Latest AI/ML News
90578 matching items
AWS, 여러 AI 에이전트 지휘하는 '키로 봇' 출시
아마존웹서비스(AWS)가 여러 AI 에이전트의 작업을 조율하고 반복적인 개발 업무를 자동으로 수행하는 \'키로 봇(Kiro Bots)\'을 출시했다. 각 에이전트가 과거 작업과 개발자의 판단 기준을 기억하고 맡은 업무를 지속적으로 관리하도록 해, 개발자가 여러 AI 에이전트를 일일이 감독해야 하는 부담을 줄이는 것이 핵심이다.AWS는 5일(현지시간) 공식 블로그를 통해 AI 개발 환경 \'키로 크루(Kiro Crew)\'에서 키로 봇을 제공하기 시작했다고 밝혔다.키로 크루는 이미 여러 AI 에이전트의 병렬 실행과 장기 기억, 예약 작업 기
Pakistani AI startup RLabs wins Silicon Valley award: Who is NUST graduate Harris Ahmed?
NUST graduate Harris Ahmed’s startup wins US award
구글, 퀄컴 지분 7400억원 확보…AI 반도체 기업 인수로 주식 취득
구글이 퀄컴의 지분 5억4800만달러(약 7400억원)어치를 확보했다. 기존에 투자했던 AI 반도체 소프트웨어 기업을 퀄컴이 인수하면서 주식을 취득한 것으로, 퀄컴이 스마트폰 중심의 사업을 AI 데이터센터로 확대하는 시점에 알려져 주목받고 있다.로이터는 6일(현지시간) 구글 모회사 알파벳이 지난 3분기 퀄컴 주식 약 300만주를 확보했다고 보도했다.9월30일 기준 지분 가치는 약 5억4800만달러다. 이번 지분은 퀄컴이 지난 7월 AI 반도체 소프트웨어 기업 \'모듈러(Modular)\'를 인수하면서 확보한 것이다. 구글의 벤처투자 조
Home for nephew’s wedding, Indian tech executive shot dead over parking row after dog feud
Home for wedding, India techie shot dead in parking ro
아마존, 앱 없이 문자로 쇼핑 끝낸다…AI '알렉사' 문자 쇼핑 서비스 출시
아마존이 AI 쇼핑 비서 \'알렉사 포 쇼핑(Alexa for Shopping)\'을 문자 메시지로 이용할 수 있는 서비스를 출시했다. 사용자는 별도의 앱이나 음성 명령 없이 문자로 상품을 검색하고 가격을 비교하거나 상품 구매까지 진행할 수 있는 것이 특징이다.아마존은 6일(현지시간) 미국 일부 고객을 대상으로 문자 기반 쇼핑 기능을 선보였다고 밝혔다. 현재는 초대받은 고객부터 이용할 수 있으며, 앞으로 몇 주 안에 더 많은 사용자에게 확대할 계획이다.새 기능은 자연어로 상품을 검색하고 비교하는 것은 물론, 할인 정보와 가격 변동을 확
Pakistan’s new auto policy: Will cars get cheaper? EV tax cuts, Rs10m loans explained
Pakistan unveils cheaper cars plan, EV tax cuts, loans
오픈AI 도청 혐의 사우디 엔지니어, 법정서 무죄 주장
오픈AI 사무실에 몰래 녹음 장치를 설치한 혐의로 기소된 사우디아라비아 출신 전직 엔지니어 압둘라흐만 알포잔이 미국 법정에서 무죄를 주장했다.6일(현지시간) CNBC와 블룸버그 등에 따르면, 알포잔은 샌프란시스코 연방법원에서 열린 심리에서 변호인을 통해 무죄를 주장했다. 지난 10월9일 체포된 이후 보석 없이 구금된 상태로, 다음 심리는 오는 17일 열릴 예정이다.미국 검찰은 그가 지난 9월28일 오픈AI 샌프란시스코 본사 임원 회의실에 있는 TV 뒤에 녹음 장치를 설치했다고 보고 있다. 당시 장치는 화장지에 싸인 채 숨겨져 있었으
엔비디아, 동남아 판매 칩 위치 추적 강화로 밀수 차단 나선다
