural upgrades, and training improvements that deliver gains in coding, instruction following, math, and knowledge recall. The model is available on Inception's API platform and through partners OpenRouter and Models.dev, with pricing unchanged at $0.25 per 1M input tokens and $1.00 per 1M output tokens. The company also announced a $50M financing round led by Menlo Ventures.

### Amp — Rush Mode

- Company: Amp (ampcode.com)
- Announced: 2025-11-07
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/rush-mode
- Record: https://forck.live/items/7785-rush-mode
- Subject: Amp

Amp released a new mode called 'rush' that is 67% cheaper and 50% faster than smart mode token-by-token, but less capable. It is designed for small, well-defined tasks and uses Haiku 4.5 as the underlying model.

### Baidu — ERNIE-5.0-Preview-1022, ready for testing in LMArena!

- Company: Baidu (baidu.com)
- Announced: 2025-11-07
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://ernie.baidu.com/blog/posts/ernie-5.0-preview-1022-release-on-lmarena/
- Record: https://forck.live/items/4651-ernie-5-0-preview-1022-ready-for-testing-in-lmarena
- Subject: ERNIE

Baidu announces that ERNIE-5.0-Preview-1022 ranks #2 on LMArena Text leaderboard and plans to officially release it soon.

### Google Research — DS-STAR: A state-of-the-art versatile data science agent

- Company: Google Research (research.google)
- Announced: 2025-11-06T17:50:36+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/ds-star-a-state-of-the-art-versatile-data-science-agent/
- Record: https://forck.live/items/1394-ds-star-a-state-of-the-art-versatile-data-science-agent
- Subject: Research

Google Research introduces DS-STAR, a new data science agent that automates tasks like statistical analysis, visualization, and data wrangling across diverse data formats. It features a data file analysis module, an LLM-based verifier for plan assessment, and iterative sequential planning. DS-STAR achieves state-of-the-art performance on the DABStep, KramaBench, and DA-Code benchmarks, outperforming existing methods like AutoGen and DA-Agent.

### Harvey — The Brief: November 2025

- Company: Harvey (harvey.ai)
- Announced: 2025-11-06T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/the-brief-november-2025
- Record: https://forck.live/items/17653-the-brief-november-2025
- Subject: Harvey

Harvey announced November 2025 updates including an Outlook add-in, iManage integration, redact and template workflows in Word, multi-source querying, agentic search, and GPT-5 as the default model for chat and draft responses. New localized legal knowledge sources for Austria, Germany, and Canada were added, along with improved admin controls, user groups, and project ownership transfer.

### Vercel — What we learned building agents at Vercel

- Company: Vercel (vercel.com)
- Announced: 2025-11-06T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/what-we-learned-building-agents-at-vercel
- Record: https://forck.live/items/20793-what-we-learned-building-agents-at-vercel
- Subject: v0 / AI SDK / AI Gateway

Vercel describes a methodology for finding internal AI agent projects that have a high likelihood of business impact, focusing on tasks requiring low cognitive load and high repetition. The post details two internal agents: a lead processing agent that reduced a team from 10 people to 1, and an anti-abuse agent that led to a 59% reduction in time to ticket closing. Vercel also open-sourced agent templates and offers a hands-on program for customer support.

### Vercel — How AI Gateway runs on Fluid compute

- Company: Vercel (vercel.com)
- Announced: 2025-11-06T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/how-ai-gateway-runs-on-fluid-compute
- Record: https://forck.live/items/20792-how-ai-gateway-runs-on-fluid-compute
- Subject: v0 / AI SDK / AI Gateway

Vercel's AI Gateway runs on Fluid compute, which uses Active CPU Pricing to charge only for active CPU time. During its first month, AI Gateway paid CPU rates for less than 8% of runtime hours, as most time was spent waiting for AI provider responses. Fluid also enables in-function concurrency and in-memory caching to reduce latency and costs.

### ElevenLabs — Building the First Agentic Government with Ukraine

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2025-11-06T12:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/building-the-first-agentic-government-with-ukraine
- Record: https://forck.live/items/3382-building-the-first-agentic-government-with-ukraine
- Subject: Voice AI

Announcement of building the first agentic government with Ukraine.

### Factory AI — Welcome Chris Degnan

- Company: Factory AI (factory.com)
- Announced: 2025-11-06
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://factory.com/news/degnan
- Record: https://forck.live/items/17828-welcome-chris-degnan
- Subject: Droid

Factory announced that Chris Degnan, formerly Snowflake's first sales hire and later CRO, has joined the company as an advisor to help scale its enterprise deployments.

