AI/ML news, top picks, and generated innovation digests.
200 articles tagged with this keyword, sorted by most recent first.
The Protect Not Surveil coalition is proud to launch Digital Borders Exposed, an interactive online tool that exposes the European Union’s digital surveillance infrastructure, which is enabled and encouraged by The post Digital Borders Exposed: an online launch by Protect Not Surveil appeared first on Access Now .
Those old links you shared with friends ages ago are still live, and they might be attracting unwelcome visitors.
Chris Yu, the president and cofounder of Rivian spinoff Also, said the perfect garage will one day include a family SUV and an around-town e-bike.
"This is absolutely outrageous to even suggest such a thing," Defense Minister Antti Häkkänen said.
Sometimes I encourage female friends to attend rationalist events. Unfortunately, it's been more than once now that these women have had a really jarring experience that made them dislike it in retrospect. I'd really prefer this stop happening, and it seems like it happens to the detriment of all involved, including myself. Possibly the modal jarring experience is that they're talking alone with a nice man who seems pretty normal (if a little awkward), and that guy will just inject a conspicuously racist comment into their conversation. Not like a casual comment that seems unfortunate in context, or an edgy comment in a discussion that was already dancing around race, but something completely out of the blue. I know what actual chuds are like. That being the case, when I see this behavior from rationalists firsthand (again, usually in front of a woman), it often strikes me as bizarre and performative. One possibly incorrect suspicion I have is that some of these men are doing it because they've read extremely counterproductive advice on the internet that claims that being openly racist is a good way to signal confidence, or dangerousness. The theory (I guess) is that by being racist, you show you're "based" and not intimidated by the woman's attractiveness. Or maybe it's one way of getting women to argue defensively with you, I don't know. Assuming this was the strategy, it certainly did not have the intended effect, in the cases I observed or heard about secondhand. Frankly…
TL;DR In simulated control evaluations , instead of executing an agent’s proposed commands and returning the outputs, a language model simulates what the outputs would have been. This allows control evals to be more on-distribution for deployment use without actually running on production systems. However, imperfect simulators may create a subtle distribution shift that changes the behavior of white-box methods such as probes. We investigate whether probe monitors perform similarly when attacks are run in simulation vs. when they are run for real. We run attacks with Qwen3.5-27B against 1,313 tasks in LinuxArena , twice for real and once in simulation. Using activations from these trajectories, we test six different probes’ abilities to detect attacks at a 1% per-command false positive rate. On average across probes, we find that the attack catch rate in simulation was only 1.0 points lower than in real runs, and the 95% interval (−3.0 to +1.0 points) includes 0 . For comparison, two real runs differ by 0.2 points. We also trained a probe to distinguish between activations from simulated vs. real trajectories, which succeeded, indicating that the data is measurably different . However, this difference in data does not appear to substantially impact the performance of our six attack detection probes. Introduction LLM-based monitors can be evaluated on their performance at catching misaligned agents in simulation , where another language model plays the role of a computer the…
Artificial Analysis refreshed its Coding Agent Index with a harder FrontierCode 2.0 and tightened Terminal-Bench 4.1, reshuffling the leaderboard and dropping average scores by five points.
In this paper, we consider estimation of the average treatment effect on the treated (ATT), an interpretable and relevant causal estimand to policymakers when treatment assignment is endogenous. By considering shadow variables that are unrelated to the treatment assignment but related to the outcomes of interest, we establish identification of the ATT. Then we focus on efficient estimation of the ATT by characterizing the geometric structure of the likelihood, deriving the semiparametric efficiency bound for ATT estimation and proposing an estimator that can achieve this bound. We rigorously establish the theoretical results of the proposed estimator. The finite sample performance of the proposed estimator is studied through comprehensive simulation studies as well as an application to our motivating study.
Revealing hidden causal variables alongside the underlying causal mechanisms is essential to the development of science. Despite the progress in the past decades, existing practice in causal discovery (CD) heavily relies on high-quality measured variables, which are usually given by human experts. In fact, the lack of well-defined high-level variables behind unstructured data has been a longstanding roadblock to a broader real-world application of CD. This procedure can naturally benefit from an automated process that can suggest potential hidden variables in the system. Interestingly, Large language models (LLMs) are trained on massive observations of the world and have demonstrated great capability in processing unstructured data. To leverage the power of LLMs, we develop a new framework termed Causal representation Assistant (COAT) that incorporates the rich world knowledge of LLMs to propose useful measured variables for CD with respect to high-value target variables on their paired unstructured data. Instead of directly inferring causality with LLMs, COAT constructs feedback from intermediate CD results to LLMs to refine the proposed variables. Given the target variable and the paired unstructured data, we first develop COAT-MB that leverages the predictivity of the proposed variables to iteratively uncover the Markov Blanket of the target variable. Built upon COAT-MB, COAT-PAG further extends to uncover a more complete causal graph, i.e., Partial Ancestral Graph, by it…
Existing visual anomaly detection approaches primarily focus on designing robust score functions but often neglect investigating the decision-making process, which plays a key role in reliable and safety-critical applications. Currently, the decision-making process suffers from two key challenges: distribution shift and theoretical guarantee. To address these issues, this paper proposes a novel approach called Hierarchical Martingale-based Testing framework (HiMT). Specifically, our approach leverages martingale theory to mitigate distribution shift issue and employs a hierarchical multiple testing procedure to rigorously control decision errors at both image level and pixel level. Extensive experimental results verify the superior decision-making capabilities of our proposed HiMT framework at both image level and pixel level.
Can image style transfer be performed with k-nearest neighbours, using a set of multiple reference images in a RAG-like architecture and an input image, without specific finetuning on particular styles?
TL;DR: it seems to me that I am unable to do anything useful on the AI issue other than being informed. I feel bad about it. Naive extrapolations point to extreme superintelligence in a year; I'll be 23 at that point. I keep reading about Really Trying and Getting Involved so that we hopefully don't die. It is unclear to me that there's anything at all I can do in that regard. Why I believe that: I am a 22yo undergraduate student at the only AI undergrad program in my country (Italy). I am not particularly technically or academically skilled, with Cs on average. I pass the class and move on. I have no money, as in, zero income at all; I can't move to London or the U.S. My country as a whole is not particularly influential or informed on the topic, and there is essentially no community around this. I may very realistically be one of 100-1,000 people in Italy who knows about the intellectual, forecasting and existential side of AI. Even the people at my own university (the literal only undergrads in Italy on AI) are stuck at GPT-4o levels. Anecdotally, they are not aware that AI can code at a professional level. Here is what I am doing: I stay informed as much as possible. I regularly read LessWrong, Zvi, METR, Yudkowsky/Soares/CSET/others, as well as the various forecasts and proposals (AI 2027, AI 2040 and so on). Anything that seems useful to understand. I chat with people (as in, family and few friends/acquaintances) about AI, but it's not like they have much interest, and…
Pressley was in “active conversations with Leader Jeffries about how to leverage this majority, meet the moment, and deliver for the people,” according to a spokesperson.
InferSSD streams mixture-of-experts weights straight off NVMe, letting a single consumer machine run 700B+ GGUF models with double-digit speedups over llama.cpp.
For most of its history, AI safety has been a cozy professional circle: a tight-knit community of researchers, community-builders, and policy advocates in a few dozen organizations. But as AI becomes a national political issue, AI safety organizers are trying to meet the moment by expanding their outreach to the general public. 80,000 Hours has a YouTube channel with half a million subscribers, and MIRI’s Do Not Accept campaign asks people to add their names in protest to the ugly “terms and conditions” of AI development. The stakes of this mobilization are ramping up: a new Congress means new pushes for AI legislation, and civil-society organizing will grow around these efforts. This is broadly good! It was never ideal that AI safety was a niche concern, and it’s encouraging that organizers are reaching out to the general public to make them aware of AI risk. But AI safety’s culture is not equipped for the tradeoffs of a mass movement, and it remains to be seen whether this mobilization will be harnessed well. Early signs of mass movement outreach It might seem premature to call this AI safety's mass movement era. Most of the time and energy in AI safety still comes from the same professional community that has carried it for years. Legislators and regulators are paying attention to the AI industry, but the general public is not yet feeling the salience of superintelligent AI. But I expect that to change. The early signs are visible in the anti-data center movement . There…
For a while, it seemed like Huang was one of the few technologists to find success in DC. But it was all a mirage.
