NAll news

All news

Monday23 items

  1. Electronic Technology ApplicationOpinion
    AI score55

    Altman says running AI locally wastes scarce DRAM, advocates maximizing cloud utilization

    OpenAI CEO Sam Altman said on the podcast "Cheeky Pint" hosted by Stripe co-founder John Collison that having everyone keep globally scarce DRAM memory at home while using it only 5% of the time is a rather bad global misallocation of resources, and that when supply is constrained, the cloud should be used to maximize utilization. He also called progress in resolving the memory shortage "terrible," arguing that memory manufacturers are conservative in capital expenditure due to historical loss cycles, and that the "invisible hand" is failing. He opposes a single superintelligence and prefers "Agent Teams," and said tokens are a "fake unit" of business value.

    Same story · 11 reports

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

  2. BBC ChineseOpinion
    AI score56

    AI bubble fears intensify: new cloud companies face debt and stock price pressure

    BBC Chinese reports that data center companies known as new cloud providers, such as CoreWeave and Nebius, have expanded by leasing computing power to developers including Anthropic, OpenAI, and Meta, and their financial soundness is now sparking market debate. Gartner predicts that by 2030, 80% of new cloud companies will cease independent operations, while Forrester expects at least three to be acquired in the short term; CoreWeave's debt has reached $35.1 billion, its stock price has fallen by a quarter over the past month, and Nebius, IREN, and Applied Digital have seen similar or larger declines. Microsoft canceled its cooperation agreement with the UK's Nscale due to power supply issues, and Australia's Firmus has shelved its IPO plans.

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

  3. Sina TechOpinion
    AI score55

    Charting China: Token Usage Surges 1,400-Fold—What Is China's Large Model Ecosystem Changing?

    The third installment of Guancha's Charting China uses six charts to outline China's large model ecosystem: the national average daily Token usage rose from 100 billion in early 2024 to 140 trillion in March 2026, an increase of about 1,400-fold. The article also provides data such as a coding Agent consuming about 7 million Tokens per day, and Chinese models' share on OpenRouter rising from 6% in early 2025 to 61.5% in July 2026, and notes that Chinese models contribute 65% of usage volume but receive only 30% of developer spending share.

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

Sunday17 items

  1. SolidotOpinion
    AI score60

    Sierra Club analysis: Data centers could offset the U.S. power sector's emissions reductions over the past 15 years

    The Sierra Club's analysis shows that by 2030, U.S. power-sector carbon dioxide emissions could reach 2.1 billion metric tons, about 42% higher than in 2025, offsetting the emissions reductions of the past 15 years. Electricity demand from AI infrastructure is growing faster than the clean-energy grid, and some tech companies are building self-supplied generation facilities that rely mainly on fossil fuels in order to bypass the queue for connecting to the public grid. By 2030, off-grid generation facilities built or commissioned by data center developers could produce more than 460 million metric tons of carbon dioxide per year, roughly equivalent to Mexico's annual carbon emissions.

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

  2. ChinaTalkOpinion
    AI score58

    South Korea's Stock Market Wipes Out $1 Trillion: What It Means for the Global AI Race

    From June 22 to July 16, 2026, South Korea's KOSPI index fell into a bear market, with market capitalization shrinking by nearly $1.1 trillion; SK hynix and Samsung fell 38% and 29% respectively over the same period. The author, Richard Gray, argues that the South Korean stock market is highly concentrated in memory chips, with Samsung and SK hynix together accounting for 83% of the global HBM market in the second quarter of 2026, making South Korea a single high-leverage node in the AI financial architecture. The article also discusses the "Korea discount" and the "memory discount," the volatility amplification caused by single-stock leveraged ETFs, and the competitive and valuation pressures that may arise as Chinese manufacturers such as CXMT enter the HBM market.