### Microsoft AI — Towards Humanist Superintelligence | Microsoft AI

- Company: Microsoft AI (microsoft.ai)
- Announced: 2025-11-06
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://microsoft.ai/news/towards-humanist-superintelligence/
- Record: https://forck.live/items/10970-towards-humanist-superintelligence-microsoft-ai
- Subject: MAI

Microsoft AI announces its vision for Humanist Superintelligence (HSI), emphasizing AI that is problem-oriented, domain-specific, and designed to serve humanity while remaining controllable. The company has formed the MAI Superintelligence Team to research and build this practical technology, rejecting narratives of an AGI race and focusing on safe, tangible benefits.

### Microsoft AI — Towards Humanist Superintelligence

- Company: Microsoft AI (microsoft.ai)
- Announced: 2025-11-06
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://microsoft.ai/news/towards-humanist-superintelligence
- Record: https://forck.live/items/8436-towards-humanist-superintelligence
- Subject: MAI

Microsoft AI announces the formation of the MAI Superintelligence Team to research and build Humanist Superintelligence (HSI), an AI designed to serve humanity while remaining controlled and domain-specific.

### LG AI Research — [AI Ethics Seminar 2025 EP.1] How AI Is Changing Human Critical Thinking

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-11-06
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=594
- Record: https://forck.live/items/4774-ai-ethics-seminar-2025-ep-1-how-ai-is-changing-human-critical-thinking
- Subject: EXAONE

LG AI Research announces its AI Ethics Seminar series, summarizing research on how AI affects critical thinking and introducing ChatEXAONE's design philosophy to support appropriate AI reliance.

### Moonshot AI — Kimi K2 Thinking

- Company: Moonshot AI (moonshot.ai)
- Announced: 2025-11-06
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.kimi.ai/blog/kimi-k2-thinking
- Record: https://forck.live/items/4628-kimi-k2-thinking
- Subject: Kimi

Moonshot AI announces Kimi K2 Thinking, an open-source thinking model that reasons step by step while using tools, achieving state-of-the-art scores on HLE, BrowseComp, and SWE-Bench Verified benchmarks. It can execute 200–300 sequential tool calls and supports INT4 quantization for faster inference.

### OpenAI — AI progress and recommendations

- Company: OpenAI (openai.com)
- Announced: 2025-11-06
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/ai-progress-and-recommendations
- Record: https://forck.live/items/394-ai-progress-and-recommendations
- Subject: GPT / ChatGPT / API

OpenAI publishes a general statement about AI progress and recommendations for shaping it toward discovery, safety, and a better future.

### OpenAI — Introducing the Teen Safety Blueprint

- Company: OpenAI (openai.com)
- Announced: 2025-11-06
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-the-teen-safety-blueprint
- Record: https://forck.live/items/393-introducing-the-teen-safety-blueprint
- Subject: GPT / ChatGPT / API

OpenAI introduces the Teen Safety Blueprint, a roadmap for building AI responsibly with safeguards, age-appropriate design, and collaboration to protect and empower young people online.

### OpenAI — How CRED is tapping AI to deliver premium customer experiences

- Company: OpenAI (openai.com)
- Announced: 2025-11-05T21:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/cred-swamy-seetharaman
- Record: https://forck.live/items/395-how-cred-is-tapping-ai-to-deliver-premium-customer-experiences
- Subject: GPT / ChatGPT / API

CRED is using OpenAI's GPT-powered tools to improve support accuracy, reduce response times, and increase customer satisfaction for premium customers in India.

### Harvey — Harvey In Practice: How to Speed up Diligence Review

- Company: Harvey (harvey.ai)
- Announced: 2025-11-05T17:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-in-practice-speed-up-diligence-review
- Record: https://forck.live/items/17654-harvey-in-practice-how-to-speed-up-diligence-review
- Subject: Harvey

Harvey published a guide showing how its review tables can extract and summarize data points across a set of contracts for due diligence, with results linked to source language and exportable to Excel, Word, or CSV.

### Google DeepMind — Mapping, modeling, and understanding nature with AI

- Company: Google DeepMind (deepmind.google)
- Announced: 2025-11-05T16:59:46+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/mapping-modeling-and-understanding-nature-with-ai/
- Record: https://forck.live/items/1281-mapping-modeling-and-understanding-nature-with-ai
- Subject: Gemini

AI models can assist in mapping species, protecting forests, and listening to birds globally.

