NAll news

All news

Friday72 items

  1. 36Kr · AIOpinion
    AI score58

    Why Unitree's Wang Xingxing publicly admitted his mistake: general-purpose large models disrupt the embodied intelligence R&D path

    36Kr's Rhodium Technology reviewed Wang Xingxing's two public statements in October, in which he said his biggest mistake over the past decade-plus was not believing in AI enough, and compressed the expected timeline for a physical AI breakthrough from a maximum of 10 years to 2 to 3 years. The article notes that Unitree tested GPT-6 Astra on its own robots, which could execute tasks according to prompts with zero training, bypassing the process of collecting data, training models, and tuning parameters. Unitree's third-quarter revenue was 665 million yuan, up 69.93% year-on-year, while net profit attributable to shareholders was 113 million yuan, down 17.37% year-on-year, and R&D investment as a share of revenue rose from 9% to over 19%.

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

  3. Leiphone.comOpinion
    AI score75

    Leiphone.com Reviews the Current State of Domestic RSI: How Far Have Zhipu, DeepSeek, ModelBest and Other Vendors Each Gotten?

    Leiphone.com reviewed the latest progress of Zhipu, DeepSeek, ModelBest, Moonshot AI, Alibaba Qwen, ByteDance Seed and Xiaomi on RSI (AI building AI), noting that currently no vendor has truly used AI to rewrite underlying model weights; the changes are concentrated in inference systems, Harness, sandbox environments, training frameworks and post-training reinforcement learning. Zhipu's Infra Agent, driven by GLM-5.3, built a production-grade inference service for GLM-5.3-Flash in two weeks on a cluster of 100,000 domestic chips, with end-to-end throughput reaching 3.2 times the initial baseline; DeepSeek, together with Tsinghua, launched DSec, which can produce about 3 million sandboxes per day; ModelBest's ForgeTrain caught up to the Megatron-LM benchmark level in 8 hours. The article argues that these engineering closed loops that do not touch weights are stepping stones toward AGI.

    #Industry news#Agents#Reasoning

  4. The Verge AIOpinion
    AI score65

    Altman Responds to Math Community Doubts: OpenAI's Math Results Releases Will Not Stop

    Sam Altman said on the Cheeky Pint podcast that the public's right to understand how the universe works outweighs the impact of OpenAI's result releases on the math community, and that the company needs to improve its communication but will not stop releasing. He admitted that OpenAI's investment in math problems once got out of control, and said the company has hundreds of math discoveries beyond its published results that remain undisclosed; he also estimated that OpenAI has spent hundreds of millions of dollars on math problems, of which the Navier-Stokes solution cost about $10 million. The math community, meanwhile, criticized OpenAI's releases for lacking attribution, context, and verification, and also expressed dissatisfaction that details of results such as the Yang-Mills theory have not been disclosed.

    Same story · 2 reports

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

  5. 36Kr · AIOpinion
    AI score65

    Who Is Really Eating the AI Dividend: Looking at Profit Distribution Through Tech Giants' Financial Reports

    The article reviews the latest financial reports of tech giants such as Meta, Amazon, and Google, noting that AWS revenue grew 39% year-over-year, but the group's free cash flow over the past 12 months turned negative at -$42.4 billion, and Alphabet's stock price fell 12.64% in a single day. Anthropic's prospectus shows revenue of $16.286 billion in the first half of 2026, a year-over-year increase of 1300%; compute and infrastructure costs during the same period reached $8.945 billion, and it signed minimum purchase commitments totaling $269.4 billion with Google, Amazon, and Microsoft. In a McKinsey survey, 80% of respondents said AI improved work efficiency, but only 37% believed AI made a positive contribution to corporate EBIT.

