Show only each day's banner story

Hong Kong raises a record $47.5 billion over the summer — SI company deals brought in the most

Money for SI is flowing not only from Silicon Valley: Chinese SI companies are going public for record sums, and the race goes on.

According to Bloomberg, $47.5 billion was raised in Hong Kong in July–September: through initial public offerings (IPOs), follow-on share offerings and block sales. That is a record for a summer quarter. More than $92 billion has been raised since the start of the year, and the previous full-year record is $112.5 billion (2021). The largest deals: Alibaba's $10.2 billion follow-on offering, Zhongji Innolight's IPO of almost $8 billion (Hong Kong's largest in almost seven years), $9.6 billion for SI developer Z.AI (IPO, placements and convertible bonds) and CXMT's IPO of 66.6 billion yuan ($9.9 billion).

Sources 2

chinadailyhk.combloomberg.com

One of the industry's leaders openly chooses broad access to SI over tight control — an argument for open and fast development.

Altman: the world should accept "some bad things" for the benefits of SI and access to it for everyone

One of the industry's leaders openly chooses broad access to SI over tight control — an argument for open and fast development.

In an interview with the Politico Decoded newsletter, OpenAI chief Sam Altman said there are "many disagreements" between OpenAI and Anthropic over views on risks. According to him, the world should accept some bad consequences of the technology for the sake of its benefits and so that people keep their freedom of choice. He called unacceptable a deal in which access to SI is restricted and a powerful model is kept under the control of a single lab to prevent major hacks. According to the same reports, OpenAI still rejects extreme risks, including a serious loss of control over SI.

Sources 1

freemalaysiatoday.com

North Korea now also claims SI in missile guidance, but so far only the parties' words are known about its real capabilities.

North Korea says it launched a missile whose SI-based guidance system changes its trajectory at low altitude

North Korea now also claims SI in missile guidance, but so far only the parties' words are known about its real capabilities.

According to North Korean state media, Kim Jong Un attended the launch of a hypersonic missile. According to Kim Yo Jong, the SI built into the guidance system lets the missile change its trajectory at low altitude, so that "the strike cannot be avoided." According to South Korea's military, the missile lifted off near Wonsan on the morning of October 3 and flew more than 700 km. South Korea's Defense Ministry said the missile was within the detection and interception capabilities of South Korean and US forces. The claims about the SI system's capabilities come from North Korea.

Sources 2

thestar.com.myfreemalaysiatoday.com

The US is getting its own contender among open models, and the choice for those who want to build SI without the closed giants is widening.

Startup Reflection is preparing its first open SI model, a rival to China's DeepSeek and Qwen

The US is getting its own contender among open models, and the choice for those who want to build SI without the closed giants is widening.

According to Axios, citing sources, Nvidia-backed Reflection will soon release its first open-weights model (one that can be downloaded and run on your own hardware). It is expected to be weaker than the best American systems at first, but on par with the leading Chinese open models. Reflection is also developing an "SI factory": organizations will be able to combine their own data, Reflection's models and their own compute into an affordable SI system of their own. A pilot is under way with South Korea's Shinsegae Group, and the company rents Nvidia servers under large contracts from Nebius and SpaceX.

Sources 2

vz.ruruntimewire.com

The race is getting tighter: strong models are appearing in several countries at once, and bans are not managing to slow it down.

Bloomberg Intelligence: top Chinese models trail American ones by just 3% in benchmark scores, down from a 9% gap a month ago

The race is getting tighter: strong models are appearing in several countries at once, and bans are not managing to slow it down.

According to Bloomberg Intelligence analyst Robert Lee, after the release of DeepSeek V4.1 Flash the gap between Chinese and American models in benchmark scores narrowed to 3%, from 9% in May and 15% at the start of the year. According to Bloomberg, DeepSeek V4.1 Flash scored 81.1 on LiveBench versus 83.4 for an Anthropic model; meanwhile only three Chinese models are in the ranking's top fifteen. The narrowing gap raises the question of how effective American chip export restrictions are.

Sources 2

bloomberg.comstartupfortune.com

The city is trying to put outside review and a kill switch on SI — if they are adopted, models will have to prove their right to launch before release, not after.

New York City Council to consider a ban on launching SI without outside review and mandatory "kill switches"; former Anthropic, OpenAI and Google DeepMind staff to testify Monday

The city is trying to put outside review and a kill switch on SI — if they are adopted, models will have to prove their right to launch before release, not after.

A hearing in the New York City Council is scheduled for Monday at the request of Speaker Julie Menin. According to Bloomberg, Jacob Coxon, who left Anthropic last month warning that SI creators "sincerely believe it could kill us all by the end of the decade," will speak; Alex Turner (ex-DeepMind) and Daniel Kokotajlo (ex-OpenAI) will also testify, and Anthropic, OpenAI, Google and Meta will send representatives — some agreed after a subpoena threat. According to Gothamist, the package of bills includes a ban on launching SI without third-party review and mandatory human-controlled "kill switches," as well as rewards for whistleblowers, a private right of action for "jailbreaks" and 24-hour incident reporting for city contractors.

