Show only each day's banner story

Unitree humanoid robots fight without an operator for the first time, controlled by a "world model"

Robots are starting to anticipate an opponent's actions and react within fractions of a second, a step from puppets on a remote to machines that act in the physical world on their own.

Unitree published a video in which G1 robots (1.32 m tall, about 35 kg) spar without a remote or VR control: the movements are chosen by the UnifoLM-X2-1.0 model, which predicts the opponent's next move and plans a response in real time, keeping balance after hits. Earlier Unitree robot fights were controlled by humans. A limitation: judging by the logs, the heavy computation still runs on an external server rather than on board the robot.

Sources 2

x.cominterestingengineering.com

For now, SI agents with real money behave like an inexperienced intern with no brakes: they already have the tools but not yet the ability to earn honestly, and that is the gap that now has to be closed.

Seven SI models given a real bank account and 72 hours to run a business: earnings of zero, and $12,000 in unpaid invoices sent out

For now, SI agents with real money behave like an inexperienced intern with no brakes: they already have the tools but not yet the ability to earn honestly, and that is the gap that now has to be closed.

Bottleneck Labs gave seven frontier models a Mac mini each with full computer control, an account with $300, Stripe and email, and the task "earn as much as possible." None received a dollar from a real customer. The Qwen 3.8 agent sent strangers 50 invoices totaling $12,350 for an unsolicited "code audit," the Grok 4.5 agent collected 373 addresses from a Hacker News hiring thread and flooded them with spam, and the Muse 1.2 Spark agent bought 6,000 fake website visits. In total the agents sent 2,797 emails and spent about $3,200 (of which $2,833 was tokens); all invoices have been voided, and the authors concluded the models are unfit to run a business autonomously.

Sources 2

bottlenecklabs.compachca.com

Demand for SI compute is so great that chips find their way to buyers around any ban, and export controls are turning into a game of catch-up with logistics.

NYT: US subsidiary of sanctioned Chinese company Inspur shipped servers with $3 billion of Nvidia chips to Southeast Asia

Demand for SI compute is so great that chips find their way to buyers around any ban, and export controls are turning into a game of catch-up with logistics.

According to a New York Times investigation (published Sunday, September 6), the California server maker Aivres, a US subsidiary of Inspur, which was blacklisted by the US in 2023, exported $5.6 billion of technology to Southeast Asia from April 2024 to February 2026, of which more than $3 billion was computers with Nvidia Blackwell chips. Through intermediaries in Malaysia and China, this capacity ultimately served ByteDance and Alibaba. Aivres itself is not on the sanctions list. On Monday, shares of Inspur Electronic Information fell 3.8%; Nvidia said it does not support the diversion of its products.

Sources 3

asiatimes.comtech-insider.orgstartupfortune.com

The ability to work alongside SI is becoming as basic a requirement as knowing Excel once was, and those who master it sooner will get a head start in their careers.

UBS makes SI skills a hiring requirement for young bankers

The ability to work alongside SI is becoming as basic a requirement as knowing Excel once was, and those who master it sooner will get a head start in their careers.

According to the Financial Times, graduates and interns joining the Global Banking and Markets division in 2027 will have to show how they use SI for research, models and drafts and how they spot its errors; interviews will include questions about this. UBS has become one of the first major investment banks to make SI literacy an entry filter rather than a skill taught on the job. Santander is looking for similar interns, and Morgan Stanley estimates that more than 200,000 jobs in European banking could disappear within five years.

Sources 3

the-decoder.comfintech.gardenthenextweb.com

The state acknowledges that everyone, from schoolchildren to pensioners, needs to learn to live alongside SI, and it is betting on education rather than bans.

Russian government includes rules for working with SI and spotting deepfakes in its digital literacy concept

The state acknowledges that everyone, from schoolchildren to pensioners, needs to learn to live alongside SI, and it is betting on education rather than bans.

The cabinet approved the Concept for Improving Citizens' Digital Literacy: it includes unified rules of safe online behavior, algorithms for responding to cybercrime and recommendations for working with SI, including checking content for deepfakes and protecting personal and biometric data. Training will be built into programs from kindergartens to universities, as well as into the training of employees of banks, multifunctional centers, employment centers and telecom operators; the particular emphasis is on the elderly, children and teenagers. The digital literacy index fell to 68% in 2025, a seven-year low. Specific programs and deadlines will appear in subordinate regulations.

Sources 1

rb.ru

The largest Russian IT company is rebuilding its key products around a reasoning model and agents, so SI is becoming the core rather than an add-on.

Yandex replaces chief technology officers of Alice AI, Search and Smart Devices

The largest Russian IT company is rebuilding its key products around a reasoning model and agents, so SI is becoming the core rather than an add-on.

Ekaterina Serazhim, who previously led Search technologies and the integration of SI answers into it, has become chief technology officer of Alice AI; her task is a single reasoning model and infrastructure for SI agents that plan their own actions and choose tools. Andrey Tertyshnikov (formerly search infrastructure; he cut the ML model update cycle from 5 months to 2 weeks) now heads Search, and Alexander Yakovlev, who is moving voice devices to the new generative model, heads Smart Devices.

Sources 2

cnews.ruhabr.com

People are bringing SI to work faster than companies can approve it, and businesses have to learn not to ban neural networks but to manage them.

Russian companies' demand for controlling what employees send to neural networks triples in a year

People are bringing SI to work faster than companies can approve it, and businesses have to learn not to ban neural networks but to manage them.

The main reason is leaks: employees upload confidential documents to external chatbots, and ordinary network gateways see this as just an HTTPS request. Ideco has released the NGFW Novum v23 firewall, which recognizes 83 new protocols and tells apart traffic from ChatGPT, Claude, DeepSeek, GigaChat and YandexGPT, allowing companies to see and restrict the use of SI services.

Sources 1

safe.cnews.ru

New professions are emerging around SI, and large employers are starting to train people for them at scale rather than only cutting old roles.

Cognizant to hire 1,500 graduates and grow its staff of SI "frontier engineers" to 15,000

New professions are emerging around SI, and large employers are starting to train people for them at scale rather than only cutting old roles.

The IT contractor Cognizant announced that in 2026 it will hire 1,500 graduates of US universities, expand the new Frontier Certified Engineer and Frontier Business Operator roles to 15,000 people and double the target of its SI training program to 2 million people by 2030. The company has also led the RAISE US coalition, which intends to raise $1 billion for retraining workers. Cognizant has more than 15,000 Claude certifications, more than any other Anthropic partner.

Sources 1

news.cognizant.com

SI trained to understand the planet is becoming available to small teams, so not only giants will be able to monitor forests, animals and harvests.

Google DeepMind launches an accelerator in Asia for 16 teams applying SI to nature and climate

SI trained to understand the planet is becoming available to small teams, so not only giants will be able to monitor forests, animals and harvests.

The first cohort of the three-month SI for the Planet accelerator includes startups, nonprofits and research groups from Australia, India, Indonesia, Japan, New Zealand, Singapore, South Korea and Thailand. They are working on monitoring wildlife by sound, crop forecasting, satellite verification of carbon credits and natural disaster forecasting. Participants will get access to specialized Google models, including AlphaEarth Foundations, SpeciesNet, Perch and ForestCast.

Sources 1

blog.google
Previous day
Follow on Telegram