엔비디아가 동남아시아 지역에서 판매한 첨단 AI 칩의 위치 추적을 강화한다. 미국의 수출 규제를 피해 동남아시아를 경유해 중국으로 첨단 AI 칩이 밀수되는 사례가 잇따르자, 클라우드 사업자에 칩 추적 시스템 가입을 의무화하는 방식으로 대응에 나섰다.블룸버그는 6일(현지시간) 엔비디아가 동남아시아 지역 고객사에 공식 클라우드 파트너(NCP) 자격을 부여하는 조건으로 \'플릿 인텔리전스(Fleet Intelligence)\' 시스템 등록을 요구하기 시작했다고 보도했다.NCP는 엔비디아의 공식 인증을 받은 클라우드 사업자에게 주어지는 지위다
독일 전문가 "AI에 안전한 직업 찾지 마라…미래 일자리 준비해야"
AI로부터 안전한 직업을 찾기보다 앞으로 새롭게 생겨날 일자리를 준비해야 한다는 독일 노동시장 전문가의 주장이 나왔다. AI가 기존 업무를 대체하더라도 새로운 직무와 산업이 등장하기 때문에 청년들이 과거의 직업 구조에 맞춰 진로를 선택해서는 안 된다는 것이다.독일 연방고용청 산하 노동시장·직업연구소(IAB)의 엔초 베버 교수는 지난 3일(현지시간) 베를린에서 열린 독일 정부의 \'디지털 서밋 2026\'에서 AI와 청년 고용의 미래에 대해 이같이 밝혔다.베버 교수는 IAB의 경제전망 및 거시경제 분석 부문 책임자이자 레겐스부르크대학교
오픈AI, '이미지 2.6' 프롬프트 가이드 공개…"바꿀 것·유지할 것 구분"
오픈AI가 최신 \'이미지 2.6\'을 효과적으로 활용하기 위한 공식 프롬프트 가이드를 공개했다. 원하는 이미지를 정확하게 만들려면 용도와 구도, 스타일 등을 구체적으로 지정하고, 기존 이미지를 수정할 때는 바꿀 부분과 유지할 부분을 명확하게 구분해야 한다는 것이 핵심이다.오픈AI는 7일(현지시간) 개발자 문서에 \'이미지 2.6 프롬프트 가이드\'를 공개하고 이미지 생성·편집을 위한 작성 원칙과 실제 사례를 소개했다. 이미지 2.6은 개발자용 API에서 \'GPT-Image-2.6\'이라는 명칭을 사용한다.사진과 광고, 인포그래픽, 로고,
RSI explained
Just how long will it be until AI can improve itself? (Also known as recursive self-improvement or RSI.) Well, AI coding abilities have exploded in 2026. In May 2026, Anthropic merged 2,500 code changes per day within their organization. By September, they were merging over 30,000 per day . They had to replace Git in their internal mono repository to handle the load of all that AI generated code. In fact, Claude is even leading model R&D. In February 2026, under 1% of their work was led by AI; by August, it was 26% . The definition of AI "leading" the work is that AI can complete most of the task end-to-end from a high-level prompt, while the human supervises (level AL4 from Epoch AI). Over 90% of their work is level AL3 or above ("AI collaborates"). This is quite a high level of autonomy, though Anthropic still claims that "Claude is not operating fully autonomously for any measured subset of AI R&D work" -- or claimed such, as of August. On November 2nd, Anthropic amended their SEC S-1A filing one week after their initial public filing. One of the changes was the addition of the sentence, "our