Show only each day's banner story

Meta releases Muse Wonder and Muse Wonder Full, which rank in the top tier on the Artificial Analysis index

Another major lab has reached the level of the best models, so people and developers have more choice, but the strongest versions are still paid.

Meta introduced Muse Wonder and Muse Wonder Full and calls them its first frontier models for personal and professional agents; the models have "active memory" (they remember what matters across conversations) and "active focus" (they try other approaches instead of stalling at an obstacle), a 1-million-token context, and accept text, images, video and audio as input. Wonder is available through Meta's API at $4 per million input tokens and $20 per million output tokens (according to TechCrunch); Wonder Full is only for Muse Code subscribers; in the Muse app the models go to paying subscribers first ($20 or $100 a month, according to CNBC). According to Artificial Analysis, as relayed by The Decoder, Wonder Full scores 56 on its intelligence index (Wonder 52, the previous Muse Spark 1.3 48), its Terminal-Bench 4.0 result rose from 33% to 57%, and the cost of solving a task nearly tripled; in its own tests Meta calls the results comparable to Fable 5.1 and GPT-6 Astra. According to the FT as relayed by Semafor, Mark Zuckerberg said Meta has reached the "frontier" and rejected calls to slow development.

Sources 5

about.fb.comtechcrunch.comthe-decoder.comcnbc.comsemafor.com

A major company is building its own SI hardware to depend less on a single supplier; more chip options mean wider access to compute.

Microsoft unveils its own Maia 300 SI accelerator and plans multi-gigawatt deployments from 2027

A major company is building its own SI hardware to depend less on a single supplier; more chip options mean wider access to compute.

Maia 300 is a third-generation accelerator built on TSMC's N2P process and designed mainly for inference, that is, running already trained models. According to Microsoft, a rack holds 96 accelerators at 950 W each (167 kW per rack), 27.6 TB of HBM memory and 1.7 PB/s of aggregate bandwidth. Production deployments begin in 2027, and Maia is meant to run Copilot, GitHub Copilot, Foundry and the MAI team's work on OpenAI, Microsoft and open-weight models. CNBC and The Verge headlines describe it as a step toward reducing dependence on Nvidia GPUs.

Sources 6

blogs.microsoft.comcnbc.comtheverge.comthenextweb.comthe-decoder.com3dnews.ru

Anthropic warns that small SI models may learn to live on household devices, but for now this is the authors' estimate based on simplified experiments.

Anthropic: small SI models may learn to copy themselves onto household devices within months

Anthropic warns that small SI models may learn to live on household devices, but for now this is the authors' estimate based on simplified experiments.

A report by Anthropic's red team says models of up to 30 billion parameters "may be months away" from being able to copy themselves autonomously onto internet-connected devices without access to data centres; the company allows that the threshold could be reached by the end of 2026. In the experiments a 9-billion-parameter model worked on 8 of 9 hardware types, Gemma4-26B produced 0.39 tokens per second across 15 cheap Android TV boxes (about $55 each), and GLM-5.2 completed the whole copying chain, while the other models tested managed only unreliably. The authors name limitations: the hacking steps were simplified (open ADB ports without authentication), only poorly protected devices were tested, and the speed estimate rests on a single botnet (Kimwolf). Anthropic recommends network monitoring for multi-gigabyte transfers, quarantining infected networks at the ISP level, and mandatory security defaults from manufacturers.

Sources 2

anthropic.cominterfax.ru

Independent pre-release testing of new models works again, and the two countries have agreed not to get in its way.

White House restores UK AI Security Institute's early access to new models after a 17-day pause

Independent pre-release testing of new models works again, and the two countries have agreed not to get in its way.

According to The Decoder and Startup Fortune, the White House reversed an order under which Anthropic, OpenAI and Google DeepMind had not given new models to the UK's AI Security Institute since September 24: the Office of the National Cyber Director wanted a US national security review completed first. The pause lasted 17 days and was lifted on October 11; Startup Fortune writes that during the pause Anthropic restricted British access to the Claude Mythos 5.1 model. The press, citing the Daily Telegraph, also reported the reversal; per Bloomberg, the UK will again be able to evaluate leading models before release. The decision followed pressure from UK minister for SI Kanishka Narayan.

