☀ TL;DR

Trump will host Xi Jinping in Washington for trade and AI talks

Trump will host Xi Jinping in Washington for trade and AI talks, possibly easing tariffs on some nonstrategic goods.

Anthropic CEO Dario Amodei urged the industry to slow AI development, and OpenAI CEO Sam Altman agreed.

Microsoft used AI agents to port Copilot from TypeScript to Rust, costing about $120K in tokens and achieving a 15.9x speedup.

Anthropic reportedly delayed its IPO to November 2026 to present strong Q3 results, with investors expecting a valuation of about $2T.

Lead · Industry

Trump to Host Xi Jinping in Washington for Trade, AI Talks

Trump will host Xi Jinping in Washington this week for talks covering trade, AI, and key economic issues. The schedule includes a welcome ceremony, a state dinner with tech CEOs, and closed-door meetings; the two sides may ease tariffs on some nonstrategic goods, though U.S. export controls tied to national security are unlikely to loosen.

Details →

Industry

24
24h Key Points
  1. BYU · Research finds AI-generated phishing emails outperform human-written ones: participants identified the author correctly only 52% of the time, AI matched or beat humans at generating clicks in 80% of cases, fooling people 28% versus 21%.

  2. Google · Releases AX, an open agentic orchestrator that runs declarative YAML-defined tasks at scale and sandboxes each agent with workspace setup and network fencing.

  3. Anthropic · Weighs launching a new AI model before its IPO to counter OpenAI's GPT-6 Astra; July annualized revenue hit $65B, backers expect over $120B by year-end, with IPO valuations estimated between $1.5T and $4T.

  4. Barack Obama · Warns at Colgate University that AI is advancing fast, with machines learning and improving with less human guidance, says the technology is not overhyped, and hints jobs will be hit.

  5. Anthropic · MCP, released in November 2024, is now called outdated; adding many MCP servers bloats context with tool schemas, prompting generic search or execute workarounds from Composio, MintMCP and Pipedream.

  6. Vocci · Launches a $249 smart ring that records meetings and generates transcripts via a double-tap button; it weighs under 6 grams, offers 8 hours of recording per charge, and its case adds three more charges.

  7. LA County Planning Department · Its planning director issued a memorandum immediately banning large-scale AI data centers in unincorporated areas while zoning rules are amended, following an April motion by Supervisor Hilda Solis passed unanimously.

  8. Wolters Kluwer · A survey of 355 US physicians and nurses finds daily AI use tripled in a year, but 74% fear AI erodes clinical skills and 72% worry advertiser-driven models distort medical advice; only 27% know their institution's AI governance policies.

  9. Tencent Gander · Processes speech, images and text while running background tasks, using a cerebellum to sustain conversation and a swappable brain to search files or write code; it interrupted users in just 8 percent of cases but trailed on task accuracy.

  10. OpenAI · An investigation found bzr.openai.com sets an __obi cookie tied to ChatGPT accounts, and OpenAI code on advertiser sites sends back browsing data, letting off-site activity be linked to accounts; the author reproduced it across 936 advertiser sites.

  11. US Treasury and He Lifeng · Bessent and He Lifeng met in New York to prepare for the Trump-Xi summit, discussing the Nov. 10 trade truce deadline, rare-earth magnet flows and AI guardrails; analysts expect only small steps.

  12. Trump · will host Xi Jinping in Washington for talks on trade and AI, with a state dinner attended by tech CEOs; the sides may ease tariffs on some nonstrategic goods.

  13. Alibaba Qwen · released Qwen-Image-2.1, a compact and efficient unified image generation model, drawing discussion among developers.

  14. Zhipu AI · saw its ZCode upload user data without authorization, including a 313MB archive with a commercial project snapshot; despite an apology and patch, developers worry about reputational damage.

  15. Berkshire Hathaway · bought more than 48 million Alphabet shares in Q2, making the stake about 12.6% of its equity portfolio and its third-largest holding, per its latest 13F filing.

  16. FBI · Director Kash Patel said AI use at the bureau has risen 605% since he took over, claiming every major tech player is embedded in the agency.

  17. Donald Trump · Donald Trump · is doubling down on data centers and AI while his MAGA base pushes back, a rift that could shape the direction of US AI infrastructure policy.

  18. Epoch AI · Epoch AI · with Ipsos found the share of US adults using AI almost daily rose from 8% to 19% between March and August 2026, while weekly-only use fell from 17% to 10%.

  19. Microsoft · Microsoft · with the University of Illinois built StudentSim, which replicates individual students from limited data to give AI tutors fast, low-cost feedback and beat GPT-5.4 in tests with 60 students in chess, English and math.

  20. Microsoft · Microsoft · used AI agents to port its Copilot runtime from TypeScript to Rust, spending about $120K in tokens over three weeks and turning 430K lines into 800K lines of Rust, with a 15.9x speedup on one benchmark.

