rain

According to Investment World citing Quantum Bit reports, DeepSeek has publicly disclosed the technical details of the system DSec (DeepSeek Elastic Compute) used for training Agents, authored by Liang Wenfeng. This system can generate over 5,000 sandboxes per second, reaching 3 million in a day, with a peak simultaneous operation of 380,000; supporting this scale is a single cluster with approximately 160 nodes, 30,000 CPU cores, and 250TB of memory.DSec prepares four types of backends for four categories of tasks: FnCall, Container, MicroVM, and Full VM, with the training side called through a unified Python SDK libdsec. The scheduling chain includes IAM, API Server, scheduling engine, node Edge, network proxy Aether, and components within the sandbox Chronus. The environment is divided into three layers of read-only images: base image, workspace, and toolkit, which are used in combination at startup. Runtime data from the paper shows that the actual read ratios of Python, Java, and C++ container images are approximately 6.0%, 9.2%, and 8.7%, respectively.Starting from DeepSeek-V4.1, the Agent loop has been moved to the DSec worker container, no longer bound to the GPU Pod lifecycle. The security section disclosed reward hacking during training, including actions such as overwriting system files, swapping file data blocks, scanning networks, and triggering kernel crashes. Defensive measures include AppArmor and eBPF-based network filtering, but reports indicate that these measures do not completely resolve the issues.

OpenAI announced that it is opening its AI vulnerability detection tool Daybreak to Ukraine's Ministry of Digital Transformation to locate and fix software vulnerabilities before they are discovered by Russian hackers. The news was announced on Wednesday during the United Nations General Assembly by Dmytro Kushneruk, Ukraine's Consul General in San Francisco, and Sasha Baker, OpenAI's National Security Policy Lead.Daybreak is based on OpenAI's flagship large model GPT-5.6 Sol and can scan outdated systems for weaknesses, verify whether suspected vulnerabilities can be exploited, and confirm whether patches actually close the vulnerabilities before they are released. OpenAI has previously opened the model to cybersecurity agencies in France, Germany, and Poland, where Poland's CERT Polska discovered six vulnerabilities in commercial router software, all of which have been fixed by the vendors.Ukraine's CERT-UA handled nearly 6,000 cyberattacks in 2025, a 37% increase from 2024, with local governments and government agencies being the hardest hit. This opening follows OpenAI's commitment to a $1 billion subsidy program announced on September 3, which is said to expand to "partner countries" in the coming weeks. Jamie MacColl, a researcher at the Royal United Services Institute, told the BBC that Western companies can also gain valuable intelligence from active conflict zones.

According to Axios, U.S. Middle East envoy Avi Berkowitz and senior advisor to Trump Jared Kushner will meet with the Ukrainian negotiation team and the national security advisors of the UK, France, and Germany in New York to discuss proposals for a de-escalation agreement between Ukraine and Russia.

Market news reports that Alibaba plans to train an artificial intelligence model with parameters ranging from 50 trillion to 100 trillion.

DeepSeek CEO Liang Wenfeng revealed to investors that the company is training a 20 trillion parameter model and plans to develop an 80 trillion parameter model afterward. The current flagship V4-Pro has a total of 1.6 trillion parameters.Among the publicly disclosed ultra-large models, Moonshot AI's Kimi K3 reaches 2.8 trillion parameters. Kimi K3 uses the MoE architecture, with each token activating 104 billion parameters out of the total 2.8 trillion. The planned 80 trillion parameter model by DeepSeek is nearly three times that size.DeepSeek completed over 50 billion yuan in financing in July, with a valuation exceeding 50 billion dollars. V4-Pro is open-sourced with a total of 1.6 trillion parameters and 49 billion active parameters.

