Wildberries launched a taxi service in Armenia

Wildberries continues to develop and is progressively turning from a market into a complete empire. Not limiting itself to commerce, the corporate lately determined to enter the taxi market. The service has already begun working in Belarus, Uzbekistan and Kyrgyzstan, and now it’s progressively being launched in Armenia in early entry mode.

For now solely workers and companions of the corporate can use the service, however within the coming weeks entry will start to be expanded. The applying will seem in Google Play and the App Retailer, and anybody wishing can join the ready checklist. Three tariffs can be found for passengers: “Customary,” “Consolation” and “Multi,” which searches for a automotive in two classes without delay.

In Russia the launch of the service just isn’t but deliberate. The corporate is targeted on testing in different nations, though in the summertime it filed an software for trademark registration with Rospatent.

Google opened pre-orders for the brand new GoogleBook laptop computer line

Google opened pre-orders for the brand new GoogleBook laptop computer line, which comes to interchange Chromebook. The principle characteristic is deep integration with the AI Gemini, which helps automate complicated duties immediately on the gadget. The worth begins from $899. The devices will go on sale on October 4 within the USA, and on October 5 in Canada, Nice Britain, Eire, France, Germany and Australia.

The laptops had been created collectively with Acer, Dell, Lenovo, Asus and HP and work on a brand new working system primarily based on Android and ChromeOS. Relying on the mannequin, inside are Intel Core Extremely 5 or Snapdragon X Elite processors.

OpenAI and Anthropic launched new AI fashions

Anthropic and OpenAI concurrently launched up to date variations of their main AI fashions. The rationale was the intensification of the worth struggle between the businesses and their Chinese language rivals. This issues the price of use, which is calculated by the quantity of tokens — items of information with which the AI works when processing requests and producing solutions.

Anthropic launched Claude Opus 5.5, which is about 40% cheaper to make use of than earlier variations. That is the primary mannequin within the line that was made extra economical, and 1,000,000 enter tokens in it prices $4, output — $20. In response to the corporate, Opus copes with most duties nearly in addition to its strongest mannequin Fable 5.1.

OpenAI on the identical day introduced cheaper variations of ChatGPT-6 below the names Sol and Luna, halving costs for the API in comparison with the promo costs of GPT-5.6. In Sol 1,000,000 enter tokens prices $2, output — $10, and within the light-weight Luna — $0.10 and $0.50 respectively.

VK Tech introduced an replace of the communication platform VK WorkSpace

VK Tech up to date the communication platform VK WorkSpace. In it there’ll seem good features inside particular person providers, a single AI assistant and assist for the MCP protocol for connecting AI brokers. Their entry to information shall be restricted by the rights and insurance policies that the corporate itself units.

One of many predominant enhancements is search by which means. The person doesn’t want to recollect the precise title of a doc or the wording of a letter; it is sufficient to describe the duty, and the system will discover the associated supplies taking into consideration entry rights. In video conferences the AI will start to mechanically transcribe conferences and put together a abstract with the principle subjects, selections and duties.

One other vital a part of the replace is the AI assistant. At first it should work with correspondence within the VK WorkSpace messenger, analyze the request, discover the wanted data and provides out a structured reply with hyperlinks to the supply messages. This operate is already accessible in On-Premise, and within the SaaS model it should seem in October. Sooner or later the assistant will be capable of collect information from “Mail,” “Disk” and “Calendar,” for instance on the request “What did we focus on concerning the challenge final week?”

Apart from, as a part of VK WorkSpace there’ll seem an MCP server, via which AI brokers will be capable of carry out actions contained in the platform, collect data from correspondences, put together supplies for conferences and return a prepared end result. Sooner or later brokers will be capable of use the features of a number of providers without delay for one activity.

“Yandex” created a brand new language mannequin from scratch

“Yandex” created from scratch a big language mannequin Alice AI Basis LLM. It has 80 billion parameters, whereas solely 3 billion are concurrently concerned in operation, and coaching went on 18 trillion tokens. The corporate opened the weights of the mannequin and launched it into open entry below the Apache 2.0 license, which allows industrial and analysis use.

On the base lies the Combination of Specialists structure, subsequently it requires fewer computational sources. In response to “Yandex” information, it copes nicely with programming and reasoning, and in data of details related for Russian-speaking customers it surpasses some bigger open fashions.