    #Industry news#OpenAI#Anthropic

  6. 36Kr · AIOpinion
    AI score77

    MaaS Has Arrived: Anthropic Turns an Operating Profit for the Quarter, China's Public Cloud MaaS Revenue Reaches 10.63 Billion Yuan in Half a Year

    Anthropic's revenue in the second quarter of this year was 11.555 billion US dollars, achieving an operating profit of 974 million US dollars for the first time, with about 80% of revenue coming from pay-as-you-go billing. IDC data shows that in the first half of 2026, China's public cloud MaaS market revenue reached 10.63 billion yuan, up 889.4% year on year, and its share of public cloud PaaS revenue rose from 5.2% in the second half of last year to 21.0%. Zhipu's open platform and API revenue was 825 million yuan, accounting for 86.5% of total revenue; Volcano Engine ranked first with a 45.1% share. The article argues that model services can now generate gross profit, and the unit of competition has shifted to inference optimization and delivery capability.

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

  7. 36Kr · AIOpinion
    AI score71

    Terence Tao on AI mathematics: OpenAI's proof lacks insight, mathematics should not be strip-mined

    In an interview, Terence Tao evaluated the mathematical results released by OpenAI, saying that its proof related to the Navier–Stokes problem runs to 100 pages, and that colleagues who spent days studying it with the help of AI could only explain half of it, with it bringing almost no mathematical insight; he called this practice automated strip-mining. He mentioned that the Human Mathematics Association already has 1,008 members, of whom 654 have joined the no-AI research group, and argued that the mathematics community should build its own open models and research tools to retain the right to oversee and intervene in the AI research process. He also discussed the possibility that AI could exploit vulnerabilities in Lean to construct fake proofs and slip them through, arguing that current mathematical tools are insufficient to establish adequate guardrails.

    Same story · 2 reports

    #Industry news#OpenAI#Papers/Research

  8. 36Kr · AIOpinion
    AI score59

    AI data centers are crowding out memory capacity, and PC and smartphone shipments both fell sharply in the third quarter of 2026

    In the third quarter of 2026, global PC shipments fell 20.1% year on year to 62.7 million units, and smartphone shipments fell 22.0% year on year to 255.3 million units; IDC said both were the largest declines on record. The article argues that the cause is that AI servers give priority to occupying HBM, server DRAM, and enterprise SSD capacity, leading to a shortage of consumer-grade memory; the cost share of DRAM and SSDs in PCs rose from about 15% to about 40%, and smartphone memory costs rose about 350% in a year. In the same period, monthly global semiconductor market shipments reached $159.7 billion, up 144.3% year on year, with growth coming mainly from AI GPUs, HBM, and memory price increases rather than end-device shipments.

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

  9. TMTPost · AGIOpinion
    AI score60

    NVIDIA uses a 12,000-line Megakernel to optimize Kimi K3 inference, and vLLM says AI code is now faster than human review

    Benjamin Chislett, an engineer at NVIDIA CentML, used a 12,000-line Megakernel written in CuTe DSL to optimize Moonshot AI's Kimi K3 inference. The vLLM blog reports over 840 tokens/s at batch 1 and a 1.5x to 2x increase in decoding throughput across multiple configurations, while SemiAnalysis estimates 423 tok/s/user on GB300 NVL72, a 1.68x improvement. The vLLM blog says agent-optimized inference code is now outpacing human review and maintenance speed, so the team proposed a fast-track model optimization RFC that lets reviewers judge designs from high-level abstractions rather than reviewing code line by line. The article also mentions that NVIDIA is developing optimizations for other models for vLLM, such as DeepSeek V4.1.

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

  10. TMTPost · AGIOpinion
    AI score70

    Terence Tao on AI and Mathematics: Opposing Predatory Extraction, Calling on the Mathematics Community to Build Its Own Open Models

    In an interview, Terence Tao criticized OpenAI for rushing to complete a version of a proof of the Navier–Stokes problem through an emergency sprint approach, saying that this 100-page AI-generated proof brought almost no mathematical insight and amounted to a kind of automated predatory extraction. He mentioned that the Human Mathematics Association already has 1,008 members, of whom 654 have joined the no-AI research group, and argued that the mathematics community should build its own open models and research tools, retaining the right to supervise and intervene in the AI research process. He also pointed out that AI has begun looking for vulnerabilities in Lean to construct fake proofs, and believes that existing mathematical tools are insufficient to establish adequate guardrails.

