WorkBuddy to Refund the Points Difference After Failing to Announce the Return to the Original Price in Time
要闻
WorkBuddy said that after DeepSeek-V4.1 Flash returned to its regular price at 14:00 on November 5, it would settle usage from November 5 through 8 at half price and automatically refund the difference because the change was not announced in time.
WorkBuddy said that, because the return of DeepSeek-V4.1 Flash to its regular price was not announced in time, all points consumed from 14:00 on November 5 through 23:59 on November 8 will be settled at 50% of the regular price. The difference, equal to 50% of the total points actually consumed, will be refunded automatically by the system without an application, with the credit expected to arrive within seven business days.
DeepSeek-V4.1 Flash launched on September 10, alongside a two-week promotion offering points at half price. The promotion was originally scheduled to end on September 24, but WorkBuddy extended it through November 5 and restored the regular price at 14:00 that day. The company said it apologized because the adjustment was not announced to users in a timely manner.
OpenAI Employee Asks Users Whether a Pro 50 Tier Should Be Added
要闻
Tibo, a member of OpenAI’s Codex and ChatGPT teams, said that Pro subscriptions have no five-hour limit. After outlining Pro 100, Pro 200, and Pro 500 usage relative to Plus, he asked users whether OpenAI should add a $50 Pro 50 tier.
In the clarification, Tibo, a member of OpenAI’s Codex and ChatGPT teams, said that none of the Pro subscription plans has a five-hour limit. According to his explanation, Pro 100 provides 10 times the usage of Plus, Pro 200 provides 20 times, and Pro 500 provides 50 times; only Pro 500 includes ultrafast access.
Tibo then quoted the related post and asked users whether adding Pro 50 would be useful. The proposed tier would provide Pro-level benefits at a price of $50. The source does not state that the launch of this tier has been decided.
NVIDIA Founder 黄仁勋 Admits to Infidelity and Apologizes to His Wife
要闻
NVIDIA founder Jensen Huang admitted that he had been unfaithful to his wife and apologized publicly, while saying the company’s board had opened an independent investigation into national security concerns raised in news reports.
In a public message to NVIDIA employees, its board, partners, and shareholders, NVIDIA founder Jensen Huang admitted that he had been unfaithful to his wife and said his actions had hurt her and their children. He accepted full responsibility for his choices and the harm caused, and apologized to his family and people connected to the company. Huang said that leading NVIDIA was the responsibility of his life and that he would continue serving the company and work to regain trust. He also asked for privacy for his family and stated that they were not responsible for his choices. Regarding national security concerns raised in news reports, NVIDIA’s board of directors has initiated an independent investigation. Huang said he supports the investigation and will cooperate fully, and stated that he is committed to protecting NVIDIA and supporting U.S. national security.
DeepSeek Harness Lead Responds to Plugin Author: Built-in Features Can Be Replaced by Plugins
开发生态
Tianyi Cui, head of the DeepSeek Harness team, told plugin authors that common capabilities will be built in, while its “everything is a plugin” design lets third-party plugins replace built-in features; as of November 4, daily active users stood at 1.54 million and plugins at about 35,000.
Tianyi Cui, head of the DeepSeek Harness team, explained the relationship between built-in features and third-party plugins, stating that the product aims to become an operating system for the Agent era. According to his explanation, DeepSeek Harness will provide stable foundational capabilities and cover common functions needed by most users, while allowing third-party plugins to replace functions included by the platform.
Cui used smartphone operating systems as an example, saying that an operating system typically provides basic applications such as an alarm and a calculator while also seeking third-party applications that perform better than built-in versions. An ecosystem progress report released earlier showed that, as of November 4, daily active users stood at 1.54 million based on official API user statistics, while the number of third-party plugins was about 35,000.
GPT-6.1 Sol Participates in Three Consecutive Live Mouse Brain Surgeries, Halving Localization Error
技术与洞察
OpenAI’s GPT-6.1 Sol took part in three consecutive live-mouse brain surgeries, cutting its localization error for Bregma and Lambda from 0.6 mm to 0.3 mm and reducing identification time from 169 seconds to 86 seconds.
