US-Iran truce faces toughest test after fresh strikes: Why it matters
Truce tested by fresh escalation as Trump warns Iran
AI/ML news, top picks, and generated innovation digests.
35507 matching items
Truce tested by fresh escalation as Trump warns Iran
In this Sunday edition of Business Insider Today, we're talking about the ungainly quality of Slate's new EV truck and why some people would buy it anyway.
메타가 최근 폭발적으로 성장 중인 ‘예측 시장’에 도전하기 위해 자체 앱 개발에 나섰다는 소식에 이어, 이 분야 선두주자들과의 협력에 나선 것으로 알려졌다.26일(현지시간) 뉴욕타임스에 따르면, 마크 저커버그 CEO는 최근 참모진에게 폴리마켓(Polymarket)과 칼시(Kalshi) 등 대표 플랫폼과의 제휴 방안을 모색하라고 지시했다.이에 앞서 저커버그 CEO가 사내 소규모 팀을 통해 예측 시장 플랫폼을 모델로 한 독립형 모바일 앱 ‘아레나(Arena)’를 비밀리에 개발해 왔다는 소식이 전해졌다. 구체적인 협력 형태는 알려지지 않
It was 1977, two years after Franco’s death, when a group of people made history for Spain’s LGTBIQ+ community by marching on Barcelona’s Ramblas after years in the shadows.
From fish and chips to fine dining, New York City is in the middle of a British restaurant boom.
Researchers at Princeton University built CEO-Bench, a test where AI agents have to run a fictional software company for 500 simulated days. Most current models go broke, and a simple rule-based heuristic with no AI beats nearly all of them. The article Only three AI models finished above starting capital in a 500-day startup survival test appeared first on The Decoder .
Moving to Japan presented challenges in practices like gift-giving and limited sick leave, and reshaped what I expected from work.
Sun hoodies are the greatest new garment since the original hoodie.
Catch up with the most important stories from around Europe and beyond this June 28th, 2026 - latest news, breaking news, World, Business, Entertainment, Politics, Culture, Travel.
Time to switch back to paper and harvest that suddenly valuable RAM
The Spanish Ministry of Finance has updated its list of non-cooperative jurisdictions, removing Gibraltar after 35 years and adding Russia, in line with the European Union's position
"I don't live in a van, eat rice and beans, and I've never given up traveling," said one early retiree. "That's not what the movement is about."
Eliana Jordan left office life, became a scuba instructor, and later taught herself to code to build something of her own.
Ukraine's air force said local air defences had shot down six Russian ballistic missiles, one anti-ship missile and 125 drones overnight.
"We wanted to take our revenge against Austria": Algeria fans react in Algiers after the draw against Austria secured the Fennecs' place in the World Cup round of 32.
360 founder Zhou Hongyi presents two AI security tools designed to compete with Anthropic's Mythos. One has already flagged 3,432 vulnerabilities. Zhou admits Chinese models trail Western ones by 20 to 30 percent, but compares Mythos to "cyber nuclear weapons" and calls for China to build its own strategic deterrent. The article Chinese cybersecurity firm builds AI tools to rival Mythos and frames the race as cyber-nuclear deterrence appeared first on The Decoder .
The Euclid space telescope's stunning photo of our galaxy's “crowded heart” captures more than 60 million stars.
Eating nutritious food can get relegated to the bottom of your to-do list when you're busy, but this can be counterproductive. Try these tasty recipes, instead.
A product designer share how embracing her inner "mad scientist" and experimenting with AI helped her land a job at Adobe, with no formal background in machine learning.
Germany is groaning under the year's first heatwave. Berlin police are resorting to an unusual tactic, sending two water cannons onto city streets normally reserved for protests.
An unadorned ebike is a blank canvas. Here, get tips for maximizing its cargo-hauling and person-carrying capabilities.
The Chinese supercomputer LineShine was ranked as the fastest in the world, despite not using any GPUs.
Der Einsatz von KI verändert den Arbeitsmarkt. Der Chef der Regionaldirektion der Bundesarbeitsagentur warnt allerdings vor Panikmache.
A preliminary estimate from the United Nations Development Programme put the cost of direct physical damage caused by the quakes at $6.7 billion.
Several events have been canceled or postponed in the country to avoid overloading hospitals which are already near their breaking point, especially in the Paris region.
Abu Dhabi's G42, Core42 and MGX join Washington Pax Silica Summit
Doctrinal chief in Rome says the European Union applies international law selectively, sanctioning some military invasions but not others.
