Newshot 英伟达 CEO 黄仁勋 Announces Resignation
On November 12, the 英伟达 (NVIDIA) board of directors announced that company co-founder 黄仁勋 (Jensen Huang) has resigned from the positions of president, chief executive officer, and board member, effective immediately. Current Executive Vice President and Chief Financial Officer 科莱特 克雷斯 (Colette Kress) will serve as interim CEO while continuing in their existing role until the board selects a permanent successor. At the same time, 英伟达 appointed longtime director Harvey (Harvey C. Jones) as the company's first chairman in its history.
According to the Form 8-K filed by 英伟达 with the U.S. Securities and Exchange Commission (SEC), the board accepted 黄仁勋's resignation on November 11 and approved the related personnel appointments the same day. The board has now initiated the formal CEO selection process, which will consider both internal executives and external candidates. 英伟达 did not disclose the specific reason for 黄仁勋's resignation in its official announcement or regulatory filings.
According to the introduction, 克雷斯 joined 英伟达 in September 2013 as chief financial officer, after nearly 25 years in finance-related roles at technology companies including Cisco and Microsoft. New chairman Jones has served as a 英伟达 director since 1993 and has more than 30 years of experience in technology company management, entrepreneurship, and venture capital. He previously served as CEO and executive chairman of Synopsys, and has been a managing partner at Square Wave Ventures since 2004.
Jones said that 克雷斯 has a deep understanding of 英伟达's business, customers, operations, and financial condition, and that the board has full confidence in them and the management team. He expects the company's day-to-day operations, customer commitments, and product roadmap to remain unaffected. 克雷斯 said they will continue to build on the business foundation 黄仁勋 established for 英伟达 over the past 33 years and drive the company's development in accelerated computing and artificial intelligence.
黄仁勋 co-founded 英伟达 in 1993 and served as the company's CEO continuously for the past 33 years, leading the company's expansion from graphics processors into data centers, accelerated computing, and artificial intelligence. In its announcement, the 英伟达 board thanked 黄仁勋 for their contributions to the company and the technology industry. The resignation also marks the first CEO transition at 英伟达 since its founding.
In addition, 英伟达 confirmed that it will still report financial results for the third quarter of fiscal year 2027 at approximately 1:20 p.m. Pacific Time on November 17 as originally planned, and will hold an earnings conference call at 2:00 p.m. Source
2026 Golden Joystick Awards Winners Announced
On November 11 local time, the 44th Golden Joystick Awards was held in London, UK, revealing the winners of 25 awards. Capcom's Resident Evil Requiem won the Ultimate Game of the Year award with 17% of the public vote, while the Chinese-developed action game Phantom Blade Zero placed second with 16%, a difference of only 1 percentage point, and Crimson Desert placed third with 15%. The Phantom Blade Zero team responded on November 12 that the game had only 40 hours from its official release on October 29 to the Golden Joystick Awards voting deadline, and that finishing second greatly exceeded the team's expectations. They also expressed thanks to the players who supported the game.
In addition to Ultimate Game of the Year, Resident Evil Requiem also won Best Console Game, Best Game Trailer, and Best Lead Performer, making it one of the most-awarded games at this year's event. The indie game Super Chameleon also took home four awards, including Best Multiplayer Game, Best Self-Published Indie Game, Streamers' Choice, and Developers' Choice.
In other major awards, Dispatch won Best Narrative and Best Soundtrack, Crimson Desert won PC Game of the Year, Big Walk won the Critics' Choice Award, and The Witcher 4 won Most Wanted Game. Capcom won Studio of the Year, Onimusha: Way of the Sword won Best Audio Design, Control: Resonant won Best Visual Design, Subnautica 2 won Best Early Access Game, Valheim won Best Indie Game, and the Steam Controller won Best Gaming Hardware. In addition, Balatro and Minecraft won the Still Playing awards for mobile and PC/console, respectively. Source
Apex Compute Launches Cross-Platform AI Frame Generation Technology ANR
On November 11, Apex Compute announced the launch of its cross-platform AI neural rendering technology suite Apex Neural Rendering (ANR), covering frame generation, upscaling, and ray reconstruction, aimed at competing with 英伟达 DLSS. The company has now opened the beta of ANR Frame Generation (ANR FG), supporting 英伟达 GeForce GTX 16 and RTX 20 series and newer graphics cards, AMD Radeon RX 5000 series and newer graphics cards, as well as Intel Arc A and B series graphics cards and the Steam Deck handheld.
According to test data published by Apex Compute, at 4K resolution using a GeForce RTX 5090 graphics card, ANR's per-frame processing time in 2x frame generation mode is 1.01 milliseconds, lower than DLSS's 1.26 milliseconds; in 4x mode, the average per-frame processing time is 0.61 milliseconds, lower than DLSS's 0.82 milliseconds. Its VRAM usage is 214 MB, lower than DLSS's 562 MB. The company also claims that ANR offers better image quality and showed image comparisons from games such as The Last of Us Part II. However, the above performance and image quality claims have not yet been verified by independent testing.
