All news

PPhoenix Tech17:49

AI score48

Goldman Sachs Tightens Access to AI Coding Tools Such as Claude Code, Some Traders Unhappy

In internal guidance updated in October, Goldman Sachs tightened access to Claude Code and Microsoft GitHub-related tools, mainly restricting them to use by engineers, which has drawn complaints from some traders. Previously, some traders relied on these tools to monitor client trading flows and simulate market and position changes; "strats" employees whose work spans trading, quantitative research and software engineering are not subject to this restriction. Goldman Sachs' own AI assistant, GS AI Assistant, integrates models from OpenAI, Anthropic and Alphabet.

Same story · 3 reports

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

PPhoenix Technology17:49

AI score62

Oracle Raises $20 Billion for AI Expansion, Shares Fall Nearly 9%

Oracle announced that it would raise approximately $20 billion through bond issuance and stock sales to support AI and cloud computing infrastructure construction. After the news was announced, the stock price fell 8.8% to $119.22, and the cumulative decline for the year widened to about 39%. The offering consisted of 86.2 million shares priced at $116 per share, a discount of about 11.2% to last Friday's closing price; the bond portion consists of five tranches of fixed-rate notes with maturities ranging from 2029 to 2046 and interest rates between 7% and 8.65%. As of August 31, Oracle's notes payable and other borrowings reached $125.34 billion, and the company expects revenue to exceed $225 billion by fiscal year 2030.

Same story · 2 reports

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

PPhoenix Tech17:49Research

AI score73

MIT CSAIL uses GPT-6 Astra to directly rewrite the weights of Qwen3.5-0.8B

The MIT CSAIL team had OpenAI's GPT-6 Astra directly read and modify the weights of Qwen3.5-0.8B. By changing only about 0.016% of the parameters, they raised the model's score on MATH-500 from 7.9% to 48.1% and improved IFEval by about 3x, all without gradient descent or backpropagation. The team calls this method Weight-Space Programming (WSP) and points out that it removes obstacles in the process of the model exercising capabilities it already has, rather than creating new capabilities; applying the same method to Gemma 3 270M only pushed MATH-500 from 6.8% to 7.7%. The experiment has so far been released only as a blog post and has not yet produced a peer-reviewed paper.

Same story · 14 reports

#Papers/Research#Reasoning#Open-source ecosystem

AAnt inclusionAI: New GitHub Repository17:49First-party

AI score46

Ant inclusionAI Open-Sources dInfer: An Efficient Inference Framework for Diffusion Language Models

Ant inclusionAI has open-sourced dInfer, an efficient inference framework for diffusion language models (dLLMs) that splits inference into four modules—the model, the diffusion iteration manager, the decoder, and the KV-cache manager—and supports batched inference. It supports the LLaDA, LLaDA-MoE, and LLaDA2 series; on a single 8×H800 node, LLaDA-MoE exceeds 1,100 TPS on HumanEval at batch size 1 and averages 800+ TPS across six benchmarks, 10× faster than Fast-dLLM while maintaining accuracy.

#English text is being prepared.#Reasoning#Open-source ecosystem

SShanghai Observer17:42

AI score37

Li Jiaqi's livestream sells Tokens: 100,000 Qwen Office monthly cards sold in half an hour

On November 2, during Tmall's "Double 11" shopping festival, Li Jiaqi's livestream sold AI computing power products for the first time, selling 100,000 Qwen Office individual membership monthly cards in half an hour, totaling about 2 trillion Tokens, at a unit price of just 1 cent. A report by the China Academy of Information and Communications Technology shows that in June this year, domestic daily average token consumption was about 500 trillion, nearly four times that of December last year, and the normalized average price rose from 6.8 yuan to 13.46 yuan. The article argues that what the livestream is selling is low-cost customer acquisition and the psychological comfort of "already being on board," rather than scarce computing power.

