Show only each day's banner story

Physicist Nathaniel Craig describes GPT-6 Astra's role in work on a lattice formulation of the Standard Model

The November publication explains the model's contribution to the October preprint. The assessment of that contribution came from the author of the work himself.

In a guest post on OpenAI's science blog, physicist Nathaniel Craig described work on a possible nonperturbative definition of the Standard Model on the lattice. According to him, he proposed the approach, while GPT-6 Astra helped build the construction and played a leading role in developing the proof. The preprint appeared on arXiv on October 6, 2026. Craig stresses that the definition remains a candidate for now: the check of reflection positivity at finite lattice spacing is still ongoing.

The market has stopped paying for mere membership in SI infrastructure: from here capital will go to those with real orders and stable financing, not just participation in the data center theme.

Industrial and energy stocks of the SI boom have lost $4.2 trillion in market value since June

The market has stopped paying for mere membership in SI infrastructure: from here capital will go to those with real orders and stable financing, not just participation in the data center theme.

An FT analysis puts the fall in market value of global manufacturers, power companies and miners tied to data center construction at $4.2 trillion since June — almost half of the $9.4 trillion gain over the previous 12 months. Caterpillar has lost $214 billion in market value (down 44% from its June 30 record, according to FactSet), Oklo 49%, and Comfort Systems and Corning about half their value. The FT attributes the selloff to doubts about debt financing of hyperscaler capex and to political resistance to new data centers, and Paul Kedrosky of MIT tells the FT that investors mistakenly valued old industrial companies as if they were semiconductor companies. At the same time, US data center spending totaled $96 billion in the 12 months to September, Caterpillar's order backlog stood at a record $78 billion, and Siemens' orders for electrical equipment for data centers rose 60% to €7 billion.

The package under discussion covers both chip shipments and remote access to them. There is no final decision on the bills yet.

US Congress advances restrictions on SI chip exports and cloud access to them

The package under discussion covers both chip shipments and remote access to them. There is no final decision on the bills yet.

According to the Financial Times, Republicans and Democrats backed a package of bills restricting foreign sales of sensitive chips and giving Congress the right to review and block export deals already approved by the Trump administration; among the measures is the Remote Access Security Act, which shuts Chinese companies out of remote access to advanced chips via cloud services. On November 13, House committee chairs John Moolenaar, Brian Mast and Rick Crawford demanded hearings with Jensen Huang and representatives of the Commerce Department and urged passage of four bills that could be included in the annual defense budget (NDAA). Commerce Secretary Howard Lutnick agreed to testify before the Senate Banking Committee on December 10, The Hill reports.

The scandal around Nvidia is moving from the personal to the corporate level: the companies through which large chip orders flowed are being forced to explain themselves publicly and launch internal reviews — supply-chain transparency is becoming a condition for working with accelerators.

Megaspeed: Alice Huang left the company in August 2025, internal review launched

The scandal around Nvidia is moving from the personal to the corporate level: the companies through which large chip orders flowed are being forced to explain themselves publicly and launch internal reviews — supply-chain transparency is becoming a condition for working with accelerators.

On 15 November Megaspeed said that senior executive Alice Huang, who has been linked to former Nvidia head Jensen Huang, left the company in August 2025, and announced an internal investigation into reports about her business, CNBC, BBC, SCMP and Korea Times report. The company rejected allegations of supply-chain violations and said it is distancing itself from her. According to The Decoder, the firm linked to Alice Huang reportedly brokered almost $12 billion in Nvidia chip orders for China; Megaspeed disputes these claims.

Following the November 11 announcement, the ministry and Sber signed the agreement. German Gref called GigaChat the basis of the model.

Russia's Ministry of Education and Sber sign agreement on school model built on GigaChat

Following the November 11 announcement, the ministry and Sber signed the agreement. German Gref called GigaChat the basis of the model.

On November 15, at a strategic session on SI in education, the Ministry of Education and Sberbank signed an agreement to create a specialised model for the general education system; according to German Gref, it is being built on GigaChat. Deputy Prime Minister Dmitry Chernyshenko said it will be up to teachers, lecturers and administrators how quickly the sovereign model gains the experience education requires. Education Minister Sergei Kravtsov said more than 250 thousand teachers have completed the course "Artificial Intelligence in a Teacher's Work", and according to smotrim.ru the tool is planned to be presented before the end of the current academic year.