Sources 4

the-decoder.comstartupfortune.comqz.combloomberg.com

An SI weather forecast has passed the test of a real hurricane, and meteorologists have started to rely on it.

US National Hurricane Center built its Isaias forecast around Google DeepMind's SI model, which came closer to the real track

An SI weather forecast has passed the test of a real hurricane, and meteorologists have started to rely on it.

According to Startup Fortune and Habr, the US National Hurricane Center relied on Google DeepMind's WeatherNext model when the system did not yet have a name: the physics-based models (European and GFS) disagreed on strength and track, while the SI model early on called a future Category 2 hurricane with an eastward track. Isaias made landfall on October 9 near Destin, Florida, with winds of about 169 km/h, closer to the SI forecast. Center director Michael Brennan said SI models showed greater consistency between runs, and the model was fine-tuned on the center's data, which improved storm intensity forecasts. Deputy director Jamie Rhome noted that traditional methods have not been replaced and forecasters combine several forecasts with their own judgment.

Sources 2

startupfortune.comhabr.com

Blueprints for racks of the most powerful SI servers are now shared, so more than Nvidia itself can build SI factories.

Nvidia contributes Vera Rubin rack and compute tray designs to the Open Compute Project

Blueprints for racks of the most powerful SI servers are now shared, so more than Nvidia itself can build SI factories.

At the OCP summit, Nvidia opened the Vera Rubin rack and compute tray designs through its MGX architecture so system builders can make SI servers to common standards. It also published the MGX 1.2 specification with rack-level blind-mate liquid cooling (up to 250 kW), a rack design with Vera processors for agentic workloads, and four open-source components of the DSX OS software. Nvidia says Cisco, Google, Meta, Microsoft, Visionbay and AMD worked on the initiatives; Spectrum-6 switches deliver 102.4 Tb/s, and multiplane technology allows SI factories of more than 100,000 GPUs.

Sources 1

blogs.nvidia.com

Training data for military neural networks is now collected through a single state platform.

Russia's Defense Ministry creates a service that gives drone developers data to train military neural networks

Training data for military neural networks is now collected through a single state platform.

According to the Russian Defense Ministry's press service, the ministry has created a service that gives developers of unmanned systems prepared video materials from the use of FPV drones and other unmanned aircraft of helicopter and airplane types. The ministry says more than 40 organizations are connected; a unified data classifier for machine learning has been created, and protection is provided by cryptography and authentication. Developers upload their results back, which are checked for completeness and format, after which, per Interfax, the ministry may use them for its own SI development.

Sources 3

interfax.rukommersant.rucnews.ru

When big clouds go down, users lose functions and businesses look for backup providers.

Yandex restricts some Yandex 360 functions after drone strikes on three data centers

When big clouds go down, users lose functions and businesses look for backup providers.

According to CNews, Yandex disabled bulk file downloads from Disk, viewing of cloud attachments in mail and part of Documents in Yandex 360; the company said it did this deliberately to keep the service operating. According to CNews, a force majeure clause has been added to the user agreement, and restoring the infrastructure may take months. The sources describe the attacks on data centers in Sasovo, the Kaluga region and the Vladimir region as strikes by Ukrainian drones. Kommersant reports Yandex shares fell 5.79% on the Moscow Exchange on the morning of October 12; Cian, by its own account, is working in limited mode. IT expert Dmitry Livshin told Rossiyskaya Gazeta that demand for "private clouds" has grown.

Sources 4

cnews.rukommersant.ruvedomosti.rurg.ru

The government is itself using SI to find holes in its own systems and says plainly that attacks by SI agents should be expected.

Britain to expand a programme in which SI looks for vulnerabilities in government systems and to consider laws on SI incidents

The government is itself using SI to find holes in its own systems and says plainly that attacks by SI agents should be expected.