### Google Research — Forecasting the future of forests with AI: From counting losses to predicting risk

- Company: Google Research (research.google)
- Announced: 2025-11-05T15:41:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/forecasting-the-future-of-forests-with-ai-from-counting-losses-to-predicting-risk/
- Record: https://forck.live/items/1395-forecasting-the-future-of-forests-with-ai-from-counting-losses-to-predicting
- Subject: Research

Google Research announces ForestCast, a deep learning model for forecasting deforestation risk using pure satellite data, and releases a public benchmark dataset for training such models. The model uses vision transformers and achieves accuracy comparable to or better than methods relying on specialized input maps.

### OpenAI — How Chime is redefining marketing through AI

- Company: OpenAI (openai.com)
- Announced: 2025-11-05T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/chime-vineet-mehra
- Record: https://forck.live/items/396-how-chime-is-redefining-marketing-through-ai
- Subject: GPT / ChatGPT / API

Chime's CMO discusses how AI is reshaping marketing into an agent-driven model and why leaders who prioritize AI literacy and thoughtful adoption will drive growth.

### Harvey — Scale Securely With Finer Admin Controls

- Company: Harvey (harvey.ai)
- Announced: 2025-11-05T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/finer-admin-controls
- Record: https://forck.live/items/17655-scale-securely-with-finer-admin-controls
- Subject: Harvey

Harvey released new admin control features including a centralized settings experience, user groups for mirroring organizational structure, and automatic permission updates. The company also announced that full SCIM support for identity provider synchronization will enter early access later this year.

### OpenAI — 1 million business customers putting AI to work

- Company: OpenAI (openai.com)
- Announced: 2025-11-05T05:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/1-million-businesses-putting-ai-to-work
- Record: https://forck.live/items/397-1-million-business-customers-putting-ai-to-work
- Subject: GPT / ChatGPT / API

OpenAI announces that over 1 million business customers worldwide are using its AI services.

### Factory AI — Snyk and Factory: Embedding Security into Agent-Native Development

- Company: Factory AI (factory.com)
- Announced: 2025-11-05
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/snyk-partnership
- Record: https://forck.live/items/17872-snyk-and-factory-embedding-security-into-agent-native-development
- Subject: Droid

Factory partnered with Snyk to integrate Snyk's security scanning into Factory's Droids, enabling automated vulnerability detection, fixing, and verification within the agent-native development workflow. The integration, built on the Model Context Protocol (MCP), is available to all Factory customers.

### Google Research — Exploring a space-based, scalable AI infrastructure system design

- Company: Google Research (research.google)
- Announced: 2025-11-04T16:58:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/
- Record: https://forck.live/items/1396-exploring-a-space-based-scalable-ai-infrastructure-system-design
- Subject: Research

Google Research announces Project Suncatcher, a moonshot exploring the design of a space-based AI infrastructure system using solar-powered satellite constellations equipped with TPUs and free-space optical links. The post describes a preprint paper detailing progress on challenges such as high-bandwidth inter-satellite communication, orbital dynamics, and radiation tolerance of TPUs.

### Sakana AI — Petri Dish Neural Cellular Automata

- Company: Sakana AI (sakana.ai)
- Announced: 2025-11-04T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/pd-nca
- Record: https://forck.live/items/4704-petri-dish-neural-cellular-automata
- Subject: Sakana models

Sakana AI introduces Petri Dish Neural Cellular Automata (PD-NCA), a new ALife substrate where multiple neural cellular automata learn to self-replicate and grow via ongoing gradient descent, leading to complex emergent behaviors like cyclic dynamics, territorial defense, and spontaneous cooperation. The work was presented at ALIFE2025.

### Vercel — Build and deploy data applications on Snowflake with v0

- Company: Vercel (vercel.com)
- Announced: 2025-11-04T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/build-and-deploy-data-applications-on-snowflake-with-v0
- Record: https://forck.live/items/20794-build-and-deploy-data-applications-on-snowflake-with-v0
- Subject: v0 / AI SDK / AI Gateway

Vercel announced an integration between v0 and Snowflake that lets users connect to Snowflake, ask questions about their data in natural language, and build data-driven Next.js applications that deploy directly to Snowflake. The integration uses a split architecture where compute runs inside the user's Snowflake account while the application and authentication layers are managed by Vercel, ensuring data never leaves Snowflake. The integration is announced at Snowflake BUILD and is available via a waitlist.