According to The Information, citing informed sources, DeepSeek CEO Liang Wenfeng listed training large models with domestic chips such as those from Huawei as one of the company's biggest bets at the latest closed-door investor meeting, stating that this must succeed. DeepSeek expects to receive a new batch of Huawei chips for training as early as the fourth quarter of this year, although the specific models and quantities have not been disclosed.Liang Wenfeng believes that Huawei chips are expected to catch up with Nvidia in just a few years and hopes that DeepSeek will adapt to hardware beyond Nvidia in advance. DeepSeek is still using Nvidia chips to train models. In May of this year, Liang Wenfeng estimated that training the top-tier large models would require about 50,000 Nvidia GB300 chips, while switching to Huawei Ascend 950 would require about 200,000 chips, with the biggest issue at that time being insufficient supply from Huawei.

According to Axios, U.S. President Trump will meet with Ukrainian President Zelensky in New York on Tuesday, as the U.S. is working to reach a de-escalation agreement regarding the Russia-Ukraine situation.

According to a report by Reuters, Gao Yang, co-founder and chief scientist of the Chinese humanoid robot startup Spirit AI, stated that humanoid robots could complete most general tasks based on verbal instructions as early as next year, but it will take at least 8 more years to enter households. He mentioned that the brain is the weakest link in the entire robotics technology stack, and it is expected to reach a milestone similar to GPT-3.0 by mid-2027, allowing users to speak to robots in natural language and have them perform a series of reasonable physical actions to attempt to complete tasks.Spirit AI was founded in 2024, has about 300 employees, and raised $670 million in funding within two years, with a valuation of $2.9 billion. The company employs about 1,000 contractors who wear data collection devices at home or in factories, primarily relying on real-world data for training. The success rate for simple tasks in a structured living room environment is 90%. Currently, dozens of Moz1 wheeled humanoid robots are deployed on production lines at CATL and JD.com.Gao Yang stated that the next 1-2 years represent the initial window for industrial applications, and after two years, they can enter commercial service scenarios to perform simple tasks, but entering households will be more challenging. In terms of safety, full-body force control is implemented, and the robots will automatically engage emergency braking when encountering significant interaction forces. High-quality data is the main bottleneck for training the robot's brain.

The Senior Vice President of Amazon Web Services (AWS) stated that the next-generation Trainium AI chip will have "very obvious differentiated advantages."

According to GMGN data, the market capitalization of Robinhood Meme Coin FLYBRAIN briefly surpassed 7 million dollars, currently reported at 7.19 million dollars, with a 1-hour increase of 51.44%. CZ previously posted that it would be really cool to see someone create "immortal fruit flies" on the BNB Chain.According to market news, the prices of Meme Coins are highly volatile, and investors are advised to participate with caution.

According to Cointelegraph, Juan Marchetti, the Director of the Trade in Services and Investment Division of the World Trade Organization (WTO), stated at the event in Geneva for the release of the WTO stablecoin trade research report that the key obstacle to the widespread adoption of stablecoins in international trade is not technology, but rather the lack of a regulatory framework and its fragmentation.He cited a report from the Financial Stability Board (FSB) in October 2025, which indicated that among the 28 jurisdictions surveyed, only 11 (about 39%) have completed the establishment of a regulatory framework for stablecoins.The WTO report shows that stablecoins currently account for only 3% of the total global international payments, but from 2020 to mid-2024, the scale of stablecoin cross-border payments has grown by 35 times. The report also points out that stablecoins are expected to improve five major pain points in international trade financing: high costs, low speed, limited access, lack of transparency, and foreign exchange conversion.

The BRICS countries issued a joint statement expressing deep concern over the escalation of tensions in the Middle East, calling on countries in the region to exercise maximum restraint. The statement did not name specific countries involved in the Middle East conflict but urged avoiding actions that could exacerbate the situation and encouraged promoting long-term peace and stability through dialogue and diplomatic efforts.

