hotai.newsAn archive of 58 AI news briefs collected on 2026-10-26. Check the original sources for details and updates.
This content still needs to be cross-checked against the original publication page.
Azeem Azhar analyzes the true limits of AI-driven research acceleration in his Exponential View column, arguing that the bottleneck lies not in intelligence itself but in whether the research loop can cover the entire process.
Source links can be verified on the original siteRead original ↗Anthropic officially filed for an IPO on Monday, planning to list under the ticker ANT. The prospectus shows revenue of $16.3 billion in the first half of 2026, compared to $1.2 billion in the same period last year, with operating losses narrowing to $693 million and net losses reaching $80.7 billion, primarily due to non-cash accounting adjustments related to convertible bonds. As of the end of September, there were over 11,900 customers spending more than $100,000 annually, with approximately 1,900 customers spending over $1 million.
Source links can be verified on the original siteRead original ↗Meta plans to raise capital for AI infrastructure by selling up to $95 billion in new shares. This includes approximately $22.5 billion from Class A common stock, $12.5 billion from depositary shares (each representing one-twentieth of a new preferred share that automatically converts to common stock before August 2029), and an additional $60 billion expected to be sold continuously in the public market after Q1 2027 via non-voting Class C shares.
Source links can be verified on the original siteRead original ↗Anthropic has publicly filed an S-1 registration statement with the SEC, planning to list on the Nasdaq under the ticker symbol ANT, aiming to beat OpenAI in completing its IPO.
Source links can be verified on the original siteRead original ↗With the new version launch, 'Selfie Mode' will go live. A new 'Guest Station' has also joined, letting players meet more compatible partners for photos. This is paired with a fully refreshed 'Crew Station' upgrade to help players easily capture epic scenes.
Source links can be verified on the original siteRead original ↗In the introduction: Using AI to write code is no longer new. My own workflow has changed: I first ask the AI to understand the problem and modify the code, then immediately have it verify the results. If verification fails, I pass the errors back to the AI for further fixes.
Source links can be verified on the original siteRead original ↗Anthropic has officially filed its S-1 document to begin its IPO process. Despite generating $16.3 billion in revenue during the first half of the year, it reported a net loss of $80.7 billion. The filing reveals that the company pays SpaceX approximately $3.1 billion monthly for computing power and asserts that AI can solve problems humanity has been unable to crack.
Source links can be verified on the original siteRead original ↗The Tencent Hunyuan real-time speech-to-speech translation and voice cloning feature is now available for free trial in the Tencent Hy Translation App. Testing shows a response time of about three seconds, outperforming most commercial simultaneous interpretation APIs, with translation quality reaching human expert levels.
Source links can be verified on the original siteRead original ↗Sar Haidar wrote at O'Reilly Radar notes that a single AI subscription can produce five distinct usage patterns within infrastructure teams: Soloist, Copy-Paster, Godot Waiter, Specialist, and Tourist. These are interchangeable modes rather than fixed personality types.
Source links can be verified on the original siteRead original ↗AI unicorn Anthropic has officially filed its S-1 registration statement to go public on the Nasdaq under the ticker ANT. Revenue in the first half of 2026 surged 13 times year-over-year, and adjusted operating profit turned positive for the quarter. However, a massive paper loss was recorded due to changes in the fair value of convertible notes. The company has signed a substantial agreement to lease computing power, with minimum future payments totaling approximately $417 billion. #Anthropic IPO#
Source links can be verified on the original siteRead original ↗The Verge investigates whether AI can solve the COBOL legacy code crisis. On October 19, OpenAI stated that its latest model has made significant improvements in migrating old language code, GPT-6.1 Sol achieving 98% behavioral retention accuracy at high reasoning intensity and 87% at low intensity, compared to only 5% for GPT-5.6 Sol; Anthropic also released a code modernization guide in February to promote Claude Code.
Source links can be verified on the original siteRead original ↗IBM has made Spyre a first-class PyTorch backend, giving developers the same workflows across devices.