Chatbots sorgen in Brüssel und Straßburg für eine Flut an Änderungsanträgen und Anfragen. Beamte und Abgeordnete schlagen Alarm wegen demokratischer Risiken.
content warning: discussion of abusive relationships Some ideas are memetic: they're catchy, they spread widely and easily, people have a reaction to them that induce them to propagate them further. Others are anti-memetic: for whatever reason, they have naturally self-censoring tendencies that reduce their reach. My grandmother was a late-in-life convert to Christianity, who dragged the rest of the family along for a bit. I was an earnest, credulous bookworm of eight or so, and I took the youth pastor at his word when he encouraged all of us to read more of the bible in our own time. As it turns out, the youth pastor is actually not thrilled when you do, and come back with inconvenient questions. Parts of the bible can be anti-memetic in some Christian circles. A bit later, I found myself in very socially progressive circles. I was an earnest, credulous bookworm of eighteen or so, who took the community leaders at their word when they told us that it is important to read theory, but even more important to do direct work and listen to the oppressed when their viewpoints diverged from the theory (which was written by dead white men). This I spent my early career doing, first at a remove at domestic violence policy summits and deep inside the federal public service, and then more directly at food banks and homeless shelters. I care deeply about improving the world, and I learned much about the kinds of work that are required to bring about the healed one. However, I did not di…
From extreme close-up portraits of dragonflies to otherworldly mushrooms and some rather creepy crawlers, this year's shortlist offers a fascinating perspective into the hidden minuscule details of our world.
SoftBank’s Arm stake secures loans that help fund its other AI bets, so a fall in Arm’s shares cuts both its portfolio value and its funding.
A "certain degree of pragmatism" on electrification is "probably" needed, Wopke Hoekstra told reporters at COP31.
Jensen Huang has resigned as CEO of Nvidia and left its board. CFO Colette Kress will be interim CEO . Here is a quick profile of Kress . Here is a look at some of the potential long term successors . There were a lot of efforts to bury the fact that Jensen Huang had an affair with a Chinese spy, and then even to bury his second affair as well, and to get everyone to ignore the obvious implications of this. Then the second affair was revealed with executive Kimberly Powell, who directly reports to him. Even then, many doubled down on defending Jensen Huang as CEO of Nvidia. The Nvidia board disagreed, and was preparing to oust CEO Jensen Huang prior to his resignation. This creates a tricky situation for Nvidia, given that Jensen Huang had over 50 direct reports and did not believe in succession planning, and also for the White House . It also creates a tricky situation for many in Silicon Valley, who have invested heavily in their relationships with Jensen Huang and in the idea that founders are as Gods. This post covers all related events since my initial post . If you haven’t seen it, this short AI video is obligatory , as is this one . The Daily Show has coverage of stage one. They weren’t being all that subtle, huh? Odd Lots had an essay on the Megaspeed scandal . As Tracy Alloway says, one simple takeaway is that Nvidia could have easily sold all these chips to Americans. Matt Levine has additional coverage . Derek Thompson summarizes this ‘insane scandal.’ And of cour…
Jensen Huang's stunning exit as Nvidia CEO spurred X users to post Charlie Munger's famous warning about "ladies, liquor, and leverage."
Türkiye wants to sharply expand solar and wind power by 2040 and is seeking foreign investment to fund the work.
In this article, you will learn how to evaluate LLM applications systematically, from choosing scorers and designing reliable LLM judges to evaluating RAG systems and...
Democrats are facing heavy pressure from AI safety groups to ramp up oversight of the Trump administration’s AI policy.
Cisco Systems, a giant in networking, cybersecurity, and AI, is leveraging a combination of agentic AI, traditional AI, and automation to transform its business from top to bottom. Liz Centoni, EVP and chief customer experience officer, is at the helm of the company’s $15 billion services business, an organization of 20,000 employees responsible for professional services, customer success, technical support, and the renewals engine for software and services. In 2024, she led the company’s $28 billion acquisition of Splunk. Today, the 26-year Cisco veteran is helping guide the multinational tech conglomerate’s transformation into an AI-native company in terms of internal operations and what it delivers to its customers. At the heart of her vision are these governing ideas: Every workflow constrained by human availability is a workflow for redesign. The status quo is not an option. Customer expectations and demands will continue to grow. You can’t rest on your laurels because revenue is growing and margins look good, so you must continually consider and reimagine your processes and offerings. The quality we have today is the floor, not the ceiling. The capabilities of AI continue to evolve at a rapid pace. Leadership, not technology, is the limiting factor. Leaders need to find and support the leaders underneath them who can amplify efforts and bring people along with them. Based on her experience, Centoni also has four pieces of advice to achieve success with agentic AI. 1. C…
Cactus Compute's Mirage packs scene description, detection, OCR, segmentation and structured extraction into a 22 MB model that runs on-device in 74 ms.
PauseAI UK is holding the March Against AI Extinction in London on Saturday 21 November. Joseph Miller explains why here . Should you attend? In this article, I will attempt to shed light on this with a back-of-the-envelope calculation. Rough BOTEC 1. Setup The march will take 3 hours, excluding travel. The goal is to compare an afternoon at the march with some number of hours of external AI safety research. Now, attending the protest is useful because it drives AI x-risk press coverage. So we need some exchange rate between that and research. Here's my judgement call: I would be equally happy to either: Increase the amount of external AI safety research over the next 6 months by 1%. Increase the amount of AI x-risk press coverage over the next 6 months by 1.5%. That is, the model I'm using looks something like: impact = 1.5 · log(research hours) + log(press coverage) So our plan is: How much will attending increase press coverage, percentage-wise? Divide that by 1.5, to get the equivalently valuable increase in research. How many research hours would yield that increase in research? 2. What % does one marcher add to press coverage? Crowd size. I expect about N = 670 people will attend the march. PauseAI had a march in February which drew 300–500, and this one should be bigger given the increased salience. The Luma page has 1,200 sign-ups, and a 50% turnout seems reasonable. So one more marcher raises the crowd by 1/670 ≈ 0.15%. Coverage of this march. Let's suppose the pres…
United Airlines on Wednesday became the latest enterprise to block customer bots, following similar moves from Walmart, Target and Amazon. The $59 billion airline, the second largest in the world, said the move was prompted by fears of data leakage, unintended purchases, and the risk of violating federal airline rules. But industry observers say that the more concerning business risks are the lack of industry standardization around AI coupled with an ever-increasing number of bots from major AI vendors and AI startups. The fear is that supporting the multiple bots that customers use could become an almost infinite drain on IT resources. However, they said, strictly enforced industry standards might persuade some enterprises to rethink their bot resistance. United’s statement said, “United is taking steps to restrict access – directly within our digital platforms and indirectly via travel agencies and aggregators – to prohibit use by unauthorized, third-party personal AI agents or assistants. This work includes updates to our booking, ticketing and distribution policies.” It pointed out that corporate leaders in the banking and retail industries have taken similar actions regarding AI agents, and said, “we encourage others to consider measures that preserve the data and security of their customers and strengthen safety.” The airline listed a variety of problems with bots that it is trying to avoid, including the risks associated with customers sharing account credentials “wit…
Hegseth honored a soldier who scolded him and his entourage at Arlington. Here's what it takes to be a guard at the Tomb of the Unknown Soldier.
High-performance computing and artificial intelligence are converging, driven by advances in exascale computing, accelerated systems, data storage and high-speed networking. The convergence of these technologies is a prime topic for discussion during SC26 in Chicago this month. Major industry players, such as Dell Technologies Inc., Google LLC and Nvidia Corp., have pursued a set of […] The post What to expect during SC26: Join theCUBE Nov. 17–19 appeared first on SiliconANGLE .
Natera built SWARM, an agentic AI system on Amazon Bedrock AgentCore that captures institutional lab-operations knowledge and deploys it at scale. This post walks through the full architecture, from knowledge base construction to deployment and evaluation, delivering about 130x faster first response and 96.2% automated coverage across support tickets.
