新云社 update: OpenAI is testing a new ChatGPT mobile interface, initially for Pro users who have dot or are eligible to create it, according to Tibo (@thsottiaux). The redesign gives dot its own tab and lets users pin features like Codex, Projects, and Images to the bottom bar. Some users will see the new design by default, and it can be toggled in Settings > New ChatGPT Design. dot is also being tested with a small group of Plus users.
新云社 reports: Polymarket’s contract on whether Jensen Huang will publicly announce the founding of a new company before 11:59 p.m. ET on Dec. 31, 2027, saw its implied probability rise to 27%, up 19% over 24 hours. The contract requires a first public announcement naming Huang as founder or co-founder and identifying the company; statements that he is considering, exploring or in talks to start a company do not qualify.
新云社 reports: Morgan Stanley analysts said some large telecom operators have indicated they will deploy their own AI agents to counter third-party AI agents. The analysts believe AI agents could prompt users to switch carriers, with operators leading on network quality or offering more competitive pricing likely to gain market share, while those lacking an edge in both areas may come under pressure.
新云社 update: OpenAI introduced “completion window” options for the background mode of its Responses API, letting developers choose 2-hour or 8-hour windows that cut token rates by 30% and 40% respectively versus Standard pricing. The default soonest window gets no discount. The feature is in public beta on GPT-6.2 Astra, GPT-6.2 Sol, GPT-6.1 Terra and GPT-6 Luna, and discounted windows require Standard processing, do not support streaming and share rate limits with the Batch API.
新云社 update: Signal said it will block screenshots by default to stop AI agents built into mobile operating systems from capturing and storing user conversations. The encrypted messaging app will block screenshots on iOS 27 and automatically enable its “Screen Security” setting on Android versions where it detects an AI assistant trying to read the screen. President Meredith Whittaker wrote that Google’s Gemini Intelligence and Apple’s Siri AI Onscreen Awareness scan screen content, moving conversations into local databases or remote servers beyond Signal’s control.
新云社 update: Tesla CEO Elon Musk said on November 20 that the company’s AI, running on Tesla’s AI4 computer, can now reach Torment II difficulty in Diablo. Torment II is the sixth of 16 difficulty levels in Diablo 4 and the second of 12 Torment tiers, marking the entry level for endgame content.
新云社 update: China National Data Group has launched its first round of central-local cooperation pilots with Ningxia and Hebei to accelerate the cultivation of the data industry. In Ningxia, cooperation covers data resource development, computing hub interconnection, AI industry cultivation and data security systems. In Hebei, the group will leverage its headquarters economy effect to cluster the data industry chain and deploy data-element innovation products and application scenarios rooted in Xiong’an and serving the province.
新云社 reports: Semafor reporter Rohan Goswami said on a Monday program that former Nvidia CEO Jensen Huang may return to the AI industry after a break and launch a new semiconductor company, potentially backed by friends and industry allies with about $50 billion in funding. Goswami said Huang is unlikely to return to Nvidia and that the board’s separation from Huang was deliberate and unlikely to change. He stressed the prediction is not an announced plan and there is no information that Huang has raised funds or founded a new company.
新云社 update: China’s intelligent computing capacity is expected to reach 10,000 EFLOPS by 2030, accounting for more than 90% of the country’s total computing capacity, according to Liu Wenqiang, party secretary of the China Center for Information Industry Development, speaking at the 2026 World Computing Conference in Changsha on September 20. Inference demand will make up 85% of intelligent computing, and nearly 90% of that capacity will be hosted by the eight national computing hubs and three computing-power coordination zones.
新云社 update: HTX will open spot and grid trading for PIEVERSE/USDT at 21:00 (UTC+8) and VVV/USDT at 21:30 on November 20. Deposits for both tokens open at 18:00 the same day, with withdrawals available from 21:00 and 21:30 on November 21. Pieverse is a Web3 AI agent platform built on TEE and the ERC-8004 identity standard, while Venice is a privacy-focused AI platform whose VVV token can be staked for compute credits.
新云社 update: WeChat’s team released a beta desktop client for ‘Make a Small Tool’ on November 20, supporting macOS and Windows. Users must scan a QR code on the page with WeChat to sign up for the trial. The official tagline is ‘Let Ideas Happen,’ and the product is described as ‘Simple, complex, all possible.’
新云社 update: Hangzhou Chengyan Enterprise Management Consulting Partnership became a shareholder of DeepSeek on Nov. 17, ranking as its third-largest shareholder, with CITIC Securities, CICC, CSC Financial and Shenwan Hongyuan among the brokerages contributing capital behind the entity. Founder Liang Wenfeng’s direct stake declined but he remains the second-largest direct shareholder. The stake was disclosed by Chinese brokerage media.
新云社 update: Meta’s personal AI agent Muse launched a Chrome browser extension on November 20, allowing users to open Muse in the Chrome sidebar, resume prior conversations, and delegate tasks such as filling out forms, comparing prices and following up on refunds. Each action requires user approval before execution, the extension logs every operation, and users can revoke access to connected services at any time.
新云社 reports: Anthropic’s third-quarter revenue rose about $5.4 billion from the prior quarter while compute and infrastructure expenses increased about $4.3 billion, according to its prospectus, implying roughly 80 cents of new compute cost per $1 of added revenue. Revenue reached $17 billion versus $14 billion a year earlier, but adjusted operating margin fell to 15% from 19%, and the company swung to a $567 million operating loss under US accounting rules.
新云社 update: OpenAI has introduced GPT-Live-1.1 to its API, adding a standby option priced at $0.02 per minute. The standby mode also works with GPT-Live-1, but developers must enable it for WebRTC sessions. Backend models and tools are billed separately.
新云社 reports: Valuations for tech stocks, particularly AI-linked names, have fallen sharply, with the Nasdaq 100, S&P Information Technology Index and Philadelphia Semiconductor Index all trading at multi-year lows on forward earnings. Navellier & Associates founder Louis Navellier said the AI sector is set for a rebound, calling Micron and Super Micro Computer — trading at less than four times expected earnings — “excellent buying opportunities.” Candriam’s Johan Van der Biest said recent earnings show cloud and AI computing demand remains strong.
新云社 update: Shanghai Guotou Xiandao said on Nov. 20 that its AI master fund has preliminarily selected three sub-funds for investment: vehicles managed by Primavera Capital, Hillhouse Venture Capital and Oriental Fortune Capital, registered as Shanghai Huajun Private Equity Partnership, Shanghai Puling Private Investment Fund Partnership and Shanghai Fuhai Xiandao Sci-Tech Industry Investment Fund Partnership.
新云社 update: Anthropic’s seven co-founders sold a combined roughly $1.96 billion of stock in the company’s IPO, or about $280 million each before tax at the $83 offer price. Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Chris Olah each sold about 3.4 million shares. At the $74 midpoint of the original range, each would have realized about $250 million. Anthropic said each founder has pledged 80% of their personal equity to charity, though the pledges are not legally binding.
新云社 reports: A comparison of AI agent builders n8n, Dify and LangGraph evaluates the tools by coding requirements, connections, hosting, human approval steps and model billing, rather than by a single ‘no-code’ label. The review focuses on operational differences that affect how teams build and run AI agents.
新云社 update: Google has expanded Gemini’s computer-use capabilities on Mac to let US subscribers on Plus, Pro and Ultra plans automate app control, including background clicking and typing. Users can choose which apps Gemini is allowed to access. The rollout brings the automation feature to paid tiers in the US.