The ANR FG beta is now available for download on GitHub, compatible with Windows 10/11, Linux, and SteamOS running via Proton, and the official listing also includes support for Apple Silicon Macs. At this stage, PC games require the OptiScaler tool to install and run it, and specific compatibility varies by game. ANR's upscaling and ray reconstruction models are still in development and are expected to be released later. Source
Tencent Launches AI-Based Yuanbao Search
Tencent has officially launched "Yuanbao Search," an AI search product for general users that supports searching web pages across the internet, WeChat public account articles, Channels videos, Tencent News, and academic literature, among other content, and offers features such as AI summaries and follow-up questions. Users can currently access it through the standalone website so.qq.com, or upgrade the QQ Browser to the latest version and set it as the default search engine in the search engine settings.
According to the introduction, Yuanbao Search is based on the large model capabilities of Tencent Yuanbao and supports the full search process from search intent understanding and content retrieval to result ranking and answer generation. It has a hundreds-of-billions-scale index and connects to 157 authoritative sources. In addition to text search, the product also supports voice and photo search, and can provide search results in the form of cards, charts, and other formats based on the query content.
Regarding the issue of search advertising, 李锐章, product lead for the Tencent Yuanbao Search web version, said that the product will not reject advertising commercialization, but will remain restrained to avoid ads interfering with users' access to information. Yuanbao Search uses independent ad modules that are separated from organic search results, clearly labeled "Related Ads," and users can collapse ad content. Tencent said that commercial content will not affect the ranking and display of organic search results. Source
Volcano Engine Launches Seedance 2.5 flash Video Model
Volcano Engine, under ByteDance, has officially launched the AI video generation model Seedance 2.5 flash. The company said that while the model's generation quality is close to the flagship Seedance 2.5, its generation speed is doubled and its price is about 34% lower, mainly targeting high-frequency creation and batch video production scenarios.
According to the introduction, taking text-to-video and image-to-video at 720P resolution as examples, Seedance 2.5 flash's generation price is about 0.99 yuan per second, on par with the previous-generation Seedance 2.0. The model can generate videos up to 30 seconds long in a single run, supports input of up to 10 images, 5 videos, and 5 audio clips, and offers two output resolution tiers: 480P and 720P. Volcano Engine said that Seedance 2.5 flash's overall generation quality is better than the previous-generation flagship model Seedance 2.0 and that it retains video editing capabilities.
Seedance 2.5 flash is currently rolling out to platforms including Jimeng, LibTV, Updream, Creao AI, and TapNow, and its API is scheduled to go live on Volcano Ark on November 15. With the launch of the flash version, the Seedance 2.5 series will form two product lines: a flagship model focused on high-quality video production and a flash model focused on generation efficiency and cost control. Source
Article title: Morning Pai: 英伟达 CEO 黄仁勋 announces resignation, and more
Body:
Google AI News Roundup
Google releases Gemini 4 Hydrogen model
On November 12, Google officially launched the Gemini 4 Hydrogen model, focused on fast execution, everyday programming, and AI agent workflows, and for the first time extended Gemini 4 series model capabilities to free users. The model supports a context window of up to 1 million tokens, with an API input price of $1.50 per million tokens (about 10 yuan) and an output price of $7.50 per million tokens (about 50.1 yuan), with unified pricing across the entire context window.
In terms of performance, Gemini 4 Hydrogen offers four thinking-depth levels: low, medium, high, and xhigh, with high used by default. According to data published by third-party evaluation agency Artificial Analysis, the model's intelligence index at the high level reaches 51 points, surpassing Gemini 4 Argon's 50 points, and the xhigh level further raises it to 54 points. Google said that compared with the previous generation Gemini 3.8 Flash, the new model's task completion speed at the default level is 2.5 times faster, its Terminal-Bench 4.1 test score rose from 18.2% to 57%, and fabricated errors decreased by more than 5 times.
Gemini 4 Hydrogen supports native 1 million token input and up to 1 million token output, and can process multimodal content such as PDFs, spreadsheets, charts, audio, and video, making it suitable for scenarios including coding, software debugging, and complex document analysis. At the same time, Google also launched the Orchestrator intelligent scheduling feature for its AI development platform Google Antigravity, which can automatically switch between models such as Gemini 4 Carbon and Hydrogen within the same session according to task requirements. According to Terminal-Bench 4.0 test results published by Google, this feature reduces average quota consumption by 24% and improves task completion speed by 19% while approaching Carbon's task success rate.