Same story · 10 reports

#English text is being prepared.#Industry news

SShanghai Observer17:42

AI score22

China Eastern Airlines APP's "Yanbao Intelligent Agent" gets a refreshed upgrade, with digital and intelligent services covering the entire travel process

The intelligent travel assistant "Yanbao Intelligent Agent" on the China Eastern Airlines APP has completed a refreshed upgrade. Travelers can handle the full chain of needs—including itinerary planning, flight booking, airport waiting services, in-journey inquiries, and benefits usage—in a one-stop manner within the same conversational interface. After the upgrade, hotel and attraction ticket booking have been added, allowing free combination of flight tickets, accommodation, and sightseeing products. Services such as VIP lounges, expedited security screening, and airport parking can also be booked directly through intelligent conversation. From now until December 31, booking domestic all-route flight tickets through the "Yanbao Intelligent Agent" offers a direct discount of up to 66 yuan.

#Product updates#Agents

SShanghai Observer17:42

AI score42

Li Jiaqi's livestream sells about 2 trillion Tokens in half an hour; Qwen Office membership card becomes a Double 11 hit

On the evening of November 2, Li Jiaqi's livestream began selling the Qwen Office personal membership monthly card, selling 100,000 units in half an hour, totaling about 2 trillion Tokens, marking the first time an e-commerce livestream has sold AI computing power products. This Double 11, Tmall for the first time made AI Tokens an independent category, and manufacturers such as Zhipu, Kimi, MiniMax, and DeepSeek have already joined the Token recharge center, allowing consumers to purchase monthly card packages just like topping up phone credit. As of July this year, China's average daily Token calls exceeded 500 trillion, an increase of more than 5,000 times compared with 100 billion at the beginning of 2024.

Same story · 10 reports

#English text is being prepared.#Industry news

4404 Media17:42

AI score53

AI scrapes AO3 fanworks to build a site for sale, authors send 2,000 takedown requests in one day

A website called ao3.life, which used AI agents to scrape AO3 fanworks and convert them into paid audiobooks, shut down one day after launching; the creator said it received more than 2,000 takedown requests from authors that day. The site claimed it used a group of AI agents controlling 40 virtual machines to scrape works from the internet, offered the first 3 minutes free, then charged for AI-generated audiobooks, and claimed authors could receive 91% of the revenue. The creator described himself as a 17-year-old developer from India and said the site had 126,592 unique users within 24 hours, but actually sold only about 11 audiobooks.

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

Tt3n17:37

AI score67

OpenAI claims AI solved nearly 400 math problems; mathematicians criticize it for prioritizing results over process

OpenAI claims that with the help of an as-yet-unreleased AI model, it published more than 700 manuscripts in October, covering more than 370 math problems, with each taking an average of three hours, including work related to the Riemann Hypothesis. Institutions such as the German Mathematicians Association criticized this approach as a results-oriented "power display," arguing that the value of mathematical research lies in the proof path and new methods rather than the answers themselves, and expressing concern about the use of unverified knowledge and issues of academic norms. Stuttgart professor Christian Hesse, on the one hand, called OpenAI's demonstration "stunning," and on the other hand, believes that doctoral training and the definition of mathematical understanding need to be readjusted.

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

Tt3n17:37

AI score60

Anthropic's internal AI usage spending exceeds labor costs for the first time

Anthropic has published data for the third quarter of 2026, stating that since January 2026 its internal AI usage has grown 250-fold, and that at list prices, internal AI usage spending has exceeded all labor costs—though this comparison excludes stock-based compensation. The cost is calculated at external list prices rather than the actual cost of self-used infrastructure, and Anthropic has not separately disclosed the total spending on internal AI usage. Anthropic also says that in September 2026, 90% of contributions to its codebase were completed by AI, per-developer code output grew 45-fold compared with 2024, and headcount over the same period grew more than fivefold compared with the end of 2024.

Same story · 5 reports

#Industry news#Anthropic

336Kr · AI17:29

AI score20

Anthropic Fable 5.5 rumor: 14 trillion parameters, animation generated purely from code, to replace video and image models within 6 months?

