# forck.live — 19–25 January 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-W04, 19–25 January 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-W04
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Alibaba — Pushing Qwen3-Max-Thinking Beyond its Limits

- Company: Alibaba (alibaba.com)
- Announced: 2026-01-25T20:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen3-max-thinking
- Record: https://forck.live/items/4598-pushing-qwen3-max-thinking-beyond-its-limits
- Subject: Qwen

Alibaba's Qwen team announces Qwen3-Max-Thinking, a new flagship reasoning model with adaptive tool-use capabilities and a test-time scaling strategy, now available in Qwen Chat and via API.

### Google Research — Introducing GIST: The next stage in smart sampling

- Company: Google Research (research.google)
- Announced: 2026-01-23T17:46:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/introducing-gist-the-next-stage-in-smart-sampling/
- Record: https://forck.live/items/1374-introducing-gist-the-next-stage-in-smart-sampling
- Subject: Research

Google Research introduces GIST, a novel algorithm for data subset selection that balances diversity and utility with provable guarantees, presented at NeurIPS 2025.

### OpenAI — Unrolling the Codex agent loop

- Company: OpenAI (openai.com)
- Announced: 2026-01-23T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/unrolling-the-codex-agent-loop
- Record: https://forck.live/items/297-unrolling-the-codex-agent-loop
- Subject: GPT / ChatGPT / API

The post is a technical deep dive into the Codex agent loop, explaining how Codex CLI orchestrates models, tools, prompts, and performance using the Responses API.

### Sakana AI — Sakana AI、Googleとの戦略的パートナーシップ締結を発表

- Company: Sakana AI (sakana.ai)
- Announced: 2026-01-23T06:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/google
- Record: https://forck.live/items/4697-sakana-ai-google
- Title in English (translated by forck.live from ja): Sakana AI Announces Strategic Partnership with Google
- Subject: Sakana models

Sakana AI announces a strategic partnership with Google, including a financial investment following the Series B round. The partnership aims to leverage Google's state-of-the-art models (Gemini and Gemma) to enhance product quality and deploy reliable AI in mission-critical industries.

### Odyssey — The GPT-2 Moment for World Models Is Here

- Company: Odyssey (odyssey.systems)
- Announced: 2026-01-23
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://odyssey.systems/the-gpt-2-moment-for-world-models
- Record: https://forck.live/items/18602-the-gpt-2-moment-for-world-models-is-here
- Subject: Odyssey world models

Odyssey released Odyssey-2 Pro, its most powerful world model, and launched a new developer API. The model outputs continuous, interactive simulations at 720P and 22 FPS, with improved physics, dynamics, and visual quality. The API offers three endpoints—interactive streams, viewable streams, and simulations—along with JavaScript and Python SDKs.

### Factory AI — Signals: Toward a Self-Improving Agent

- Company: Factory AI (factory.com)
- Announced: 2026-01-23
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/factory-signals
- Record: https://forck.live/items/17848-signals-toward-a-self-improving-agent
- Subject: Droid

Factory Research introduced Signals, a system that uses LLMs as judges to analyze user sessions of its Droid agent, identifying friction and delight moments without exposing raw conversation content. Signals processes thousands of sessions daily, extracting structured facets and evolving its detection categories through embedding and clustering. When friction crosses a threshold, Droid autonomously adjusts its behavior, enabling recursive self-improvement.

### OpenAI — Think Bigger: How Small Teams Win with ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2026-01-22T21:12:55+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/business/guides-and-resources/chatgpt-business-smb-guide
- Record: https://forck.live/items/18795-think-bigger-how-small-teams-win-with-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI published a guide for small and midsize businesses using ChatGPT Business, which is powered by GPT‑5. The guide provides ready-to-use prompts and examples for tasks such as marketing, data analysis, meeting preparation, and vendor evaluation.

### Google Research — Small models, big results: Achieving superior intent extraction through decomposition

- Company: Google Research (research.google)
- Announced: 2026-01-22T16:56:44+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/small-models-big-results-achieving-superior-intent-extraction-through-decomposition/
- Record: https://forck.live/items/1375-small-models-big-results-achieving-superior-intent-extraction-through
- Subject: Research

Google Research presents a novel approach for understanding user intents from UI interaction trajectories using small multimodal LLMs. The method decomposes the task into two stages: first summarizing each screen individually, then extracting intent from the summaries. The approach shows results comparable to larger models and is suitable for on-device applications.