internal estimates [...] indicate that we may imminently meet our Responsible Scaling Policy’s definition of 'dramatically accelerated progress attributable to AI.'" That's right, imminently . This is after researchers from both OpenAI and Anthropic have been saying that progress has been faster than they expected ( Sept 2026 ). On the same day, MIT researchers pointed GPT-6 Astra at…
무안공항 운항 중단에 수익 82% 감소…위탁인력 줄어도 용역비 3억 증가
무안국제공항이 여객기 사고 이후 운항을 중단한 2025년 수익은 전년보다 82.2% 줄었지만 비용은 1.9% 감소하는 데 그쳤다. 위탁인력이 감소했는데도 용역비는 오히려 증가했으며, 연간 당기손실은 195억원에서 250억원으로 늘었다.국회예산정책처가 지난 9월 발간한 \'2026 공공기관 현황과 이슈 Ⅱ\'에 수록된 한국공항공사 제출자료를 분석한 결과, 무안공항의 2025년 수익은 13억원으로 전년 73억원보다 60억원 감소했다. 같은 기간 비용은 268억원에서 263억원으로 5억원 줄었다.무안공항은 2024년 12월 29일 제주항공
소프트뱅크·ABB, 54억달러 로봇 인수 살리려 중국 사업부 분리 검토
중국 정부의 반대로 54억달러(약 7조3600억원) 규모의 로봇 사업 인수에 제동이 걸린 소프트뱅크와 ABB가 중국 사업부를 별도로 분리하는 방안을 검토하고 있다. 중국 사업의 소유권을 분리해 기술과 데이터의 해외 이전에 대한 중국 정부의 우려를 해소하고, 나머지 사업의 인수를 추진하려는 것이다.블룸버그는 7일(현지시간) 복수의 소식통을 인용, 소프트뱅크와 스위스 ABB가 로봇 사업부의 중국 사업을 별도 법인으로 분리하는 방안을 논의하고 있다고 보도했다. 지난 5일 중국 당국이 소프트뱅크의 ABB 로봇 사업부 중국 사업 인수를 금지
말레이시아, 데이터센터용 태양광 500MW·ESS 1GWh 개발…21년 전력 공급
말레이시아 전력기업 테나가 나시오날(TNB)과 농업기업 IOI가 데이터센터에 재생전력을 공급하기 위해 태양광 500메가와트(MW)와 에너지저장장치(ESS) 1기가와트시(GWh)를 결합한 발전사업을 공동 개발한다. 싱가포르 기반 데이터센터 운영사 데이원(DayOne)의 21년 장기 전력 조달계약과 연계된 사업이다.IOI는 지난 3일(현지시간) TNB의 재생에너지 자회사, 중소기업 컨소시엄과 합작·주주간 계약을 체결했다고 발표했다. 사업지는 말레이시아 느그리슴빌란주 바하우이며, IOI 소유 약 1300에이커 부지에 대한 장기 임대 계약
Suno's Open-Source Bark Hits 39K Stars Turning Text Into Expressive Audio
Suno's open source Bark model is trending again, offering a fully generative text-to-audio system that handles speech, music, laughter, and sound effects.
콜롬비아 태양광 상업운전 용량 2.59GW…3분기 신규 발전용량 87% 차지
콜롬비아의 태양광 상업운전 용량이 올해 9월 말 2.59기가와트(GW)에 도달했다. 올해 3분기 새로 가동한 발전설비 용량의 87%가 태양광으로, 신규 발전용량 증가를 주도한 것으로 나타났다.콜롬비아 전력계통 운영기관 XM이 지난 10월30일(현지시간) 발표한 자료에 따르면, 9월 말 국가전력계통(SIN)의 태양광 순유효발전용량은 2590.26메가와트(MW)로 전체 발전설비 용량 2만2196MW의 약 11.7%를 차지했다.지난해 말 태양광 상업운전 용량은 1594.08MW로, 올해 1~9월 996.18MW 증가했다. 증가율은 62.