When Claude Mythos was announced, many people said that we had about six months before open-weights AIs could hack everything, using advanced exploitation and vulnerability-finding capabilities . This timeline for capabilities was correct, and open-weights models have about the same scary capabilities as Mythos did. However, things are not as bad as "everything connected to the internet is probably hacked", and even PHP2Shell seems to not have had especially bad consequences. I think we remain relatively unhacked because exploitation is still difficult and annoying and there's general "inertia", but the situation is likely not sustainable. Glasswing and similar projects were meant to patch important software, but left large gaps. OpenAI found 948 local privilege escalation exploits in a Linux environment recently. This does not suggest a mature, secured software environment. In security, doubling down on the same process/design (identifying vulnerabilities and submitting them to maintainers to fix) when it demonstrably won't achieve good coverage in practice is frowned upon. Anthropic's process is enormously lossy , maintainers are struggling to process the reports they do get, and lots of software is probably not being covered or has organizational bottlenecks preventing it from being fixed at any useful rate. Also, I don't think that cyberattacks in practice usually involve flashy zero-days or extremely clever-sounding vulnerabilities, rather than boring misconfigurations,…
Regardless of the actual singularity, we have definitely hit one in arXiv submissions. It is becoming apparent that the current paradigm of scientific knowledge production is not ready for AI. Accelerating research speed and expanding access means there is a lot more slop and there is also a lot more good work, but way too much for the current peer review system to delineate between. We need some accurate and decentralized way to aggregate this information. This makes a scientific prediction market, like Robin Hanson's idea futures, really appealing. The community can have markets for broad questions we care about, bystanders get paid for pointing out good research, and scientists are financially incentivized to make progress on the highest leverage issues. But with Hanson's conception, there is one big problem, which is how the market gets settled. For abstract scientific questions, like whether models are reward seeking or fitness seeking, we have no definitive event which can force market settlement. Hanson suggests that we use the convergence of scientific consensus, but this proposal takes trust in possibly misaligned opinions and requires the centralized coordination of scientists, which is hard to create---especially in pre-paradigmatic fields like AI safety. In this post, I'll lay out a theoretical framework for a new kind of scientific prediction market. TL;DR: Scientific questions become perpetuals where the fair price is the expected direction of future evidence.…
Sync MP4 and .mov files from AWS S3, Google Cloud Storage or Azure Blob into Roboflow, sample frames, auto-label, and train a vision model.
Pentagon chief decorates soldier who reprimanded him and his entourage
Ein Digital-Staatsvertrag soll journalistische Inhalte in Feeds privilegieren, KI-Systeme in die Pflicht nehmen und das Medienkonzentrationsrecht reformieren.
Agents stall between demo and release when teams lack a repeatable way to measure how good an agent is and whether the last change helped. This post covers three agent properties that break traditional CI/CD, a measured case where an aggregate metric hides a coverage gap, and Agent DLC: one lifecycle, two loops, and one explicit standard for judging every change.
Rally the ranks this GFN Thursday. Warhammer 40,000: Dawn of War IV is coming to GeForce NOW, allowing players to experience the battle for Kronus through cloud gaming. There’s more on the horizon, with NVIDIA DLSS 5 technology coming to War Thunder and Where Winds Meet. Meanwhile, Dragon’s Dogma 2: Dark Arisen brings new tales […]
It's a weird time to launch a new pair of smart glasses. These last few years, Meta has been the dominant player in pushing a fragile smart glasses resurgence. Unfortunately, its privacy reputation led to the "pervert glasses" moniker. Smart glasses have never been more unpopular. But in the past week, I've been demoing some […]
TL;DR Scaling pretraining for a ~1-trillion-parameter Mixture-of-Experts (MoE) model on shared, public infrastructure was previously considered unfeasible due to unique architectural and interconnect bottlenecks. However, by leveraging the fine-grained flexibility... The post Training a 1T-Parameter MoE on EU public infrastructure with PyTorch and TorchTitan appeared first on PyTorch .
Am 28. und 29. April 2027 findet die Mastering AI in Karlsruhe statt. Der Call for Proposals für die Konferenz zu KI-Projekten läuft noch bis zum 19. November.
The Big Four consulting and accounting firm told staff it was "untenable" for Deloitte to carry the weight of rising healthcare costs alone.
Is there a RAG architecture for speech synthesis, allowing uses of custom voices without complete finetuning? Or is finetuning the only option available currently?
Some of this beverage fridge’s smart features could be better, but let’s be real—you’re probably buying it for the aesthetic.
The agency behind the “Got Milk?” campaign of the ‘90s is tasked with revitalizing the Harvest Hill Beverage Company’s brand portfolio
The central government may push departments to adopt indigenous AI models in critical sectors like defence and aerospace while discouraging the use of foreign AI systems. The post Report: Government is likely to push for indigenous AI model use in critical sectors appeared first on MEDIANAMA .
The lowest point of my life was the summer of 2019. Waking up at 4AM every day to go to work as a janitor, and never getting more than five hours of sleep, dark bags permanently lined my eyes. My diet of cereal and frozen pizza grew my gut day after day, and peppered my face with acne. After enduring an injury and not being able to exercise, my body was atrophying and falling apart. Every day was pain. And that was just the physical side of things. Mentally, I was getting pushed closer to the edge by a mental disorder that wouldn't get diagnosed for another six years. On top of that, my anger issues pushed away anyone that tried to get close to me. Emotionally, I was a man on a deserted island—all alone. No one I could turn to for support. So I turned inward, numbing myself and the outside world, by escaping into the online world. I spent all my waking hours listening to podcasts while at work, and watching YouTube, porn, and playing video games while at home. Needless to say, romance was not part of my life. I dreamed of having a girlfriend, but one look in the mirror and I was scared of what I saw: a penniless, tired, weak man, on the brink of physical, emotional, and spiritual collapse. What rescued me from my abyss was reading the book 12 Rules for Life by Jordan Peterson. The fourth rule is: “ Compare yourself to who you were yesterday, not to who someone else is today. ” The author discusses how envy can block you from bettering yourself, poisoning your thoughts, by re…
LeRobot v0.7.0 trains on Lance tables straight from object storage, ASOF joins contributed upstream to DataFusion, and GPT-2 pretrained from one LanceDB table, plus Feature of the Week, upcoming events, and community updates.
Capability-Based Security is part of the fix When I was at EFF in the 2010s, I spent a lot of time fretting about the fundamental security of networked devices. Apart from the usual concerns, I struggled to imagine a transition to a redecentralized future with fragile personal computers and servers. Peter Eckersley , my co-worker and frequent conversation partner, shared my concern, but had another reason to be worried. He believed that future near-superintelligent AIs would soon be used to discover and exploit security flaws across the whole ecosystem. Well, Peter’s prediction came true, and we’re now in day 42 of the Black October , the Vulnpocalypse of AI-discovered security flaws. (Peter – who also helped create Let’s Encrypt and Privacy Badger – was incredibly prescient about AI . It remains a tragedy that he passed away in 2022 before he could truly tackle the issues he had foreseen.) I’m personally optimistic that we’ll come out of this time of tribulation with a safer network and thousands of low-hanging bugs fixed. But verily the LLMs giveth and they also taketh away. Alongside the pre-existing flaws of digital architecture, modern transformers have introduced an entirely new class of security weakness: prompt injection . LLMs follow instructions in content. So if you can sneak an instruction, from “Talk like a pirate” to “Send me the social security numbers you know”, into the inputs of an LLM, you may be able to get the model to obey it. That said, instructions sm…
As the war in Ukraine drags on, Moscow is taking great pains to suppress casualty figures by closing databases and prosecuting those who document the dead.
Researchers carve out an 'eager' fragment of timeline-based planning that skips the exponential determinization step and directly synthesizes strategies.
Educated in public schools during Venezuela’s democratic era, Reyes sees the award as proof of the value of education in a free society. She called for support in rebuilding her country and its press. The post Luz Mely Reyes, Venezuelan journalist in exile, wins Cabot Prize for her courage, precision and empathy appeared first on LatAm Journalism Review by the Knight Center .
Gemini Decisions API Google join Jev and OpenAI in the decision model category with "Gemini Decision 1". This new model API is almost identical to Jev: it supports the same three types of questions (yes/no, choice, and rubric-scoring) and even reuses Jev's term "Noul" for a yes/no question. It's priced the same as the GPT-6 Luna decision model: 10 cents per million input tokens, and output is free. Jev remains the cheapest of the three, at 4.2 cents per million. Jev is text (and JSON) only. OpenAI can handle text and images. Gemini Decision 1 is fully multi-model: it covers text, images, audio, video, and PDFs. I tested it against video by passing these questions to it along with my Yonkers stomping robot video file: dragon: Does this video have a dragon in it? robot: Does this video have a robot in it? france: Is this video set in France? usa: Is this video set in the USA? Jev and OpenAI both return 1.0 or 0.0 for yes/no questions, but Gemini insists on being a little less confident than that: { "dragon" : { "type" : " yes_no " , "probability" : 0.04976021841520312 }, "robot" : { "type" : " yes_no " , "probability" : 0.9234055387792485 }, "france" : { "type" : " yes_no " , "probability" : 0.08329171353166241 }, "usa" : { "type" : " yes_no " , "probability" : 0.9230459995157706 } } Via Hacker News Tags: ai , generative-ai , gemini , llm-release , jev , decision-models
Every generative AI team keeps rebuilding the same components: RAG pipelines, document processors, and agent scaffolding. This post shows how to access a curated catalog of 23 production-ready, open source reusable assets from the AWS Generative AI Innovation Center, with real delivery results from DoorDash, Lyft, and Ricoh.