### Harvey — Introducing Harvey's New Reasoning Capabilities

- Company: Harvey (harvey.ai)
- Announced: 2025-11-04T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/introducing-harvey-deep-research
- Record: https://forck.live/items/17656-introducing-harvey-s-new-reasoning-capabilities
- Subject: Harvey

Harvey introduced agentic search and deep analysis capabilities for its legal AI platform. Agentic search performs iterative, multi-source searches over documents to produce more complete answers, while deep analysis can generate citation-backed reports from large corpora of internal and external data. Both features are available now to all Harvey customers.

### Upstage — Introducing Commitment Tiers: Flexible, predictable, and built for growth

- Company: Upstage (upstage.ai)
- Announced: 2025-11-04
- Category: pricing-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/blog/en/introducing-commitment-tiers
- Record: https://forck.live/items/18459-introducing-commitment-tiers-flexible-predictable-and-built-for-growth
- Subject: Solar Pro / Solar Mini

Upstage introduced optional commitment tiers starting November 4, 2025, a prepayment system that offers bonus credits, automatic rate-limit upgrades, and tier-based support for teams using Upstage APIs including Solar Pro2, Document Parse, and Information Extract. Existing pay-as-you-go plans remain unchanged, and accounts created before November 4, 2025 retain legacy rate limits until January 6, 2026.

### Amp — Social Coding

- Company: Amp (ampcode.com)
- Announced: 2025-11-04
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/social-coding
- Record: https://forck.live/items/7787-social-coding
- Subject: Amp

Amp introduced public profiles and public thread sharing, allowing users to see how others use agents. The company also removed public thread sharing and related features.

### Amp — Workspace Settings

- Company: Amp (ampcode.com)
- Announced: 2025-11-04
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/cli-workspace-settings
- Record: https://forck.live/items/7786-workspace-settings
- Subject: Amp

The Amp CLI now supports workspace-specific settings via .amp/settings.json files, allowing users to configure the CLI differently per project with settings like permissions and MCP servers automatically picked up.

### Cognition — Windsurf Codemaps: Understand Code, Before You Vibe It

- Company: Cognition (cognition.com)
- Announced: 2025-11-04
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/codemaps
- Record: https://forck.live/items/7686-windsurf-codemaps-understand-code-before-you-vibe-it
- Subject: Devin / SWE

Cognition released Codemaps, a feature that generates a shared understanding of a code system between humans and AI, enabling the AI to explain code paths accurately and consistently.

### Together AI — How to evaluate and benchmark Large Language Models (LLMs)

- Company: Together AI (together.ai)
- Announced: 2025-11-04
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/evaluate-and-benchmark-llms
- Record: https://forck.live/items/2402-how-to-evaluate-and-benchmark-large-language-models-llms
- Subject: Inference platform

To evaluate and benchmark large language models, covering principles, datasets, and best practices.

### Together AI — Announcing the fastest inference for realtime voice AI agents

- Company: Together AI (together.ai)
- Announced: 2025-11-04
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/the-fastest-inference-for-realtime-voice-ai-agents
- Record: https://forck.live/items/2401-announcing-the-fastest-inference-for-realtime-voice-ai-agents
- Subject: Inference platform

Together AI announces expanded voice infrastructure including streaming Whisper speech-to-text, serverless open-source text-to-speech models Orpheus and Kokoro, and new Voxtral Mini transcription and speaker diarization, all optimized for low-latency realtime voice AI agents.

### OpenAI — Introducing IndQA

- Company: OpenAI (openai.com)
- Announced: 2025-11-03T22:30:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-indqa
- Record: https://forck.live/items/398-introducing-indqa
- Subject: GPT / ChatGPT / API

OpenAI introduced IndQA, a benchmark for evaluating AI systems on cultural understanding and reasoning across 12 Indian languages and 10 knowledge areas, developed with domain experts.

### Harvey — 3 Principles That Helped us Scale Agent Development

- Company: Harvey (harvey.ai)
- Announced: 2025-11-03T17:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/principles-that-helped-us-scale-agent-development
- Record: https://forck.live/items/17657-3-principles-that-helped-us-scale-agent-development
- Subject: Harvey

Harvey engineers describe three principles—no custom orchestration, capabilities as Tool Bundles, and eval gates on capabilities—that allowed them to scale in-thread feature development from one team to four after shifting to an agent framework.

### OpenAI — AWS and OpenAI announce multi-year strategic partnership

- Company: OpenAI (openai.com)
- Announced: 2025-11-03T06:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/aws-and-openai-partnership
- Record: https://forck.live/items/399-aws-and-openai-announce-multi-year-strategic-partnership
- Subject: GPT / ChatGPT / API

OpenAI and AWS announced a multi-year, $38 billion partnership for AWS to provide infrastructure and compute capacity for OpenAI's next generation of models.