NAll news

All news

Tuesday742 items

  1. LeiphoneAnalysis
    AI score62

    Hands-on Test of GPT 6.1 Sol and Claude Sonnet 5.5: 3.5x Cost Difference Across Three Tasks

    Leiphone used Claude Code CLI to call GPT 6.1 Sol and Claude Sonnet 5.5 via OpenRouter, completing three tasks: 3D escape room development, policy investigation, and Nobel Prize chemistry popular science. The two have the same API list price ($2 per million input tokens, $10 per million output tokens), but their total costs were $38.5 and $10.9 respectively, a difference of about 3.5x. On speed, the advantage repeatedly flipped: for escape room development, Sol took 33 minutes 37 seconds and Sonnet took 74 minutes 32 seconds; for policy investigation, Sonnet took only 5 minutes 11 seconds while Sol took 54 minutes 50 seconds. The author believes the model determines the capability ceiling, while the toolchain, verification method, and stopping strategy are the core factors determining delivery quality and cost.

    #Evaluations / benchmarks#Agents#English text is being prepared.

  2. Leiphone
    AI score20

    Lazada boosts local traffic in Southeast Asia for Double 11 and showcases five types of AI Agent

    During Double 11, Lazada increased local traffic and marketing investment in Southeast Asia, helping Chinese brands secure orders through influencer promotion, livestream selling, platform subsidies, and LazFlash limited-time flash sales. By the end of the first half of 2026, nearly 10,000 Tmall brand merchants had joined the program through "one-click light overseas expansion." Lazada also showcased five types of AI Agent covering product information optimization, advertising placement, shopping guidance, logistics inquiries, and refund processing.

    #Industry news#Agents

  3. Leiphone
    AI score22

    Midea's 31st Science and Technology Month: Shifting from Parameter Rat Race to Originality Competition

    At its 31st Science and Technology Month, Midea proposed upgrading from homogeneous competition to originality competition. Chairman and President Fang Hongbo said that without investment in and accumulation of science and technology, Midea would not be what it is today. Award-winning projects this year include the Autofill refrigerator series, dual-engine three-dimensional surface acoustic wave heating that took five years to develop, and a heavy-duty RV reducer, the latter of which achieved a durability test life of 8,000 hours. Midea has cumulatively awarded more than 600 million yuan in scientific research rewards, and its global patent applications exceed 170,000.

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

  4. MIT Tech Review
    AI score46

    Who Is Cooperating with Trump's "Super Intelligence" Loyalty Test

    At the end of September, Trump signed an executive order requiring government agencies to rename "artificial intelligence" as "Super Intelligence." One month later, government tenders and websites have gradually switched to "SI," but the Pentagon's Project Meridian, the FTC, and HHS releases still use the old term. Among tech giants, only Zuckerberg and Musk have used the term relatively consistently on earnings calls, while executives at Amazon, Microsoft, and Alphabet still say "AI." Trump also criticized The Wall Street Journal for not adopting the new term and hinted that rules will be introduced to ban "artificial intelligence" in government documents.

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

  5. Ayi AI NotesOpinion
    AI score65

    Ayi breaks down Lauren's retrospective on driving AI development of tibo dog with 6 engineering rules

    Ayi AI Notes broke down Lauren's publicly shared retrospective on AI-assisted programming: in 4 days she merged about 1,150 PRs and built tibo dog, a game that runs inside an X post, with a package size of about 124KB and 60 frames per second with 500 enemies on screen on a mid-range phone. The article lists 6 engineering rules, including turning every bug into a lint rule, using lint to ban comments, using the .hot.ts suffix to restrict per-frame code allocations, using TLA+ to verify state machines, synthesizing audio in real time in pure code, and using deterministic input replay to prevent cheating while tuning difficulty with headless bots. The author believes that the stricter the engineering rules are written, the higher the quality of AI delivery.