Divers have found the wreck of the Hofuku Maru off Luzon in the Philippines. The Japanese freighter sank in 1944 with over 1,200 Allied POWs; a digitised military file finally solved the mystery.
Sina Weibo's VibeThinker-3B has just three billion parameters but matches models like DeepSeek V3.2 and Kimi K2.5 on math and coding benchmarks. Those models are up to 333 times larger. The secret isn't size but multi-stage post-training. The researchers propose a hypothesis based on their findings: logical reasoning compresses well into small models, but broad world knowledge does not. The article Sina's open model VibeThinker-3B aims to show reasoning compresses well but factual knowledge doesn't appeared first on The Decoder .
Temperatures have topped 40°C across parts of Europe this week, piling pressure on hospitals and first responders amid a surge in emergency callouts.
In this tutorial, we build a stable workflow around the Fable 5 Traces dataset from Hugging Face. We avoid fragile dependencies and manually parse the merged JSONL file to keep Colab reliable. We inspect repository files, normalize tool calls, audit structure, redact secrets, and visualize key distributions. We also export safe no-CoT chat datasets and train pure-Python Naive Bayes baselines on the traces. The post Building a Stable Fable 5 Traces Workflow in Colab: Parsing Tool Calls, Auditing Data, and Training Baselines appeared first on MarkTechPost .
Die Aufsichtsbehörden der G7-Staaten verlangen einen strikten Grundrechtsschutz bei der Online-Altersüberprüfung und nehmen vernetzte Kinderzimmer ins Visier.
I've seen lots of so-called "smart" bike locks over the years, but none so far could justify the added cost. A newcomer that got its start securing ATMs for banks is trying to change that. There's nothing wholly unique about the TMD Chain Lock, but the combination of materials, performance, and insurance-friendly ART-2 certification makes […]
Drafted March 6 2025 I have thought for many years that it is a risky thing to have a partner who is into creating things, lest you be called upon to appreciate those things, and everyone’s preferred narrative about the situation slams head-on into the realistic odds that you especially love their compositions. Especially if you are a kind of curmudgeonly appreciation-prude with an apparent all-round disinclination for having any particular feeling you are meant to have. Friends share songs with me, and I awkwardly tell them that the songs seem probably good but the jury is really out for me until I have heard them a few times. It’s worse when a friend has written the song themselves, because then I feel compelled to really listen to it, rather than just having it on in the background. Yet I so quickly tire of it. Once I have heard a song a few times, I do often like it. Though this feels more like the liking of liking a good pastry, not the liking of liking God dissolving your brain in a vat of transcendence. Then sometimes there are songs that lap into my soul and pulsatingly undo me, open me onto some other world of being. My body spasms and I gasp for breath. My mouth opens. If you looked at me you’d think I was having a pretty exciting sexual time of it. And perhaps I do feel some exquisite pain of sensibility through the intricacies of my body. Far from every time, but for instance even after hearing it a lot of times, Au Fond du Temple Saint would pull me relieved fro…
Croatia finished second in Group L after securing a 2-1 win over Ghana on Saturday, with goals for Nikola Vlasic and Petar Sučić.
From lesser-known UNESCO World Heritage Sites to foodie destinations you can’t miss, these are the recommendations of people who have seen it all – quite literally.
The US also carried out strikes on Iran on Friday after a drone attack on the M/V Ever Lovely ship. US aircraft targeted Iranian missile and drone storage locations as well as coastal radar sites, CENTCOM said.
Can steering vectors drive gradient routing? Yes, but not in realistic reward hacking environments, they are not precise enough classifiers of hacky vs clean solutions. Instead, can we use a steering vector to initialise adapters so that gradient routing happens without a classifier, and we get automatic seperation of hacky and clean gradients? Partly! This init approach suppressed 70% of hacking by absorbing gradients into the hacky initialised adapter. This is not as good as the prior approaches which use labelled examples, and get near perfect suppression. However there is a place for this self-supervised approach, as strong labels may not be available for unknown reward hacks during frontier training. This approach, if improved, has the potential to be used at scale by initialising two adapters for a task with synthetic pairs, and merging the clean adapter into the model after the training has been completed. Intro Gradient routing ( Cloud et al. , 2024 ); ( Shilov et al. , 2025 ) is a fascinating technique, it lets us quarantine unwanted behaviours into a discardable part of a model. It's not adversarial because the model is blind to any conflict of incentives, which makes it promising for stable alignment. It needs some labels, but it is robust to missing 40-50% of them, because the unlabeled samples follow the path of least resistance and get "absorbed". Absorption is the most interesting part of gradient routing, and Cloud et al. ( 2024 ) described it: we posit absor…
Lance Schroyer, who has helped oversee ICE partnerships, would become the agency’s first Senate-confirmed director since 2017.