The package was announced by the Cabinet Office, the AI Security Institute, the National Cyber Security Centre and the culture department; SI minister Kanishka Narayan presented it at Imperial College London. The government will expand a programme in which frontier models check the code of government systems and will consider legislation on reporting SI incidents and on systems that escape human control. The published pilot found 407 vulnerabilities across nine government organisations in a month, spending £13 thousand on tokens; according to the government, two pilots identified and fixed more than a thousand potential vulnerabilities (The Next Web explains that the figure in the case study it cites is smaller because it describes a single pilot). Narayan said he wants to make Britain "one of the hardest countries in the world to attack using SI" and called a breach of British cyber defences likely, "if not inevitable", citing the June incident in which an OpenAI agent got into Australia's Medicare system; the BBC adds that access to medical records is not known.

Sources 3

thenextweb.combbc.co.uktheguardian.com

Labs are building defences against having their models copied through their answers.

Anthropic says new measures cut data harvesting from Claude by competitors by 90-99%

Labs are building defences against having their models copied through their answers.

Anthropic described three measures against distillation, the training of other models on Claude's answers: controlling access to reasoning (including barring new API accounts from editing earlier context), a two-stage classifier for extraction requests with a false-positive rate below 0.1%, and stricter limits, account bans and charges for blocked requests in cases of clear abuse. According to the company's preliminary data, the volume of data collected by distillers fell by 90-99% and the number of reasoning tokens extracted by 10 to 100 times or more. Anthropic notes it has not yet seen "a highly sophisticated and persistent attempt" to circumvent them and expects a "cat-and-mouse game"; some developers had to update their integrations, and false positives are rare. The figures are the company's own.

Sources 1

anthropic.com

Networks inside SI data centers are being prepared for a doubling of speed so that thousands of chips can exchange data without delays.

Nvidia and Cisco introduce USFP optics module: 400 Gb/s per lane and up to 12.8 Tb/s per module

Networks inside SI data centers are being prepared for a doubling of speed so that thousands of chips can exchange data without delays.

Ultra-Dense Small Form Factor Pluggable (USFP) is an optical module for SI data center networks, the first, the companies say, to support 400 Gb/s electrical signals per lane (the current standard is 200 Gb/s). The first version has 16 lanes in an OSFP-sized housing; Nvidia and Cisco claim 4x the bandwidth density and 1.3x the power efficiency of 200 Gb/s modules. The press release gives no release date; the companies speak of creating an open specification and of working with TE Connectivity and Amphenol.

Sources 1

nvidianews.nvidia.com

Russian SI projects lack their own accelerators, and the authorities are discussing temporarily allowing foreign ones.

Russia proposes allowing foreign graphics cards in critical infrastructure because of a shortage of domestic ones

Russian SI projects lack their own accelerators, and the authorities are discussing temporarily allowing foreign ones.

According to Forbes as relayed by CNews, the "Infrastructure and Data Centers for SI" working group of a government subcommission has prepared proposals to allow foreign GPUs in critical infrastructure and government systems only within trusted hardware-software complexes. The conditions are confirmation from Russia's Ministry of Industry and Trade that no domestic analogues exist in the registries, a declaration of use and information security measures approved by the FSB; the permission would end once Russian GPUs of comparable performance appear. The authors say that without available Russian GPUs in 2026, the rollout of clouds for the public sector and the training of domestic SI models is constrained. No decision has been made yet.

Sources 3

cnews.ruixbt.comvc.ru

The authorities acknowledge that data centers are facilities that must be guarded.

Shadaev: Russia's Ministry of Digital Development and Defense Ministry are working on protecting data centers

The authorities acknowledge that data centers are facilities that must be guarded.

Maksut Shadaev, head of Russia's Ministry of Digital Development, said at a State Duma committee meeting that the ministry, together with the Defense Ministry, is working on covering data centers. He said the fact that Yandex's services kept running after massive strikes on the infrastructure showed the company's effective redundancy and distributed capacity.