### Vercel — Testing if "bash is all you need"

- Company: Vercel (vercel.com)
- Announced: 2026-01-22T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/testing-if-bash-is-all-you-need
- Record: https://forck.live/items/20762-testing-if-bash-is-all-you-need
- Subject: v0 / AI SDK / AI Gateway

Braintrust and Vercel tested three agent approaches (SQL, bash, filesystem) on querying GitHub issues. SQL achieved 100% accuracy, while bash reached 52.7% and filesystem 63.0%. A hybrid approach combining bash and SQL also achieved 100% accuracy but used more tokens.

### Harvey — Search That Works With Firm Knowledge

- Company: Harvey (harvey.ai)
- Announced: 2026-01-22T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/firm-knowledge-in-harvey
- Record: https://forck.live/items/17623-search-that-works-with-firm-knowledge
- Subject: Harvey

Harvey introduced Firm Knowledge, a feature that lets legal teams connect company resources, embed practice-area content, and integrate proprietary workflows within the platform. Users can search across connected sources with permission-aware results, and the feature includes enterprise-grade governance such as authentication, audit logs, and role-based access.

### OpenAI — Scaling PostgreSQL to power 800 million ChatGPT users

- Company: OpenAI (openai.com)
- Announced: 2026-01-22T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/scaling-postgresql
- Record: https://forck.live/items/298-scaling-postgresql-to-power-800-million-chatgpt-users
- Subject: GPT / ChatGPT / API

OpenAI's infrastructure techniques for scaling PostgreSQL to handle millions of queries per second for ChatGPT.

### Runway — The Turing Reel

- Company: Runway (runway.com)
- Announced: 2026-01-22T10:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/research/theturingreel
- Record: https://forck.live/items/7898-the-turing-reel
- Subject: Gen / Aleph

Runway published a research study demonstrating that over 90% of participants could not reliably distinguish videos generated by its Gen-4.5 model from real video. The company also released image-to-video capabilities for Gen-4.5 and launched a website where people can try the test.

### Gradium — Why Your Voice Cloning Sounds Fake (And How to Fix It)

- Company: Gradium (gradium.ai)
- Announced: 2026-01-22
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://gradium.ai/blog/voice-cloning-sounds-fake
- Record: https://forck.live/items/18331-why-your-voice-cloning-sounds-fake-and-how-to-fix-it
- Subject: Gradium TTS / Phonon

Which voice sounds more like the original? Voice cloning technology has evolved beyond simple text-to-speech (TTS) synthesis.

### Moonshot AI — Kimi Vendor Verifier

- Company: Moonshot AI (moonshot.ai)
- Announced: 2026-01-22
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.kimi.ai/blog/kimi-vendor-verifier
- Record: https://forck.live/items/4627-kimi-vendor-verifier
- Subject: Kimi

Moonshot AI is open-sourcing the Kimi Vendor Verifier (KVV), a project that provides benchmarks and tools to verify the accuracy of inference implementations for open-source models, addressing issues of benchmark anomalies and quality control in the open-source ecosystem.

### Cursor — Subagents, Skills, and Image Generation

- Company: Cursor (cursor.com)
- Announced: 2026-01-22
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/2-4
- Record: https://forck.live/items/2635-subagents-skills-and-image-generation
- Subject: Cursor

Cursor announces new agent capabilities including subagents, skills, image generation, Cursor Blame for AI attribution, and various improvements.

### Together AI — Optimizing inference speed and costs: Lessons learned from large-scale deployments

- Company: Together AI (together.ai)
- Announced: 2026-01-22
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/optimizing-inference-speed-and-costs
- Record: https://forck.live/items/2387-optimizing-inference-speed-and-costs-lessons-learned-from-large-scale
- Subject: Inference platform

Together AI shares lessons learned from large-scale deployments on optimizing inference speed and costs, covering quantization, distillation, regional inference proxies, reducing compute stalls, decoding optimizations, and hardware choices.

### OpenAI — Inside Praktika's conversational approach to language learning

- Company: OpenAI (openai.com)
- Announced: 2026-01-22
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/praktika
- Record: https://forck.live/items/300-inside-praktika-s-conversational-approach-to-language-learning
- Subject: GPT / ChatGPT / API

Praktika uses GPT-4.1 and GPT-5.2 to build adaptive AI tutors for language learning.

### OpenAI — Inside GPT-5 for Work: How Businesses Use GPT-5

- Company: OpenAI (openai.com)
- Announced: 2026-01-22
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/business/guides-and-resources/chatgpt-usage-and-adoption-patterns-at-work
- Record: https://forck.live/items/299-inside-gpt-5-for-work-how-businesses-use-gpt-5
- Subject: GPT / ChatGPT / API

This is a report about ChatGPT usage patterns at work, not a model announcement.