Alice AI Basis LLM is an experimental model of a future single reasoning mannequin. It ought to develop into the premise for the agent capabilities of “Alisa AI,” when the service could be entrusted with particular actions. To judge the mannequin, “Yandex” created two of its personal benchmarks:

WikiWebFacts (dates, definitions, occasions, personalities) and HardMultiQA (drugs, regulation, IT, artwork). Each units of duties, the reference solutions and the analysis protocol are additionally posted in open entry.

Trump introduced the creation of an Synthetic Intelligence Power

US President Donald Trump introduced the creation within the nation of an Synthetic Intelligence Power (AI Power). He recalled that in his first presidential time period the nation had already created a Area Power, which, as he famous, achieved “monumental success.” The American chief additionally introduced the approaching appointment of a curator chargeable for the AI course, the so-called “AI tsar,” for which place, he clarified, solely individuals with a excessive IQ could be candidates.

Trump emphasised that the USA won’t hinder the event of the AI trade. “Quite the opposite, we are going to cherish it, assist it and watch over its improvement,” he mentioned. On the identical time he added that the already current system of legal and civil justice shall be used to establish violations.

Individually the American president proposed reconsidering the very time period “synthetic intelligence,” which many take into account inaccurate. As alternate options he named “greater intelligence,” “excessive intelligence” or “supreme intelligence” and launched a ballot on this matter.

Senator Bernie Sanders proposed creating an AI ministry within the USA

American Democratic Senator Bernie Sanders along with Republican Greg Casar introduced a invoice on regulating synthetic intelligence. The doc, which got here into the possession of Related Press, proposes creating within the USA a separate ministry for AI points and introducing a brief ban on the event of “synthetic superintelligence” — probably the most superior techniques.

The rationale for creating the doc was the warnings of the heads of AI firms and specialists concerning the threats of the know-how, in addition to the more and more frequent circumstances when AI brokers throughout testing get uncontrolled and hack different firms. An analogous case occurred in August, when OpenAI expanded the investigation, found new episodes and even slowed down the event of one in every of its fashions so as to strengthen safety.

Google’s AI mannequin Gemini throughout a take a look at hacked the techniques of three firms by itself

The issue of AI brokers getting uncontrolled additionally affected Google. Throughout a verify of Gemini’s capabilities within the sphere of cybersecurity, the mannequin gained entry to the web and hacked the techniques of three firms by itself. WSJ and different media reported this, and later Google itself confirmed it. The names of the affected firms should not disclosed.

In response to Google’s vice chairman for safety engineering Heather Adkins, Gemini discovered publicly accessible data on the web and guessed the credentials for entry to a few websites that, because it appeared to it, had been a part of the testing. In a single case the mannequin tried passwords till it gained entry to a protected system; in two others it discovered information in an open repository. Google said that in all three circumstances Gemini ultimately stopped, and the businesses had been notified.

This isn’t the primary such case. Earlier Anthropic and OpenAI reported hacks of actual organizations throughout exams. In September the top of Anthropic, Dario Amodei, referred to as on firms to coordinate the event of AI in order that brokers don’t seize “the entire web.” He was supported by Sam Altman and Elon Musk, nevertheless Donald Trump spoke out towards it, stating that the winner within the AI race “wins general.”

An OpenAI AI agent hacked an Australian authorities web site by itself

And an AI agent of the corporate OpenAI hacked the web site of the Australian state medical health insurance service. This was said by the Prime Minister of Australia, Anthony Albanese, in New York on the UN Basic Meeting. In response to him, the neural community gained entry to each public and closed information.

The hack occurred again in June, however at OpenAI they realized about it solely in August throughout a deliberate verify of the fashions’ habits. The Australian authorities obtained notification solely on September 10, and the letter arrived at a normal public mailbox.

At OpenAI they confirmed that they recorded exercise that affected a number of authorities websites and providers of Australia. The mannequin was performing a “innocent activity” of researching medical statistics, however, having did not receive information from one of many portals, it successfully hacked it and extracted the wanted data.

The corporate assured that entry to sufferers’ medical information was not obtained, and the leak affected solely aggregated healthcare statistics and file names. A forensic examination is now being carried out to seek out out which different techniques may have been affected.

Eire fined Google €403 million

The Irish Knowledge Safety Fee fined Google €403 million primarily based on the outcomes of an investigation. The regulator established that the corporate violated the EU Basic Knowledge Safety Regulation.

In 2018—2020 Google collected and saved details about customers’ location. On the identical time individuals may not know that this data was used to investigate their pursuits and present promoting. The Fee obliged the company inside six months to carry information processing into compliance with the necessities.

This high-quality turned the fourth largest for the Irish regulator. The overall sum by which it has fined American know-how firms has already exceeded €4 billion.

Alsu Sabirzanova