All news

336Kr · AI08:54Opinion

AI score52

Chinese large-model vendors shift from competing on models to grabbing consumer hardware entry points

In 2026, Chinese large-model vendors began making forays into consumer hardware, with StepFun and Alibaba's Qwen releasing AI phones and glasses/earbuds products on the same day, October 13. The article argues that rapid model iteration and fierce price wars have led to low user retention; QuestMobile data shows that mainstream AI apps have a 30-day retention rate of only 12.8%, and hardware is seen as an entry point for locking in users. The article also mentions ByteDance's partnership with ZTE nubia to launch the Doubao phone assistant, StepFun's STEPX Neo Pioneer Edition priced at 3,999 yuan, and the losses of vendors such as Zhipu and MiniMax.

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

HHacker News Popular08:53Opinion

AI score55

The Current State of Context Engineering: From Long Context to Agent-Driven Harness

In a talk at AGNTCon, the author reviewed the evolution of context engineering, pointing out that long context brings four failure modes—context pollution, distraction, confusion, and conflict—and offered six fixes: context augmentation, tool loading, isolation, trimming, offloading, and summarization. As models shift toward autonomously driven harnesses, two new problems have emerged: "adversarial weights," where models overfit to their training environments, and "context inertia," where models become overly fixated on a single goal. Research from Anthropic, Microsoft Research, and UC Berkeley all addresses these by periodically clearing context while retaining environment artifacts. The author predicts that the next challenge will be memory accumulation and conflict in personal agents, which will require strategic forgetting.

#English text is being prepared.#Agents#Reasoning

HHacker News Trending08:53Opinion

AI score70

OpenAI and Anthropic to Cut LLM Prices by Up to 80% Within Months; Max Woolf Analyzes the Reasons

Max Woolf traces the timeline of OpenAI's and Anthropic's model releases and price adjustments from June to September 2026, noting that the prices of the two companies' flagship models have fallen by about 75% cumulatively, with GPT-5.6 Luna seeing a single price cut of 80%. He argues that models are becoming increasingly homogeneous and switching costs are extremely low, and that, combined with the number of providers hosting DeepSeek V4.1 Flash 1017 on OpenRouter rising to 29, this creates sustained price pressure. The author also mentions that the two companies are approaching IPOs, and says the inference business itself may still have relatively healthy profit margins.

#English text is being prepared.#OpenAI#Anthropic

TTomasz Tunguz's blog08:00Opinion

AI score61

Tomasz Tunguz on why the third product-market fit for AI is security

Tomasz Tunguz believes that, after coding and customer support, security will become the third product-market fit for AI. Citing a survey of 75 enterprise security practitioners, he breaks down security budgets excluding endpoints into five categories—security operations centers, SIEM, penetration testing, threat intelligence, and vulnerability management—and notes that each category has significant token consumption. The article also cites OpenAI President Greg Brockman's judgment that security spending should far exceed software spending.

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

RRohan Paul07:35Opinion

AI score66

Microsoft AI GPUs face power shortage; OpenAI pushes for 100 gigawatts of new US power generation per year

Satya Nadella previously said Microsoft does not have enough power to supply all its AI GPUs, and chips may sit idle because they cannot be connected to power. Citing content from the Bg2 Pod, Rohan Paul pointed out that the bottleneck is not chip supply, but the lack of "warm shells"—completed data center buildings—near large power sources, and that OpenAI is pushing the US government to add 100 gigawatts of power generation per year as an AI strategic asset. Power delivery and construction require multi-year approval and construction cycles, while GPU supply expands on a quarterly basis, so GPUs that are waiting depreciate and become stranded capital.