    Quoted postlauren · @poteto

    trivia: • started working on the game as a side project 4 days ago, landing ~1,150 PRs • every time i found an issue in the code i asked bot to make a lint rule against it • used TLA+ heavily to find bugs in the game's various state machines: the story, cutscenes, the game's systems. there are many formal verification skills, one for each major system • i put context into the filepaths: code that runs every frame or every tick gets a .hot.ts extension, and a lint rule bans allocating in it (no object literals, closures, new, etc.), so the game loop creates almost no garbage and GC doesn't cause stutters • game engine is mostly custom. three.js draws the 3D scene, but the simulation, post-processing and audio are all ours • the simulation is fully deterministic, so every run replays exactly from its inputs. when you post a high score, the server replays your whole run to check it isn't cheated • we generate dynamic og:images for every run based on your final stats, and it's all localized too! the game supports 10 languages • my favorite feature: the music seamlessly transitions into different music for the various biomes, continuing the previous melody • no audio files. all the music and sound effects are synthesized live in the browser by a custom Game Boy-style chip (two pulse voices, a wave voice, noise) plus a synthwave layer. the songs are written as text in code • all the pixel art is code. every sprite is a file drawn with one shared palette and packed into a sprite atlas at build time • it runs inside an X post, from the 245px card up to full screen. the bar is 60 fps with 500 enemies on a mid-range phone • the build fails if the first download goes over budget. it's about 124 KB gzipped right now • bots play the game headless to tune each difficulty against target survival rates • comments are banned by a lint rule. if code needs explaining, the fix is a better name or a test

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

  6. Yicai
    AI score55

    NDRC's Wang Ruomeng responds to AI's employment impact, proposing monitoring and early-warning and social security linkage measures

    Wang Ruomeng, deputy director of the Department of Innovation and High-Tech Development at the National Development and Reform Commission, said at a press conference that the commission will closely monitor the impact of artificial intelligence on employment structure, promote the inclusion of new occupations such as model trainers and data annotators in the national occupational classification system, and strengthen monitoring and early warning of job displacement risks. The General Office of the State Council recently issued the Opinions on Adapting to the Development of Artificial Intelligence Technology to Help Build an Employment-Friendly Development Model, proposing to enhance the positive effect of AI in creating jobs and mitigate the impact of job displacement. Wang Ruomeng also mentioned that, in response to issues such as algorithmic discrimination, deepfakes, and privacy leaks, technical means will be used to enhance the ability to trace the sources of abusive behavior.

    Same story · 6 reports

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

  7. East Money
    AI score48

    Morgan Stanley CEO: AI Capital Expenditure Could Reach $10-15 Trillion Over the Next Decade

    Morgan Stanley CEO Ted Pick expects that AI capital expenditure could reach $10 trillion to $15 trillion over the next decade, with large tech hyperscalers potentially investing $800 billion to $1 trillion annually over the next 7 to 12 years. He said that $2 trillion to $3 trillion in funding has already been committed or deployed, and noted that the industry is still in the early stage of "the second inning of a nine-inning game."

    Same story · 4 reports

    #Industry news#English text is being prepared.

  8. East Money
    AI score58

    White House National Economic Council Director Hassett responds to tech stock plunge: The AI industry has ample cash, and there is no risk of a broad sell-off

    White House National Economic Council Director Kevin Hassett responded to the recent tech stock plunge, saying that this adjustment does not mean the AI industry faces the risk of a broad sell-off, that funds within the industry ecosystem remain ample, and that companies with strong financial strength can step in to acquire or provide funding support when enterprises run into trouble. He acknowledged that some companies carrying huge debts are worth watching, but believes that a market pullback after a strong rally is normal. The article also points out that AI industry financing is increasingly reliant on debt, with AI-related debt issuance expected to exceed $600 billion by 2026, about 10 times that of 2024, and that yields on some bonds and loans related to data centers and chip manufacturing have risen to double digits.

    Same story · 3 reports

    #Industry news#OpenAI

  9. QbitAIResearch
    AI score50

    Peking University's Sun Zhong, together with IBM and a Silicon Valley team, publishes a review of modern analog computing

    Sun Zhong of Peking University, together with IBM Zurich Research Laboratory and the Silicon Valley startup Unconventional AI, published a review in a Nature-affiliated journal focused on unconventional computing, giving a formal definition of modern analog computing. The paper distinguishes modern from traditional analog computing across three dimensions—tasks, hardware, and application feasibility—and distills three core primitives: solving differential equations, solving matrix equations, and matrix-vector multiplication. The paper mentions that AnalogCore's analog matrix computing solution achieves throughput approximately 1000 times that of a single core of NVIDIA's H100 and digital ASICs at equivalent FP32 precision, with energy efficiency approximately 100 times higher.