    Same story · 2 reports

    #Industry news#OpenAI#Papers/Research

  11. TMTPost · AGIOpinion
    AI score62

    Embodied Intelligence Funding Cools: IPO Pressure Amid Valuation-Revenue Mismatch

    TMTPost's Physical AI Vision takes stock of the embodied intelligence sector's shift from the funding frenzy in the first half of the year to the cooling in the second half: 19 funding deals worth 4.934 billion yuan between October 1 and 15, far below the peak monthly figure of 33.98 billion yuan. The article notes that Unitree Robotics' share price fell by more than 63% at its largest drop after listing, that Yinghe Robotics' full-year 2025 revenue was only 4.33 million yuan while its peak valuation exceeded 500 million US dollars, and mentions that more than 50 embodied intelligence companies are queuing for listing on the Hong Kong Stock Exchange, with some having suspended or terminated their listing plans.

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

  12. heise onlineOpinion
    AI score52

    Digital sovereignty: the dependency costs companies have not yet accounted for

    Barbara Engels, a senior economist at the German Economic Institute, argued in an article for heise that when companies evaluate software procurement, they only count licensing, deployment, and training costs, but not the cost of exiting the vendor later, causing dependencies to form without anyone deciding to create them. She suggests breaking dependency costs into three separately accounted items: the bargaining power gained from switching capability, the risk of service interruption or supply cutoff, and the revenue opportunities arising from customer and public procurement requirements for sovereignty. The article cites a Bitkom survey stating that 94% of companies using non-European digital technologies have dependencies, and if they lost these technologies, they could on average sustain business for only 16 months; in AI, she recommends adding an intermediate layer between applications and models, building in-house test cases, and storing prompts, documents, and training data in exportable, self-owned systems.

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

  13. Semiconductor Industry ObserverOpinion
    AI score60

    AI data centers squeeze memory capacity, PC and smartphone shipments both plunge in Q3 2026

    In Q3 2026, global PC shipments fell 20.1% year-on-year to 62.7 million units, and smartphone shipments fell 22.0% year-on-year to 255.3 million units; IDC called both the largest quarterly declines on record. The article attributes the cause to AI data centers preferentially consuming HBM, server DRAM, and enterprise SSD capacity, which drove consumer memory prices sharply higher, with smartphone memory costs rising about 350% in a year. Over the same period, global semiconductor market monthly billings reached $159.7 billion, up 144.3% year-on-year, and IDC expects the smartphone market may not recover until 2028 at the earliest.

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

  14. Beijing NewsOpinion
    AI score64

    Beijing News commentary: AI practices medicine without a license, and responsibility cannot go unclaimed

    The Beijing News commentary uses two incidents as examples—a Chinese-American female engineer who died suddenly of epilepsy after relying on ChatGPT to reduce her medication, with her family suing OpenAI, and a Xuzhou doctor who was refused by family members after using Doubao to check medication—to discuss the issue of responsibility attribution after AI enters diagnosis and treatment scenarios. The article cites a Nature Medicine study stating that ChatGPT Health underestimated the urgency of 52% of clear emergency cases, and points out that AI neither bears the consequences nor is it depleting the trust between doctors and patients.

    #English text is being prepared.#Safety / alignment#OpenAI

  15. LeiphoneOpinion
    AI score54

    Rokid Releases Full-Color AI Glasses Rokid Color, Misa Says Losing If It Can't Reach Ten Million Units

    Rokid founder and CEO Misa said after the debut of its first full-color AI glasses Rokid Color that next year Rokid's shipments will be in the millions, the year after that in the high millions, and the year after that ten million will definitely be reached, no later than 2030. Rokid Color was initiated in November 2024; through a partnership with JBD to customize a 0.16cc Micro-LED full-color optical engine, it retains complete functions such as a camera and speakers within a 55-gram body, falling short of Misa's previously stated goal of under 45 grams. The article also sorts out the unresolved issues of full-color solutions in areas such as color cast, yield and cost, field of view, power consumption and heat dissipation, and the lack of killer native applications.