The memory maker is becoming a full-cycle SI infrastructure contractor — the fight for compute is no longer only about chips but also about cooling, construction and data-centre operations.

Samsung invests $1 billion in data-centre operator Helix and moves into full-stack SI infrastructure

The memory maker is becoming a full-cycle SI infrastructure contractor — the fight for compute is no longer only about chips but also about cooling, construction and data-centre operations.

As the FT reported, over the past year Samsung Group has moved from supplying components to trying to become a full-stack SI infrastructure provider: in September it invested $1 billion in data-centre operator Helix Digital Infrastructure, which is backed by KKR and Nvidia, bought a European cooling-systems manufacturer and entered a partnership with OpenAI. Samsung SDS is in early talks with Anthropic and OpenAI about hosting their capacity as a tenant; analysts expect SDS's data-centre revenue to reach 6.6 trillion won ($4.9 billion) by 2031 — roughly six times the current level. The OpenAI agreement also provides for Samsung SDS to take part in designing and operating data centres, and for Samsung C&T and Samsung Heavy Industries to develop floating data centres.

The number of job postings fell in all groups, most of all in occupations with high SI exposure. The note does not establish a causal link.

Reserve Bank of Australia flags weaker demand for young workers in jobs highly exposed to SI

The number of job postings fell in all groups, most of all in occupations with high SI exposure. The note does not establish a causal link.

On November 15, The Decoder went through a research note by Mike Major for the Reserve Bank of Australia (RBA) on monitoring the most vulnerable groups in the labor market — new entrants and occupations with high SI exposure: programming, clerical work, customer support. Since late 2022, job postings have declined across all groups, but most sharply in the quintile with the highest exposure; there, workers aged 15–24 and recent graduates have lower employment and higher unemployment than their older colleagues. The author warns that causality has not been proven and that what is measured is the theoretical exposure of occupations, not companies' actual use of SI. For comparison: Stanford — employment of 22–25-year-olds in exposed occupations down 13% in relative terms, with young developers at about minus 20%; Anthropic — a 14% lower probability of finding a job, while the share of junior hiring in technical roles fell from 27% to 9% since the start of 2025.

Chip export restrictions are increasingly investigated as criminal cases against individuals rather than only as claims against companies — supply-chain managers face personal liability.

Nvidia asks Taiwan authorities to review case of manager accused of shipping chip servers to China

Chip export restrictions are increasingly investigated as criminal cases against individuals rather than only as claims against companies — supply-chain managers face personal liability.

According to Bloomberg, Nvidia has approached political authorities in Taiwan, disagreeing with the charges against a manager surnamed Chang, arrested in July and charged in August with forgery and breach of trust. Prosecutors in Keelung describe Chang as the central figure in a network that shipped 74 servers with Nvidia chips to China, with another 56 servers intercepted by the coast guard in May; according to the prosecution, he helped obtain Nvidia's approval to sell 130 servers to a buyer without sufficient data-centre floor space and concealed that the payment came from a Chinese legal entity. Nvidia said it carried out a review and found no evidence that the employee was aware of the scheme, and sought advice from TSMC, which declined to take part. The maximum penalty in the case is up to five years in prison.

An exam that tests memory and speed proves powerless against a model that answers in seconds: assessment systems will have to check not the answer but the conditions in which it was obtained.

Top scorer on South Korea's tax accounting exam admits SI use and turns himself in

An exam that tests memory and speed proves powerless against a model that answers in seconds: assessment systems will have to check not the answer but the conditions in which it was obtained.

The person who scored highest on South Korea's annual tax accounting exam has admitted using SI while sitting it and turned himself in, Yonhap, BBC and The Guardian report. The case became public a week before the nationwide university entrance exam, and the authorities have begun an investigation: according to Japan Times, the inquiry concerns how the examinee gained access to the model. The scandal has set off a debate about how well exam procedures are protected against generative models.