In terms of safety, Google claims that Gemini 4 Hydrogen reduces unauthorized actions in agent tasks by up to 81%, reduces task underservice issues by up to 79%, achieves a score of 96.7% on internal agent honesty evaluations, and improves resistance to indirect prompt injection attacks by more than 2.3 times. However, the above safety improvement data mainly come from Google's own evaluations, and the actual effect still needs further verification.
Gemini 4 Hydrogen has now begun rolling out globally. Developers can use it through platforms such as the Gemini API, Google AI Studio, and Android Studio, while enterprise users can access it through Gemini business services and Google Antigravity. Ordinary users can use the new model in the Gemini app, where free users can obtain Hydrogen capabilities through the automatic model selection feature, and Google Search has also begun integrating the model. Source
Google launches multimodal decision-making Gemini Decisions API
Google announced the launch of the Gemini Decisions API, powered by the all-new Gemini Decision 1 model and designed specifically for decision-making tasks such as tool selection, policy checks, and model routing. The API supports up to 1,000 independent judgments on multimodal content such as text, images, audio, video, and PDFs in a single request, including three types: yes/no judgments, multiple choice, and graded scoring.
Unlike traditional generative models that generate answers token by token, Gemini Decision 1 can score candidate options through a single forward computation and return corresponding probabilities and confidence scores. The API supports an input context of up to 256K tokens, with Google reporting end-to-end latency of 150 to 250 milliseconds and pricing at $0.10 per 1 million input tokens, with no output token fee.
In addition, developers can combine different types of input content in the same request and adjust the resolution of images and videos to control visual detail and token costs. Google said internal products such as Gemini Notebook and Stitch have begun adopting the API. The Gemini Decisions API is now available through Google AI Studio, with a free quota for developers to conduct prototype testing. Source
Google releases stable version of A2UI, an agent-generated cross-platform interactive interface
Google officially released the open generative UI protocol A2UI 1.0, the first stable version of the protocol since it was announced in December 2025. A2UI allows AI agents to directly generate interactive interfaces such as forms and charts, and supports interface rendering by native components on different platforms.
A2UI 1.0 also adds a bidirectional function-calling mechanism, allowing interactive interfaces to query information such as flights and bookings in real time, while expanding accessibility and custom metadata support. Google also provides open-source renderers for React, Angular, Lit, Flutter, and SwiftUI, as well as agent SDKs supporting Python, TypeScript, and Dart. The Jetpack Compose renderer for Android remains at version 0.9.1, with 1.0 support to be launched later.
In addition, A2UI can be integrated with MCP Apps, enabling generated interactive interfaces to be displayed in AI assistants such as Claude and Gemini that support MCP Apps. Google said companies including Atlassian, Alation, Bloomberg, and ByteDance have begun developing based on the protocol. Source
Doubao Work integrates third-party models such as Kimi, GLM, and DeepSeek
On November 12, Doubao announced that its AI Agent office product "Doubao Work" has officially integrated third-party models such as Kimi K3.1 Pro, GLM 5.3, and DeepSeek V4.1 Flash, and simultaneously launched limited-time quota discounts. Previously, Doubao Work mainly offered two self-developed models, Doubao 2.1 Lite and Doubao 2.1 Pro, and this update further expands users' model choices.
According to the introduction, users can switch models independently according to task requirements, and Doubao Work will indicate the quota consumption multiplier of different models in the model selection interface. The lower the multiplier, the less quota is consumed to complete the same task. Previously, Doubao Work opened third-party model configuration to team and enterprise subscription users on October 29, and this update further extends the related capabilities to more users.
In addition, while retaining free benefits for simple small-quota tasks, Doubao Work launched a limited-time offer. Users who use participating models during specified periods on weekdays and holidays can enjoy quota consumption discounts; the quota consumption multipliers of Doubao 2.1 Lite and DeepSeek V4.1 Flash have also been lowered for a limited time to reduce quota consumption for the same tasks. Source
Recent updates from SSPAI
- We have opened a special page to summarize and display the "Outstanding Author Selection" awards and results. "Outstanding Author Selection" is a mechanism implemented by the editorial team since 2023. After the end of each quarter, it selects existing authors with outstanding contributions during that quarter, as well as emerging new authors with potential. Winners receive exclusive badges and prizes. Go to view
- The Supervisor Otter replica returns. From now until November 15, enjoy a 15% discount across all platforms. Start on duty now
- Limited-time offers on SSPAI membership, genuine software, and paid columns. Don't miss out
Good articles you may have missed
- Bringing old speakers back to life: Home Assistant Connect AUX-2 first-look experience
- Mac Buying Guide 2026 Winter Edition
- City Walk Guide | Late summer in Osaka, the "transitional" season is just right
Download the SSPAI 2.0 client and follow the SSPAI official account to unlock a new reading experience 📰
Practical and useful genuine software, presented to you by SSPAI 🚀