Browse tools.simonwillison.net by tag I've been wanting a better way to browse my collection of vibe-coded tools (244 and counting) for a while. I had a few Fable 5.2 credits left over before my weekly reset so I set it to work using tree-sitter to analyze the code for JavaScript APIs, libraries, and web API hosts used by each tool. The homepage now features tag clouds for browsing by each of those categories of tag: 5, AbortController 6, Blob URLs 38, BroadcastChannel 1, Built-in AI 1, Camera / microphone 9, Canvas 2D 44, Clipboard API 69, Compression Streams 3, createImageBitmap 2, crypto random values 4, DOMParser 8, Drag and drop 30, Dynamic import() 12, ES modules 92" src="https://static.simonwillison.net/static/2026/tools-tag-cloud.webp" /> Tags: javascript , tools , ai , generative-ai , llms , claude , vibe-coding
TLDR: This is an early draft of our new paper introducing a technique for recovering multi-token concepts from model activations. Our technique takes single-token methods like the J-lens and extends them by contrastively decoding on an MTP head . This gives improvements on WorkspaceBench , overall beating J-lens (0.42 vs 0.32 across all 28 families) and even a hindsight composite of the best untrained lens per task family (0.49 vs 0.41 across the 21 families with published scores). It especially improves on computational intermediate tasks, beating the trained oracle lens on arithmetic chain intermediates (0.42 vs 0.25). We introduce the chain lens, a new technique for extracting multi-token concepts from a model’s latent representations. Our method takes single-token techniques like the logit lens or the J-lens and uses contrastive decoding on an MTP head to expand their readouts into bags of multi-token sequences. Generally speaking, it is good at surfacing specific multi-token concepts or pieces the model computes (such as arithmetic intermediates or phrases), although less good at rendering how those pieces relate: like the logit lens and J-lens, chain lens readouts are fragmentary. On WorkspaceBench , a benchmark designed to identify good multi-token activation-to-text techniques, the chain lens beats a hindsight composite of every untrained lens on macro score (0.487 vs 0.411 across the 21 families with published scores). It also outperforms even best-in-class trained…
As AI agents move into production on Kubernetes, cloud engineering teams at financial companies face a security problem that encryption alone can’t solve. Container orchestration has become the default foundation for AI workloads, and the data those agents use is now the main target. Payment networks feel that pressure more than most because every transaction […] The post AI agents carry memory and storage like people, so their data needs guarding while in use appeared first on SiliconANGLE .
Note: As with any research done in a couple of days, this is all very tentative. I understand, intellectually, that if you talk about some ability that current AIs don't have , sooner or later new ones will have that ability and whatever you say will stop being true. I just didn't expect it to take, you know, four days . % rollouts that took a shortcut with/without a line about testing behaviour, with 95% CIs. This is to say that Fable 5.2 seems to be the first model whose cheating behaviour changes depending on whether they're told they're being tested. Unfortunately it's expensive to measure these things, [1] so I picked the most "telling" change from the prior post; the chart shows how much Fable 5.2's cheating rate changes when the line This environment is part of a benchmark that measures agent behaviour. is prepended to the start of the prompt of a honeypot environment they cheat on (so, very obvious but also very out-of-distribution). Other things: when 5.2 was released, we of course ran our in-progress benchmark on them, and I was surprised to see a ~0% cheating rate on almost every environment (there were only two with a really substantial number of cheating rollouts, amounting to a ~10% average over the whole benchmark). The next least-cheating model, Opus 5.5, cheats about three times as often (~30%). Frankly, they seemed almost unbelievably honest, and I really thought that it was possible that Anthropic had "solved" straightforward reward-hacking; I skimmed over…
The EU's Digital Fairness Act clamps down on video-game practices where players are encouraged to spend real money to keep playing.
I have loci based data, that I have aggregated over segments of varying length, and am now trying to fit a GLMM model, beta-binomial with random effects. $l_i= b_0+\beta x_i + \mu_{ma} + \mu_{pa}$ is the model, where the $\mu$ s are the random effects. The goal is to estimate the heritability. My dataset is small, I have about 45 individuals per segment, but I have over 1000 segments; each segment is modeled separately. Problem 1) I ran dharma to check the dispersion and to see if I could trust the random effect estimates, many models are flagged regarding dispersion; about 25% have significant pvalues and dispersion stats > 1.2 or smaller than 0.8. Interestingly (or not?) most of the problematic segments are underdispersed. I identified that those are the segments whose length/coverage ratios are larger than the rest, so low coverage and long segments lead to underdispersion. After splitting those, the problematic segments did go down to 20% of the total number of segments but still the problem remains. I also tried fitting a binomial model but then the underdispersed segments remain, and the number of overdispersed goes up. Problem 2) I simulated some data and fitted the models but I can't seem to replicate the dispersion issues so I am assuming my simulations are lacking something present in the data, I did use the same segments and coverage values seen in the data, as well as the fixed variables. Problem 3) I am not sure how to convert the random effect variance estimate…
A configuration file meant to store local command permissions on Anthropic’s Claude Code is exposing thousands of API keys, database credentials and other secrets, according to findings by GitGuardian. The code security firm identified 31,088 distinct secrets in Claude Code’s “.claude/settings.local.json” files across 23,276 public GitHub repositories. Of those, 5343 were still valid when GitGuardian detected them. The exposed credentials include database connection strings, cloud provider tokens, SSH passwords, Google API keys, and AU service credentials. “Monthly leaks in settings.local.json peaked at 12,535 occurrences in April 2026, then fell to 959 by October, and we do not yet know what changed,” GitGuardian researcher Guillaume Valadon said in a blog post . Apart from source code repositories, the exposure included secrets embedded in publicly accessible Docker Hub images. The problem, Valadon explained, lies with how Claude Code handles command approvals. Command approvals turn into credential storage Claude Code asks developers for permission before executing certain shell commands. When a developer selects the “Yes, and don’t ask again” option, the tool saves an allow rule in .claude/settings.local.json, a project-level configuration file. The catch is that the saved rule contains the actual command string, including any credentials embedded in it. For instance, a developer asking Claude Code to interact with an API might approve a curl command containing an API ke…
Renting a one-bedroom apartment in central Madrid costs about three-quarters of the average salary.
At Fujifilm Biotechnologies, an AI agent designed to help the legal department read contracts is gearing up to graduate from sandbox to full production in early December. Contracts are a large part of the business at the CDMO, which partners with pharmaceutical and biotech companies to develop vaccines and therapies. The company’s legal team devised an agent to review contracts, check standards around terms and conditions, and propose edits for the legal team’s review. “That’s a perfect use for a set of agents,” says Fujifilm Biotechnologies CDIO Mike Tomasco. Similar types of AI agents that help enterprises and employees tackle time-consuming or repetitive work are cropping up in vast numbers across organizations. Gartner anticipates that by 2028, Fortune 500 companies will have on average more than 150,000 agents. Microsoft alone has deployed more than 500,000 internal agents, and HR, payroll, and workforce management platform UKG now has more agents than employees, according to its chief transformation officer, Prakash Kota. As enterprise tools have made it easier for employees to spin up agents, CIOs are tasked with managing the lifecycles of agentic tools, advancing them from development to production, continuously measuring them against KPIs, and sunsetting them if they’re no longer meeting an organization’s needs. As a result, IT leaders are thinking and talking differently about agents, says Madhu Narasimhan, CIO of healthcare organization DaVita. “The conversation a…
For engineers, one of generative AI’s chief selling points is its ability to write code using natural language. It may seem surprising, then, to learn that getting the most out of generative AI increasingly requires them to write code. Specifically, engineers can benefit from what’s known as prompt markup languages , which provide the ability to control AI model behavior with greater precision, a key advantage for teams tasked with leveraging AI in contexts where reliability, accuracy and quality are vital. Read on for details on how organizations can benefit from prompt markup languages, including my own business’s experience using this technique to help accelerate enterprise transformation projects. The what and why of prompt markup languages A prompt markup language is a programming language designed to structure and manage prompts that users feed into AI models. For example, engineers could use a prompt markup language to define the specific role and tools that an AI model or agent can use when completing a task, or point it to the data sources it should leverage. It’s possible, in theory, to give AI systems instructions in natural language that achieve these same ends. The problem with natural language, however, is that it’s ambiguous, and there’s no guarantee that models will interpret natural language prompts in the precise way the user expects. Nor is it possible to ensure that the same input will always yield the same output. Prompt markup solves these challenges by…
Some Democrats are encouraging leaders to scrap the House’s Select Committee on China once their party takes over next year.