Quoted postRohan Paul · @rohanpaul_ai

Elon Musk: compute will sit stranded in half-built data centers. > Nscale’s IPO filing (Sept 2026): As of Aug 31, about 25,000 active GPUs against 461,000 active and contracted. Roughly 95% of the GPUs this neocloud has committed to are not yet running. > Wood Mackenzie, July 2026 report: Generator step-up transformer lead times exceeded 160 weeks in Q1 2026, up from 143 weeks in 2024, and high-voltage circuit breakers rose to 125 weeks versus 77 weeks in 2023. GPUs ship in months. The gear that energizes them takes two to three years, so any project that ordered chips before locking in electrical equipment is exposed. > Building your own gas power no longer lets AI builders skip the power queue, because turbine makers are booked years ahead. GE Vernova (Q2 2026): Its gas equipment backlog and slot reservations grew from 100 GW to 116 GW in one quarter. A turbine ordered today won’t arrive until around 2031. On-site power, the usual way around grid queues, has its own multi-year queue. arkolith

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

TTechCrunch AI01:34Opinion

AI score52

Apollo Chief Economist Says AI Has Not Delivered Financial Returns for Most Companies

Torsten Slok, chief economist at Apollo Global Management, said in a report that the profit margins and revenue per employee of the S&P 493 companies—excluding Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla—have remained essentially flat, with no evidence that AI has improved productivity. He said that despite massive investment and intense hype around AI, the technology has delivered zero financial return for U.S. companies outside the tech industry. Slok believes it is difficult for companies to convert increasingly cheap intelligence into higher revenue growth and profit margins, and said AI may be a very powerful calculator, but intelligence in business is more about judgment.

Same story · 11 reports

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

TTechSpot00:35Opinion

AI score71

Cloudflare CEO: AI will keep unearthing critical software vulnerabilities for years

Cloudflare CEO Matthew Prince told Reuters that the PHP vulnerability that forced government websites, GOG and other sites offline last week was only the beginning; the company already knows of several other severe vulnerabilities in other critical software, and AI tools will keep finding new ones for years to come. He believes large language models have greatly lowered the time and skill threshold needed to discover and exploit severe vulnerabilities, and expects a similar sense of crisis every week in the coming years. EFF says vulnerabilities found in October alone account for 3% of all historical records, and those found this year account for nearly a quarter of all known security issues.

Same story · 3 reports

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

RRohan Paul23:20Opinion

AI score62

Rohan Paul cites data: lengthening lead times for electrical equipment may leave compute stranded in half-finished data centers

Rohan Paul cites Musk's judgment that many partially built data centers will have large amounts of stranded compute, and uses multiple data sets to illustrate the mismatch between electrical equipment lead times and GPU arrival speeds. Nscale's IPO filing shows that as of August 31, about 25,000 GPUs were running, while the total active and contracted volume was 461,000, meaning about 95% were not yet running. A Wood Mackenzie report from July 2026 states that in the first quarter of 2026, lead times for generator step-up transformers exceeded 160 weeks, compared with 143 weeks in 2024, and high-voltage circuit breakers rose to 125 weeks. In the second quarter of 2026, GE Vernova's gas equipment backlog and slot reservations increased from 100 GW to 116 GW, and turbines ordered now will not be delivered until about 2031.

Quoted postElon Musk · @elonmusk

@Yuchenj_UW I think we’re going to see a lot of stranded compute in partially complete data centers

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

AAlibaba Cloud Developer Community AI20:36Opinion

AI score51

FDE May Be the Shortest-Lived Hot Job of the AI Era: An Experiment in Conflicting Interests in Enterprise Delivery

The author argues that FDE (Forward Deployed Engineer) is more likely a transitional business strategy for AI vendors competing for enterprise customers, rather than a long-term standard model for enterprise AI delivery. The article points out that FDE relies on original vendor certification and technical support, lacks the incentive to recommend competitors' SOTA solutions, and may not have cross-platform evaluation capabilities, thus creating a delivery paradox: if delivery succeeds, the long-term necessity of FDE should decline; if it remains irreplaceable, then whether delivery is successful needs to be re-examined. The article cites a Gartner forecast that by 2028, 70% of enterprises will abandon Agentic AI built by vendor FDEs, and offers five questions for enterprises to judge whether delivery is successful.

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

TThe Verge AI20:33Opinion

AI score62

Google AI search overhaul hits the online publishing industry

The Verge reports that after Google launched AI Overviews in 2024 and AI Mode in May 2025, online publishers' search traffic has declined significantly. A June 2025 WSJ report, citing Similarweb data, said that within three years the share of The New York Times' traffic coming from search fell from 44% to 35.5%; Digiday reported that 19 DCN members saw Google traffic drop by 1% to 25% between May and June 2025; DMG Media told the UK CMA that click-through rates fell by as much as 89% from June to July 2025. Google search head Liz Reid said click volume is relatively stable, and Senior Vice President Nick Fox said the web is still thriving; the company has made sources more prominent and said 400 publishers have joined the News AI pilot.