### Alibaba — Qwen3-TTS Family is Now Open Sourced: Voice Design, Clone, and Generation!

- Company: Alibaba (alibaba.com)
- Announced: 2026-01-21T16:00:04+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwen.ai/blog?id=qwen3tts-0115
- Record: https://forck.live/items/4595-qwen3-tts-family-is-now-open-sourced-voice-design-clone-and-generation
- Subject: Qwen

Alibaba's Qwen team has open-sourced the Qwen3-TTS family of speech generation models, including two sizes (1.7B and 0.6B parameters), supporting voice clone, voice design, and natural language-based voice control across 10 languages. The models are available on GitHub and via the Qwen API.

### ElevenLabs — Introducing The Eleven Album

- Company: ElevenLabs (elevenlabs.io)
- Announced: 2026-01-21T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://elevenlabs.io/blog/introducing-the-eleven-album
- Record: https://forck.live/items/3373-introducing-the-eleven-album
- Subject: Voice AI

ElevenLabs introduced The Eleven Album.

### OpenAI — How Higgsfield turns simple ideas into cinematic social videos

- Company: OpenAI (openai.com)
- Announced: 2026-01-21T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/higgsfield
- Record: https://forck.live/items/301-how-higgsfield-turns-simple-ideas-into-cinematic-social-videos
- Subject: GPT / ChatGPT / API

Higgsfield uses OpenAI GPT-4.1, GPT-5, and Sora 2 to create cinematic social videos from simple inputs.

### Hugging Face — AssetOpsBench: Bridging the Gap Between AI Agent Benchmarks and Industrial Reality

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-21T06:25:31+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/ibm-research/assetopsbench-playground-on-hugging-face
- Record: https://forck.live/items/1560-assetopsbench-bridging-the-gap-between-ai-agent-benchmarks-and-industrial
- Subject: Platform

IBM Research introduces AssetOpsBench, a benchmark and evaluation system for AI agents in industrial asset lifecycle management, featuring 2.3M sensor telemetry points, over 140 scenarios across 4 agents, 4.2K work orders, and 53 failure modes.

### OpenAI — How countries can end the capability overhang

- Company: OpenAI (openai.com)
- Announced: 2026-01-21T01:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/how-countries-can-end-the-capability-overhang
- Record: https://forck.live/items/303-how-countries-can-end-the-capability-overhang
- Subject: GPT / ChatGPT / API

OpenAI announces a report on differences in AI adoption between countries and initiatives to help nations gain productivity from AI.

### OpenAI — Introducing Edu for Countries

- Company: OpenAI (openai.com)
- Announced: 2026-01-21T01:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/edu-for-countries
- Record: https://forck.live/items/302-introducing-edu-for-countries
- Subject: GPT / ChatGPT / API

OpenAI announced Edu for Countries, a new initiative to help governments use AI to modernize education systems and build future-ready workforces.

### Supabase — Introducing: Postgres Best Practices

- Company: Supabase (supabase.com)
- Announced: 2026-01-21
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/postgres-best-practices-for-ai-agents
- Record: https://forck.live/items/21307-introducing-postgres-best-practices
- Subject: Supabase

Supabase released a repository of 30 Postgres best-practice rules for AI coding agents, organized into eight categories by impact. The rules follow the Agent Skills format, an open standard developed by Anthropic, and are designed to guide agents to generate correct, performant, and secure Postgres code. The repository is available at github.com/supabase/agent-skills.

### Perceptron — Composing Weight and Data Sparsity in MoE

- Company: Perceptron (perceptron.inc)
- Announced: 2026-01-21
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.perceptron.inc/blog/composing-weight-and-data-sparsity-in-moe
- Record: https://forck.live/items/18618-composing-weight-and-data-sparsity-in-moe
- Subject: Perceptron Isaac / Perceptron Mk1

Perceptron's post explains how mixture-of-experts (MoE) layers can combine weight sparsity and data sparsity to vary compute per token, arguing that not all tokens need equal compute and that data sparsity lets the model skip compute where it is not needed.

### Cognition — Devin Review: AI to Stop Slop

- Company: Cognition (cognition.com)
- Announced: 2026-01-21
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-review
- Record: https://forck.live/items/7683-devin-review-ai-to-stop-slop
- Subject: Devin / SWE

Cognition released Devin Review, a code review tool that uses AI to help developers understand code diffs on GitHub PRs. The tool is currently free during its early release and works on any public or private GitHub PR.