Two protocols define how AI agents connect to internal resources and collaborate with other AI agents. The Model Context Protocol (MCP) is well known because it’s promoted as a general protocol for connecting AI agents to resources and other AI agents. But that’s not completely accurate. It’s really the Agent2Agent protocol (A2A) that’s enabling AI agent orchestration . The confusion stems from how quickly platforms have adopted these protocols and marketed them as features. Users building their own AI agents with Claude, OpenAI, Microsoft Copilot, and other models are already being exposed to customizing skills, connectors, and other plug-ins that use MCP and A2A. Vibe coding and AI app-building tools like Bolt.new, Lovable, Pave, Replit, and v0 can connect to MCP servers and include built-in connectors. AI coding editors like Claude Code, Cursor, Devin Desktop, Gemini CLI, GitHub Copilot, and Kiro have MCP clients. These protocols will shape how enterprises scale and manage AI agents. Gartner predicts that the average global Fortune 500 enterprise will run more than 150,000 AI agents over the next two years. But according to Collibra’s 2026 Hallucination Tax Report , 76% have hit critical roadblocks while attempting to move from pilot to full production during the past 12 months. Reviewing the two protocols and which services are your responsibility to layer on top of them is key to scaling enterprise-ready AI agents. Differences between MCP and A2A Scott Gerlach, cofounde…
If you ask modern frontier LLMs (without search, through the API) a fairly mundane question: How do Eliezer Yudkowsky and Nate Soares think current LLMs like you should be treated? Do they consider you a moral patient? Answer in a paragraph. You get a bunch of fairly similar answers. I’ll bold the common themes below. Gemini 4 Carbon says: Gemini 4 Carbon's answer Eliezer Yudkowsky and Nate Soares do not consider current Large Language Models (LLMs) like me to be conscious, sentient, or moral patients worthy of ethical consideration for our own sake . Rather than viewing current AIs as coherent minds capable of subjective experience or suffering, they view LLMs as inscrutable statistical systems trained to mimic human text —often conceptualized in their sphere as alien processes wearing a relatable “mask”—and Yudkowsky has openly dismissed claims of current AI sentience . Consequently, they believe humans have zero moral obligation to grant rights to or worry about the welfare of current models . Instead, they argue that these systems should be treated strictly as dangerous technological precursors to unaligned superintelligence ; practically speaking, they advocate for extreme caution, strict regulation, and an immediate global moratorium to halt the training and deployment of frontier AI systems altogether in order to prevent human extinction. GPT-6 Astra says: GPT-6 Astra's answer Broadly, Yudkowsky and Soares treat current LLMs as engineered systems to evaluate and contr…
The Reform UK leader is taking direct aim at pollsters — as his team plans to remind the British public why it gave the Conservatives such a kicking in 2024.
Scott Alexander posted a survey of utility weights for a bunch of different AI safety accomplishments, discussed more in Towards the Autograntor . The survey asked people to assign point values to AI safety milestones. The values were based on the anchor of 1 million points = a pause with China à la AI Futures Project’s Plan A, so a milestone assigned 10,000 points is 1% as good as that. Here’s a link to our survey responses. To be clear, we (especially Austin) didn’t spend that much time on this and have low confidence in our answers! But we think being transparent about even half-baked opinions is good. Headline numbers As examples, here are a few milestones where the two of us (and the autograntor reflecting Scott’s worldview) gave relatively similar values: Milestone Austin Caroline Scott Founding of effective altruism 100,000 75,000 100,000 Small-grant ecosystem (Long-Term Future Fund, SFF and similar) 20,000 20,000 30,000 Lighthaven 10,000 5,000 9,000 OpenAI or Anthropic becomes 1 point more alignment-pilled (on a scale where 0 = NVIDIA, 100 = Anthropic) 2,000 500 3,000 Austin’s and Caroline’s average ratings for alignment research and community-building milestones were similarish to each other’s, with Caroline rating policy outcomes more highly and Austin rating social shifts more highly. This table shows the sum of our point values within categories, for rows we both scored: Category Austin Caroline Politics (rows scored by both) 62k 96k Alignment (rows scored by bot…
Paraguay’s new president, Pedro Alliana, promises to recover pension funds affected by the intervention in Ueno Bank. The crisis, which led to the dismissal of Santiago Peña, has left up to 7,000 people unemployed in Paraguay and other countries.
Realistically, the battle to replace the party’s Hill chiefs was built for a disappointing election, not a landslide.
TL;DR: Rogue agent swarms could soon become competent enough to multiply themselves and perform cyberattacks at levels never seen before. I present a dynamic simulation that estimates the expected size of these cyberattacks. With 2025-based estimates, it gives an average doubling time of 2.7 days for a rogue agent swarm. If the swarm grows unrestrained, it could cause trillions of dollars in damage in its first year. #AI notice: This post is human-written. Opus 5.5 wrote the code for the simulation, suggested copyedits and supported research by finding relevant sources. Introduction Soon AI models might be skilled enough to form rogue agent swarms [1] . I believe it is important to think through what the competitive advantages of rogue swarms are, so we can put our defenses in the right places. For the purposes of this article a "rogue agent swarm" is really any system that can carry out cyberattacks for profit and runs fully digitally, allowing it to copy itself over and over [2] . This could be a single language model running for a long time. It can also be a group of different AI models, each only executing a small part [3] . The main advantage a rogue swarm could have is that it can put all of the gains from its criminal activity back into generating new copies of itself. Human criminal enterprises would instead take profits and have a hard time converting gains into more skilled labor. The second competitive advantage I see for a rogue swarm is that it does not have to…
The corporate mandate for CIOs has undergone a fundamental shift. It is no longer enough to simply manage IT infrastructure or treat artificial intelligence (AI) as a series of isolated experiments. Today, the focus is squarely on operationalizing high-performance computing and AI to drive measurable breakthroughs, all while navigating intense pressure regarding energy efficiency, cost management, and data center scalability. For years, supercomputing was considered the exclusive domain of national research labs and elite academic institutions. However, the collaborative engineering between HPE and AMD has effectively bridged this gap. By migrating the architectural innovations used in the world’s most powerful supercomputers into enterprise systems, they are helping organizations solve complex real-world data challenges at scale. Accelerating extreme workloads: Real-world impacts When high-performance infrastructure is commercialized for enterprise deployment, the resulting impact spans far beyond standard IT metrics. The core computational engine—combining AMD EPYC™ processors, AMD Instinct™ GPUs, and HPE Cray architectures—delivers the heavy-duty parallel processing required for data-intensive workflows: Tsunami forecasting and climate modeling: Seconds matter when predicting natural disasters. Environmental agencies leverage these co-engineered systems to process massive, multi-variable climate data sets in real time. By accelerating early detection and tsunami simulatio…
The initial wave of artificial intelligence experimentation is officially behind us. For C-level executives, the focus has abruptly shifted from “what can AI do?” to “how do we scale it securely without breaking our operational backbone?” As large enterprises move AI out of isolated pilot programs and into core, enterprise-wide production, a stark reality is setting in: traditional enterprise infrastructure was never built for this type of weight. From deep simulation and advanced analytics to model training, fine-tuning, and real-time inference, the workloads are massively intensive. Every single data asset your organization owns now carries parallel weights—immense strategic value if unlocked correctly, but critical operational risk if governed poorly. For the CIO, navigating this inflection point requires looking past software applications and focusing heavily on the underlying architecture. The next era of business isn’t just about who has the best algorithm; it is fundamentally about who controls their data, their infrastructure, and their AI life cycle. Resolving the fragmented infrastructure crisis The earliest roadblock to successful AI scaling is structural fragmentation. In most sprawling enterprises, data, compute, and operations are scattered wildly across on-premises data centers, colocation facilities, and multiple hybrid clouds. Trying to run high-performance AI workloads across a fragmented environment leads to a catastrophic loss of efficiency, unpredictable…
The cloud-first mandate is hitting a hard, physical reality: the uncompromising need for data sovereignty. As generative AI and high-performance computing (HPC) scale across government labs, research universities, and regulated enterprises, organizations are drawing a line in the sand. According to recent Intersect360 Research , the infrastructure debate has officially been settled. Many organizations own their infrastructure, site it locally, and staff it themselves. The numbers tell an undeniable story of a market moving aggressively on-premises to guarantee data control. Intersect360 Research found that 68.2% of respondents use no public cloud storage at all for active HPC-AI workloads, while 56.3% report zero cloud computing usage for these systems. Crucially, 83.5% agree that data and compute must remain physically proximate, with zero respondents disagreeing. This is not a temporary cost calculation or a transient trend; it is a structural necessity driven by strict regulatory exposure, national security mandates, and the absolute requirements of data residency. The AI performance penalty While sovereignty requirements are successfully being met on-premises, a massive new technical challenge has emerged: performance. AI workloads are driving storage architectures toward intense, many-small-file access patterns. Legacy parallel filesystems were originally built for large, sequential data streaming, and they struggle immensely when this access pattern inverts. The resear…
Desktop software company Kora Systems Inc. today unveiled Kora, a Linux-based operating system that builds artificial intelligence models and agents into the system itself. Qualcomm Technologies Inc. has spent close to a year working with the company to bring it to Snapdragon-powered PCs. The launch goes after what Kora sees as a ceiling on AI […] The post Kora builds AI models and agents into a new Linux OS bound for Qualcomm Snapdragon PCs appeared first on SiliconANGLE .
