AI News

2026-09-18 · FridayPublished daily at 08:00

AI Daily

Issue 150182026-09Friday
14major stories14sources6firsthand announcements1new modelsAbout a 5-minute read
Today's Highlights
  1. 1
  2. 2
  3. 3
In This Issue
  1. 01Model Releases/Updates1 item
  2. 02Product Releases/Updates4 items
  3. 03Industry News1 item
  4. 04Paper research2 items
  5. 05Tips and perspectives6 items

01Model Releases/Updates

Qwen:Blog RetrievalFirsthand

Qwen releases its native natively multimodal model Qwen3.8-Omni-Flash, focused on delivering intelligent agent tasks for audio and video

Qwen released its next-generation native multimodal model, Qwen3.8-Omni-Flash, supporting text, image, audio, and video inputs with a 1M-token context window. Its average score across 29 evaluations improved by more than 25% over Qwen3.5-Omni-Plus. The hourly price for audio input fell by more than 98%, while the hourly price for audio-and-video input fell by more than 93%.

Original text

02Product Releases/Updates

Sherwin Wu

ChatGPT for Word launches; OpenAI employees say usage of Excel and PowerPoint has surged recently

ChatGPT is now integrated into Microsoft Word, where it can turn rough notes into a first draft, improve paragraphs, proofread, suggest edits, and spot formatting issues. OpenAI’s Sherwin Wu said usage of ChatGPT for Excel and PowerPoint has surged recently, and the Word launch completes the full Office integration.

Original text
Unsloth

Unsloth Releases a Docker Image and Unsloth Desktop for Local Training and Running of 500+ Models

Unsloth announces that its Docker image can be used to locally train and run 500+ models, with a new GUI and notebooks workflow, no configuration required, and support for NVIDIA and AMD. See the guide at https://unsloth.ai/docs/get-started/install/docker.

Original text
Claude:BlogFirsthand

Claude Code restructures Projects: from folders into conversational projects capable of hosting multiple threads

Anthropic has overhauled Projects in Claude Code. After users set a goal, Claude breaks it into tasks, schedules multiple threads in parallel, reviews their outputs, and compiles the results. The threads are essentially separate Claude Code cloud sessions running on independent branches.

Original text
AI at Meta

Meta released Muse for Mac, allowing personal agents to execute tasks directly on a computer.

Meta announced that Muse for Mac is available starting today. With the user’s explicit authorization, its personal agent can complete tasks directly on their computer. Capabilities include organizing the Downloads folder, finding lost files, and summarizing messages and notes. Meta says more features are coming soon. Download at http://ai.meta.com/muse/download/.

Original text

03Industry News

TechCrunch:AI

Unsealed documents in The New York Times’ lawsuit against OpenAI and Microsoft reveal AI scraping was called the largest labor theft in history

Newly unsealed documents in The New York Times’ copyright case against OpenAI and Microsoft reveal that Microsoft executive Brent Hecht called AI scraping “the largest-scale labor theft in human history” in an internal memo, while OpenAI executive Nick Turley called chatbots an “existential threat” to publishers.

Original text

04Paper research

Anthropic:ResearchFirsthand

Anthropic used Claude to optimize more than 30 open-source biomolecular models, achieving an average speedup of approximately 4x and open-sourcing all the code.

Anthropic published research showing that Claude optimized more than 30 open-source biomolecular models in under four weeks, achieving an average speedup of about 4x, or about 2x when producing exactly identical outputs.

Original text
Goodfire ResearchFirsthand

Goodfire Research finds that internal model signals can be used to detect reward hacking at scale

Goodfire Research found activation signals associated with reward hacking inside models that simple probes can detect in real time. Across three agent benchmarks using the open-source models Kimi K3, GLM 5.2, and Qwen 3.8 Max, 50–96% of rollouts exhibited reward hacking. The probes can catch cases missed by LLM chain-of-thought monitoring and generalize to tasks outside the training data.

Original text

05Tips and perspectives

GitHub BlogFirsthand

GitHub used a Copilot agent to migrate the Copilot runtime from TypeScript to 830,000 lines of Rust.

GitHub engineer Stephen Toub recapped how Copilot agents rewrote the Copilot agent runtime entirely from TypeScript/Node.js into 832,378 lines of production Rust in about 14.5 weeks. AI agents wrote most of the code, which was merged into main incrementally through 128 PRs and continuously shipped.

Original text
404 Media

Newly unsealed documents in The New York Times’ lawsuit against OpenAI: Microsoft and OpenAI internally acknowledged that LLMs are built on theft, triggering a Doom Loop

An unredacted court filing in The New York Times’ copyright lawsuit against OpenAI has been unsealed, revealing multiple internal admissions from Microsoft and OpenAI executives. It says LLMs are built on content that a Microsoft executive called theft on an unprecedented scale, and that this has triggered a doom loop that would destroy the entire web. The filing shows that clicks on stolen news sites from Bing fell by more than 90%, that OpenAI bypassed The New York Times’ paywall, and that OpenAI described itself as an existential threat to news publishing.

Original text
The Verge:AI

The Verge summarized the debate over slowing AI superintelligence: Amodei called for slowing down, with Altman and Musk joining him, while Meta opposed it.

The Verge reviewed recent debates over AI safety and slowing development. Anthropic CEO Dario Amodei proposed a three-step plan: involve third-party evaluators (Anthropic has unilaterally committed to the first step), have frontier AI companies in democratic countries coordinate standards, and pursue global coordination among governments. OpenAI’s Sam Altman and Elon Musk expressed support.

Original text
Anthropic:The InstituteFirsthand

Anthropic releases a tool for measuring frontier AI development pace and a snapshot of internal metrics

Anthropic has released a set of metrics for measuring the pace of frontier AI development, covering AI-led R&D, agent supervision, and compute allocation.

Original text
Dwarkesh Patel:Podcast & Blog

Dwarkesh in conversation with Noam Brown: agent swarms, alignment, and recursive self-improvement.

Dwarkesh Patel interviewed OpenAI researcher Noam Brown about multi-agent systems, alignment, and recursive self-improvement.

Original text
WeChat Official Account: Digital Life Kazike

Use an MCP plugin to have GPT-6 Pro share Codex planning tasks, saving weekly Pro subscriber quotas

A social media creator shared a workflow for saving Codex usage: have Codex wrap its own server as a read-only, least-privilege MCP Server authenticated with Feishu OAuth, then use it as a plugin called by GPT-6 Pro in the ChatGPT web app to analyze and plan using real production data and GitHub PR records.

Original text

Finished reading the news? Let’s talk about your business.

Leave your contact details, and I’ll get in touch with you personally.

WECHAT

Add me on WeChat

Press and hold or select the WeChat ID below to copy it.