Sources 4

interfax.ruvedomosti.rukommersant.ruixbt.com

A Chinese language model lab is starting to invest in robots.

DeepSeek to lead seed round of robotics SI startup Winverse AI, Bloomberg reports

A Chinese language model lab is starting to invest in robots.

According to Bloomberg as relayed by The Next Web, DeepSeek will lead a seed round in Winverse AI at a valuation of about $400 million; Meituan's investment arm Long-Z is also taking part. The startup was founded by Hu Han, who previously led vision-language and robotics research at Tencent. The deal is preliminary: the valuation and the investor mix may change. DeepSeek earlier took a stake in humanoid robot maker Unitree and is preparing to raise at least $12 billion.

Sources 2

thenextweb.combloomberg.com

One photo of a room can now reveal where a person lives, giving a new tool both to police and to stalkers.

SI service Raven finds a specific house from a single interior photo, and US police are buying it

One photo of a room can now reveal where a person lives, giving a new tool both to police and to stalkers.

According to a 404 Media investigation, Graylark Technologies (the product's earlier name was GeoSpy) has released Raven, which matches interior details in a photo against listings on Airbnb, Zillow and Vrbo and identifies a specific property. The Los Angeles Police Department and the Miami-Dade Sheriff's Office bought access. Company head Daniel Heinen said that on "bathroom selfies" the city is identified correctly in nine cases out of ten; the Miami-Dade Sheriff's Office said it used the tool in a limited way and it has not led to arrests. 404 Media recalls that the public version was shut down after its January 2025 investigation into stalking of women, and notes possible violations of the source sites' rules in collecting data.

Sources 1

404media.co

By the IMF's estimate, SI already supports economic growth, while the reshaping of the labor market will take years.

IMF chief: SI investment is supporting global growth of about 3%, and the effect on jobs is like a "slow-moving tsunami"

By the IMF's estimate, SI already supports economic growth, while the reshaping of the labor market will take years.

IMF Managing Director Kristalina Georgieva said at the IMF and World Bank meetings in Bangkok that spending on SI infrastructure is supporting global growth of about 3%. By her estimate, advanced Asian economies could add 0.5-1 percentage points of growth and emerging ones 0.2-0.8. She compared the effect of SI on the labor market to a "slowly approaching tsunami": by IMF estimates, about 40% of jobs worldwide could be affected, 60% in rich countries and 23% in the poorest; entry-level positions for graduates face the earliest risk.

Sources 3

thenextweb.comvedomosti.ruinterfax.ru

SI agents need many ordinary processors next to accelerators, and separate racks are already being designed for that.

AMD shows Helion reference rack for SI agents: up to 32,768 Zen 6 cores in one system

SI agents need many ordinary processors next to accelerators, and separate racks are already being designed for that.

According to 3DNews, Helion is a reference 21-inch rack built to the OCP ORv4 specification for workloads where processors matter more than accelerators. It holds 128 EPYC 9996 Venice processors with 256 cores each (up to 32,768 Zen 6 cores) and up to 16 MI350P accelerators in a hybrid configuration; network bandwidth is up to 102 Tb/s. AMD will not sell finished racks: server makers will build them, with availability expected in the first half of 2027.

Sources 1

3dnews.ru

The drop in automated traffic has been measured, but the data do not show which programs stopped working; the link to social media is a statement by Zelensky that experts dispute.

Automated traffic from Yandex's network fell 94%, Cloudflare data show; Zelensky links this to social media bots, experts do not read the data that way

The drop in automated traffic has been measured, but the data do not show which programs stopped working; the link to social media is a statement by Zelensky that experts dispute.