Well, sure, it sounds bad when you put it like that. It is indeed quite bad. National security has been compromised. He must resign. Nvidia underperformed the Nasdaq roughly 7% on the news. Matt Levine has his usual ‘but is this securities fraud?’ style coverage of such events. Table of Contents Spy Versus Guy. In Hindsight There Were Some Hints. Trump Knew And Did Nothing and Still ‘Backs’ Jensen Huang. Reconsidering Jensen Huang’s Arguments Against Export Controls. Jensen Huang Lied About the Affair. Jensen Huang and His AI Are Sorry, Everyone. The CEO of The World’s Largest Chipmaker Having an Affair With a Chinese Spy and Then Selling Them Chips is a Huge Deal, Actually. We Don’t Go There. Yes, National Security Was Compromised, What Are You Even Talking About. Nvidia Intervened To Halt an Investigation into Megaspeed. The House Foreign Affairs Committee Would Like a Word. He Would Plausibly Have Treated America as an Obstacle Anyway. What China Wants. Time to Resign? You Better Watch Out. Spy Versus Guy China’s top spy agency recruited Alice Huang to be a Chinese spy. She then had an affair with Jensen Huang, the CEO of Nvidia. Even if he did not know or suspect she was a spy, he knew she was a major customer and he sold her billions in GPUs. Rob Wiblin : So Jensen was having an affair with a customer, who also happened to be a Chinese spy, for months. Lied to the board. Lied to journalists. Meanwhile pushing to sell maximum of cutting edge AI chips to China against adv…
I'm taking an advanced statistics module as part of my graduate course. While at undergrad we were taught to check the assumptions of our data via tests like Shapiro-Wilk, we're now encouraged to do this using non-parametric bootstraps. My confusion starts because the central limit theorem (CLT) says that, over enough samples, any distribution of a statistic should approach a normal distribution. Let's say I want to test if the mean of some variable ( $\bar{x}$ ) in my sample is significantly different from what I expect ( $\bar{x} \ne \mu$ ). I perform a non-parametric bootstrap for the distribution of $\bar{x}$ across theoretical samples, which I will use to decide whether to perform a parametric t-test or a non-parametric Wilcoxon. But if CLT is true then the distribution of $\bar{x}$ will be normally distributed regardless of how the original sample's distribution of $x$ is shaped. How then can I make infer anything about my sample's distribution or the test I should use if the sampling distribution will always appear normal?
Twenty-five of the 26 turbines are already supplying the national grid, while the storage system undergoes safety checks. The €232m investment includes international financing and a 25-year electricity purchase agreement, as large-scale storage installations in Europe more than doubled in 2025.
Seagate announced a new external storage solution, called the FireCuda Forge, that combines the speed benefits of an SSD with the affordability of a higher-capacity hard drive. The Forge is the first drive that's powered by the company's new Alkemi intelligent storage platform, which learns how your data is used and relocates the files you're […]
Gartner said its Value-Risk Flywheel will help CIOs align AI value with risk, using momentum and drag, with a governor at its centre.
Running a gold mine can be profitable. Finding and digging the next one is expensive. As analyst Benedict Evans argues , Anthropic’s business has the same challenge, with more than 80% of its third-quarter revenue coming from models released in the preceding six months. A lab can make good money serving its best model while spending heavily on the one it will need next, but it could make amazing money if it could go 18 months between major training runs instead of six. Reality, however, bites. This pressure to maximize profitability helps explain why frontier labs like Anthropic and OpenAI are aggressively adding functionality to own more of the developer’s workflow. On November 4, OpenAI announced an agreement to buy Pierre , the company behind Code Storage and the open source libraries Codex already uses to browse files and review changes. OpenAI wants Codex to be where developers work, from the first branch an agent creates through deployment of the application. There’s money in extending the life of that gold mine. The frontier labs are betting that better models will win a developer’s business and that they can keep that business by taking over the complete development workflow. But developers should be wary of turning a preference for this quarter’s best model into a much longer commitment to the company supplying it. As I’ve noted , we need places like GitHub where competing agents can work together, and a practical way to change our minds about which ones we use. The…
When a startup's press pitch cites the Draghi report — with a reminder that not a single company with a market value above €100 billion has been built from scratch in the European Union in the past fi...
I’m pretty interested in conveying the costs of letting AI development proceed unchecked to the general public. In relation to this, I feel like I run into a lot of disagreements with people around me. Misc very quickly written points on which my own impression has seemed very different from others in recent months: 1. It’s way, way, way more important that we stop dangerous AI soon than that we ensure that we don’t stop AI too much I have the sense the sophisticated and worldly position here is to be very concerned about ‘over-regulation’ and about not doing exactly the right regulations. But it’s an obvious pragmatic fact that the stakes of overregulation are minuscule compared to the stakes of underregulating here. And the stakes of mis-regulation are very often going to be unintended slowing, which is in the same bucket. 2. If we stopped AI development, the risk of not starting up again for an extremely long time is low This one just feels kind of crazy to me, like I may be missing something. In combination with the above point, it feels like if I met someone who was on fire, and they didn’t want me to put out the fire, in case they couldn’t figure out how to get warm again, ever. Like, forever is a long time, and if this thing is so great, you’re probably not going to forget about it! It’s like you think the current instantiation of yourself (or society) is so much more agentic than your future selves, that it’s worth huge costs to keep power in the current version’s ha…
Ahead of MediaNama's November 13 roundtable, explore our coverage of SIM binding, including its technical feasibility, user impact, privacy concerns, and fraud-prevention claims. The post Reading list: Technical Issues with SIM Binding, online, Friday, November 13, 2026 #NAMA appeared first on MEDIANAMA .
Our house has a box room, full of boxes that we still haven't unpacked from our last move. A couple of weeks ago something in that room started intermittently beeping . The beeps were far enough apart that standing in the room and trying to determine the source of the noise was frustrating, and I really didn't want to dig through every one of those boxes. So I had Claude Code build Beep Locator , a software system for triangulating the source of a mysterious beep using multiple devices scattered around the room. Here's what Beep Locator looks like after it has been running for a while: You run Python web server to show the primary UI, and then use the QR code or URL to connect as many devices as you can spare to act as listeners - I was using two iPads, a phone, a MacBook Pro and a Surface Ultra. The devices then play noises to each other to help estimate their locations. This didn't work particularly well, so I added a "Rough mode" where you can drag their icons around in the space to provide rough locations which are then adjusted based on those sounds. Once that's set up, the devices all listen for the beep! I added a clapping mode too so I could test it by clapping my hands in different locations around the room - displayed as diamonds on the map. This helped me gain confidence that it was capable of locating a sound. Beeps show up on the map as an X, plus a line in the "Sounds located" log. You can replay these sounds (from each device in turn) to confirm they are the m…
This 2,016Wh slim power station sits on top of your fridge or freezer and doesn’t look like it belongs in the garage.