#English text is being prepared.#Search#Google

IInterconnects20:32Opinion

AI score57

Podcast: Kimi K3.1 Pro, the new cybersecurity reality, and the Muse user experience

In episode 20 of the Interconnects podcast, Nathan Lambert and Florian Brand discussed how Kimi K3.1 Pro ranks high on Artificial Analysis but receives mixed reviews, arguing that its training leans more toward exploitation than exploration, with weaker performance on long-horizon benchmarks such as ProgramBench. The two also discussed the cybersecurity controversy surrounding open-weight models, noting that no large-scale exploitation occurred after the disclosure of the PHP2Shell vulnerability, and talked about Anthropic not hiring junior employees as well as Florian's experience using Muse to handle everyday tasks such as shopping.

#English text is being prepared.#Open-source ecosystem#Safety / alignment

RRohan Paul19:45Opinion

AI score58

Bill Gates says AI is the most dangerous technology to date

In an interview on The Verge's Decoder podcast, Bill Gates said humanity has crossed a certain capability threshold, that AI is the most dangerous technology to date, and that it will have broad impacts on the economy and the way wars are fought. This clip was cited by Rohan Paul on X, and the full video was published on the Decoder with Nilay Patel YouTube channel.

Quoted postRohan Paul · @rohanpaul_ai

Bill Gates wants powerful open-weight models to run only where the government says they can. For any model capable of sophisticated cyber attacks, he wants safeguards and monitoring kept in by law. Developers could still use these models for free and customize them, but only on certified platforms that detect and refuse any attempt to strip out the safeguards. ---- Full video on "Decoder with Nilay Patel (@reckless) and The Verge" YouTube channel, (link in comment)

Same story · 2 reports

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

AAYi AI Notes18:43Opinion

AI score52

Musk on the OpenAI lawsuit: converting from nonprofit to for-profit sets a dangerous precedent

In his latest Forbes interview, Musk publicly responded directly for the first time to the lawsuit with OpenAI, saying that OpenAI's shift from a nonprofit to a fully for-profit entity is setting an extremely dangerous precedent. He mentioned that when the parties reached a compromise back then, they agreed that once AGI was achieved, equity and control must be returned to the nonprofit organization, but this return clause was unilaterally deleted; he believes that if this matter can be gotten away with, existing charitable donation contracts will lose their foundation. The interview video is about two minutes long.

Quoted postAYi · @AYi_AInotes

English text is being prepared.

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

TThe Decoder18:31Opinion

AI score61

Former Google CEO Eric Schmidt Says OpenAI and Anthropic Are Like Modern Colonial Trading Companies

Historian Niall Ferguson, researcher Lisa Klaassen, and former Google CEO Eric Schmidt wrote in Foreign Affairs, comparing OpenAI and Anthropic to colonial-era East India Companies, arguing that they negotiate with governments but remain constrained by them. The article cites the example of the U.S. Commerce Department in June requiring Anthropic to cut off foreign nationals' access to its two most powerful models; because Anthropic could not verify users' nationalities in real time, it took Fable 5 and Mythos 5 offline worldwide. About two weeks later, export controls on Fable 5 were lifted, while Mythos 5 remains available only to certain U.S. institutions. The authors also say U.S. tech companies plan to invest more than $600 billion in 2026, much of it debt-financed, while China plans to invest $295 billion over five years, and they propose an "imperial trilemma" in which defense, technological leadership, and the welfare state are difficult to achieve at the same time.

#Industry news#OpenAI#Anthropic

TThe Decoder18:31Opinion

AI score68

Anthropic says a single agent often outperforms multi-sub-agent architectures, and releases a migration toolkit

Anthropic recommends that enterprises evaluate whether to consolidate sub-agents, because a single agent that loads context on demand often outperforms multiple specialized sub-agents in quality, cost, and latency. Starting with Opus 4.5, as well as newer models such as Opus 5.5 and Fable 5.1, complex plans can be managed directly within a single loop; in one comparison, the single-agent architecture ran at roughly twice the speed of the multi-agent approach. Anthropic released a migration toolkit on GitHub and cautions against consolidating blindly, because in one customer-service agent test, merging in sub-agents reduced latency but broke the existing workflow.

