NAll news

All news

Tuesday788 items

  1. East Money
    AI score31

    CITIC Securities: Trends within the technology sector are diverging, and non-AI sectors present broad opportunities for structural recovery

    A CITIC Securities research report forecasts that in 2027, trends within the technology sector will diverge, non-AI sectors will present broad opportunities for structural recovery, and the main process of valuation contraction was already completed in 2026. The research report estimates that in 2026 and 2027, the net profit attributable to parent companies of all A-share companies will grow by 15.5% and 9.5% year on year, respectively, and the net profit attributable to parent companies of all non-financial A-share companies will grow by 20.1% and 14.7% year on year. In terms of allocation, it is recommended to moderately diversify around four threads: steady returns, excellent overseas-expanding manufacturing, domestic demand recovery, and new AI technology themes.

    Same story · 2 reports

    #Industry news#Agents

  2. East Money
    AI score53

    Grok Bot users report quotas running out early; Musk says service costs are too high to sustain

    Musk responded on X to complaints from Grok Bot users that their quotas were running out early, saying the product's service costs are too high relative to revenue and unsustainable. Multiple subscribers reported that weekly quotas are being consumed noticeably faster, with some using up their quota with 4 days still left before the reset. Grok Bot is an AI agent launched by SpaceX in August this year, offering tiers ranging from a free version to a $200-per-month plan, with each tier having usage limits that reset weekly; reports say its daily active users have grown from fewer than 100,000 in early September to more than 600,000 as of October 26.

    Same story · 4 reports

    #Industry news#Agents#xAI

  3. Phoenix Tech
    AI score35

    OpenAI tells 2028 US election candidates: AI will be a key question they must answer

    OpenAI Chief Global Affairs Officer Chris Lehane privately sent a memo to about six potential candidates interested in running in the 2028 US presidential election, saying that AI will become a major issue that shapes the next presidential race. He warned Democrats that if they oppose building the infrastructure needed for AI development, they could be at a disadvantage in the policy debate, and urged candidates to give voters a complete governance vision on issues such as taxation in the AI era, child protection, and human-led decision-making.

    Same story · 3 reports

    #English text is being prepared.#OpenAI

  4. Phoenix Technology
    AI score41

    Tencent TenPay Global and Mastercard Sign Memorandum of Understanding to Advance Inbound QR Code Payments, Remittances and Agentic Payments

    Tencent's cross-border payment platform TenPay Global and Mastercard have signed a memorandum of understanding that will combine Mastercard's global network with the WeChat Pay ecosystem to enable inbound tourists to make QR code payments within China using their existing digital wallets or banking apps, extending to cross-border remittances and agentic payments. From January to August 2026, the number and value of transactions made in China by overseas e-wallets via WeChat Pay grew by 81.8% and 80.6% year-on-year, respectively. On agentic payments, the two parties will explore trust, security and interoperability around Mastercard Agent Pay, Agent Connect and Virtual C-Suite.

    #Industry news#Agents#English text is being prepared.

  5. Phoenix Technology
    AI score48

    Alphabet Plunges 12% After Earnings: The Great AI Bubble Debate and Lessons from History

    Alphabet's third-quarter revenue was $127.577 billion, up 25% year over year, and Google Cloud revenue was $30.709 billion, up 103% year over year, but free cash flow was negative $9.07 billion and capital expenditure was about $51.4 billion; after the earnings report, the stock fell 12% overnight. The bull-bear divide centers on whether the returns on new AI capital expenditure can cover the cost of capital, and on the opaque risks brought by circular financing. The article uses the railway and telecommunications bubbles as reference points, and compares them with three types of mismatches in China's computing power buildout: construction versus use cases, hardware versus software, and general capabilities versus industry demand.

    #English text is being prepared.#Industry news

  6. Shao Meng
    AI score10

    Shao Meng helps forward: Boyi seeking platform and AI Agent infrastructure roles

    Shao Meng forwards Boyi's job-seeking information: looking for opportunities in platform engineering, cloud infrastructure, or AI Agent infrastructure, located in Hangzhou or remote. Boyi has over 8 years of backend and systems engineering experience, formerly served as platform engineering lead at Zeabur, with Go as the primary language.

    Quoted postca110us · @ca110us

    English text is being prepared.

    #English text is being prepared.

  7. Garry Tan
    AI score48

    Garry Tan: Capy Captains Boost Issue-Fixing Throughput by About 5x

    Garry Tan said that Capy Captains boosted his issue-fixing throughput by about 5x overnight, and called the ability to coordinate across multiple captains and sub-threads a huge breakthrough. The Capy announcement he cited shows that Captains and Issues in Capy 0.7.0 are now generally available (GA).