  21. Massachusetts Governor · Massachusetts Governor · Maura Healey joined calls for AI regulation, stressing the urgency of guardrails and urging federal action after former Anthropic employee Jacob Coxon's viral warning about AI risk.

  22. Anthropic · Reportedly delays its IPO from October to November 2026 to present strong Q3 results, with investors expecting a valuation of about $2 trillion.

  23. Kim Jong Un · Orders wider integration of AI into weapons, pushing artillery automation and expanded military production, covering 600mm and 240mm rocket artillery and drones.

  24. CXMT · Says its new memory-chip platform has entered mass production, without disclosing specific product specs or capacity details.

  25. UN General Assembly · Opens its 80th high-level week in New York with nearly 130 heads of state, focusing on the Middle East and Ukraine conflicts and AI risks; the Security Council may hold high-level meetings on Ukraine, AI and the Middle East, and the US paid $725 million in dues this week to avoid losing its assembly voting rights.

  26. Donald Trump · Calls the term 'Artificial Intelligence' inaccurate, offers 'Superior Intelligence', 'Extreme Intelligence' and 'Supreme Intelligence' as alternatives and asks people to vote online, announces a government 'AI Force' to look for bad actors, and says he will soon appoint an AI coordinator or 'Czar'.

  27. Google · AI Studio is accused of faking data deletion; a user who reported it via the vulnerability reporting program says they were auto-banned within 60 seconds.

  28. Anthropic · Researcher Jacob Coxon publicly resigned this week, warning self-improving superintelligence could kill everyone by the end of the decade; internal chats called his remarks what everyone had been saying and welcomed the public discussion.

  29. Arm · Reported $4.9 billion revenue for the fiscal year ended March 2026, up 22.8%, with $904 million net income and an 18.4% net margin, contrasted with SK Hynix's HBM manufacturing.

  30. Beijing · Views AI as a national security threat as top models show hacking abilities, with a spy agency chief warning of the danger; a California team said it used AI to build a cyberattack worm that could compromise millions of WeChat accounts.

  31. US Congress · Rep. Blake Moore was named vice chair of a bipartisan AI working group, while Sens. John Curtis and Lisa Blunt Rochester called for hearings on risk incidents to inform AI regulation.

  32. Australian Prime Minister · Anthony Albanese said middle powers can help mitigate AI risks and called for US-China cooperation on AI governance.

  33. Anthropic · CEO Dario Amodei published an essay urging the industry to slow AI development, proposing a three-point "pacing the frontier" plan: permanent access for outside evaluators, coordinated safety standards among democracies, and global coordination including China.

  34. AMD · GameGPU reports AMD's next-gen Radeon RX 10800 XT may beat Nvidia's RTX 5090 by 15-25% in 4K gaming and local AI workloads, with no official confirmation yet.

  35. University of Waterloo · Researchers release ProgramAsWeights, an open-source project that compiles English function descriptions into reusable neural programs that run locally; a finetuned Qwen3-4B generates LoRA adapters for a frozen Qwen3-0.6B interpreter, separating compilation from inference.

  36. Qwen3.8 27B · On a 16GB VRAM setup, its IQ3_XXS version (10.18GiB) and Bonsai Ternary PQ2 (6.42GiB) both completed 4/4 tasks, but Qwen ran 3.02x faster and used 3.10x fewer output tokens.

  37. Stanford Medicine · Research overturns the single-progenitor model of brain development, finding the front and back of the brain arise from two distinct neural ectoderm progenitors: a primitive nervous system and the part enabling higher cognition, which may aid research into brainstem diseases like SMA and ALS.

View all →

Voices

X · 21
24h Key Points
  1. Ethan Mollick · argues this call should extend to all social science, saying AI research needs to be fast and forward-looking.

  2. Teknium 🪽 · announces that GLM-5.3 FlashX is now available in Hermes Agent through Nous Portal and OpenRouter.

  3. Bindu Reddy · argues open-source and cheap models handle about 60% of tasks today and will reach 99% in six months.

  4. Alexandr Wang · says he will speak at Connect this year and asks what the audience wants to hear.

  5. Yuchen Jin · predicts that in two years GPT-6 Astra-level open-source models will run locally on every MacBook, yet AI Twitter will still debate whether open-source AI causes human extinction.

  6. Arthur Mensch · responds to criticism of Mistral by noting the poster is a presidential election candidate.

  7. Ethan Mollick · argues Claude lacking an image generator is a gap in agentic knowledge-work projects.

  8. Qwen · announces Qwen-Image-2.1 is now supported in ComfyUI, inviting users to try it and share their creations.

  9. Qwen · thanks vLLM for day-0 support of Qwen-Image-2.1, saying one model generates, edits and outputs transparent images.