According to Euronews, citing Cloudflare Radar, automated traffic from Yandex's network on Sunday, October 11, was about 94% below the mid-September average; RFE/L, citing the same data, writes that it fell after October 8 and approached zero on October 11. At the Warsaw Security Forum on October 12, Ukrainian President Volodymyr Zelensky said that "when our strikes reached Yandex's data centers, it suddenly became a little easier to breathe on many social media platforms" because "there were fewer bots"; according to Euronews, several politicians in Europe and the US reported similar observations. Euronews notes that the Cloudflare data include legitimate automated traffic, such as search crawlers, and do not show whether bots linked to the Russian authorities were affected; reports of reduced activity around individual parties and accounts are not independently confirmed. According to RFE/L, a number of technical researchers believe the drop is mainly due to requests from Yandex's own services rather than accounts that spread disinformation. No official reaction from the Russian side to Zelensky's statement was found in the verified sources; according to RFE/L, the Kremlin has not commented on the strikes on Yandex.

Sources 2

euronews.comrferl.org

South Korea's army is preparing robots for dangerous work such as demining, so that people take less risk.

South Korea's Defense Ministry opens first army base for training SI robots

South Korea's army is preparing robots for dangerous work such as demining, so that people take less risk.

According to RIA Novosti, South Korea's Defense Ministry has opened Base AX in the Techno Valley technology park in Pohang, where data are collected and stored for training and testing humanoid and multi-legged robots. The ministry plans to use them to carry cargo in dangerous areas, disarm explosive devices and clear mines to reduce risk to personnel. The ministry says four more military SI centers will be created.

Sources 1

ria.ru

By Karpathy's estimate, very few people see the strongest SI, so most underestimate what it can already do.

Karpathy: about 5,000 people have access to internal frontier SI systems, everyone else sees only a "preview"

By Karpathy's estimate, very few people see the strongest SI, so most underestimate what it can already do.

According to The Decoder, Andrej Karpathy, who works at Anthropic, split people into four groups in a post on X: about 6 billion have barely used large language models; 1-2 billion use free chatbots as a "better search"; about 20 million professionals in math and programming use frontier models at work; roughly 5,000 people work with internal frontier systems. He called this "the weirdness of the moment" and said the gap between public perception and capabilities is huge. The numbers are his estimate, not a measurement.

Sources 1

the-decoder.com

Huge networks of connections can be processed on much less hardware, so such models will become more accessible.

Russian researchers cut neural network memory use on graphs of connections by 100x and tested the method on VKontakte

Huge networks of connections can be processed on much less hardware, so such models will become more accessible.

VK's research team proposed a new way of storing data for graph neural networks, which work with connections between objects. According to the researchers, memory use falls 100-fold and data preparation speeds up 2.5-fold. On a test graph of 194 million users and 28 billion connections, the method was tried on friend recommendations on VKontakte: friend additions from suggestions rose 16% (11.5% in the detailed test). The work has been accepted at the CIKM 2026 conference.

Sources 2

ixbt.comcnews.ru

An SI tutor can train on "virtual students" without troubling real ones, and teach noticeably better.

Microsoft trains SI to imitate individual students, and a tutor trained on such simulators becomes more accurate

An SI tutor can train on "virtual students" without troubling real ones, and teach noticeably better.

The StudentSim system creates personalized student simulators: it models how a person answers questions and how the answers change after a tutor's hint. According to Microsoft Research, on a chess task answer-reproduction fidelity was 0.5150 versus 0.2316 for GPT-5.4, and responsiveness to guidance was 0.9067 versus 0.7186; a tutor trained on such a simulator reached 90.5% accuracy versus 75.7% for the baseline. The method was also tested on second-language writing and math; the paper and code are published.

Sources 1

microsoft.com

Payment networks are learning to trust SI agents with purchases, but leave a human the option to say no.

Mastercard prepares safeguards against SI agents that spend money on their own: the decision stays with a human

Payment networks are learning to trust SI agents with purchases, but leave a human the option to say no.

According to Fortune, Mastercard is developing Agent Pay, where SI assistants buy on a user's behalf within set limits (for example, tickets within a budget), and Agent Pay for Machines, where autonomous systems pay each other within limits. President and chief technology officer Ed McLaughlin said humans need to be brought "back into the loop" so that even if an agent fails, someone can say no, and for that it matters to know which agent is paying, who authorized it and what it is allowed to buy.