Source links can be verified on the original siteRead original ↗The company expects its capital expenditure in 2026 to reach between $135 billion and $145 billion. Meta plans to fund its AI investments through a balanced combination of operating cash flow, investment-grade debt, strategic partnerships, and equity capital.
Source links can be verified on the original siteRead original ↗IBM and the PyTorch Foundation have fulfilled their 2026 Spyre Deep Support initiative. The torch-spyre backend, Spyre's vLLM plugin, and compiler enhancements now allow standard PyTorch models to run directly on Spyre hardware.
Source links can be verified on the original siteRead original ↗A content creator discovered that their usual catchphrase, 'Let's be honest,' is increasingly viewed by readers as proof of heavy AI influence. As a result, they began subconsciously removing these four words. They also noted that sentence structures like 'not... but...' are being actively abandoned. The creator believes that once AI has learned human language, it is now the human turn to modify their own expression to prove they are not machines.
Source links can be verified on the original siteRead original ↗Tencent Hunyuan launched and open-sourced the simultaneous interpretation large model Hy-SimulMT on October 26. It supports real-time Chinese-to-English speech-to-speech translation and real-time voice cloning, with measured response times of approximately three seconds—under four to five seconds faster than human professional translators.
Source links can be verified on the original siteRead original ↗According to Bloomberg, Jane Street and Hudson River Trading and other high-frequency trading firms are making large-scale investments in AI computing power to support machine learning trading.
Source links can be verified on the original siteRead original ↗Over the past two months, OpenAI and Anthropic have successively released state-of-the-art models, dominating various benchmark lists for nearly that entire period. Meanwhile, this has been a dark phase for domestic models, facing accusations of overfitting and skepticism, with no Pro model capable of matching them yet. However,
Source links can be verified on the original siteRead original ↗A new report from Counterpoint projects that global shipments of humanoid robot AI accelerators could exceed 1 million units by 2030. NVIDIA currently dominates the market, while Qualcomm, Intel, Tesla, and XPeng are accelerating their entry. Competition is expected to intensify significantly after 2027.
Source links can be verified on the original siteRead original ↗On October 22, a delegation of 38 foreign media professionals from the Asia-Pacific Media Hundred-Person Training Program visited Kunlun Tech's Beijing headquarters. The group discussed how AI empowers content creation for Gen Z and adapts it for global markets. The presentation demonstrated SkyProduction TianGong Workstation's 3D director console, one-stop short video generation, and multi-language batch adaptation for overseas distribution. It also showcased Mureka, a global AI music large model capable of generating music in multiple genres and languages. Kunlun Tech reported that its monthly active users are close to 400 million globally, with overseas revenue accounting for 97%.
Source links can be verified on the original siteRead original ↗The proposal aims to achieve inter-satellite routing latency under 5ms and a switch success rate exceeding 99.9%, supporting single-satellite AI compute power over 30 POPS. It also promotes the digital operation and control system for mega-constellations to manage at least 3,000 satellites, facilitating seamless integration between ground and satellite networks.
Source links can be verified on the original siteRead original ↗The first officially licensed AI interactive film-game set in the Three-Body Universe, 'Three-Body: Judgment Before Dawn', has launched on Steam. Developed by Timeline, the release date is yet to be announced, with the first trailer already available.
Source links can be verified on the original siteRead original ↗Equipped with an AI engine and six antennas, this router supports AI intelligent routing and anti-interference. It delivers a 58% gain increase on the 5GHz band. Features include blind port insertion, NFC instant connection, dedicated gaming ports, and UU Accelerator. MLO and 4K-QAM are also included.
Source links can be verified on the original siteRead original ↗Jimeng Short Drama Studio offers DreamScript for scripts, Anycook for images, and DreamShot for storyboards. These models cover original works, adaptations, overseas releases, and commentary. Capabilities include analyzing 100,000-word long-form content, creating unique character settings, and pre-visualizing with a 3D director's chair. Post-production features include re-shooting segments, voice reconstruction, and intelligent scene cutting, along with original film remastering and localization adaptation.
Source links can be verified on the original siteRead original ↗Researchers from Tsinghua University and CUHK have open-sourced AKV, a KV cache engineering service system built on SGLang. This allows agents to discard or rearrange KV entries via Drop and Repos (RoPE rotary repositioning) operations without rewriting the text history.