    Quoted postCapy · @capydotai

    Captains and Issues are GA! Capy 0.7.0 • Captains: persistent agents, orchestrate hundreds of threads across devices • Issues: filters, full-text search, labels, and due dates • Dock CLI: run Capy headless on any Linux, macOS, or Windows machine https://t.co/JvEmE7I33c

    #English text is being prepared.#Agents

  8. Tibo
    AI score50

    Tibo responds to OpenAI's self-developed Git backend Pando

    OpenAI Developers announced Pando, a self-developed Git backend for its monorepo, stating that it reduces push latency by about 80%, serves nearly 1 PB of Git data per day, and can scale Git capacity in step with growing agent usage. Tibo reposted it, saying the team "knows some Git kung fu" and is glad to finally be able to reveal a little about it, and gave a shout-out to @tnm and the team.

    Quoted postOpenAI Developers · @OpenAIDevs

    We built Pando, a new Git backend for our monorepo, to handle more pushes and fetches without changing how our engineers and agents use Git. Pando cut push latency by ~80% and now serves nearly a petabyte of Git data a day. We can grow Git capacity alongside agent usage. https://t.co/4c0YXOfKqK

    #English text is being prepared.

  9. Rohan Paul
    AI score87

    Anthropic's revised IPO filing discloses up to $1.07 trillion in compute commitments

    Anthropic's revised IPO filing lists $895.4 billion in compute commitments, with another up to $175 billion not yet priced, for a combined total of up to about $1.07 trillion—roughly 32 times its $33.2 billion in revenue as of September. These payments extend through 2045, with $69.8 billion due within one year, while the company holds $131.9 billion in cash and short-term investments. Quarter-over-quarter growth slowed from 144% to 47%, revenue reached $17 billion, and compute costs rose to 58% of sales.

    Same story · 20 reports

    #Industry news#Anthropic#OpenAI

  10. Simon Willison
    AI score34

    Simon Willison releases LLM CLI version 0.37

    Simon Willison has released version 0.37 of the LLM CLI and Python library, adding the gpt-6.1-sol and gpt-6.1-terra models and the llm openai endpoint add command, and changing the default model to gpt-6-luna, at half the price of the previous default, gpt-5.6-luna. Embeddings now support binary content of different content-types, to accommodate Gemini Embedding 2's text, audio, image, and video embeddings. This release includes fixes from 12 authors, almost all completed with AI assistance.

    #Product updates#OpenAI#English text is being prepared.

  11. 36Kr · AIProduct release
    AI score73

    Karpathy Built the HackerTube Video Site with Claude Code in Two Hours

    Karpathy used the Claude Code desktop app with Opus 5.5 at Medium reasoning effort to build HackerTube in two hours; the page layout replicates YouTube, the videos come from YouTube, and the filtering rules are drawn from videos recommended on Hacker News over the years, 6,454 in total. He personally did only three things: create a Google Cloud API key, start a VPS, and buy a domain; coding, data processing, deployment, and security hardening were all left to Claude. The site does not require login, watch status is stored in the local browser, thumbnails are taken from a middle frame of the video, and the underlying data can be exported to one's own AI agent to remake a version.

    Same story · 10 reports

    #English text is being prepared.#Agents#English text is being prepared.

  12. 36Kr · AI
    AI score35

    ThredUp Uses AI to Price Secondhand Clothing, Achieving a Gross Margin of 79.5%

    Secondhand clothing platform ThredUp uses machine learning and reinforcement learning to handle identification, pricing, and automatic markdowns, selling about 17 million items of clothing each year. In the second quarter of 2025, its gross margin reached 79.5%, far above the under-30% level common among traditional apparel retailers. In August 2024, it launched AI features such as image-based search and Style Chat; consumers who use image search are 85% more likely to place an order, and in 2024, new customers brought in by AI grew 32% year over year.

    #Industry news#Agents#English text is being prepared.

  13. 36Kr · AIAnalysis
    AI score62

    Karpathy built HackerTube with Claude in two hours, featuring only videos recommended on Hacker News

    Karpathy had Claude sift through 43 million comments from nearly 20 years of Hacker News to extract nearly 40,000 comments recommending YouTube videos, organizing them into 6,454 videos across 150 topics, and launched a YouTube-like site called HackerTube within two hours. Throughout the process he did not read or write any code, only handling server setup and domain purchase; Opus 5.5 oversaw the work, Sonnet batch-screened comments, and deployment and security hardening were also completed by Claude. He disclosed that over 90% of model usage went to comment annotation, app development and deployment used about 1 million Opus tokens, and the comment analysis data has been packaged and released publicly.

    Same story · 10 reports

    #English text is being prepared.#Agents#English text is being prepared.

  14. 36Kr · AI
    AI score57

    Anthropic Updates Prospectus: Admits It Previously Failed to Timely Identify AI R&D Acceleration, and Bans Claude from Fully Autonomously Designing Next-Generation Models

    Anthropic submitted an updated prospectus (S-1/A) to the U.S. SEC, changing the first version's "may be unable to identify before R&D significantly accelerates" to "has previously failed to identify," acknowledging that AI R&D acceleration has already genuinely occurred internally. The filing discloses that by September of this year, more than 90% of code in the internal codebase was generated by Claude, the average daily code commits per engineer rose from 35 times the 2024 average to more than 45 times, and internal spending on calling Claude, at public list prices, grew more than 250-fold from January to September. Page 188 explicitly states that Claude will not be allowed to fully autonomously design and build next-generation frontier models without human oversight, and that it is considering requiring key decisions to be reviewed and signed off by humans.

