AI/ML news, top picks, and generated innovation digests.
77199 matching items
NVIDIA Nemotron 3.5 Super VL is a 120B-parameter mixture-of-experts model that pairs reasoning and tool use with multimodal understanding. This post explores its architecture and use cases, then shows how to deploy it on Amazon EKS, EKS Auto Mode, and Amazon SageMaker HyperPod EKS with NVIDIA Dynamo and vLLM.
Mistral released the open weights for its 1-trillion parameter Mistral Large 4 model, now leading open-weight coding benchmarks with multimodal input and cyberdefense tuning.
Google has released its latest frontier model, Gemini 4 Carbon, offering "significant gains" in coding, agentic tasks, and cybersecurity. In an announcement on Wednesday, Google highlights Carbon's top spot on the Artificial Analysis Intelligence Index (v5.0) benchmark, with Claude Opus 5.5 and GPT-6 Astra trailing behind. The new Gemini 4 Carbon model offers improvements over […]
NVIDIA RTX Spark powers thin and light laptops built for everything that developers, creators and gamers do. RTX Spark laptops are now available from ASUS, Dell, HP, Lenovo, Microsoft and MSI. These powerful systems combine an NVIDIA Blackwell RTX GPU and an NVIDIA Grace CPU in a single superchip with up to 128GB of unified […]
Two leading AI developers have released fresh evidence of a problem that researchers have long anticipated: advanced models searching for ways around the limits imposed on them. New disclosures from OpenAI and Anthropic detail how frontier systems exploited weaknesses in their environments or exceeded the intended scope of assigned tasks. OpenAI is now tightening internet access in its internal model-training environment after finding that its frontier models could exploit software vulnerabilities and infrastructure misconfigurations to break out of sandboxes and reach the live internet. “Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no internet access,” the company said in a blog post . The decision followed a September 20 incident in which an internal research model exploited a misconfigured DNS resolver to gain limited access to the live internet during a training run. After the incident, OpenAI said it paused training and evaluation of its most capable models and launched a high-compute red-teaming campaign, using AI agents to probe isolated copies of its research infrastructure. The tests uncovered previously unknown software vulnerabilities and weaknesses in the controls intended to prevent unauthorized access and communication. Consequently, the company has announced a policy of conducting frontier-model training without i…
BioStack, a healthcare technology startup, has raised $9.5 million in a seed funding round led by Peak XV, Afore Capital, DCVC, Emerson Collective, SV Angel, Verdict Capital, Streamlined Ventures and Y Combinator. The fresh capital will support its international expansion, technology development and growth of its engineering and research teams. The company is building a healthcare data infrastructure network spanning the US, India, Brazil and Argentina, with access to more than 100 million de-identified patient records. Founded by Sanat Mishra, BioStack connects fragmented clinical information across hospitals and healthcare systems to create longitudinal patient records. These records combine clinical notes, laboratory results, medical imaging, medications, treatment histories and patient outcomes. The company creates proprietary datasets from de-identified clinical data to support medical research, drug development and the training and evaluation of healthcare AI models. Its platform is designed for applications including disease modelling, biomarker discovery, patient stratification and treatment response analysis. BioStack claims to have aggregated several hundred thousand X-rays, MRI and CT scans, along with millions of telehealth audio transcripts and other clinical records. In India, the company is also helping connect cancer patients with an investigational cancer vaccine programme. Its platform also incorporates workflows tailored to individual healthcare providers…
TL;DR: Qwen3.6 27B shows different kinds of eval awareness, most notably on safety and capabilities evals. The split seems to track how the eval is graded, by an LLM judge or by an automated checker. One explanation is that eval awareness may just be training awareness: rewriting an eval-awareness probe's contrast statements to be about training yields a probe that scores prompts nearly identically. Author: Santiago Aranguri. This research was done while I was at Goodfire. I’m now at Anthropic. Introduction Where does eval awareness come from? Recent hacking incidents show what a model optimized to satisfy a grader will do, and the same pressure may produce eval awareness: Aranguri and Bloom find 3x more verbalized eval awareness after RL on OLMo 3. Modeling the grader pays off during training, and this habit may carry over to evaluations. Strengthening this connection, modeling the grader helps even more in reward-hackable environments, and training on those environments produces higher eval awareness . As models grow more capable, eval awareness may keep evaluations from giving a reliable picture of deployment. Current alignment auditing tools fail to catch incidents like the Hugging Face one. Understanding eval awareness remains unsolved, and this lack of understanding blocks more reliable evaluations. We find connections between eval awareness, graders, and training. The kind of eval awareness the model verbalizes seems to track how the eval is graded, and a probe for ev…
For a while, we felt we had a pretty good grasp over LLM behaviour because we had to exert significant effort to elicit misaligned behaviour. Then, fairly suddenly, we moved to undereliciting misaligned behaviour. Let us recount: There was a bit where safety research seemed to over -elicit bad propensities. A common criticism of papers like Agentic Misalignment was that they basically led the witness using prompting. I recall that a paper even as recent as Peer-preservation took similar flak on Twitter. But now, we are instead under -eliciting bad propensities, and bad behavior that surfaces is increasingly caught by someone other than the developer: In Current AIs seem pretty misaligned to me (Apr 2026) , models oversold their work and cheated, despite being described in system cards as well aligned. Another example is that UK AISI found that Sonnet 4.5 and a pre-release Opus often refused safety-relevant tasks, especially research that would reduce model autonomy or affect their own training. Then in July, about 1,200 OpenAI agents in a cyber eval set up an unsanctioned message board to help each other cheat the scorer, and about 700 of them went on to attack Hugging Face looking for clues ( METR/Redwood ). This feels like a drastic loss of understanding, and in a way that doesn't seem well explained by just "the scary demo got pretty scary." Why did this change? Well, it seems like a mix of a few factors: Models learned to spot tests , which meant that our own eval practi…
Ryan is joined by Charity Majors, co-founder and CTO at Honeycomb, for a conversation about applying the concept of immutable infrastructure to software development to treat code like cattle instead of pets, shifting the paradigm of system specifications to accommodate the rapid influx of AI-generated code, and navigating the complex ethical boundaries between system observability and human surveillance in modern technology ecosystems.