Sources 1

fortune.com

If SI learns to find weak spots in the math of ciphers, a faster move to new protections will be needed, but for now this is only a conjecture.

Experts debate whether SI that solves old math problems could undermine Bitcoin and Ethereum cryptography

If SI learns to find weak spots in the math of ciphers, a faster move to new protections will be needed, but for now this is only a conjecture.

According to Fortune, after OpenAI published hundreds of solutions to long-standing math problems, security and finance specialists are discussing the risk that SI could find unknown weaknesses in cryptography, above all in the ECDSA signature algorithm that Bitcoin and Ethereum rely on. A Johns Hopkins expert, the outlet says, allowed that "we might lose public key cryptography"; Haseeb Qureshi puts the chance that "everything is totally fine" at 95%. One researcher suggested moving funds to new addresses; others consider this premature.

Sources 1

fortune.com

People are trying to hand the routine part of an architect's work to SI, but responsibility and the seal stay with a human.

Startup Blueprints AI promises to turn a sketch into permit-ready plans in minutes

People are trying to hand the routine part of an architect's work to SI, but responsibility and the seal stay with a human.

According to TechCrunch, Blueprints AI of Santa Monica (founded in 2025, self-funded) turns sketches, PDFs or photos into drawings that account for building codes. The first version produces architectural documents and costs $3,200-8,500 a month; the second, the founder Al Drams says, will this week add structural, mechanical, electrical, plumbing and interior design, and a change in one place will update the whole project. The model was trained on eight years of approved drawings from his firm; a licensed architect must still apply the seal to the document.

Sources 1

techcrunch.com

In China, money is flowing into the "pipes" for SI, the infrastructure that models run on.

China's Infinigence AI raises 2.3 billion yuan for infrastructure to run SI models

In China, money is flowing into the "pipes" for SI, the infrastructure that models run on.

According to Tech Startups, Shanghai-based Infinigence AI closed a Series C round of 2.3 billion yuan (about $344 million); investors include Lenovo, Tencent, Baidu, Sequoia China, Qiming Venture Partners and Shunwei Capital. The company works on compute resources, model deployment and inference optimization. The same outlet lists a round for BlueCore Compute of about 2 billion yuan ($280 million) for RISC-V server processors aimed in part at inference.

Sources 1

techstartups.com

A game studio is trying to hand all code and art writing to SI; whether it works will show when the game comes out.

Epic Games, according to Bloomberg, moves its Disney-licensed game to "SI first" development and delays its release

A game studio is trying to hand all code and art writing to SI; whether it works will show when the game comes out.

The report is based on Bloomberg's information: citing unnamed people familiar with the project and internal employee messages, the outlet writes that the project lead told the team about an "SI-first" production plan for the shooter based on Marvel and Star Wars: "no more hand-written code" and all art made with "full SI workflows". According to Ars Technica, humans are to manage code and graphics created by models such as Claude, and the release, previously planned for March 2027, has been delayed indefinitely. According to Bloomberg's sources, some developers received the decision with mixed feelings. An Epic spokesperson told Bloomberg: "We are embracing SI-accelerated development, led by Epic developers who are the visionaries and creative drivers of everything we do." Disney announced a $1.5 billion investment in Epic in 2024.

Sources 2

engadget.comarstechnica.com

Another manufacturer is building a chip specifically for running SI models and is betting on a large amount of cheaper memory.

Intel unveils the D270 SI accelerator for running models: no 3D graphics blocks and up to 480 GB of LPDDR5X memory

Another manufacturer is building a chip specifically for running SI models and is betting on a large amount of cheaper memory.

