Search “Video”

Amazon CEO Jassy on $220 Billion in Capital Expenditure: Compute Is Still in Short Supply, Inference Costs Need to Fall by Another Order of Magnitude

Amazon CEO Andy Jassy said in an interview on the a16z podcast that the company's capital expenditure is expected to reach $220 billion in 2026, but compute capacity is already sold out for 2026, essentially fully booked for 2027, and a large share of 2028 quotas has been locked in. He described enterprise AI demand as dumbbell-shaped: a large number of production systems in the middle have yet to be deeply integrated because inference costs are too high, and costs need to fall by another order of magnitude before an explosion in adoption occurs. Jassy also noted that 99% of AWS's top 1,000 customers use Graviton, which offers about 40% better price-performance than x86, while Trainium has a 30% to 40% advantage over comparable hardware.

Claude engineering lead Felix on how to use Claude: fewer role-based Agents, use adversarial review to constrain output

In an interview, Claude engineering lead Felix Rieseberg said he barely writes code anymore; his core approach is to hand crash data, business tools, and a computer to the model and let it decide what to fix itself, rather than having a human act as a relay repeating things item by item. He suggests opening multiple threads on the same output dedicated to proving it wrong, then opening another as a judge to filter which criticisms hold up; rather than giving each Agent a tiny slice of data, it is better to give a panoramic business view and well-documented tools. He also mentioned that the half-life of Skills is extremely short, that both official guidance and his own advice are to use them cautiously, and that self-evaluation within the same session remains unreliable, so external signals should be sought as much as possible.

Why AI Audit Tools Are Supplied by the Audited Party: Verification of Four Security Incidents and Seven Papers in One Week

Using four security incidents within the week of October 16 to 23 as its thread—including two PHP remote code execution vulnerabilities discovered by AI, the backdoored LiteLLM version v1.107.0 surviving on PyPI for about two hours and being downloaded roughly 49,000 times, the OpenCode console leaking third-party model keys for 1.4% of users, and Microsoft reporting that a human organization used the open-source multi-agent penetration framework ARTEX to attack the supply chain—the article discusses the common problems facing AI auditing. After examining seven papers one by one, the author points out that the faithfulness of latent reasoning self-decoding reaches at most 0.4788, sparse autoencoders are nearly blind to backdoor shifts, probes fail when targeted by the object being optimized, and all three mainstream audit tools originate from the audited party's ecosystem or require borrowing the audited party's permissions. The article also cites the NARCBench multi-agent collusion benchmark from Oxford and New York University, as well as the report by Anthropic and Theorem using Claude to write roughly one billion lines of Rocq formal proofs for 239 entry points across 123 open-source libraries.

Baidu Dazi reaches 9.77 million MAU as the AI office flanking battle faces a phased major test

The QuestMobile Q3 2026 report shows that Baidu Dazi reached a combined 9.77 million MAU and 83.81 million total multi-device usage sessions, up 5,810% from June, ranking second on the AI office Agent list, behind only Tencent WorkBuddy's 32.353 million. The author traces its three-stage leap over half a year: in March it entered as a desktop agent, in August it launched 96 professional skills and 15 industry suites, and in September it launched an enterprise edition and integrated with Xiaodu hardware. The article argues that Baidu avoids head-on competition at the IM entrance and bets on the delivery standard DuMateBench, the data context of search and cloud storage, more than 30 expert suites, and the Xinchuang channel, but whether retention, renewals, and the rankings can become a general benchmark remains an open question.

Exiting after just 188 days, big-company AI products are accelerating their reshuffle

36Kr sorted through the product moves of eight companies—Tencent, Alibaba, Baidu, ByteDance, and Moonshot AI, Zhipu, MiniMax, DeepSeek—across ten AI tracks over four years, finding that competition cycles have shortened, product lifecycles have grown shorter, and renaming, integration, and shutdowns have become the norm. Tencent QClaw went from public beta on March 20 to announcing service shutdown arrangements on September 24, just 188 days, with user data to be migrated to WorkBuddy; Alibaba's Qwen announced the upcoming release of its personal assistant product Qwen Go!. Alibaba and ByteDance cover all ten tracks, Tencent and Baidu each cover nine, and DeepSeek covers only two.

Gary Marcus: US Banks Are Under Attack by AI Agents

Gary Marcus, citing an Axios report, said that US banks are tightening system access permissions due to AI agent attacks, and that bank executives are alarmed by these findings. He also mentioned that the attack on France's refrigeration system on October 11 may have been facilitated by agents, as well as the continuously increasing cybersecurity attacks in Japan. He believes that attackers do not need the most advanced frontier models to cause damage, and therefore regulation should not focus only on frontier models and future risks.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

Ethan Mollick reviews 30 months of AI development in charts

On the release day of his new book Co-Existence, Ethan Mollick used charts to review the AI progress in the 30 months since his previous book Co-Intelligence was published. He said that since GPT-4, 20 models have successively taken the top spot, and 180 models were stronger than GPT-4 at release, 19 of which can run on modern smartphones; the METR long-task benchmark showed that the amount of human work AI can complete increased from 4 minutes in 2024 to 17 hours in April 2026, and the benchmark was subsequently discontinued because it could no longer measure accurately. He proposed a third mode of collaboration beyond centaur and cyborg: conjurer, that is, summoning agents, assigning tasks, and evaluating results. The charts in this article were completed by Opus 5.5 in Claude Projects orchestrating multiple agents.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.

Morgan Stanley says Nvidia will depend on AI infrastructure spending for 91.7% of its 2027 revenue

A report released by Morgan Stanley on October 15 estimates that 91.7% of Nvidia's 2027 revenue will come from AI infrastructure demand, the highest share among the companies it covers, above Vertiv's 90% and Broadcom's 81%. The report projects that hyperscale cloud providers and AI cloud service providers will spend $1.58 trillion in 2027, with nearly $700 billion flowing to Nvidia, while SpaceX and AWS together account for about 54% of Nvidia's hyperscale customer revenue. The report also lists risks: 2027 will see the largest cash flow gap in the forecast period, U.S. data centers will face a 32-gigawatt power shortfall by 2028, and data center restrictions have also become a midterm election issue.

English translation is being prepared.

Qin Shuo: All 243 2B software companies on the A-share market are losing money industry-wide—how much longer will China's software industry keep losing money?

Citing data, Qin Shuo says that the revenue of 243 2B software companies on the A-share market grew only 7.9% from 2021 to 2025, while non-recurring net profit swung from a profit of 17.89 billion yuan to a loss of 17.68 billion yuan, with 64.7% of companies operating at a loss. The article attributes the causes to a cultural habit of valuing hardware over software, a low-margin model dominated by project-based work, lowest-price bidding (a project in one province with a budget of 7.672 million yuan was awarded for 10 yuan), and repeated construction by the digital technology subsidiaries of central and state-owned enterprises squeezing out market-oriented vendors. The author proposes breaking the deadlock in four areas: comprehensive bid evaluation, productization plus subscription models, fair market access, and cost reduction through AI agents, and gives the example of an enterprise that used agents to shorten custom delivery cycles to about one week and cut costs to 1/5 to 1/10 of the original.

English translation is being prepared.

English translation is being prepared.

English translation is being prepared.