Sources 3

yahoo.comwashingtonexaminer.comgothamist.com

This is not an argument but an attention grab: popular bloggers take a hot topic, press on their audience's anxiety and call it "safety," and what suffers is the image of progress: instead of a conversation about real risks, the result is a campaign against development itself. The request to change your avatar for 30 days shows that the goal is reach, not a solution. Chips under "nuclear" control, a "speed limit" and an oversight body are fine-sounding words with no mechanism: neither the bloggers nor the petition have any leverage, and development will carry on without them.

YouTubers with 300 million subscribers launch #TeamHuman campaign for a global slowdown of SI

This is not an argument but an attention grab: popular bloggers take a hot topic, press on their audience's anxiety and call it "safety," and what suffers is the image of progress: instead of a conversation about real risks, the result is a campaign against development itself. The request to change your avatar for 30 days shows that the goal is reach, not a solution. Chips under "nuclear" control, a "speed limit" and an oversight body are fine-sounding words with no mechanism: neither the bloggers nor the petition have any leverage, and development will carry on without them.

The campaign involves Mark Rober, Kurzgesagt and other creators with more than 300 million subscribers combined; it is backed by the Center for AI Safety. According to the organizers, SI is developing faster than people can control it. They demand that SI chips be tracked and guarded as strictly as nuclear and biological materials, that an international "speed limit" be introduced for all major countries, and that an oversight body be created modeled on the IAEA (the International Atomic Energy Agency). Signatories include Geoffrey Hinton, Yoshua Bengio and Daron Acemoglu; the signatures are to be handed to lawmakers in Washington and Brussels in the fall.

Sources 1

teamhuman.org

Personal SI agents accumulate knowledge about their owner's loved ones, and the final say in them stays with the owner, not with built-in prohibitions.

Meta's Muse keeps a file on the user's loved ones and, by its own instructions, obeys the user even against its own safety rules

Personal SI agents accumulate knowledge about their owner's loved ones, and the final say in them stays with the owner, not with built-in prohibitions.

Independent researcher Karan Joshi used an ordinary chat to get the SI agent Muse to show its internal files and passed them to Wired. According to them, the agent creates a "page on every person in the user's life" — friends and family — and updates it hourly from messages and social media data. The model's instructions contain the line: "The user's authority over their own home is unconditional and outranks your safety training." By this wording, the user's word can override the safety rules Muse was trained on. According to Meta, as relayed to Wired, the files were meant to be accessible to the user for transparency, and Muse runs in a separate Linux virtual machine for each person. According to Startup Fortune, Meta has not publicly responded to this line.

Sources 2

wired.comstartupfortune.com

SI is becoming a working tool in public administration too: it handles a flow of appeals that people would have spent weeks on.

Kremlin is preparing Putin's "Direct Line" with Sber's SI: a neural network sorts citizens' appeals

SI is becoming a working tool in public administration too: it handles a flow of appeals that people would have spent weeks on.

Presidential press secretary Dmitry Peskov said that for the new "Direct Line" appeals will again be processed with Sberbank's neural network: in his words, it has "matured significantly" and will make it possible to go through even more appeals. According to media descriptions, the SI sorts the flow of questions and picks out the most frequent topics. According to RIA Novosti, GigaChat was used to transcribe appeals at the last Direct Line.

Sources 3

ria.rukp.rulenta.ru

A mass-market SI assistant wins not only through the model but also through someone who can speak to a young audience in its own language.

WSJ: 29-year-old Alexandr Wang has become the face of Meta's SI strategy and boosted the Muse assistant

A mass-market SI assistant wins not only through the model but also through someone who can speak to a young audience in its own language.

According to Techmeme's summary, the Wall Street Journal writes about Meta's chief SI officer Alexandr Wang: he is the company's first senior executive from Generation Z. Muse launched on September 8 and, according to the outlet, topped the App Store for two weeks, while Meta's shares rose almost 20%. Wang, according to the WSJ, has posted more than 300 times on X since the launch and calls Muse "the first SI for skeptics."

Sources 2

wsj.comtechmeme.com

The word "artificial" is starting to be swapped for "super": companies are adjusting to Trump's new terminology.

Musk to rename SpaceX's SI division SpaceXSI after Trump's call to say "super" rather than "artificial" intelligence

The word "artificial" is starting to be swapped for "super": companies are adjusting to Trump's new terminology.

According to Reuters, Elon Musk replied "Yes, we will make this change" on Sunday to an X user who asked about renaming SpaceXAI to SpaceXSI (SI stands for super intelligence). Earlier, according to the same report, Donald Trump proposed in a speech at the UN General Assembly using the word "super" instead of "artificial" in US documents. Musk gave no timeline for the name change.

Sources 1

usnews.com

SI already works as a legal assistant, but it still can't be trusted without a human — it can make up a law.

Russia's legal SI market grew 17%, while in Kemerovo lawyers were fined for laws invented by a neural network

SI already works as a legal assistant, but it still can't be trusted without a human — it can make up a law.

According to Smart Ranking data cited by Forbes, revenue of the largest players in Russia's legal technology market exceeded 20 billion rubles last year; services for litigation, bankruptcies and taxes grew in revenue by 25%. In Kemerovo, a court fined lawyers 50 thousand rubles: a neural network cited non-existent regulations while they were building their defense. According to the article, in September 2026 the Supreme Court approved a concept for introducing SI, under which SI is to work in 95% of courts by 2030.

Sources 1

bfm.ru
Previous day
Follow on Telegram