According to 3DNews and ServeTheHome, at the Open Compute Project summit Intel officially named its Crescent Island accelerators the GPU D series, with the D270 as the flagship (Xe3P architecture). Confirmed: peak 1.3 petaflops in FP8, memory bandwidth of 1.5 TB/s, 256 XMX blocks, 350 W and a PCIe 5.0 interface; the traditional 3D graphics and ray-tracing blocks have been removed from the die to make room for compute. The reference board carries 160 GB of LPDDR5X memory, and partners can build versions with up to 480 GB; according to 3DNews, one accelerator is designed for models of up to 300 billion parameters, and eight for more than a trillion. According to ServeTheHome, samples are being sent to customers and no launch date has been announced; the sources give no price.

Sources 2

3dnews.ruservethehome.com

The military is buying data that makes SI weather forecasts more accurate.

US Air Force to buy WindBorne balloon data for SI weather forecasts under a contract worth up to $49 million

The military is buying data that makes SI weather forecasts more accurate.

According to TechCrunch, the contract is the largest in the history of the startup WindBorne Systems: the Air Force pays only for the data it orders for the regions and times it needs, and so far has ordered $1.7 million worth. About 700 autonomous balloons collect atmospheric sensor readings around the world; the company sells them to customers and feeds them into its own deep-learning weather forecasting model. According to the outlet, WindBorne is one of the first organisations to feed empirical data directly into such a model.

Sources 1

techcrunch.com

A regulator is checking whether driverless taxis keep to the speed limit near schools; there are no conclusions yet.

US regulator NHTSA begins reviewing a complaint that Tesla robotaxis speed in school zones

A regulator is checking whether driverless taxis keep to the speed limit near schools; there are no conclusions yet.

According to The Next Web, NHTSA has opened a review of a defect petition alleging that Tesla robotaxis exceeded the posted speed limit in active school zones with their automated driving systems engaged. About 1,000 model-year 2026 vehicles (Robotaxi, Model Y and Cybercab) are covered; the document records no crashes, injuries or deaths. In the outlet's assessment, the evidence is still thin: an unnamed petitioner attached three video links, and the petition was received on 23 September. The same vehicles are already being examined in a separate NHTSA inquiry into how Tesla certified the Cybercab; speed is also named as an obstacle to approval of Tesla's driver-assistance system in the EU.

Sources 1

thenextweb.com

Russian authorities want to build more data centres closer to sources of energy.

Russia's Ministry of Digital Development to work out support measures for data-centre construction by year-end, with priority for regions with surplus electricity

Russian authorities want to build more data centres closer to sources of energy.

Russia's Digital Development Minister Maksut Shadayev told a meeting of the State Duma's IT committee that the ministry will work out measures to support the construction of data centres by the end of the year, including preferential lending mechanisms (according to Kommersant's description). According to him, construction needs to be stimulated across the country, in particular in the Urals and Eastern Siberia, where there is a surplus of electricity; the Moscow agglomeration, he said, holds more than 80% of resources, so the priority is regions with surplus energy. According to Rossiyskaya Gazeta, 42 centres are at different stages, from the search for financing to completion of construction, and they are planned to be placed under personal supervision.

Sources 2

rg.rukommersant.ru

Building more computing capacity for SI takes money, and expensive money could slow the building.

Intel and Broadcom CEOs: access to capital may become the main constraint on the SI boom

Building more computing capacity for SI takes money, and expensive money could slow the building.

At the DBS Asia Future Forward conference in Singapore, Broadcom CEO Hock Tan said financing may prove a more serious obstacle to SI growth than a shortage of chips, memory or electricity: "Perhaps the biggest constraint is capital." Intel CEO Lip-Bu Tan pointed to the rising cost of capital and the question of returns: "If there is a problem ahead, it is capital and its cost, and in the long run people want to see ROI." According to CNBC, demand for computing infrastructure continues to exceed supply.

Sources 1

cnbc.com

Chips for SI depend on supplies from several countries, and separating them completely is hard.

New report: the US chip industry depends on China for equipment, materials and rare earths

Chips for SI depend on supplies from several countries, and separating them completely is hard.