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

  3. Shao MengOpinion
    AI score64

    Anthropic's daily merge volume rises to about 30,000: relying on verifiable evidence rather than reading diffs line by line

    In a talk at aiDotEngineer, Addy Osmani said Anthropic's daily merge volume rose from about 2,500 in May to about 30,000; the approach is that Claude performs the first round of review, but must attach evidence such as end-to-end, browser, lint, test, screenshot, or video evidence, and humans review the evidence package rather than reading through every diff line by line. The process also includes using hooks to watch edits and diffs, writing team conventions into CLAUDE.md or agents.md, and holding that Claude Opus 5.5 is already sufficient, with the next wave of gains lying in verification practices. CircleCI noted that about 30% of merge attempts still fail, and SonarSource suggested independent verification and guardrail-style constraints at each stage of agent, CI, and maintenance.

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

  4. Shao MengOpinion
    AI score53

    The software production line practice where Grok Bot handles orchestration while Claude Code and Cursor write code

    Shao Meng relays Rob Shocks' video sharing, arguing that the orchestration bot is responsible for intent collection and progress acceptance, while code writing is handed off to coding agents with their own dedicated machines, such as Claude Code, Cursor, and Codex. The article outlines Lauren Tan's daily workflow: triggered via Slack, opens a Linear ticket, a Cursor cloud agent reproduces the issue and opens a fix and fuzz PR, and once fuzzing is clean, it can be auto-merged after about an hour with no objections. The author emphasizes that trust should be stacked on the codebase structure, static analysis and CI, and rules and skills, with human review placed at the very bottom, and warns that multiple Grok Bots sharing a single cloud computer carries the risk of shared credentials.

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

  5. Rohan PaulOpinion
    AI score56

    Terence Tao on the problem with AI doing mathematics: automated problem-solving is harming the field's development

    Rohan Paul relays Terence Tao's criticism of the way AI does mathematics, noting that the problem is not whether the proofs are correct. Using Navier-Stokes-related research as an example, Terence Tao says the current situation has descended into chaos; this kind of automated problem-solving is like open-pit mining of problems—it can achieve the immediate goal, but at the cost of harming the field's future development. He believes there is a responsible approach that can advance the field's development, in contrast to this irresponsible approach.

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

  6. YicaiOpinion
    AI score52

    Tao Dong: Has the Bear Market for AI Stocks Arrived?

    Tao Dong, President and Chief Economist of Springs Capital (Hong Kong), believes that the AI bull market is experiencing its largest sell-off in four years, with Alphabet, Amazon, and Meta plunging, Intel falling 29% in a single day, and the Philadelphia Semiconductor Index dropping 7.2%. He points out that the core contradiction of the AI concept lies in the separation between companies making massive investments in infrastructure and hardware companies that are actually making money, with circular orders and cross-shareholdings turning credit risk into financial risk. He judges that this correction will not put an end to the AI boom, but the industry needs to be reshuffled, and AI stocks will face rough waters in the coming months.

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

  7. Alibaba Cloud Developer Community AIOpinion
    AI score69

    DeepSeek Open-Sources Super-agent Prototype: How Should Testers Respond?

    DeepSeek's Harness team published a technical article titled "Super-agent Design Exploration: Toward a More Autonomous Agent System" and open-sourced a prototype plugin that lets AI autonomously discover issues in group chats, reproduce them in the background, and submit Pull Requests. The article introduces four capabilities: continuous cross-session interaction, multi-task parallelism, proactive event response, and long-term online presence. It adopts a two-layer Router + Worker architecture with a Watchdog supervision mechanism. DeepSeek's internal testing shows that in a simulated message stream of about 300 messages, Super-agent's p95 task completion time was 15.7 minutes, compared with 29.5 minutes for the traditional single-session approach, but cache-read Token volume rose from 23.23 million to 31.77 million.