### World Labs — Announcing the World API

- Company: World Labs (worldlabs.ai)
- Announced: 2026-01-21
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.worldlabs.ai/blog/announcing-the-world-api
- Record: https://forck.live/items/7343-announcing-the-world-api
- Subject: Marble / Atlas

World Labs launched the World API, a public interface for generating explorable 3D worlds using its multimodal world model, Marble. The API accepts text, images, panoramas, multi-view inputs, and video, and returns navigable 3D environments that can be rendered on the web, exported, or integrated into interactive systems.

### Mistral AI — Heaps do lie: debugging a memory leak in vLLM.

- Company: Mistral AI (mistral.ai)
- Announced: 2026-01-21
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://mistral.ai/news/debugging-memory-leak-in-vllm
- Record: https://forck.live/items/3306-heaps-do-lie-debugging-a-memory-leak-in-vllm
- Subject: Mistral / Le Chat

This engineering blog post by Mathis Felardos describes the process of debugging a memory leak in the vLLM library.

### Replit — Decision-Time Guidance: Keeping Replit Agent Reliable

- Company: Replit (replit.com)
- Announced: 2026-01-20T21:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/decision-time-guidance
- Record: https://forck.live/items/17920-decision-time-guidance-keeping-replit-agent-reliable
- Subject: Replit Agent

Replit described techniques for guiding Replit Agent on long trajectories, using the execution environment to provide intelligent feedback that helps the agent course-correct while keeping a human in the loop. The company said these techniques significantly improved the agent across building, planning, deployment, and overall code quality while keeping costs and context in check.

### OpenAI — Horizon 1000: Advancing AI for primary healthcare

- Company: OpenAI (openai.com)
- Announced: 2026-01-20T21:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/horizon-1000
- Record: https://forck.live/items/304-horizon-1000-advancing-ai-for-primary-healthcare
- Subject: GPT / ChatGPT / API

OpenAI and the Gates Foundation announced Horizon 1000, a $50 million pilot initiative to advance AI capabilities for healthcare in Africa, aiming to reach 1,000 clinics by 2028.

### OpenAI — Stargate Community

- Company: OpenAI (openai.com)
- Announced: 2026-01-20T19:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/stargate-community
- Record: https://forck.live/items/305-stargate-community
- Subject: GPT / ChatGPT / API

OpenAI announced Stargate Community plans, a community-first approach to AI infrastructure with locally tailored plans shaped by community input, energy needs, and workforce priorities.

### Hugging Face — One Year Since the “DeepSeek Moment”

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-20T15:02:10+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/huggingface/one-year-since-the-deepseek-moment
- Record: https://forck.live/items/1561-one-year-since-the-deepseek-moment
- Subject: Platform

The article reflects on the one-year anniversary of DeepSeek's R1 model release, analyzing how it catalyzed the growth of China's open source AI ecosystem by lowering technical, adoption, and psychological barriers, leading to strategic realignments among Chinese AI companies.

### OpenAI — ServiceNow powers actionable enterprise AI with OpenAI

- Company: OpenAI (openai.com)
- Announced: 2026-01-20T05:45:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/servicenow-powers-actionable-enterprise-ai-with-openai
- Record: https://forck.live/items/306-servicenow-powers-actionable-enterprise-ai-with-openai
- Subject: GPT / ChatGPT / API

ServiceNow is expanding access to OpenAI's frontier models to enable AI-driven enterprise workflows, summarization, search, and voice capabilities on its platform.

### Microsoft — Differential Transformer V2

- Company: Microsoft (microsoft.com)
- Announced: 2026-01-20T03:20:57+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/microsoft/diff-attn-v2
- Record: https://forck.live/items/1562-differential-transformer-v2

Microsoft introduces DIFF V2, a differential attention mechanism that doubles query heads without increasing key-value heads, enabling faster decoding and eliminating the need for custom attention kernels, while removing per-head RMSNorm to improve numerical stability.

### Isomorphic Labs — Isomorphic Labs Enters into a Research Collaboration with Johnson & Johnson

- Company: Isomorphic Labs (isomorphiclabs.com)
- Announced: 2026-01-20
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.isomorphiclabs.com/articles/isomorphic-labs-enters-into-a-research-collaboration-with-johnson-johnson
- Record: https://forck.live/items/19134-isomorphic-labs-enters-into-a-research-collaboration-with-johnson-johnson
- Subject: Drug Design Engine / AlphaFold

Isomorphic Labs announced a cross-modality, multi-target research collaboration with Johnson & Johnson, combining Isomorphic Labs' AI-first drug discovery approach with Johnson & Johnson's drug development expertise.