OpenAI’s GPT-6.1 Sol identified the anatomical landmarks Bregma and Lambda on the skull surface during three brain surgeries on live mice. After each procedure, professional surgeons corrected the locations identified by the model, which stored the images and measurement data for reference in subsequent procedures. WIRED reported the results, and Gabriel Ocana-Santero, co-founder and CEO of Exin Therapeutics, also disclosed the details on X.
During the third procedure, GPT-6.1 Sol identified the landmarks in 86 seconds, down from 169 seconds in the first procedure. The number of manual corrections fell from 4 to 2, while the localization error declined from 0.6 mm to 0.3 mm. According to the official explanation, the change did not involve retraining the model or updating its parameters; instead, it adjusted its judgments using experience accumulated from earlier procedures and expert feedback, representing an application of in-context learning. The study lead stated that the sample size of three procedures is very limited and that the results have not yet undergone independent validation or peer review.
Anthropic Reportedly Bought $3.4 Billion Worth of Gas-Fired Generators
行业动态
The Financial Times reported that Anthropic took over a $3.4 billion order for heavy gas generators totaling 1.7GW in late September to power data centers. Based on known records, it is the largest power-generation equipment purchase by an AI company to date.
The Financial Times reported that Anthropic took over an order for heavy gas generators with a total capacity of 1.7GW from OnePWR Solutions, an equipment supplier in San Antonio, Texas, in late September. The transaction was valued at $3.4 billion, and the generators will supply power to data centers. Based on known records, it is the largest power-generation equipment purchase by an AI company to date.
People familiar with the matter said OnePWR Solutions had ordered the generators from Caterpillar two years earlier but still needed financing to complete the purchase. Anthropic initially planned to buy services from OnePWR Solutions before changing course and taking over the equipment order directly. Because the order was placed before equipment prices increased, some generators can be delivered soon, while the remaining units will arrive over the next two years.
DeepMind Executive Teases Scientific Breakthroughs Coming in the Next Few Weeks
前瞻与传闻
Pushmeet Kohli, a Google DeepMind vice president, said releases over the next few weeks will show the “real shift” in the pace of scientific discovery observed across several projects, without naming specific projects, dates, or metrics.
Pushmeet Kohli, a vice president at Google DeepMind, said the company’s releases over the next few weeks will present the change in the pace of scientific discovery observed across multiple projects. In a follow-up post, he said the company’s focus as this AI-driven acceleration unfolds will be enabling a broader scientific community to use these tools responsibly. The post did not identify any specific release, provide a date, or define a metric for measuring the “shift.”
Users Report Faster Grok Bot Quota Usage; Elon Musk Says He Will Investigate Usage Levels
前瞻与传闻
Multiple users reported on X that Grok Bot subscription allowances were being consumed faster, and Elon Musk said he would investigate usage. He also said fixes to improve token efficiency and behavior are underway; plans, prices, and allowances remain unchanged, and token-based billing has not launched.
Elon Musk responded to reports from Grok Bot users about a change in the rate at which subscription allowances were consumed and said that usage would be investigated. User Derek Neighbors said the $200 plan lasted 7 days under full-speed use last week, but lasted only 14 hours under a medium load this week. Another user said the Ultra plan allowance had reached its limit with about 3 days remaining before the weekly reset.
Musk also asked whether users would prefer switching to token-based billing and said that fixes intended to improve token efficiency and make the behavior more intelligent were in progress. Grok Bot’s plans, prices, and allowances have not changed, and token-based billing has not been launched.
OpenAI Developer Account Posts a Teaser for Tomorrow
前瞻与传闻
OpenAI’s official developer account released a retro game-style video previewing a black box marked with a handheld game console outline and the OpenAI logo, without disclosing a product, features, or release plans.
On November 9, 2026, OpenAI’s official developer account for developers posted a retro game-style video on X. The video moved through a game menu, user comments, and a loading animation before showing a black box printed with the outline of a handheld game console and the OpenAI logo. The account’s caption read “Tomorrow.”
The final frame displayed “November 9,” “On Supply Co.,” and the logos of OpenAI and ModRetro Chromatic. The video did not show the device itself and did not announce a related product name, features, or release schedule. The video remained a preview and did not explain what the displayed elements specifically referred to.