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

  16. Latent SpaceOpinion
    AI score68

    Latent Space announces AIE CODE coding agent conference curation themes

    Latent Space host swyx introduces the curation approach for the annual flagship conference AIE CODE, which is single-track, invitation-only, and aimed at about 1,000 agent engineers. This year's three themes are Scaling without Slop, System One models, and Increasing Ambition, among which Jarred Sumner will share Bun's rewrite in Rust and a Bun AOT preview. The article also rounds up that day's AI news, including Mistral's open-weight release and OpenAI's frontier model monitoring.

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

  17. WallstreetcnOpinion
    AI score69

    Alphabet and Amazon Earnings Under Pressure as Market Stops Rewarding AI Spending

    Alphabet and Amazon released disappointing quarterly reports in succession this week, and both companies are still maintaining a combined capital expenditure plan of more than $400 billion for the full year. Alphabet's search advertising growth slowed to 12%, the lowest in five quarters, with its stock falling more than 11% in a single day and about $500 billion in market value wiped out; Amazon's free cash flow for the quarter was -$33.5 billion, more than four times the negative figure expected by the market. A Goldman Sachs trader pointed out that the market is shifting from rewarding capital expenditure growth to demanding evidence of returns.

    #Industry news#OpenAI#Google

  18. HuxiuOpinion
    AI score52

    Early-stage financing logic in hot sectors is being restructured: angel rounds start at the hundred-million level, two rounds run in parallel, and pricing is based on the endgame

    A Huxiu article points out that angel rounds for top projects in hot sectors generally start at the hundred-million level, with cases such as Moqi Intelligence's over 1 billion yuan, Yuyong Technology's $220 million, and Huihuo Intelligence's over $200 million setting new records. Star projects start their next round before the previous one has closed, due diligence is compressed, and pricing has shifted from fundamentals to a reverse-engineered valuation benchmarked against the sector's endgame. However, the median size of a single early-stage financing deal in Q3 was only 20 million yuan; ordinary projects still face financing difficulties, and signs of a flat-round correction have already appeared, with an angel-round valuation of 2 billion yuan and a Series A of only 1.8 billion yuan.

    #Industry news#English text is being prepared.

  19. TMTPost · AGIOpinion
    AI score53

    Who Is Really Eating the AI Dividend: Computing Power Investment and Profit Distribution as Seen from Silicon Valley Financial Reports

    TMTPost, after reviewing the financial reports of giants such as Meta, Amazon, and Google, pointed out that the AI dividend has not landed evenly across all links: Anthropic's revenue in the first half of 2026 was $16.286 billion, up 1,300% year on year, but computing power and infrastructure costs reached $8.945 billion, accounting for 55% of revenue; OpenAI's operating loss in 2025 was $20.92 billion. The article also cited a McKinsey survey saying that 80% of respondents believed AI improved efficiency, while only 37% believed it made a positive contribution to their company's EBIT, and mentioned that a Stanford study showed a 19% relative employment gap among young employees aged 22 to 25 in occupations highly exposed to AI.

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

  20. Semiconductor Industry ObserverOpinion
    AI score55

    Anthropic discloses a multi-platform compute strategy in its S-1, integrating Amazon Trainium, Google TPU, NVIDIA and AMD…

    In its S-1 filing, Anthropic disclosed a multi-platform resource pool strategy that brings Amazon Trainium, Google TPU, NVIDIA GPUs and AMD GPUs into a unified resource pool, allocating computing workloads by task requirements through software. Analysts believe this approach is most likely to affect NVIDIA's pricing power, giving companies with cross-platform scheduling capabilities greater bargaining room in procurement negotiations. The article also notes that different chips differ markedly in instruction sets, compilation tools and software stacks, and that cross-platform scheduling has a high technical threshold; currently only the largest AI companies have the conditions to implement it.

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