The cost of electricity and grid connection is becoming a separate risk item in SI infrastructure: it now has to be paid upfront, which makes new capacity more expensive and slower to bring online.

Applied Digital demands multibillion-dollar guarantees from Oracle and Meta for grid connection payments

The cost of electricity and grid connection is becoming a separate risk item in SI infrastructure: it now has to be paid upfront, which makes new capacity more expensive and slower to bring online.

Data center operator Applied Digital has asked Oracle and Meta for multibillion-dollar guarantees before signing lease agreements, The Next Web reports; according to the outlet, guarantees were also requested from Nvidia and AMD. They cover the cost of connecting sites to power grids, shifting that risk from the operator to future tenants. According to CEO Wes Cummins, the target tenant structure is 70% investment-grade paper and 30% unrated, and the share of space for lower-rated neocloud providers is capped; Barclays estimates hyperscaler debt issuance this year at $300–320 billion.

Japan is codifying military SI in official rules: the autonomy of systems is growing, but the ministry leaves responsibility for their use with humans.

Japan's defense ministry to issue guidance on SI use in the armed forces by late November

Japan is codifying military SI in official rules: the autonomy of systems is growing, but the ministry leaves responsibility for their use with humans.

Japanese Defense Minister Shinjiro Koizumi said on November 15 in Yokosuka that at the end of the month the ministry will publish guidance on the use of SI for ministry employees and the Self-Defense Forces. The document will update the basic SI policy adopted in July 2024; according to the minister, SI is an area where everything changes quickly, so the revision will be done without hesitation. The guidance rests on three principles: decisions and responsibility for the results of SI use remain with humans, data must be turned into strength for the country's defense, and the spread of SI must be accompanied by improvements in its safety.

Invasive neural interfaces in China are moving from isolated experiments to paid clinical practice — which speeds up data accumulation for the next generations of BCIs.

About 120 patients in China live with an implanted neural interface

Invasive neural interfaces in China are moving from isolated experiments to paid clinical practice — which speeds up data accumulation for the next generations of BCIs.

Hong Bo, chief scientist at the Lingang laboratory, said that about 120 patients in China have been wearing an implanted brain interface for more than a month, and called 2026 the "year of clinical application" for Chinese BCIs. Neuracle's NEO system, approved by the NMPA in March 2026 as the world's first invasive BCI medical device cleared for sale, has already been prescribed under 12 commercial prescriptions; the epidural "Beinao-1" has passed 40 implantations and a full registration cohort, with documents planned for submission in 2027; StairMed has performed 42 long-term fully invasive implantations. More than 20 hospitals have opened BCI clinics, Shanghai's Hu Huibao program has included the Neuracle product, and a conference in Shanghai on December 3–6, 2026 is set to launch the formation of a national EEG data center. The figure of about 120 was obtained by summing across several teams and does not equal the number of commercial users.

Open weights on a Chinese cloud with its own chip chain and mobile OS give HarmonyOS and Ascend developers a local alternative to Western stacks and a cheap entry point for agentic applications.

Huawei Cloud releases openPangu-2.1-Pro open model on its MaaS platform

Open weights on a Chinese cloud with its own chip chain and mobile OS give HarmonyOS and Ascend developers a local alternative to Western stacks and a cheap entry point for agentic applications.

On November 15 Huawei Cloud announced the launch of the open model openPangu-2.1-Pro on its model-as-a-service (MaaS) offering. The update covers three areas: agentic development, where Skills triggering and tool selection are tuned on more than 100 thousand specialized examples; SI coding — generating ArkTS code for HarmonyOS applications and operators for Ascend accelerators; and inference optimization with higher throughput in typical deployment scenarios. The model connects to Huawei Cloud's AgentArts, CodeArts and OfficeAce products, and pricing is 0.8 yuan per million input tokens, 3.2 yuan per million output tokens and 0.016 yuan per million cached input tokens.

When the public market will not provide the money, data center builders turn to customer prepayments — compute is increasingly financed upfront by those who will consume it.