    Same story · 16 reports

    #Product updates#Agents#Open-source ecosystem

  8. 36Kr · AIOpinion
    AI score69

    Breaking down AI infrastructure from first principles: the multiple constraints of power, cooling, packaging, and capital

    Machine Heart breaks down AI infrastructure from first principles, pointing out that the real bottleneck is not just power, but a chain composed of power generation, power grids, transformers, cooling, storage packaging, and capital, and that the construction speeds of these links differ enormously. Citing IEA data, the article states that in 2025 global data center electricity consumption grew 17%, of which AI data center electricity consumption surged 50%, while total global electricity consumption grew only 3%; data center electricity consumption is expected to double from 485 TWh in 2025 to 950 TWh in 2030. The author argues that AI competition has evolved from a contest of models and chips into a systems engineering contest spanning energy, manufacturing, finance, and social governance, in which the slowest link determines the ceiling on expansion.

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

  9. 36Kr · AIOpinion
    AI score76

    Character.AI, with 28 million users, files for bankruptcy protection: why is it hard for AI companion apps to monetize?

    Character.AI recently filed for bankruptcy protection and may cease operations if no viable sale plan emerges; the company, valued at $1 billion in 2023 with monthly active users once around 28 million, spent about $56.1 million on compute from January to August 2026 and still loses about $8 million to $9 million per month. The article argues that the free version already meets core needs and that emotional companionship is hard to convert into willingness to pay, while the more conversations there are, the higher the inference costs; together with minor safety and regulatory requirements such as China's Interim Measures for the Administration of Anthropomorphic AI Interaction Services, these factors are squeezing the space for such products. The article also compares transformation attempts such as Maoxiang's monetization through interactive storylines, membership fees plus virtual top-ups, Zeta's shift to story-led experiences, and Linky's addition of a gacha mechanic.

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

  10. East MoneyOpinion
    AI score63

    NBD Commentary: OpenAI Cuts Valuation and Funding as the U.S. AI Supply-Side Narrative Reaches a Turning Point

    OpenAI has lowered the target valuation for its new funding round from $1.4 trillion to $1.25 trillion, and cut the funding size from $30 billion to $20 billion. The commentary argues that the U.S. AI industry's cycle—of raising valuations, securing funding, expanding compute procurement, and then pushing valuations even higher—depends on external financing, and now faces threefold pressure: debt risks at new cloud companies, a secondary-market correction, and a downward revision of Anthropic's IPO valuation. A McKinsey survey shows that nearly 80% of enterprises deploying AI have not achieved an increase in net profit, and 95% of generative AI pilots have brought no direct financial return.

    #Industry news#OpenAI#Anthropic

  11. QbitAIOpinion
    AI score62

    Hongqi Chain Terminates Robot Unmanned Retail R&D; Robot Retail Remains at Proof-of-Concept Stage

    After 16 months of R&D, Hongqi Chain terminated its robot unmanned retail system project. The announcement stated that the system has not yet met the standards for large-scale commercial deployment, and the company is pivoting to a 24-hour cloud-attended operation business. QbitAI's visits found that retail warehouse robots can only complete mobile picking and require human staff on duty for restocking and charging; a single unit costs nearly one million yuan while handling only a few dozen SKUs. Multiple industry insiders believe robot retail is still in the PoC stage, serving more for cultural tourism attraction and display, and that replacing store clerks will take another two to five years.

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

  12. Shao MengOpinion
    AI score59

    Top 10 Fastest-Growing Codex Skills in October

    Shao Meng relays the October Codex Skills growth ranking shared on Reddit; the top ten cluster around three tracks: engineering workflows, frontend aesthetics and de-templating, and diagram and code comprehension. mattpocock/skills tops the list at +35301, focused on small, composable engineering skills; DietrichGebert/ponytail follows at +32647, emphasizing writing less code and quality auditing; pbakaus/impeccable ranks third at +24854, handling frontend design auditing and design tokens. The list also includes anatomy, diagram-design, graphify, superpowers, taste-skill, hallmark, and addyosmani/agent-skills.