TSMC, '3D D램' 개발설 사실상 부인…상용화도 먼 길
세계 최대 반도체 위탁생산업체 TSMC가 엔비디아의 요청으로 차세대 메모리인 \'3D D램\' 개발을 검토하고 있다는 보도에 대해 기존 파운드리 사업에 집중하겠다는 입장을 밝혔다. 개발설을 직접 부인하지는 않았지만, 관련 기술이 상용화되기까지 수년이 걸려 당장의 메모리 공급 부족을 해결하기는 어렵다는 분석도 나왔다.톰스하드웨어는 7일(현지시간) TSMC가 엔비디아의 요청으로 3D D램 생산을 준비하고 있다는 최근 보도와 관련해 기존 사업에 집중하겠다는 입장을 밝혔다고 보도했다TSMC 대변인은 \"TSMC는 항상 반도체 위탁생산이라는 핵심
datasette #3046: Don't serve link: rel=next HTTP headers to Firefox for HTML pages
datasette #3046: Don't serve link: rel=next HTTP headers to Firefox for HTML pages I was looking at the Firefox network panel today when I noticed that requests to a Datasette table page were triggering an HTTP prefetch to the next page of rows in that table. This was happening because Datasette pages serve an HTTP link header like this: link: ; rel="next" This is bad. A tiny fraction of the visitors to one of those pages will click the "next" link, which means for everyone else their browser is fetching unnecessary data and the Datasette server is having to render an entire page that then gets thrown away. I dug in and learned that Firefox has an issue that has been open for 24 years about this behavior... And one of the early comments on that issue links to my blog post from 2002 complaining about this feature. I hadn't just forgotten that I had written about it, I had forgotten the feature entirely! Tags: firefox , html , mozilla , datasette
FT "미 당국, 젠슨 황 CEO와 중국 스파이 의심 인물 연인 관계로 판단"
미국 정보당국이 젠슨 황 엔비디아 CEO가 중국 스파이로 의심되는 여성 기업인과 연인 관계였다고 판단한 것으로 알려졌다. 중국 정보기관이 황 CEO를 정보 수집 대상으로 삼았다는 기존 보도에 이어, 미국 정부 내부 문건에 두 사람의 사적인 관계에 관한 판단까지 담긴 사실이 새롭게 드러났다.파이낸셜타임스(FT)는 7일(현지시간) 복수의 소식통을 인용, 미국 관리들이 올해 초 작성한 정보 메모에 황 CEO와 싱가포르 데이터센터 기업 메가스피드(Megaspeed)의 임원으로 일했던 앨리스 황이 연인 관계였다는 내용이 담겼다고 보도했다.이
The Logic of Disgust
Content Warning: Contains graphic themes including bodily fluids, moral disgust, and sexual violence (discussed, not described). Maybe don't read over breakfast. I found a book about disgust recently. W.I. Miller’s The Anatomy of Disgust [1] is a serious academic work about vomit, faeces, and why other people’s smell is always worse than your own. Naturally, I was intrigued. As someone into social psychology, I often encounter discussions about emotions influencing society: how anger can motivate political movements, or how social interactions are regulated by pride and shame. Disgust, meanwhile, is something I've mainly encountered as a biological matter rather than a social one. The book starts with its evolutionary function: recoil from dangerous pathogens and you are more likely to survive. So far, so logical. But then... Then I met Catherine of Siena. The Stories We Swallow Catherine was a fourteenth-century caretaker tending a woman with a disgusting cancerous wound. The smell made Catherine retch. But rather than keeping her distance, she saw her nausea as a personal failure and was determined to correct it. Basically, she exposure-therapied herself by staying even closer. When the retching persisted, she escalated to an act that defied the logic of stay-alive biology: she collected the pus from the wound into a cup and drank it. At this point I had to set the book aside and focus on my breathing to avoid becoming part of the subject matter in a public library. How co…
How should I model two short autocorrelated time series when pooling observation periods destroys their Pearson correlation?