장시간에 걸친 실제 컴퓨터 업무 수행 능력을 평가하는 새로운 벤치마크가 공개됐다. 기존의 한계를 넘어 실제 전문가 수준의 업무 환경을 반영한 평가 도구다. AI 연구소인 X랭(XLANG)은 26일(현지시간) 장기 컴퓨터 사용(Long-horizon Computer Use) 능력을 평가하기 위해 \'OS월드(OSWorld) 2.0\'을 공개했다.기존 벤치마크가 수십 단계 안에 끝나는 단순 작업을 평가했다면, OS월드 2.0은 숙련된 사람이 평균 1.6시간이 걸리는 실제 업무를 기반으로 AI 에이전트의 장기 작업 수행 능력을 측정하도록 설
In recent years, Portugal has seen a sharp rise in the number of ultra-wealthy residents, fuelled by property investment, international capital flows and the appeal of the Portuguese way of life.
Catch up with the most important stories from around Europe and beyond this June 28th, 2026 - latest news, breaking news, World, Business, Entertainment, Politics, Culture, Travel.
Donald Trump’s efforts to push polluting coal onto Americans appear to be failing, as more households opt for rooftop solar.
As the global artificial intelligence (AI) boom puts intense pressure on data centre energy grids, some Chinese chipmakers are betting on highly efficient silicon carbide (SiC) semiconductors to help solve the technology sector’s power problem. Shenzhen-based Basic Semiconductor is the latest contender looking to bankroll its expansion after it passed a listing hearing earlier this week in its path to an initial public offering (IPO) in Hong Kong. Founded in 2016 by graduates from Tsinghua...
I decided to combine my need to top the leader table with my daily step count – which is how I found myself walking 10 miles a day while reading out sentences in Japanese, German, Spanish and French Hugh and I were driving from Washington, DC, to the Sea Section, our house on the coast of North Carolina, when I noticed a dot with legs traversing the hem of my untucked shirt. “There’s a tick on me!” I said. He looked down at my lap. “Well, throw it outside. It’s nothing to get hysterical about.” Continue reading...
Dilip Asbe said that newer UPI apps could be more competitive with a viable commercial model
Liquid AI released LFM2.5-230M, its smallest model yet. The 230M-parameter, open-weight model runs on-device at 213 tok/s on a Galaxy S25 Ultra and 42 on a Raspberry Pi 5. Built on the LFM2 architecture, it targets tool use and data extraction, beating larger models like Qwen3.5-0.8B and Gemma 3 1B on instruction following. The post Liquid AI Ships LFM2.5-230M with llama.cpp, MLX, vLLM, SGLang, and ONNX Support for On-Device Inference appeared first on MarkTechPost .
에포크 AI가 AI 모델의 장기간 소프트웨어 개발 능력을 평가하는 새로운 벤치마크 \'미러코드(MirrorCode)\'를 공개했다. 기존 소프트웨어 엔지니어링 벤치마크가 버그 수정이나 단일 기능 구현 등 단기 작업에 초점을 맞췄다면, 미러코드는 인터넷이나 원본 소스코드 없이 프로그램 전체를 처음부터 다시 구현하도록 요구해 AI의 실제 소프트웨어 개발 역량을 측정하도록 설계됐다.에포크 AI는 26일(현지시간) AI 안전성 연구기관 METR와 공동으로 미러코드를 개발했다고 밝혔다.AI가 기존 프로그램의 소스코드를 보지 않은 상태에서 프로그
마이크로소프트(MS)가 \'깃허브 코파일럿(GitHub Copilot)\'에 자체 AI 코딩 모델 적용을 확대하며 기업용 AI 코딩 시장 공략에 속도를 내고 있다. MS는 26일(현지시간) 자체 개발한 AI 코딩 모델 \'MAI-코드-1-플래시(MAI-Code-1-Flash)\'를 깃허브 코파일럿 비즈니스와 깃허브 코파일럿 엔터프라이즈 고객에게 정식 출시했다.빠른 응답 속도와 낮은 지연시간을 앞세워 대규모 개발 환경에서 반복적인 AI 코딩 작업을 지원하는 모델로, 기업용 깃허브 코파일럿에 MS의 자체 AI 기술을 본격적으로 확대 적용했다는