Firmus Grid raising $3 billion privately after pulling its IPO

When the public market will not provide the money, data center builders turn to customer prepayments — compute is increasingly financed upfront by those who will consume it.

According to Bloomberg, Australian Firmus Grid Ltd., which builds data centers for SI workloads and has Nvidia among its shareholders, plans to raise $3 billion through new shares, debt and customer prepayments, including from Nvidia and Meta, the Australian Financial Review reported (Bloomberg summary of November 15). Talks on prepayments are the most advanced: customers may pay about 12.5% of contract value upfront to finance new capacity. In October the company withdrew its IPO on the ASX, where it had expected to raise about $5 billion at a valuation of roughly $30 billion; the project pipeline is 912 MW, of which 46 MW has been built, and revenue for the 2026 financial year is $51 million.

If the round closes, SI drug discovery will definitively become one of the most capital-intensive areas of technology: money is going where a model predicts the behavior of molecules in the body, not just text.

Isomorphic Labs in talks on $5 billion at a $50 billion valuation

If the round closes, SI drug discovery will definitively become one of the most capital-intensive areas of technology: money is going where a model predicts the behavior of molecules in the body, not just text.

Isomorphic Labs — Alphabet's London-based company spun out of Google DeepMind that develops drugs using SI — is in talks to raise about $5 billion in new funds at a valuation of around $50 billion, the industry aggregator vbdata reported on November 15; the deal is not closed. Bloomberg previously reported talks at a valuation of at least $40 billion. Counting the $600 million first external round of 2025 and the $2.1 billion Series B, the company's external financing could reach about $7.7 billion in less than two years. Partnerships with Eli Lilly and Novartis (2024) and Johnson & Johnson (January 2026) provide for milestones of up to $1.7 billion and $1.2 billion respectively.

Sources 1

A major memory maker no longer considers itself merely a component supplier: if computing is sold as tokens, the winner is whoever owns chips, data centers and the energy for them all at once.

SK Group adopts "SI business growth cycle" strategy: from HBM chips to SI factories and trading computing

A major memory maker no longer considers itself merely a component supplier: if computing is sold as tokens, the winner is whoever owns chips, data centers and the energy for them all at once.

On November 15, SK Group announced that at a management seminar on November 12–14 in Icheon (about 50 participants led by Chairman Chey Tae-won) it adopted an "SI business growth cycle" strategy: strong positions in SI chips expand the SI factory business, that business creates demand for electricity and SI services, and their spread accelerates investment in advanced chips. Chey said every subsidiary must step outside the framework of its existing business and redefine itself as an SI company, and set the goal of taking part in the "token economy", where computing and SI services are traded. The group will continue "Portfolio Rebalancing 2.0"; SK hynix will upgrade its "smart factories" on the basis of SI and digital twins, and SK Innovation raised output at its Ulsan CLX complex through an ontological process map.

Humanoid robots are moving from demonstrations into serial production: once assembly becomes reproducible, price and volume start growing faster than customers' ability to deploy them.

Zhejiang humanoid robot center scales output to thousands, then 10,000 robots a year

Humanoid robots are moving from demonstrations into serial production: once assembly becomes reproducible, price and volume start growing faster than customers' ability to deploy them.

On November 15 the China Securities Journal reported that during a visit to the Zhejiang Humanoid Robot Innovation Center site in Ningbo, the center's vice president Liu Yun said the annual capacity of the plant in Ouchi has been raised from several hundred to several thousand humanoid robots, a new plant is under construction nearby in Jishigang, and through partner production bases the center expects to reach 10,000 units a year. According to her, this year became the "pivotal moment for mass production": last year shipments were small-batch, now the product has become standardized and reproducible, and the general-purpose wheeled manipulator robot NAVIAI WA2 is already being shipped to customers in batches. This year's orders are several times higher than last year's — some from customers that have passed the proof-of-concept stage, some from new buyers.

For the first time a bank account is being designed not for a person but for their SI agents: if the "limits instead of trust in the agent" model works, autonomous payments will become an ordinary banking service.

Starling founder Anne Boden is preparing a bank account with spending limits for SI agents

For the first time a bank account is being designed not for a person but for their SI agents: if the "limits instead of trust in the agent" model works, autonomous payments will become an ordinary banking service.