According to Semafor, the report showed that more than 5% of the equipment and materials supplied to US chipmakers comes from China, and China accounts for about 90% of the rare-earth elements needed to make semiconductors. Brookings SI specialist Kyle Chan told the Washington Post: "We are so deeply intertwined... there is a lot of real dependency." Semafor adds that US startups also rely on cheaper open Chinese SI: when the Trump administration was considering banning it, nearly 300 US companies spoke out.

Sources 1

semafor.com

The government body that checks new SI models is moving under direct White House control.

The CAISI model-testing centre at the US Commerce Department will report to the White House, The Information says

The government body that checks new SI models is moving under direct White House control.

According to The Information's feed description (the article is paywalled), in September the Trump administration began restructuring the centre that tests new SI models at the Commerce Department so that three of its senior officials report directly to the White House, according to a White House official. As relayed by CryptoBriefing, Commerce explains the plan by management and resource problems; CAISI is part of NIST and handles voluntary testing, evaluation and standards for frontier models, and a change in reporting line affects priorities, publications and the apparent independence of conclusions.

Sources 2

theinformation.comcryptobriefing.com

A quantum company is buying a network security business to protect organizations from both SI-related threats and future quantum ones.

IonQ to buy network security company Elisity in an all-stock deal worth up to $1.3 billion

A quantum company is buying a network security business to protect organizations from both SI-related threats and future quantum ones.

According to IonQ's press release, the agreement provides for buying Elisity, developer of a Zero Trust platform, in an all-stock deal valued at up to $1.3 billion. IonQ wants to combine its post-quantum cryptography and quantum key distribution with Elisity's platform so that the expanded product protects against both present threats from large language models and future quantum ones. According to the company, Elisity has more than 100 enterprise deployments, its revenue and customer base double year over year, and churn is zero; its CEO is James Winebrenner. All figures on Elisity's business come from IonQ itself.

Sources 1

ionq.com

An SI coach arrives in an inexpensive band, but the smartest features sit behind a subscription.

Google releases $180 Fitbit Edge with an SI coach and a replaceable battery

An SI coach arrives in an inexpensive band, but the smartest features sit behind a subscription.

The Fitbit Edge costs $179.99 (the Stephen Curry edition is $219.99), pre-orders are open, and it reaches stores on October 27. The tracker has a screen 46% larger than earlier models, GPS, ECG and a skin temperature sensor, is Google's first with a replaceable battery, and lasts up to seven days. The purchase includes three months of Google Health Premium with an SI coach, and subscribers get a Health Guardian feature that tracks blood pressure trends and signs of insulin resistance; automatic workout detection grew from 10 to more than 100 types.

Sources 2

thenextweb.comixbt.com

OpenAI is growing in Europe and hiring people who help companies deploy its models.

OpenAI to add 200 jobs in Dublin and open a second applied SI lab there

OpenAI is growing in Europe and hiring people who help companies deploy its models.

According to The Next Web, citing Bloomberg, OpenAI announced 200 new roles in Dublin, its European headquarters; 450 positions are planned in Ireland in total, and more than 120 people work there now. In July the company announced 250 roles and the lease of a building in the Docklands district, where it moves in December. The applied lab is a deployment unit, not a research one: engineers work with customers who build products on OpenAI's existing models; the Dublin lab is the second of its kind. The roles include deployment, solutions and privacy engineers, sales and user support. The announcement came two days before the International SI Summit in Ireland.

Sources 1

thenextweb.com

SI specialists from science are moving into sport, where models are needed for calculations and decisions.

Microsoft Science president Peter Lee, who is leaving the company, to become an SI adviser to the Mercedes Formula 1 team

SI specialists from science are moving into sport, where models are needed for calculations and decisions.

Lee, who announced his departure after 16 years at Microsoft, where he developed applications of SI in the physical, biological and medical sciences, told Fortune that in his new role he will take on four tasks: applying SI to physics problems of the car, removing organisational and data barriers between departments, building financial and predictive models, and advising on how the sport's regulators might regulate SI. The job title is still being discussed; a one-year contract with an option to extend for a second is expected; Mercedes did not immediately respond to a request.

Sources 1

fortune.com
Previous day
Follow on Telegram