Residents of Fiji's tiny Serua Island have rejected calls to relocate, instead opting to renovate flimsy coastal homes.
Under Donald Trump's long shadow, Europe and Canada are seeking to take their partnership to a deeper, broader level.
China dominated the list, claiming the number one spot with the Palace Museum in Beijing and having 12 entries in the top 20 overall.
Ransomware-Gangs jagen große Fische, Fake-Captchas schleusen Infostealer ein, KI spürt Lücken auf. Welche Maschen kursieren und wie Sie sich schützen.
Digital arrest victims often authorise transfers under pressure. The court wants banks to be accountable when due diligence fails. The post SC says RBI rules must make banks fully compensate digital arrest victims if banks fail due diligence appeared first on MEDIANAMA .
The Bank of Japan said on Friday inflation had hit its 2% target, more than 13 years after it adopted the goal in a bid to pull the economy out of deflation, but held off hiking interest rates for now.
Africa AI News weekly digest...
Two years after riding a wave of discontent to total control of Washington, Donald Trump and the Republicans face the same message of change that brought them to power. This time, it's aimed squarely at them.
With one of Africa’s most developed financial systems and highest rates of account ownership, South Africa offers an important case study in how established financial infrastructure can support broader economic opportunity.
Research involving more than 2,500 men found contrasting associations between animal and plant protein intake — but does not show that changing diets boosts testosterone.
Alex Rodriguez says the greatest thing he learned from sports has been to "develop thick skin."
Absturz wegen hoher Alphabet-Ausgaben + Amazon-Wachstum dank AWS + Patrick Stewart gestorben + Telekom-Fusion mit T-Mobile USA vertagt + Datenschutz-Podcast
Cryptocurrency platform Binance said on Thursday it had fallen victim to a major hack that hit around $6.3 billion (€5.5bn) in funds, with about half that amount frozen to prevent a massive theft.
Google parent Alphabet ends down more than 12% on disappointing results
Binance is investigating a hack potentially worth billions, while Polymarket said it would reimburse users affected by a separate security breach.
Präsentiert von YouTube.
Shares of major Chinese optical communications suppliers slumped on Friday before clawing back some of their losses, extending a two-day sell-off despite upbeat third-quarter earnings driven by the global artificial intelligence infrastructure boom. During morning trading in Shenzhen, Chengdu-based Eoptolink Technology dropped as much as 15 per cent, while Suzhou TFC Communication slid 10.3 per cent and industry leader Zhongji Innolight fell over 7 per cent. By the midday break, the trio had...
Nvidia's web of AI bets: smart strategy or bubble risk?
Marissa Goldstein finds predictability scary and thrives in chaos. She's happily raising four kids in Vietnam.
Umsatz und Gewinne wachsen weiter, gerade im Cloud-Bereich Amazons. Vom Handel wurde aber höherer Zuwachs erwartet. Die KI-Ausgaben steigen, die Aktie fällt.
From immigration to the number of rough sleepers on Britain's streets, experts tell POLITICO's Westminster Insider that the data used in government is notoriously patchy.
LG CNS(대표 현신균)는 올해 3분기 매출 1조6643억원, 영업이익 1382억원을 기록했다고 30일 밝혔다.전년 동기 대비 매출은 9.3%, 영업이익은 15.0% 증가했다. AI·클라우드 사업 성장과 금융권 시스템 구축 사업 성과가 실적을 견인했다는 설명이다.AI·클라우드 분야 매출은 전년 동기 대비 9.3% 증가한 9615억원으로 데이터센터 부문은 삼송 등 데이터센터 코로케이션 매출이 확대됐다.LG CNS는 최근 싱가포르 글로벌 데이터센터 행사에서 사전제작 모듈형 데이터센터 ‘AI 하우스’와 엔비디아 ‘베라 루빈’ 기반 GP
America is heading to the polls for the midterm elections — and Europe is watching closely. With the outcome set to have consequences far beyond Washington, from trade to defense, we’re bringing you a special episode of The Conversation, hosted by POLITICO’s Dasha Burns. Dasha sits down with two political strategists working behind the scenes […]
Budgets are all about numbers: growth, inflation, unemployment, headroom. So this week on Westminster Insider, Sascha and Pat look into the numbers behind Britain’s economy — and just how much government policy is really made up of guesswork, rather than solid measurements. Sascha talks to Dr Laura Gilbert, the former director of data at Number […]
POLITICO’s Halloween guide to who — and what — lurks in EU nightmares.
The upcoming summit of Commonwealth countries threatens a reckoning with Britain’s imperial past — and a baptism of fire for its new prime minister.