A rumored Anthropic model, Fable 5.5, has had real-world tests leak on X: by autonomously writing Python code to generate a single HTML file, it played an animated short in the browser, and some users say it autonomously completed a 66-second short film in 2 hours and 40 minutes. Rumors say it belongs to the Mythos tier, has a parameter scale of 14 trillion, and uses a hybrid architecture combining GDN with MLA, but the weight leak on Hugging Face has not yet been independently verified. The article also mentions that on October 13, LiteLLM briefly showed a claude-fable-5-5 entry in the Azure catalog, which was withdrawn the next day, and that the words Claude Mythos 5.5 appeared in the source code of Anthropic's official website.

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

336Kr · AI17:29

AI score22

SheNicest Women's Hackathon: Projects such as the toilet-finding app TOILER and the cognitive observation tool Jianwei showcased

At the "SheNicest Fission" hackathon, where women made up 70% of participants, teams presented products including the toilet-finding app TOILER, "Jianwei," a home cognitive health observation tool for adult children, and "Hemian," a smart mattress for late pregnancy. TOILER can filter toilets by conditions such as stall doors, hooks, and hand dryers; it currently includes only some toilets in areas such as Phase II of the China National Convention Center, 798, Guomao, and Sanlitun, and does not yet update in real time. Most projects are still in the prototype stage; for example, the "Taxian" input method cannot display the corresponding Chinese characters when more than four characters are entered.

#English text is being prepared.#Agents#Speech

336Kr · AI17:29Research

AI score63

MIT CSAIL proposes weight-space programming: GPT-6 Astra directly rewrites neural network weights

The MIT CSAIL team proposes a weight-space programming method that lets GPT-6 Astra directly read and modify neural network weights without training or computing gradients. On Qwen3.5-0.8B, Astra modifies only about 0.016% of the weights, raising MATH-500 from 7.9% to 48.1%, GSM8K from 13.8% to 67.1%, and IFEval from 21.0% to 62.3%, mainly because it fixed the model's failure of getting stuck in reasoning loops. The research was led by doctoral students Rikiya Takehi and Jayden Teoh, with participation from Omar Khattab and others, and the results were published on the research blog "Sparks of Weight-Space Programming"; additional experiments on Gemma 3 270M yielded improvements of no more than about 1 percentage point, and the team says more independent replications are still needed.

Same story · 14 reports

#Papers/Research#Agents#Reasoning

336Kr · AI17:29Opinion

AI score63

AI data companies are fierce: UniPat reportedly backed by Alibaba and Tencent, valued at $2.5 billion

According to a report by 36Kr's Silicon Star Pro, UniPat, a data company founded less than a year ago, is reportedly in talks to raise about $300 million, led by Alibaba with participation from Tencent and Sequoia China, at a post-money valuation of $2.5 billion. The article says a single piece of data with user behavior trajectories can be quoted at 1,000 to 3,000 yuan, while work-scenario data with a complete operation process can sell for around 10,000 yuan, and Mercor estimates that only 2% to 5% of the data frontier labs need has been completed so far. As model vendors tighten procurement, calculate ROI and run ablations on vendor data, the industry has already seen problems such as blacklisting data vendors and data quality being hard to verify.

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

KKling AI17:21First-party

AI score38

Kling 4.0 Hands-On Test: Eight Creators Discuss Dynamic Motions and Native 30-Second Generation

Kling 4.0 invited eight creators to share their hands-on creative experiences. Key capabilities include the physical performance of Dynamic Motions in highly difficult action sequences, more stable camera scheduling, and Native 30s Generation, which supports native 30-second generation. Omni Reference can fuse multiple visual references, and character consistency and lip-sync have both improved in long shots and Cantonese dialogue scenes.