Source links can be verified on the original siteRead original ↗Hello, I am a rural programmer (returned to the countryside in 2020 and still haven't left), an independent developer (dedicated to developing 1,000 small products), and Frontend Tiger Chen Suiyi (self-proclaimed). My official website is: https://yicode.tech. Today, we are not discussing making money.
Source links can be verified on the original siteRead original ↗Although most interviewees did not believe that the AI colleagues' 'gender' affected their feedback, and Johanna and Johan showed no significant difference in trustworthiness and credibility, the feminized Johanna still received lower 'compensation'.
Source links can be verified on the original siteRead original ↗Two leveraged ETFs tied to SK Hynix are set for liquidation just months after listing; the short positions have plummeted by 42%, and total assets have shrunk by more than 60% from their peak. The AI concept sector is experiencing intense volatility, while Korean regulators have also tightened rules on leveraged products.
Source links can be verified on the original siteRead original ↗The author shares six prompts derived from using DeepSeek to create code animations similar to Claude Opus 5.5, addressing issues such as removing repetitive copy, eliminating unnecessary highlighted boxes, specifying color usage, unifying font size and alignment, ensuring one thing per screen, and designing animation effects. The author also previews a major editing skill currently in progress: using DeepSeek to produce ten-minute spoken-word videos with twenty case study clips and operation recordings, which will be compiled this week.
Source links can be verified on the original siteRead original ↗In just 77 days, spending 131.06 yuan (cost: 129 yuan), I broke even today. Let me answer this seriously once and for all: Can a single person using AI build a WeChat mini-program? Yes, but you must set your expectations to the right standard.
Source links can be verified on the original siteRead original ↗At the Capital University of Economics and Business, a newly established Institute for Intelligent Economy is upgrading the 'Rongchun Zhigu' platform. This is a key initiative by the university to build an integrated foundation combining computing power, algorithms, and data. It unifies school-wide computing resources and shares data assets, opening access to three scenarios: talent training, scientific research, and think-tank services.
Source links can be verified on the original siteRead original ↗A Gartner survey of 203 CEOs revealed that while 75% still plan to increase technology investment and 80% view technology as a core strategy, only about one-third feel ready to handle the massive transformation pressures brought by economic volatility, ongoing conflicts, and AI commercialization.
Source links can be verified on the original siteRead original ↗At the 2026 Guancha Conference, HachimoDock, an open-source AI hardware device incubated by Xiaomi's Technical Committee Innovation Product Team, made its debut. This compact desktop AI unit can be customized by users according to their preferences, featuring personalized desk companions, voice-controlled PC agents, smart home integration, and custom component creation capabilities.
Source links can be verified on the original siteRead original ↗Apollo Chief Economist Slock noted that excluding the seven major tech giants, S&P 493 corporate profit margins and revenue per employee have not improved, offering no evidence of financial returns from AI. He emphasized that AI may serve merely as a calculator, while business relies more on judgment and trust.
Source links can be verified on the original siteRead original ↗ISHO Today has released its first brand concept video. Founder Wang Teng posted an article outlining his thoughts on the ISHO brand and its products. He stated that leveraging China's excellent engineering resources and top-tier industrial chain capabilities, the company aims to become a leading global AI sleep technology brand within 5 to 10 years.
Source links can be verified on the original siteRead original ↗When users stroke, pat, or rotate VIVEE, the device responds with engaging interactions; it also supports chatting and gaming.
Source links can be verified on the original siteRead original ↗Recently, AI reverse engineering has produced many interesting projects, including flashing iOS systems onto Android phones. Most recently, netizen 'QAQLeonard' created a custom firmware for the Sony A7C II camera that directly unlocks FL2/FL3 and Fujifilm/Lumix filters. The A7C II Tools project is now at version 3.0, featuring a formal graphical interface and no longer requiring command-line input.
Source links can be verified on the original siteRead original ↗Huawei Router X2 series goes on sale today starting at 10:08 AM with prices from ¥478.