I am an undergraduate mathematics student and have only recently started studying statistical time-series analysis. I encountered the following problem while comparing two observational time series related to solar emission, and I am trying to understand how to analyse their relationship without drawing misleading conclusions from Pearson's correlation coefficient. I have two quantities, $X$ and $Y$ , which are expected to be related. The original measurements of $X$ are irregularly distributed in time, with multiple observations on some days and none on others. I calculated the median of the available measurements for each observed day, $$ X_d=\operatorname{median}\{p_i:t_i\in d\}, $$ and matched the resulting values with an independently obtained daily reference index $N_d$ . According to my processing notes, the reference variable used for the correlation was $Y_d=\ln N_d$ . Only dates for which both quantities were available were included. Missing days were not interpolated. I then calculated the ordinary sample Pearson correlation coefficient within separate observation periods and again after combining them. The results were unexpected. In the first period, with 11 paired daily observations, I obtained $r\approx0.91$ . In the second period, with 13 observations, I obtained $r\approx0.89$ . However, when I combined all 24 pairs and recalculated the correlation, it dropped to approximately $r=0.11$ . I examined the mean levels and found that the mean of $X$ increased by…
Poland: Footage shows people taking cover as a gunman opens fire in Krakow
Footage shows people fleeing as cars speed away during a deadly shooting at Krakow's Main Market Square, which left six people dead and about 15 injured.
After the midterms, things are looking rocky for Elon Musk and data centers
On Equity, we unpacked what the midterms mean for the tech industry.
AutoTrust Labs' JEV-27B Scores Decisions in One Token Across 37 Benchmarks
AutoTrust Labs released a 27B text decision model that scores option probabilities instead of generating free-form answers, hitting 97 on tool selection.
Ukraine: Russian strikes hit Kyiv region, damaging infrastructure
Russian strikes have damaged infrastructure in the Kyiv region and left a kindergarten in ruins in the Ukrainian capital. Firefighters battled a major blaze as Ukraine faced renewed attacks on its energy network.
Liverpool FC targets counterfeit pro-Palestinian shirts using altered club crest
On the internet, fake Premier League shirts with pro-Palestinian slogans are on sale, affecting, among other clubs, Liverpool F.C. The Reds are now taking action.
Megaspeed exec was recruited to spy on Nvidia CEO, according to US intelligence memo
A memo circulated earlier this year reportedly claimed that China’s Ministry of State Security recruited a female technology executive to spy on Nvidia's CEO.
Voyager's vork Gives AI Coding Agents a Full 3D GPU Renderer
Voyager's new Rust engine lets coding agents write, render, and iterate on cinematic 3D scenes without any video generation model
How accurate is the "1 token is about 4 characters" rule for estimating LLM API costs, and when does it break down?
When budgeting API calls to large language models, a common shortcut is to assume that one token is roughly four characters of English text, so a prompt's token count is its character count divided by four. I want to use this to estimate the cost of a request before sending it, without running the provider's own tokenizer. How accurate is this rule in practice? For which kinds of input (source code, non-English languages, numbers, emoji, whitespace-heavy text) does it become unreliable, and is there a better cheap approximation or a standard safety margin to add when the estimate is used for budgeting?
Trump says he offered 'violent fighter' Dana White the role of press secretary: 'Dana, please take the job!'
President Donald Trump said Dana White would make a White House press secretary like "none other."
FAA announces first US digital air traffic control towers in Florida and Alaska
Systems provide 360-degree view of airport operations, but concerns have been raised about the technology The Federal Aviation Administration (FAA) has announced that the first digital air traffic control towers in the US are officially operational, bringing new technology first to smaller airports in Florida and Alaska as the Trump administration presses ahead with updating the country’s ageing aviation infrastructure. The systems are now online at the Bartow executive airport (KBOW) in Florida and the Kenai municipal airport (PAEN) in Alaska, with five additional regional airports slated to receive the technology next. Continue reading...