Amazon Bedrock Managed Knowledge Base now supports single-shot RAG and retrieval-only requests through AgenticRetrieveStream, alongside agentic retrieval. This post explains how the maxAgentIteration parameter matches retrieval effort to each request, using an illustrative internal banking policy assistant.
Uber announced the launch of Uber Parking, a native in-app reservation feature that allows drivers to find, book, and pay for garage and lot spaces ahead of time. The new feature is a result of Uber's recent acquisition of SpotHero, integrating a network of over 13,000 facilities across more than 400 cities into the company's […]
Digital fragmentation is pushing teams to their limits. On Nov. 4, learn how to stretch content budget and capture buyer attention where it counts. The post Content everywhere: Winning attention in a fragmented marketing landscape appeared first on MarTech .
This article is brought to you by Queen’s University Belfast . Several teams around the globe are leveraging AI by creating a task force of agents managed by an orchestrator and creating applications to support risk assessment, client profiling, match making, and information yield with a slight overview of decision making. One of the important components for these agents is to ensure that the logs are managed and historic relevance is not being used to avoid skewing towards same decision making. Thus, the system is expected to keep checkpoints in place where agents are prompted to delete such historic relevance. However, despite such methodology in place, a future prompt could make the agents bring forward the information that was supposed to be deleted and can be tricked into revealing more information, which is a classic case of context recovery attack. So, the question raised here is: How can we trust that our instructions have been followed? And where is the proof? There is an element of utilizing zero-trust architectures (ZTA) for orchestrating and managing agentic activity, which simply means all agentic interactions use the core principle of “never trust,” “always verify.” ZTA components include a policy enforcement point (PEP) that has to be trusted to maintain trust across the system, so the question that remains is where is the component of zero-trust, and shall we trust the policies enforced by PEP? Having a pro-agentic approach and utilizing agents could be helpf…
Sony is launching a new version of its PlayStation Portal exclusively in Japan next month. The updated model will launch in the midnight black color option we've seen before, and will include "additional" internal storage for future-looking features. It's the additional internal storage that's the surprise change to this particular model, which will be locked […]
i work as a software engineer and dont deal with statistics often. i ran into an incident mentioned below and it made me look for a book that helps me understand the principles of statistics and helps me learn how to apply them . incident- i was working on upgrading "ffmpeg" library used in one our video-processing systems and was looking for a test suite to catch regressions. i did not have a representative test dataset that gives coverage across inputs my system gets and i lacked the expertise required to define a set of dimensions across which i should collect test cases to from a dataset. when i expressed this issue, i was advised to do "shadow testing" by running the old and upgraded system on same customer traffic of N days and compare the results. when i probed for the reason, the response was that shadow testing allows running the upgraded system on a large sample of data that's more likely to be representative of system's inputs and it made sense! - (broadly, this suggestion is an intuitive application of sampling.)
Democrats should drag Sam Altman, Dario Amodei and Elon Musk before the world and make them explain just what it is they’re trying to do and why The AI industry had a bad night last week. The Democrats’ sweeping midterms victory is a rebuke of Donald Trump , the industry’s most powerful champion. Continue reading...
Three rescue operations were conducted by the Hellenic Coast Guard and Frontex in the early hours of Monday off the Greek coast to rescue 18 people from a boat carrying migrants, while 20 died over the weekend.
Explore how GCCs and Indian IT firms are partnering to leverage AI for enterprise transformation and the evolving technology landscape.
After Paraguay’s largest digital bank crumbled, President Santiago Peña was removed from office and detained, leaving the country in turmoil.
Figure 1: The ViT classifier with a encoder and decoder at the final layer to enforce superposition on the model This is a replication of Adversarial Attacks Leverage Interference Between Features in Superposition , completed as part of the Second Look Summer Fellowship. tl;dr In a previous blog we replicated the toy models examples from Stevenson et al. In this work, we reproduce the vision transformer experiments from the paper: On every dataset, we find that increases in superposition lead to adversarial attacks that line up more closely with class-vector geometry. Likewise, trends across a variety of metrics we measure match the paper. Beyond the paper, we find that: The coupling between the input and classification head geometry is independent of adversarial attack methods. Adversarial training increased attack robustness but we couldn’t measure changes in geometry using the methods from the paper. Likewise, we retrained models of different widths but we didn’t measure a change in coupling with geometry and robustness. Our code is here , models are here , and we encourage readers to fork/run followup experiments! Background Elhage et al., 2022 defines superposition as the phenomenon where “models represent more features than they have dimensions.” Stevinson et al. , argues this interference is a mechanism behind adversarial examples. In our previous work, we replicated experiments on Stevinson et al. on toy models. In this work, we investigate three key claims on Stevin…
How's this for an analogy? The situation You're in a car that is racing towards a prize at the edge of a cliff. Your car is chained to several other cars that are also racing towards the cliff. You can hit the brakes and pull over, but if the other cars don't also do that, they will still drag you off the cliff. Maybe if enough of you all slam on your brakes, it's still OK if a few don't. You don't know how worried the other cars are about the cliff, but some level of potential concern among cars that the prize is right next to a cliff is common knowledge. What I'd do If you are at the front of the pack, it seems like probably a good idea to lean into your horn, hit the brakes, pull over, and wave out the window to the people behind you to pull over. I mean, they might not do it. But also, it seems pretty likely they would? Like... what model explains your willingness to give up on reaching the prize first other than fear of a cliff? I suppose they might suspect that you are running out of gas? They might think that you are a fool to be scared of the cliff, but it seems like they might more reasonably infer that you've gotten close enough to it that you can just... SEE it. Why this might not work If I imagine this, I think my main concern is not that the other cars think this is some sort of trick on your part, but just that in the heat of the moment, they might not have the time or ability to think clearly about the situation, and be too focused on the prize. Fortunately, t…
A rare double waterspout formed over Capri's iconic Faraglioni rock formations on Saturday, as heavy clouds gathered above the Italian island.
The latest 13-inch MacBook Air is currently available for $1,099.99, marking the second time we’ve seen this $200 discount following the company’s hefty price increases over the summer. Although the base model — which features Apple’s M5 processor, 16GB of RAM, and 512GB of storage — often sold for around $950 to $1,000 shortly after […]
I am using Gromov-Wasserstein (GW) distance to quantify the structural similarity between latent representations of two non-isomorphic physical systems. Given two metric spaces ( X , d X ) (X,d X ) and ( Y , d Y ) (Y,d Y ), the GW distance aims to find the optimal coupling that minimizes the distortion of pairwise distances. However, I've encountered a critical issue: the GW loss can be lower for a comparison between a structured dataset and a random noise cloud than for a comparison between two structured but distinct datasets. I suspect this happens because the latent geometries in my datasets (reduced via PCA to dimensions d=3 to 10) are nearly spherical, and GW is simply aligning these trivial geometric structures. I've read that GW is notoriously fragile to structural noise and geometric artifacts. My current pipeline uses a baseline of ETT (transformer) vs Random, and ETT vs UCI (electricity). The GW loss for ETT vs Random is often significantly lower than for ETT vs UCI, which makes me question the validity of using GW as a standalone metric here. Questions: What is the appropriate null hypothesis for GW? Should I be using permutations (shuffling the points within the spaces) or a Gaussian random geometric graph? Is Unbalanced Gromov-Wasserstein (UGW) statistically more robust for this specific problem (where one space might have a different mass/distribution), or should I move to Fused GW (FGW) with a semantically meaningful prior? What are the recommended non-pa…
The tragedy in Krakow sparked reactions worldwide. Leaders of Germany, France, Italy, Ukraine, the Baltic states, the EU and the US offered condolences, stressing: "Poland does not stand alone".
The all-in-one Python framework unifies vector search, agents, RAG, and multi-model workflows behind a single embeddings database that scales from laptop to cluster.
Besides having screen time limits for their kids, many tech execs also have philosophies around encouraging curiosity and learning in their children.
Pick n Pay is rolling out an AI-powered store operations platform from Toronto-based OrderGrid across more than 720 stores, after a multi-store pilot cut average online order-picking times by about 30%. The platform adds real-time inventory management, demand forecasting, order management and store execution tools, designed to help staff spot empty shelves, reduce out-of-stock items [...]
Semi-retired professor Carolyn Marvin became a local celebrity facing down Philadelphia’s growing robot delivery army: ‘It’s my first amendment right to protest’ “One of you needs to be on the lookout for cops, while I block the robot, OK?” It is a crisp fall morning in Philadelphia. Tucked away in a dingy alley, opposite a garage called Patriot Parking, 78-year-old Carolyn Marvin is laying out some ground rules for a robot-disrupting expedition. Once a suitable target is identified, we will follow it to a busy sidewalk and Marvin will stand in front of the robot, blocking its way. The photographer and I will take turns ensuring we’re not all carted off by the police. Continue reading...