Same story · 5 reports

#English text is being prepared.#Agents#Anthropic

IEnglish text is being prepared.18:30Opinion

AI score64

Cybersecurity experts: Vulnerabilities discovered by AI may give rise to new security risks

Seven cybersecurity experts interviewed by Reuters believe that as AI continues to uncover hidden security vulnerabilities in software, incidents like this week's successive service disruptions at government websites in many parts of the world are almost certain to happen again. Research soon to be released by VulnCheck shows that AI has driven growth in the number of vulnerabilities discovered, with the number found in the past year doubling year over year. Cloudflare CEO Matthew Prince said that for the next two years it will feel every week as if the sky is falling, but in the long run AI discovering vulnerabilities still does more good than harm.

#Safety / alignment#Industry news#English text is being prepared.

SSemiconductor Industry Observer17:25Opinion

AI score63

Behind AI compute centers, a competition in the liquid cooling supply chain worth hundreds of billions

The article outlines the trend of AI compute center cooling shifting from air cooling to liquid cooling. Nvidia's Vera Rubin racks have eliminated fans, exceed 190 kilowatts of power per rack, and adopt 100% full liquid cooling, while air cooling can generally only handle about 20 to 30 kilowatts of heat dissipation per rack. The article points out that cold plates, CDUs, and quick connectors together account for about 85% of the cost of a liquid cooling system, that the high-end market has long been controlled by Taiwanese and European/American manufacturers, and it summarizes three ways Chinese manufacturers are entering: cross-sector restructuring, single-point efficiency, and trading time for market.

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

RRohan Paul16:38Opinion

AI score64

Bill Gates says AI gives small groups nation-state-level attack capabilities

In an interview on the Decoder podcast, Bill Gates said that AI transfers large-scale destructive capabilities once held by nation-states to independent groups—in the cyber domain, profit-seekers, and in the biological domain, people who want to cause harm. He also noted that there is currently no mandatory review of models, nor is there progress on building safeguards. The full video of the interview was published on the Decoder with Nilay Patel and The Verge YouTube channel.

Quoted postRohan Paul · @rohanpaul_ai

Bill Gates wants powerful open-weight models to run only where the government says they can. For any model capable of sophisticated cyber attacks, he wants safeguards and monitoring kept in by law. Developers could still use these models for free and customize them, but only on certified platforms that detect and refuse any attempt to strip out the safeguards. ---- Full video on "Decoder with Nilay Patel (@reckless) and The Verge" YouTube channel, (link in comment)

Same story · 2 reports

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

RRohan Paul16:07Opinion

AI score62

Bill Gates argues that capable open-weight models should only run on government-approved platforms

Bill Gates argues that for open-weight models with advanced cyberattack capabilities, safety protections and monitoring should be legally mandated, and developers can still use and customize them for free, but only on certified platforms. These platforms must detect and reject any attempt to strip away safety protections. These views come from The Verge's Decoder podcast interview video with Nilay Patel.

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

WWallstreetcn15:27Opinion

AI score57

SemiAnalysis Founder Analyzes AI Investment, Nvidia's Moat and the Memory Shortage

At CASPA's 35th Annual Conference, SemiAnalysis founder Dylan Patel and Bernstein analyst Stacy Rasgon discussed the outlook for AI infrastructure investment, arguing that with the U.S. 10-year Treasury yield breaking above 5% and unsecured bonds from emerging cloud service providers yielding close to 15%, financing capacity is becoming a key constraint on expansion. The two judged that substrates and memory are bottlenecks harder to scale than power, that HBM requires roughly 3 to 4 times as many wafers as DDR5, and that if new capacity is concentrated in releases after 2028, memory prices and profit margins could come under pressure. Nvidia still supports its pricing through performance gains, but Google TPU, Amazon Trainium, and the custom chips of OpenAI and Anthropic are eroding its share and bargaining power.

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