Anne Boden, founder of Starling Bank (she stepped down as CEO in 2023 and in October 2026 sold her last 2.7% stake, with the bank valued at £1.7 billion), has announced the creation of Lockwood — a current account designed for transactions by SI agents. The customer sets limits by spending category, and payments outside those limits are declined by the infrastructure: in an interview with The Times she gave an example in which a £100 monthly limit on groceries blocks a £200 payment. Authorization is initially planned in EMI status (target — 2027), with a full bank launch in 2028; at the start Boden is funding the project herself, with an external round next year. According to Corriere della Sera, the name Lockwood was chosen by the SI agents themselves, and the goal is 1 million customers in the first 12 months.

Generation order turned out to matter more than scaling the model: filling in masks by diffusion delivered a new best result in lip reading, which means speech recognition will work where there is no sound or it is unintelligible.

KAIST: diffusion language model reads lips with 19.4% error rate on LRS3

Generation order turned out to matter more than scaling the model: filling in masks by diffusion delivered a new best result in lip reading, which means speech recognition will work where there is no sound or it is unintelligible.

Professor No Yong Man's team at KAIST announced on November 15 a visual speech recognition system that, instead of generating words sequentially, first decodes the positions with the clearest articulation and uses them as context for ambiguous spots. It runs on a diffusion large language model that repeatedly fills in masked positions; training is two-stage — first aligning lips with text, then modeling sentence length. On the LRS3 dataset (433 hours of video with reference transcripts) it achieved a word error rate of 19.4% — according to the university, the best result among methods trained only on LRS3 with transcripts. The paper "Diffusion Large Language Models for Visual Speech Recognition" received an Outstanding Paper Award at EMNLP 2026; the first author is Yeo Jeong-hun.

The largest contract chipmaker is starting to invest directly in its ecosystem: that speeds up the emergence of new suppliers and startups that otherwise would have nowhere to get money for their first silicon samples.

TSMC sets up $500 million fund to invest in chip startups and suppliers

The largest contract chipmaker is starting to invest directly in its ecosystem: that speeds up the emergence of new suppliers and startups that otherwise would have nowhere to get money for their first silicon samples.

TSMC plans to create a $500 million fund to invest in chip startups, suppliers and other supply-chain partners, Nikkei Asia reported. According to the publication, the company explained the fund's creation by a desire to "support the future Nvidia" — that is, to help new players around its manufacturing grow.

Sources 1

The assistant is turning from an interlocutor into an executor: if Siri learns to act inside apps, wearables will get a screenless interface — and that changes competition in wearable electronics.

Apple prepares Siri integrations with apps as the basis for smart glasses features

The assistant is turning from an interlocutor into an executor: if Siri learns to act inside apps, wearables will get a screenless interface — and that changes competition in wearable electronics.

According to Bloomberg, a Bloomberg newsletter reports that Apple is developing Siri integrations with third-party apps that are meant to become the basis for smart glasses features, and the iPhone Duo 2 plan includes Face ID support. App integrations mean the assistant will be able to perform actions in third-party services, not just answer questions.

High-quality video is no longer proof that a public figure said something: what has to be checked is not the face on the screen but the registration of the scheme and the recipient's account.

Deepfake of India's finance minister costs Kerala businessman ₹1.45 crore

High-quality video is no longer proof that a public figure said something: what has to be checked is not the face on the screen but the registration of the scheme and the recipient's account.

Kerala police are investigating the case of a businessman from Naduvannur (Kozhikode district) who lost ₹1.45 crore after believing an SI video featuring India's finance minister Nirmala Sitharaman: in the clip posted on Facebook she promoted a high-yield investment scheme, The Hindu reported on November 15. According to police, the money was transferred in 30 separate payments; the case was registered with the Kozhikode Rural Cyber Police. The newspaper notes that earlier in Vadakara a couple lost about ₹3 lakh in an almost identical scheme using a fake video of the same minister; police urged people to check whether any investment scheme is registered.

Voice and face stop being a lock: since they can be faked, identity verification will rely on independent channels and multi-factor confirmation, and biometrics as the sole argument loses its force.