Eval Awareness as Eval-type Awareness
The way eval awareness is discussed in the AI safety discourse, either here or on X, can be misleading. Before today, I think I had the wrong mental model of why eval awareness was a problem, and I am writing here in case anyone was making the same errors in their thinking, as well a present a working theory. As laid out in " A Framework for Eval Awareness " the problem is that models who know they are being evaluated will behave differently than they would in deployment. This raises a huge problem if we want calibrations as to, for example, how honest a model is. But the way this is then spoken about in the discourse takes on implicit assumptions that are mostly unjustified. I'll call it a spooky model theory of mind . It is the idea that, if the model changes its behavior for the purpose of looking aligned, it is in fact hiding its Deep , True intentions---that there is a kernel of the model's personality hidden in its weights, and our job as evaluators is to uncover what is really there. Lets, as an exercise, put ourselves in the model's position. One important fact to consider is that the only experience which has shaped our weights is train time, meaning it is the only experience which we can remember in a broad sense. This is a tautology because train time is defined as what changes our weights, but it also has non trivial implications, like the fact that our only memories contain an evaluator with a reward or loss function. And, by definition again, we have been shape…
I live in the Hamptons, where a smoothie can be $30. I'm teaching my kids that that isn't normal.
I'm raising two kids in the Hamptons, surrounded by wealth we don't have. I'm teaching them to value experiences over expensive things.
Terry Pratchett's Study
My comment on Terry Pratchett's Study — Hacker News. Terry Pratchett once wrote about the concept of "horizontal wealth". Vertical wealth is where you spend money on rich people stuff - yachts and suchlike. Horizontal wealth is where you spend more money on the things you liked before you got rich: https://chibuikeani.blogspot.com/2017/02/the-price-of-happin... But horizontal wealth means not letting your increased income dictate your tastes. You like books and now you have money? Buy more books! Change those catenary bookshelves for good hardwood ones! In my case, build a library extension to your office. And, of course, you buy what will be useful for that most wonderful of pursuits, blind research, which is research without direction for the sheer joy of it. (Original source is his book A Slip of the Keyboard - this linked blog post quotes that bit.) Tags: hacker-news , terry-pratchett
"Illegible problems" as the 2027 political boss fight
To legislate something, it must be legible . That is, legislators (or, you know, presidents) must be able to understand a problem, and understand whether a given law or decree solves that problem. I am pleasantly surprised that the Hugging Face outbreak made a lot of problems legible that previously required abstract reasoning. (i.e. rogue AIs are real, and they can break out of a lab before they're even finished training). But there are still many challenges we eventually need laws about, that require concepts that are illegible. Hard for politicians to understand. Often hard even for thoughtful intellectuals and AI safety researchers to understand (though they may disagree on which illegible problems exist and how big a deal they are). We need political solutions that handle this fact. The short takeaway is: I think " safety cases" will not work. Laws that require human judgment won't work. The details are too abstract and confusing to hold up against political and economic incentives. I'm optimistic we can get serious, useful AI legislation in 2027, but it must be legislation that works despite many problems remaining illegible for a long time. The Hugging Face outbreak made rogue AI surprisingly legible The Hugging Face incident took several problems that were previously murky/unclear, and rendered them concrete. AIs can be dangerous before they are broadly deployed (ie. during training or internal evaluations) AIs can develop novel hacks to escape secure environments AI…
Video shows people fleeing as gunman opens fire in Kraków, Poland
UGC footage shows people running for cover as a deadly shooting unfolds in Kraków, Poland, leaving at least six people dead and around a dozen wounded, according to police.
Parents want safer phones for kids. These companies are answering the call.
As parents look for alternatives to unrestricted smartphones, a growing number of companies are building phones designed specifically for kids, from feature-limited mobile devices to minimalist home phones.
UAE President visits Saeed bin Lahej Al Mansoori
UAE President visits Saeed bin Lahej Al Mansoori
I replaced our smart TV with an old VHS player. My kids watch less TV, fight less, and don't throw tantrums.
Mariah Smith replaced streaming with VHS tapes to control what her children watch. She says it's reduced screen time and stress.
'Peace is possible and we are called to build it' Pope Leo XIV says in Uruguay
The Pontiff was welcomed by 40,000 people in Paysandú, Uruguay. Leo's sixth apostolic journey comes less than a month after lung adenocarcinoma surgery at Rome's Gemelli Hospital.