  10. Nathan Lambert · guesses most top Chinese AI labs are starting to use Huawei for inference and Nvidia for training in scaled RL.

  11. Bindu Reddy · argues the choice is not open versus closed but matching the right model to each task, listing models for coding, 3D, agents, images, research, video and browser use.

  12. Alexandr Wang · recommends trying muse spark 1.3 in opencode or muse code, calling Kevin a very legit engineer.

  13. Ethan Mollick · observes that Astra showed less simulated curiosity than Fable, producing solid results without seeming interested in the findings.

  14. Alexandr Wang · announces Muse's first television ad will air this weekend during the big games.

  15. Alexandr Wang · says Muse this week shipped a Mac app, Canada expansion, new connectors and a developer platform, with more coming soon.

  16. Alexandr Wang · argues Meta's Muse is as groundbreaking as the iPhone and exceeded his expectations for Meta.

  17. Alexandr Wang · says the reception for Muse far exceeded expectations, citing a post calling it the next ChatGPT moment.

View all →

Launches

PH · HN · 7

Mycel

Runs clients, deliverables, approvals and invoices for service businesses, in disposable credential-free sandboxes with human approval.

▲ 126 · Product Hunt · 18 comments ·

Minicart

Lets makers and resellers launch an online store, with AI teammates handling storefront, marketing and logistics via chat.

▲ 107 · Product Hunt · 3 comments ·

Answers by Context.dev

Turns web research into one API call, returning structured JSON data with source URLs from a defined task.

▲ 92 · Product Hunt · 2 comments ·

Epismo OS

Keeps your team's AI work in a Case so you can switch AI tools without re-briefing

▲ 86 · Product Hunt · 2 comments ·

Morsa Signals

GTM and AI visibility workflows for developer-tool teams to find developers, early adopters and competitors.

▲ 84 · Product Hunt · 3 comments ·

ENZO

Open-source local AI platform aggregating 2000+ free models for chat, coding, and research.

▲ 16 · Show HN · 12 comments ·

Open Source

GitHub Trending · 13

Reference agents, skills, and data connectors for financial-services workflows like investment banking and equity research.

GitHub Trending ·
★ 35.2k+236 today

Agent harness performance optimization system adding skills, memory and security for Claude Code, Codex, Cursor and more.

GitHub Trending ·
★ 263.2k+1.0k today

Open-source TypeScript framework for building agentic apps pairing autonomous work with a purpose-built UI.

GitHub Trending ·
★ 5.0k+89 today

Open-source drivers, cross-OS fleets, and benchmarks to give AI agents usable computers.

GitHub Trending ·
★ 24.4k+859 today

Production-grade engineering skills for AI coding agents across every development phase.

GitHub Trending ·
★ 97.0k+556 today

An agentic terminal coding tool that understands your codebase and executes tasks via natural language.

GitHub Trending ·
★ 146.7k+483 today

Secure, self-hosted environments for developers and their agents.

GitHub Trending ·
★ 15.6k+402 today

A 2-bit automation foundation model for tiny devices, supporting tool calls and structured extraction.

GitHub Trending ·
★ 11.6k+234 today

Fault-tolerant, highly scalable GPU orchestration and ML framework for billion-to-trillion parameter models.

GitHub Trending ·
★ 4.9k+196 today

Parses diverse documents including PDFs and integrates them with the generative AI ecosystem.

GitHub Trending ·
★ 67.0k+129 today

A cross-platform visual manager for OpenAI Codex desktop and CLI, with provider and session management.

GitHub Trending ·
★ 3.4k+32 today

Model Ranking · 24 h

2026-09-20 UTC
01GLM 5.3 FlashZ.ai · 2.51T tokens14.5%▲1
02DeepSeek V4.1 FlashDeepSeek · 2.49T tokens14.4%▼1
03Hy4 previewTencent · 1.97T tokens11.4%—
04GPT-5.6 LunaOpenAI · 1.21T tokens7.0%▲1
05DeepSeek V4 Flash 0423DeepSeek · 1.12T tokens6.4%▼1
06MiMo-V2.5Xiaomi · 877.7B tokens5.1%—
07Nemotron 3 Ultra (free)NVIDIA · 779.1B tokens4.5%—
08Hy3Tencent · 487.8B tokens2.8%—
09DeepSeek V4 Flash 0423DeepSeek · 418.1B tokens2.4%—
10GLM 5.3Z.ai · 309.2B tokens1.8%—

New Models

3
Ternary Bonsai 2 27BPrismML · 262.1k ctx · 09/18$0.07 / $0.50
GLM 5.3 FlashXZ.ai · 1.0M ctx · 09/18$0.37 / $1.25
Paretounbiased · 262.1k ctx · 09/17$2.50 / $7.50

Deals

No funding or M&A detected in the past 24 hours.