Russia's Interior Ministry: deepfakes and voice cloning are used to bypass biometric identification

Voice and face stop being a lock: since they can be faked, identity verification will rely on independent channels and multi-factor confirmation, and biometrics as the sole argument loses its force.

Materials from Russia's Interior Ministry reviewed by RIA Novosti and TASS say that remote scammers use SI to create convincing deepfakes, imitate voices and swap video images, posing as relatives, executives and employees of organizations. The ministry calls the use of such technologies in attempts to bypass identification and biometric authentication systems a particular threat, noting that images and voices in the digital environment are no longer unconditional proof of identity. The ministry's recommendation is to re-verify the person on the other end through an independent communication channel when transfers, confirmation codes or personal data are involved.

An attack on an intermediary hits the whole chain: while one hoster recovers, access to Ozon, Wildberries and Sber services falters — meaning infrastructure recovery has itself become a target and will have to be defended like a combat mission.

Beget fends off new DDoS attack while recovering from November outage

An attack on an intermediary hits the whole chain: while one hoster recovers, access to Ozon, Wildberries and Sber services falters — meaning infrastructure recovery has itself become a target and will have to be defended like a combat mission.

On November 15 hosting provider Beget reported on Telegram a large-scale DDoS attack in which attackers cycled through the company's subnets, estimating its power at up to 850 Gbps; IA Regnum reports up to 2.1 Tbps, which differs from the provider's statement. Beget assesses that the nature of the attack points to an attempt to derail recovery from the November 2 incident, when its cloud infrastructure was destroyed and data from VPS, S3 object storage and Kubernetes clusters was lost; by November 6 the company reported restoring more than 60% of virtual hosting servers and 99% of mail servers. Some users may have had trouble reaching individual resources, including Ozon, Wildberries and Sber services; the provider restricted SSH connections from foreign IPs. Beget serves 540.7 thousand domains in the .RU zone (14.7% of the Russian internet).

If work done by SI in minutes is graded as a student's own, written assignments stop measuring knowledge — universities will have to change the format of assessment rather than rely on detectors that also err against international students.

Students in England submit unsupervised work done with SI in minutes and get the same grades

If work done by SI in minutes is graded as a student's own, written assignments stop measuring knowledge — universities will have to change the format of assessment rather than rely on detectors that also err against international students.

Issue No. 899 of University World News, dated 15 November, describes a new study of how SI is changing the experience of higher-education students in England: unsupervised assessed assignments are being done with the help of superintelligence, sometimes in a matter of minutes, with grades close to those of students who did the work themselves. The same issue notes that SI detection systems are often wrong and particularly often penalise students for whom English is not a native language, so a detector's verdict should not be the only evidence of SI use.

Transcribing and summarising meetings is no longer a niche function: a single service earns $100 million a year from it, which means millions of working teams are ready to pay for automatic notes and digests.

Otter.ai lifts annual revenue to $100 million on meeting transcription and summaries

Transcribing and summarising meetings is no longer a niche function: a single service earns $100 million a year from it, which means millions of working teams are ready to pay for automatic notes and digests.

Revenue at Otter.ai — a service that records meetings, transcribes speech and turns them into summaries — has reached $100 million a year; the figure was cited by the company's CEO Sam Liang in a Fortune interview published on 15 November. Over the years Otter.ai has grown from a narrow transcription tool into an SI note-taking platform that not only converts speech into text but also pulls decisions and action items out of a conversation.

Sources 1

Generative models have quietly become part of how political offices work, and that is now becoming a public risk: trust in an official's text will increasingly hinge on who wrote it.

UK prime minister's aide accused of using ChatGPT to write a speech

Generative models have quietly become part of how political offices work, and that is now becoming a public risk: trust in an official's text will increasingly hinge on who wrote it.

Andy Burnham, an aide to the British prime minister, has been accused of using ChatGPT to write the head of government's speech, the Daily Mail reports (as relayed by Rossiyskaya Gazeta) along with RIA Novosti on 15 November. The report concerns the use of a generative model in the work of the government press service.