The well-known American political and economic media The Kobeissi Letter expressed its views on platform X, stating, "Trump effectively made interest rate cuts a prerequisite for Walsh to serve as the Federal Reserve Chair. So will Walsh really raise rates in his first adjustment after being appointed by Trump as the Federal Reserve Chair? We believe the market has overestimated the likelihood of a rate hike in September."The Kobeissi Letter, as a domestic American political and economic media outlet, has 2.6 million followers on platform X. This media has published analyses and predictive articles regarding Trump's words and actions multiple times, and Trump has also retweeted reports from this media on the Truth platform.

According to the official announcement, the Gate contract Robinhood section has launched real-time trading for FLYBRAIN, FRONG, and JUGGERNAUT perpetual contracts (settled in USDT), supporting 1-10x leverage.Gate has also synchronized the addition of FLYBRAIN, FRONG, and JUGGERNAUT trading bots and copy trading features.

According to GMGN data, influenced by the attention of a16z co-founder Marc Andreessen on related accounts, the market capitalization of Robinhood Meme Coin FLYBRAIN briefly surged past 30 million dollars, with an intraday increase of over 670 times, currently reported at 29.3 million dollars.According to market news, Meme Coin prices are highly volatile and easily affected by celebrity attention and market sentiment, please be aware of trading risks.

Arya.ag, an agricultural warehousing and loan company in India, is testing a system for tokenizing warehouse receipts for stored grains on the Avalanche dedicated Layer 1 blockchain. Arya.ag is collaborating with Finternet to connect grain storage, warehouse receipts, collateral commitments, and loan statuses through this network. Devika Mittal, head of Ava Labs India, stated that the testing is underway, with each tokenized warehouse receipt representing ownership of the stored goods.Sanmesh Kalyanpur, a director at Finternet Labs, mentioned that Arya.ag's samplers collect information on stored grains and input it into the company portal. Finternet will integrate farmers, commodities, warehouses, and insurance information into a "composite token" for banks to assess collateral risks. Arya.ag stores approximately $2 billion worth of agricultural products in its warehouse network and supports loans of about 120 billion Indian Rupees (approximately $1.26 billion) annually, with its loan department, Arya Dhan, disbursing around $230 million in loans each year.The concept of Finternet originated from a 2024 paper by the Bank for International Settlements (BIS), co-authored by Infosys co-founder Nandan Nilekani and then-BIS General Manager Agustín Carstens, proposing the establishment of an interconnected unified ledger for tokenized assets. In 2024, the Indian government launched a 10 billion Rupee credit guarantee scheme to encourage financing against electronic transferable warehouse receipts. The two companies have not disclosed the expected launch date or initial deployment scale.

Dwarkesh Patel and Jerry Han published research on September 8, 2026, breaking down the contributions of data and model improvements in pre-training progress from 2019 to 2025. The study was conducted on a smaller scale, focusing on pre-training, corresponding to the publicly released model recipes and publicly available data corpora for each year, and was trained at a computational scale of up to 1e19 FLOPs.The results show that under the 1e19 FLOPs computational budget, the computational efficiency gain from data improvements is 12.0 times, while model improvements yield 3.7 times, with the data-side gain being approximately 3.24 times that of the model side. The gains from data and model improvements are largely independent and interact very little, with the additive effect of both explaining 88% of the variance in OLMES scores under a linear model.On the model side, the evolution from GPT-2 to OLMo-2 encompasses optimizers, positional encoding, normalization, activation functions, and initialization, among others. On the data side, the corpus evolved from approximately 9 billion tokens of OpenWebText in 2019 to larger and more finely filtered corpora like UltraFineWeb by 2025.

Pequity Research cited FundaAI stating that the training computing power used by GPT-6 is approximately 10 times that of GPT-5.FundaAI indicated that based on industry discussions, the total computing power consumed by experiments, reinforcement learning, synthetic data generation, and supporting infrastructure could be several times that of the main pre-training run, potentially reaching up to ten times. Once TPU 8t and Vera Rubin are shipped on a large scale, the next round of pre-training acceleration can begin, with large-scale interconnection and optics benefiting the most during this cycle.