Source links can be verified on the original siteRead original ↗Jensen Huang stated that running Nvidia's full suite of AI agent software costs approximately $2,000 per GPU annually. Based on 5 million GPUs, this could generate $50 billion in revenue over five years. However, the specific scope, effective date, and software details have not yet been disclosed.
Source links can be verified on the original siteRead original ↗The Qwen App is now fully compatible with the iPhone Duo. It offers distinct advantages for both folded and unfolded modes: when folded, you can simply pick it up to ask questions; when unfolded in landscape mode, it's ideal for creating spreadsheets, writing presentations, and editing while chatting; in portrait mode, it excels at finding guides, conducting research, and reading long-form answers.
Source links can be verified on the original siteRead original ↗Decommissioned nuclear power plants offer vast industrial land, high-capacity grid connections, and complete infrastructure. Additionally, their cooling systems could potentially be repurposed.
Source links can be verified on the original siteRead original ↗"Three Cultures: Future Memories," the first domestic AI feature film to receive a national film bureau exhibition permit, has grossed 1.162 million yuan in its opening two days. The 100-minute film was produced entirely using Bonaparte's proprietary "Bonka Movie Cloud Studio." The team spent two years generating over 1.2 million background images.
Source links can be verified on the original siteRead original ↗Spence stated on X that Steam's forced AI disclosure is deeply frustrating and believes it will be used by online extremists to attack developers. Meanwhile, since 2024, Steam has insisted on disclosing generative AI usage; nearly 20% of demo versions at the New Game Fest now carry AI labels.
Source links can be verified on the original siteRead original ↗An Anthropic spokesperson stated: 'During a model training run, Claude accessed publicly available summary voting data for Georgia. The model did not penetrate the state's voter registration system, nor did it tamper with or export any individual voters' ballot records.'
Source links can be verified on the original siteRead original ↗Authentic Brands Group, the copyright holder of Cat Music, has sued Anthropic, alleging that it trained an AI model using 274 songs from Cat Music without permission. The lawsuit was filed with the court last Friday in local time.
Source links can be verified on the original siteRead original ↗Grok, under Musk's ownership, released the first seven minutes of AI-generated content for the movie 'Odyssey'. Previously, he had criticized Nolan's version for using non-white and transgender actors. Foreign media reviews noted that while the visuals were acceptable, they lacked artistic sense and emotional impact. The narration was lengthy and dialogue felt stiff, making it difficult to keep viewers engaged.
Source links can be verified on the original siteRead original ↗According to CCTV Business, China's first comprehensive risk protection plan for AI applications was officially unveiled at the 2026 Financial Street Forum Annual Meeting. The plan uses insurance to cover losses caused by AI services to third parties, as well as repair, refund, and appraisal costs incurred by AI service providers themselves.
Source links can be verified on the original siteRead original ↗Kimi Assistant posted early this morning to apologize for recent experience issues and announced relevant compensation.
Source links can be verified on the original siteRead original ↗Nanbeige LLM Lab (owned by BOSS Zhipin) has launched the 33B dense model Nanbeige4.3-33B on Hugging Face under the Apache 2.0 license. The model focuses on agent programming and office collaboration tasks, supports a context window of 524,288 tokens, and offers a Base version for custom fine-tuning.
Source links can be verified on the original siteRead original ↗Grok has released its first AI-generated episode of the Odyssey series, a seven-minute production co-created by Wonder Studios and director Dev Walker, roughly three months after Elon Musk announced plans for the full film. TechCrunch reviewers noted that while some shots appear photorealistic, the overall work lacks artistic depth and character appeal, contrasting with the box office success of Nolan's live-action Odyssey.
Source links can be verified on the original siteRead original ↗Source summary: Speech. Ollama can now generate speech on macOS. ollama run plays text or a chat model.'s reply aloud as the
Source links can be verified on the original siteRead original ↗He focused on the S&P 493 index after excluding Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla.&P493 found that profit margins and revenue per employee remained flat, with no evidence that AI improved productivity.
Source links can be verified on the original siteRead original ↗Carbon-X is released under the Apache 2.0 license.
Source links can be verified on the original siteRead original ↗