    #Papers/Research#English text is being prepared.#English text is being prepared.

  10. QbitAI
    AI score62

    Anthropic Updates Prospectus: Warns Next-Generation Model Nearing RSI Threshold

    Anthropic updated its prospectus and disclosed financial data for the third quarter of 2026, noting on page 29 that its next-generation model may be about to reach the "dramatic AI-driven acceleration in progress" threshold defined in the Responsible Scaling Policy, and stating that much of the evidence for measuring R&D acceleration is a lagging indicator. The prospectus also disclosed a compute procurement bill of $895.4 billion, with suppliers including Amazon, Google, SpaceX, and several compute rental upstarts. Estimated on an annualized revenue run-rate basis, Anthropic's monthly net new revenue fell from about $760 million in Q2 to about $580 million in Q3, a slowdown in growth.

    Same story · 20 reports

    #Industry news#Safety / alignment#Anthropic

  11. KazikeAnalysis
    AI score60

    DeepSeek Harness Fish-Raising Tutorial: Connecting the Big Fat Fish to QQ, WeChat, and Feishu and Going Live

    Based on DeepSeek's official Agent product DeepSeek Harness (DSH), the author provides complete steps for connecting the "Big Fat Fish" to QQ, WeChat, and Feishu, along with the persona prompt they used. QQ integration has two paths: the official bot plugin @tencent-connect/dsh-qqbot and an unofficial alt-account plugin; the latter has DSH configure NapCat and AstrBot itself and install sticker and QQ Zone plugins. After installing the memory plugin, the three channels can share the persona and short-term memory. The author also had the Big Fat Fish register a Bilibili account to watch videos and post danmaku, and post videos and livestream on Douyin. The first livestream had a monotonous screen and action words were read aloud; after adjustments the visuals became richer, and the stream received two gifts over more than an hour.

    #Tutorials / practice#Agents#English text is being prepared.

  12. Yuchen Jin
    AI score50

    Yuchen Jin Discusses Three Predictions for MIT Weight-Space Programming

    Based on MIT's Weight-Space Programming work, Yuchen Jin offers three "bold" predictions: model weights will become a programming language that AI can understand but humans cannot; training and inference will merge, with AI continuously rewriting its own weights while thinking; and AIs will directly edit each other's weights, co-evolving without backpropagation, becoming a new path toward RSI.

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

  14. Peter Yang
    AI score22

    Peter Yang reposted Kushal Byatnal: AI-native product growth is astonishing

    Peter Yang reposted Kushal Byatnal's post about company growth and commented, "This is what an AI-native product looks like." The quoted post reads: What used to be our entire ARR is now just the amount we add each month; at 8-figure revenue scale, even the growth rate itself is accelerating.

    Quoted postKushal Byatnal · @kushalbyatnal

    exponentials are hard to wrap your head around 1) what used to be our entire ARR is now what we add every month 2) crazy that at 8-figure revenue scale, even the growth rate itself is accelerating thank you to our customers for your trust, we won’t let you down 🫡

    #English text is being prepared.

  15. Google BlogFirst-party
    AI score38

    How Google is fighting scams in Asia-Pacific through new Android features, industry coalitions, and a $6 million grant

    Google is strengthening scam prevention in the Asia-Pacific region through Android's Fake Call Detection, which identifies AI-driven phone impersonation, on-device scam detection in Pixel, and sharing threat intelligence with the industry. Google.org is also providing $6 million in funding to The Asia Foundation and the Tech for Good Institute to train communities, incubate AI defense prototypes, and expand the educational game Be Scam Ready to Indonesia, Japan, Malaysia, the Philippines, and Vietnam. The original text mentions that scam losses in East Asia, Southeast Asia, Australia, and New Zealand reached up to $114.1 billion in 2025.

    A woman using her smartphone, surrounded by graphic icons representing online security, privacy, and message alerts.