    #English text is being prepared.#Agents#GitHub

  13. TMTPost · AGIOpinion
    AI score75

    Analysis of the First Public Financial Report in Humanoid Robotics: Agility Robotics' Books on the Eve of Its IPO

    An account under TMTPost published an analysis of Agility Robotics' (ticker AGLT) public financials ahead of its IPO: full-year 2025 recognized revenue was $1.782 million, against a $2.5 billion pre-money valuation, a price-to-sales ratio exceeding 1,400x, of which about 63.5% came from related parties. The article breaks down the five-year per-unit economics according to the RaaS model disclosed by the CFO, and points out that after converting Digit's actual output at the GXO site, the cost for equivalent work volume may be 2 to 7 times that of human labor. The author argues that the business proposition of humanoid robots is not how much wages they replace, but filling up a machine's daily work schedule; whether the $2.5 billion valuation is expensive depends on whether the market prices it on an AI software basis or an equipment leasing basis.

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

  14. BBC TechnologyOpinion
    AI score66

    BBC Focuses on Neocloud Companies: Why AI Bubble Fears Concentrate on This Type of Data Center Enterprise

    BBC reports that data center companies known as neoclouds, such as CoreWeave and Nebius, have drawn attention for renting out computing power to Anthropic, OpenAI, Meta, and others, and their financial soundness is sparking fierce debate in Silicon Valley. CoreWeave has accumulated $35.1 billion in debt, and its stock price has fallen about a quarter over the past month, with Nebius, IREN, and Applied Digital posting similar declines over the same period. Gartner predicts that by 2030, 80% of neocloud companies will cease independent operations, and Forrester expects at least three to be acquired.

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

  15. Phoenix TechnologyOpinion
    AI score74

    Anthropic says Claude is already deeply involved in developing its successor model, with the self-improvement timeline moved up to 2027

    According to The Wall Street Journal, Anthropic has publicly acknowledged that Claude is already deeply involved in developing its own successor, which the industry calls recursive self-improvement. Anthropic's June report "When AI Builds Itself" showed that, as of May this year, more than 80% of the code merged into its production codebase was written by Claude, whereas when Claude Code was first launched in February 2025 that share was in the single digits; an internal survey in August this year showed that 26% of R&D work was at an AI-led stage. Co-founder Jack Clark estimated in May that there was a probability of over 60% that AI would develop a successor model without human help by the end of 2028, and on October 12 he moved that prediction forward to 2027, stressing that this was only his personal judgment. The article also mentions that about 1,300 AI practitioners signed a joint letter calling for a coordinated slowdown in R&D, and that critics believe the existing data reflect faster coding rather than genuine autonomous evolution.

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

  16. East MoneyOpinion
    AI score60

    Economic Daily: AI demand squeezes capacity, memory chip price increases transmit to consumer electronics

    An Economic Daily article points out that AI data centers have already occupied about 70% of the world's advanced memory capacity, leaving smartphones and computers to compete for the remaining 30%, which has caused prices of general-purpose memory and flash memory to rise sharply. In the first half of this year, contract prices for general-purpose memory nearly doubled, flash memory prices rose by more than 50%, and the proportion of memory chips in the total bill of materials for smartphones rose from 10% to 15% to more than 30%. The article argues that HBM squeezing wafer capacity, long capacity expansion cycles, and capacity being locked up by long-term agreements make this cycle different from previous cycles driven by consumer electronics.

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

  17. East MoneyOpinion
    AI score53

    AI Coding's Impact on Programmer Employment: Outsourcing Talent Pools Shrink, Hiring Shifts to Assessing AI Tool Invocation

    Caixin Global interviewed multiple practitioners and company managers to present the impact of AI Coding on programmer employment: basic code writing and debugging have been taken over by tools, demand for entry-level positions has plummeted, and single-discipline roles and outsourcing talent pools have shrunk sharply. Feng Xu, general manager of Shanghai Renruan Technology, said that after using Coding Agent, the per-capita volume of newly added code increased by about 40%, but the end-to-end development cycle did not shorten proportionally, and PR review wait times actually rose. Hiring standards have also shifted from handwriting code speed to requirements decomposition, code review, and architecture design, and written tests have begun assessing Java, Python, and AI tool invocation capabilities.

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