    Same story · 20 reports

    #Industry news#Safety / alignment#Anthropic

  15. 36Kr · AI
    AI score34

    Altman Confirms: The Next-Generation GPT That Cracked the Navier-Stokes Equations Is About to Be Released

    At ADIPEC 2026, Altman confirmed that the internal model that solved the Navier-Stokes equations is the next-generation GPT, which is about to be released. The model once used 10,000 flat (non-hierarchical) Agents running continuously for 88 hours to complete the proof, then took another 17 hours to complete Lean formal verification, consuming about 2.7 million messages and 130 billion output tokens. By October, similar problems could already be completed in about 3 hours with a single prompt.

    Same story · 4 reports

    #Model releases#Agents#Reasoning

  16. 36Kr · AIOpinion
    AI score59

    From Nvidia to Pika, Why Silicon Valley Companies Build on MiniMax H3

    After MiniMax opened the weights of its multimodal model H3 in August this year, Nvidia's research team released Open-H3-Regenerate-2K based on H3, Pika launched the post-trained version H3 Flex on October 29, and fal trained H3 Max. The article states that within one month of H3's open-sourcing, downloads exceeded 24 million, spawning more than 300 public models, and in September it ranked first on the Arena image-to-video leaderboard with a score of 1495. The author argues that open weights turn partners into nodes through which model vendors enter more industries, and cites the H3 community license agreement, noting that commercial products or services with annual revenue exceeding $20 million must obtain prior written authorization from MiniMax.

    #Industry news#Multimodal#Open-source ecosystem

  17. 36Kr · AIOpinion
    AI score64

    Behind Character.AI's Bankruptcy: Why AI Companionship Still Struggles to Make Money Despite Revenue Growth

    Character.AI's parent company filed for bankruptcy protection on October 12, and Disney bid $20 million for a non-exclusive perpetual license to part of its intellectual property. The article compares cases such as Chai's annualized revenue of $120 million and Maoxiang's cumulative iOS in-app purchases exceeding $22 million, points out that AI companionship products generally face problems such as high computing power costs, general assistants replacing basic needs, and weakening reasons to pay, and outlines the different paths of three types of players: large-model vendors, internet platforms, and independent startups.

    Same story · 3 reports

    #Industry news#Agents#English text is being prepared.

  18. 36Kr · AI
    AI score66

    Altman Previews Next-Generation GPT: 10,000 Agents Collaborate to Complete Navier–Stokes Proof

    At the ADIPEC 2026 energy conference in Abu Dhabi, Sam Altman disclosed that the Navier–Stokes mathematical breakthrough previously announced by OpenAI was accomplished by 10,000 Agents driven by the upcoming next-generation GPT, which worked collaboratively and continuously for about 4 days, with no hierarchy and no unified leader. OpenAI disclosed that the experiment started on September 1, with the Agent scale ultimately reaching about 10,000; it completed the core mathematical exploration in about 88 hours, exchanging 2.7 million messages and generating about 130 billion Tokens, after which GPT-6 Astra used 17 hours to complete formalization and verification through Lean. Altman said the research cost about $10 million, and he expects that by September 2027 it could be redone at about one percent of the cost, but he did not disclose the model name or release date of the next-generation GPT.

    Same story · 4 reports

    #Industry news#Agents#OpenAI

  19. 36Kr · AIOpinion
    AI score62

    First financial reports from China's four little GPU dragons: after revenue doubles, profitability and the CUDA ecosystem become the real thresholds

    Moore Threads, MetaX, and Enflame have successively delivered their first quarterly reports since going public, with year-over-year revenue growth of 319.15%, 128.45%, and 331.74% respectively. Moore Threads and MetaX turned positive on a non-recurring-gains basis, while Enflame's single-quarter loss narrowed by 56.12%. After comparing the four companies' roadmaps and books, the article points out that the gap in hardware specifications is narrowing, but the CUDA ecosystem and the scale of R&D investment remain hard to catch up with. The focus of domestic computing power demand is shifting toward inference, while the utilization rate of intelligent computing resources remains low; the industry is moving from scrambling for cards and building clusters to competing on utilization rates and payback speed.

    #Industry news#English text is being prepared.#English text is being prepared.

  20. 36Kr · AI
    AI score42

    The tug-of-war over account bans between Chinese users and Claude: paying for accounts, getting banned repeatedly, yet still not leaving

    Anthropic's official report shows that from January to June 2026, a total of 11.4 million accounts worldwide were banned, and of 398,000 appeals, only 42,000 succeeded; its IPO prospectus discloses that in the first nine months of 2026, about 5% to 7% of revenue came from accounts later determined to be in violation. Chinese users are frequently banned due to factors such as proxy nodes, third-party top-ups with virtual cards, time zones, and language habits, yet they still top up monthly and repeatedly buy new accounts to continue using Claude, mainly because they value its irreplaceability in logical reasoning, generation of complex code architectures, and long-text comprehension.

    #English text is being prepared.#Anthropic