### Factory AI — Introducing Agent Readiness

- Company: Factory AI (factory.com)
- Announced: 2026-01-20
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/agent-readiness
- Record: https://forck.live/items/17812-introducing-agent-readiness
- Subject: Droid

Factory introduced Agent Readiness, a feature that evaluates how well a codebase supports autonomous development. It assesses repositories across eight technical pillars and five maturity levels, providing specific recommendations for improvement. The evaluation uses LLMs to check 60+ criteria, with a grounding mechanism that reduced score variance from 7% to 0.6%.

### Krea — Krea Realtime Edit: Paint and Watch AI Respond Live

- Company: Krea (krea.ai)
- Announced: 2026-01-20
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.krea.ai/blog/realtime-edit
- Record: https://forck.live/items/15092-krea-realtime-edit-paint-and-watch-ai-respond-live
- Subject: Krea 2 / AI creative platform

Krea released Realtime Edit, a canvas where each brushstroke instantly updates the image without a generate button or loading bar. Users can also stream a webcam or screen into the canvas for live restyling.

### Hugging Face — Introducing Waypoint-1: Real-time interactive video diffusion from Overworld

- Company: Hugging Face (huggingface.co)
- Announced: 2026-01-20
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/waypoint-1
- Record: https://forck.live/items/1563-introducing-waypoint-1-real-time-interactive-video-diffusion-from-overworld
- Subject: Platform

Overworld announced Waypoint-1, a real-time interactive video diffusion model controllable via text, mouse, and keyboard. The model is trained on 10,000 hours of diverse video game footage and weights are available on Hugging Face Hub, including Waypoint-1-Small with Waypoint-1-Medium coming soon.

### OpenAI — Our approach to age prediction

- Company: OpenAI (openai.com)
- Announced: 2026-01-20
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/our-approach-to-age-prediction
- Record: https://forck.live/items/307-our-approach-to-age-prediction
- Subject: GPT / ChatGPT / API

ChatGPT is rolling out age prediction to estimate whether accounts belong to users under or over 18, applying safeguards for teens and refining accuracy over time.

### Sakana AI — An Unofficial Guide to Prepare for a Research Position Application

- Company: Sakana AI (sakana.ai)
- Announced: 2026-01-19T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://sakana.ai/unofficial-guide
- Record: https://forck.live/items/4698-an-unofficial-guide-to-prepare-for-a-research-position-application
- Subject: Sakana models

Sakana AI published an unofficial guide on what they look for when interviewing research candidates, emphasizing understanding over implementation, asking questions that distill the problem space, prototyping what matters, going deep rather than broad, communicating clearly, and balancing engineering with creativity.

### Factory AI — Factory Expands to New York City

- Company: Factory AI (factory.com)
- Announced: 2026-01-19
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/factory-new-york
- Record: https://forck.live/items/17843-factory-expands-to-new-york-city
- Subject: Droid

Factory opened a New York City office in January 2026 to support its growing East Coast customer base, particularly in financial services and large enterprises adopting agent-native development. The company already works with customers in New York including Bilt, Profound, Rogo, and two global banks.

### LG AI Research — [AI Ethics Seminar 2025 EP.6] Finding the Future of AI in Philosophy

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-01-19
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=621
- Record: https://forck.live/items/4762-ai-ethics-seminar-2025-ep-6-finding-the-future-of-ai-in-philosophy
- Subject: EXAONE

LG AI Research discusses its AI Ethics Seminar 2025 EP.6, a forum where diverse members explore ethical challenges in AI. The post covers user survey findings on generative AI interaction, philosophical perspectives from Nietzsche and Fromm, and the goal of building trustworthy AI that empowers human creativity and autonomy rather than fostering dependency.

### Z.ai — GLM-4.7-Flash

- Company: Z.ai (z.ai)
- Announced: 2026-01-19
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://docs.z.ai/release-notes/new-released#2026-01-19
- Record: https://forck.live/items/4453-glm-4-7-flash
- Subject: GLM

Z.ai launched GLM-4.7-Flash, a lightweight and efficient free-tier version of GLM-4.7, with strong performance on coding, reasoning, and generative tasks, low latency, and high throughput.