#Industry news#Video#Multimodal

JJiwei.com17:21

AI score18

The China Electronics Standardization Association held a health measurement standards seminar in Shenzhen, discussing two group standards: "Bioelectrical Impedance Chip" and "Four-Electrode Body Fat Scale"

The Home Health Auxiliary Management Device Branch of the China Electronics Standardization Association held a "Chip Health · Smart Future" health measurement standards seminar in Shenzhen, with nearly 40 representatives from industry, academia, research and application attending. The seminar discussed the draft group standards for "Bioelectrical Impedance Chip" and "Four-Electrode Body Fat Scale," and solicited standardization needs for health electronic devices from the industry. The drafting groups for the two standards, represented by Chipsea Technology and Kangyou Health respectively, presented their reports. The former specifies technical indicators such as four-electrode measurement, the dynamic range and measurement error of impedance measurement, and common-mode rejection ratio; the latter standardizes metrological performance, impedance measurement performance and body fat percentage measurement performance, and classifies products into single-frequency, dual-frequency and multi-frequency according to measurement frequency.

#Industry news#English text is being prepared.

JJiwei.com17:21

AI score45

CXL Specification Advances to 4.0, QLC Orders Already Booked Through 2028: The Two Storage Roadmaps Diverge in Pace

The CXL Consortium has released the 4.0 specification, doubling link speeds from 64 GT/s to 128 GT/s, but Intel's Diamond Rapids platform supporting CXL 3.x has been delayed to a 2027 launch, and Samsung, SK hynix, and Micron have successively halted or scaled back their in-house CXL controllers. On the QLC side, Micron's 245TB 6600 ION has officially shipped, with per-drive capacity roughly quadrupling from 61.44TB over three years, and Seagate's nearline drive shipment allocations are already booked through 2028.

#Industry news#English text is being prepared.

JJiwei.com17:21

AI score13

Wanrun Technology Q3 Report: Non-GAAP Net Profit Up 214.72%, Storage Business Rides AI Boom

Wanrun Technology's revenue for the first three quarters of 2026 was 3.383 billion yuan, down 8.90% year-on-year, while non-GAAP net profit was 27.1435 million yuan, up 214.72% year-on-year, marking a clear recovery in the profitability quality of its core business. Revenue from the semiconductor memory business in the first half of the year was 490 million yuan, up 18.06% year-on-year, with gross margin rising to 9.81%; the company's inventory at the end of the period increased to 365 million yuan for stockpiling, and it completed a 130 million yuan capital increase in Wanrun Semiconductor with iSoftStone. AI servers' demand for DRAM is 8 to 10 times that of ordinary servers, and SEMI China expects the global HBM market to grow 58% to 54.6 billion US dollars in 2026.

#English text is being prepared.

SSenseTime17:16

AI score48

SenseTime Releases SenseNova 6.8 Flash, Focusing on Office Intelligence and Dynamic Design

SenseTime Releases SenseNova 6.8 Flash, Focusing on Reliable Workflows and Design Output: Office intelligence supports long-term memory and proactive interaction, while dynamic design can generate animated PPTs, dynamic posters, interactive websites, and editable Blender 3D projects. The model is available for free on the SenseNova Token Plan, offering 1,500 free API requests every 5 hours, with the model ID SenseNova 6.8 Flash.

Same story · 5 reports

#Model releases#Agents#Multimodal

CCailian Press17:10

AI score42

Goldman Sachs CEO Solomon: Compute Demand Far Exceeds Supply, AI Financing Will Remain Strong Over the Next 12 to 24 Months

Goldman Sachs Chairman and CEO David Solomon said that compute demand far exceeds supply, and that strong financing activity over the next 12 to 24 months will meet part of that demand, with investor demand for subscriptions to large AI financing deals being "very strong." He believes that AI infrastructure construction is still at an early stage, but rising financing costs will change the returns on some projects; Goldman Sachs has secured the lead underwriting role for Anthropic's IPO and is co-leading Meta's $95 billion equity financing plan.

#Industry news#English text is being prepared.