Sources 2

A "made by a neural network" tag is becoming a matter of regulation: audiences will get a formal marker to tell the generated from the real, and platforms new labelling requirements.

Russia proposes mandatory labelling of content generated by neural networks

A "made by a neural network" tag is becoming a matter of regulation: audiences will get a formal marker to tell the generated from the real, and platforms new labelling requirements.

Rossiyskaya Gazeta reported on 15 November that Russia is proposing to introduce mandatory labelling of content created by neural networks. The labelling is meant to let readers and viewers distinguish generated texts, images and videos from those made by humans.

Sources 1

The money is going into automating defence manufacturing, not only into combat systems: the ability to quickly produce complex hardware is becoming part of military power.

Nidus raises $30 million for SI robots for the US defence industry

The money is going into automating defence manufacturing, not only into combat systems: the ability to quickly produce complex hardware is becoming part of military power.

Nidus, which develops SI technologies for manufacturing robots used by US defence-industrial companies, has raised $30 million in pre-seed and seed rounds, Upstarts Media reports (via Techmeme). The money was raised at a stage when the product had not yet reached the mass market.

Sources 1

The clarification concerns covert recording with glasses that come with a visible recording indicator.

Interior Ministry clarifies rules for using smart glasses in public places

The clarification concerns covert recording with glasses that come with a visible recording indicator.

Russia's Interior Ministry has clarified the rules for using smart glasses with a camera in public places, Vedomosti and RBC reported. The ministry said the indicator may not be covered in order to conceal that recording is taking place.

The blog launch adds to the scientific results a promised channel with how-to guides and the limitations of OpenAI's models, giving scientists instructions for checking and reproducing such work.

OpenAI launches science blog with guides on using its models in research

The blog launch adds to the scientific results a promised channel with how-to guides and the limitations of OpenAI's models, giving scientists instructions for checking and reproducing such work.

According to OpenAI, on November 15, 2026, OpenAI announced the launch of a science blog for scientists and academic researchers, with sections on research results, how-to guides and views from the scientific community. The company promises to publish methodologies for using its models in scientific computing, data analysis, literature work and experiment planning, including ways to verify how a model performs and its limitations. The announcement lists OpenAI's earlier scientific results: by its account, an internal system in September 2026 produced a proof on the Navier–Stokes existence and smoothness problem, in early October a set of mathematical results including a result on the "quasi-Riemann hypothesis", and in biology SI agents built on GPT-Rosalind, together with PerturbAI and Scripps Research, analyzed data from millions of mouse brain cells in 32 minutes. The blog's first item was a guest post by physicist Nathaniel Craig.

Rules for robotaxis are reaching down to the details: the right to refuse a passenger shows that autonomous transport is moving from experiments into ordinary regulated service.

Draft rules would let driverless taxis refuse passengers

Rules for robotaxis are reaching down to the details: the right to refuse a passenger shows that autonomous transport is moving from experiments into ordinary regulated service.

RBC reported on 15 November that a draft regulation proposes allowing driverless taxis to refuse passengers. The proposal is part of a set of rules for autonomous transport, under which operators must determine in which cases a trip may not be started.

Sources 1

Computing is moving into orbit and beginning to compete with science for the sky: the more satellite data centers there are, the more important it is to agree on brightness and interference rules in advance — otherwise astronomy will pay for the growth of computing.

Space data centers could spoil observations of the night sky

Computing is moving into orbit and beginning to compete with science for the sky: the more satellite data centers there are, the more important it is to agree on brightness and interference rules in advance — otherwise astronomy will pay for the growth of computing.

On November 15 Axios reported on a paper whose authors assessed the possible interference with astronomical observations from orbital data centers.

Sources 1

An awards show of its own gives SI-generated music recognition as a distinct genre among performers and audiences.

Los Angeles hosts inaugural AIMAs with SI presenters and technical glitches

An awards show of its own gives SI-generated music recognition as a distinct genre among performers and audiences.

The first AIMAs took place in Los Angeles, with SI presenters, lip-syncing and live stand-ins. Variety, via Techmeme, reported technical glitches during the ceremony.

Sources 1