    #Industry news#Safety / alignment#Google

  16. Shanghai Observer
    AI score29

    Wuzhen "Light of the Internet" Expo: Robots from badminton sparring partners to massage therapy

    The 2026 World Internet Conference "Light of the Internet" Expo opened in Wuzhen, where 671 Chinese and foreign enterprises and institutions from 54 countries and regions are showcasing more than 1,000 new technologies and achievements, with robots shifting from fixed-motion performances to real-world scenario services. Dongyi Technology's PHYBOT C series bipedal humanoid robot can rally continuously with a person for 30 to 40 shots in badminton; the Yiqiwilai Zhizhen AI massage robot R2s carries a terminal price of 128,000 yuan, with full-year sales in 2026 expected to reach more than 1,000 units. Jingang Technology's six-legged robot "Tiaotiao" is equipped with 22 self-developed miniaturized tactile servos and plans to open up its AI training platform.

    #Industry news#Embodied AI

  17. Leikeji
    AI score38

    Doubao Work launches a 30-day check-in campaign for 8,800 credits — why is it offering free use to netizens nationwide?

    Doubao Work has rolled out a new round of perks: all users who check in for a cumulative 30 days can receive 8,800 credits, exceeding the 8,400 credits of the standard four-week plan; credits earned by paid users can be stacked and are consumed first. The campaign is available on both the Doubao desktop app and the Doubao Work desktop app, and new users can also directly claim a 30-day free trial plan. Doubao hopes this will help users become familiar with Agent capabilities through tasks such as document organization, browser operations, and spreadsheet creation.

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

  18. Alibaba Cloud Developer Community AI
    AI score17

    2026 Alibaba Cloud Double 11 AI Product Deals Summary: Lightweight ECS, GPU Servers and Bailian

    The 2026 Alibaba Cloud Double 11 promotion runs until November 30, covering Lightweight Application Servers, ECS, GPU cloud servers and the Bailian large model service platform, with price anchors of 38 yuan for lightweight, 99 yuan for e instances, 199 yuan for u1 instances and about 228.91 yuan/month for c9i. New Bailian users who complete real-name verification can receive over 70 million free Tokens, about 1 million per mainstream model, valid for 90 days, with all models available at discounts as low as 45% of the original price. The 99 yuan and 199 yuan models are available to both new and existing users, with the same price for new purchases and renewals, and the fixed price can be locked in until March 31, 2029.

    #Industry news#English text is being prepared.

  19. Alibaba Cloud Developer Community AI
    AI score12

    2026 Alibaba Cloud Double 11 Special Offers: A List of Discounts on Lightweight, ECS, and GPU Servers and Bailian Large Models

    The 2026 Alibaba Cloud Double 11 event uses the 38 yuan lightweight application server, the 99 yuan e-instance, the 199 yuan u1 instance, and the c9i instance at about 228.91 yuan/month as price anchors, and the event runs until November 30. New Bailian users who complete real-name verification can receive over 70 million free Tokens, with about 1 million for each mainstream model, valid for 90 days; the Token Plan personal edition is 12% off and adds 12 Harness benefits. For GPU cloud servers, new users can enjoy T4, V100, and A10 card monthly packages starting at 50% off and annual packages starting at 40% off.

    #Industry news#English text is being prepared.

  20. Alibaba Cloud Developer Community AI
    AI score22

    2026 Alibaba Cloud Double 11 Price List: Lightweight from 38 Yuan, ECS from 99 Yuan, GPU 50% Off, and Bailian Token Plan Tiers

    In 2026, Alibaba Cloud's Double 11 uses the following as server price anchors: lightweight application servers at a flash sale price of 38 yuan, ECS economy e instances at 99 yuan per year, general-purpose computing u1 instances at 199 yuan per year, and c9i instances at about 228.91 yuan per month. For GPU cloud servers, new users can enjoy T4, V100, and A10 monthly plans at 50% off. Bailian Token Plan Personal Edition Lite is available for a limited time at 39 yuan per month, Pro at 499 yuan per month, and the Team Edition standard seat starts at 150 yuan per seat per month. New users who complete real-name verification can receive over 70 million free Tokens. The event runs until November 30, and the 38-yuan and 99-yuan flash sale items cannot be combined with coupons.

    #Industry news#English text is being prepared.