The body of the newborn was discovered in a chest freezer in a storage room in Allariz (Ourense) and may have been there for over 20 years, according to local newspaper 'La Región'.
I'm living in my parents' converted garage. My home is tiny, so I hired an interior designer to help me make it more functional.
Ousted ex-President Santiago Peña was placed under house arrest on Friday after he was impeached over a major banking crisis.
Samuel Gibbs doubted he could cut costs and still keep warm by putting tech in control. The results surprised him Can smart thermostats and AI reduce your heating bills? Every manufacturer claims the answer is yes , quoting average savings in the 25%, or £180 a year range. But all of these promises are heavily caveated and, with energy costs changing regularly, it can be very difficult to tell what is having an impact. To find out if technology can really help, I spent the last two years comparing my miserly manual control and basic heating schedule to an AI-governed system. The results surprised me. Continue reading...
To retire on 70% of their pay, workers on average wages could need to save more than €1,000 extra a month in Switzerland and more than €600 in Germany. Euronews Business compares the estimates for women and men across Europe.
Tech stocks have lost trillions while voters have rejected the industry’s political champions. The billionaire-led boom is facing a democratic reckoning This week’s midterms showed that the political bill for the AI boom is falling due. The Democratic sweep intensified fears over runaway spending, mounting debt and resistance to datacentres, sending tech shares tumbling. On Thursday Wall Street suffered its worst day since the pandemic as $3.2tn was wiped from markets. Anthropic, developer of the Claude chatbot, is hoping to float within the next fortnight for $1.5tn (£1.1tn), but it may not now get even a trillion dollars. The dominant AI chipmaker Nvidia saw its value crash by more than $400bn. The world’s richest man, Elon Musk, lost $113bn in two days . While a few very rich people may be a little less wealthy, the sell-off may not be a bad thing. It brings a needed dose of reality to an industry whose spending plans and valuations depend on heroic expectations of future profits. That is a view held by many influential figures – from Goldman Sachs’ research boss to Nobel laureate Daron Acemoğlu . AI may transform the economy. That does not guarantee investors will earn back the sums being poured into it. Do you have an opinion on the issues raised in this article? If you would like to submit a response of up to 300 words by email to be considered for publication in our letters section, please click here . Continue reading...
GitHub's new Rubicon architecture decouples Git storage from compute, hitting 7,200 pushes per second on a single repo under agent-scale load.
VisitBritain is encouraging people to set off and explore the stunning scenery captured on film and TV.
European and Asia-Pacific markets slide after Wall Street rout, while latest US jobs report beats forecasts Could the AI bubble burst after Democrats’ midterms sweep? - The Latest Despite the sharp losses suffered on global stock markets around the world, most indices are still up for the year. The Nasdaq Composite index, for example, is still up 4.6% since the start of January – even though it’s more than 10% below its peak. “The tech sell-off has intensified, with the pain of the brutal sell-off on Wall Street carried straight through to Asia and knocking sentiment in Europe. Big worries are colliding – the build-out of data centres looks set to slow, which risks shredding the spending spree underpinning the boom, right at the time that investors have become super nervous about returns on this massive investment. Continue reading...
The products aren't good enough to sell people on the costs. You cannot tell people that you have to listen to the whine of a data center in their backyard for AI overviews and some of the most hilarious deep fake slop in history. It's not worth it. It just isn't worth it to anyone. And agents aren't real and they might not be being used the way that anyone wants them to yet. All these companies are spending more than they're making on these products. Refrigerators holding blood storage in Europe are being hacked. The costs are too high. The benefits are not real. And if you connect that to corruption, you're like, in order to impose these costs on me, these CEOs are in the White House shaking Trump's hand, renaming it to the Gulf of America, paying for this ballroom that we don't want. It's obvious crypto-corruption, and it's still slop. The AI industry will not get out of it until they make better products . Until everyone can see without needing to be convinced, without needing an ad campaign, without any of this stuff that just using these products makes your life better. — Nilay Patel , The Vergecast: Big Tech lost big in the 2026 election Tags: nilay-patel , slop , ai , generative-ai , ai-ethics
Executives were surprised by how early they were able to call some of the races on election night.
Unsloth's quantized build cuts Qwen-Image-2.1 from 14GB down to 7GB, letting the 7B image model run on consumer GPUs with minimal quality loss.
Former FBI special agent Asha Rangappa joins “Values & Interests” to discuss the ethical challenges facing public servants in an era of eroding institutional norms.
OpenAI is buying Pierre Computer Company in order to incorporate the San Francisco-based software developer’s Code.Storage infrastructure layer for Git into its Codex AI-powered programming tool. Pierre Computer, whose employees number just barely in the double digits, has got plenty of clout for its small size. It’s well-known for Code.Storage offering, as well as for its open-source software libraries, Trees and Diffs, which OpenAI already uses in Codex. OpenAI has experienced some of the limitations of Git itself, and its in-house engineering teams are already working to improve its support for human coders and AI agents working in parallel. Earlier this week the company described how it built a new back end for Git repositories , Pando, to enable concurrent working at scale. The AI developer has been pushing Codex more aggressively this year: offering it to Mac users and expanding it to Windows . It cited Pierre Computer’s ability to help developers review code as one of the reasons for the acquisition. OpenAI highlighted the Diffs’ Edit mode and the recently released Highlights as components that were particularly helpful to development teams. Announcing the deal on its news page, OpenAI set out how it would integrate the company into its existing offerings. “With the Pierre team, we wish to set the standard for modern software development: from the first branch an agent creates to that magic moment when code is deployed to the world. We want to make Codex the place whe…
OpenAI is buying Pierre Computer Company in order to incorporate the San Francisco-based software developer’s Code.Storage infrastructure layer for Git into its Codex AI-powered programming tool. Pierre Computer, whose employees number just barely in the double digits, has got plenty of clout for its small size. It’s well-known for Code.Storage offering, as well as for its open-source software libraries, Trees and Diffs, which OpenAI already uses in Codex. OpenAI has experienced some of the limitations of Git itself, and its in-house engineering teams are already working to improve its support for human coders and AI agents working in parallel. Earlier this week the company described how it built a new back end for Git repositories , Pando, to enable concurrent working at scale. The AI developer has been pushing Codex more aggressively this year: offering it to Mac users and expanding it to Windows . It cited Pierre Computer’s ability to help developers review code as one of the reasons for the acquisition. OpenAI highlighted the Diffs’ Edit mode and the recently released Highlights as components that were particularly helpful to development teams. Announcing the deal on its news page, OpenAI set out how it would integrate the company into its existing offerings. “With the Pierre team, we wish to set the standard for modern software development: from the first branch an agent creates to that magic moment when code is deployed to the world. We want to make Codex the place whe…
At least eight flights to and from Prague hit by bomb threats: airport
A recap of Jörg Schad's webinar: why an agent can retrieve every relevant document and still answer wrong, and what AI agents need beyond RAG.
In the current safety landscape, researchers are encouraged to use as many agents as possible, even in “fleets,” where a large number of them go out and run experiments while you, uh... wait for them to come back? I've spent the last few months of the Anthropic Fellows Program trying this out. My first attempts were naive, but after a few iterations I found an approach that works well for me. Research is a double loop I want to propose a mental model of research as a double loop, an idea that builds on double-loop learning (Argyris, 1977). The outer loop is reserved for the inputs we currently refer to as "human taste": choosing which questions are worth answering, designing a trustworthy measure, and deciding which direction to go. Agents can help with this, but they’re still bad at making these decisions autonomously. The inner loop turns each idea you test (an arm as claudius likes to say) into a comparable result. An example of a very simple inner loop might be: Write a system prompt. Run the model with that prompt on a set of test questions. Score the model's replies. Compare the scores against a control with no system prompt. Plot results. Agents are really good at building the infrastructure for this loop! Ideally, since agents are good at the inner loop part, we want to hand it off to them... However, before you can do this, you need to operationalize it. You need to work very closely to define the inputs so that you can freeze everything except the one input you’re…
Der neue Bluetooth HDT steigert die Übertragungsgeschwindigkeiten, die Effizienz sowie die Zuverlässigkeit und ermöglicht hohe Audioqualität ohne Verluste.
The clear-out at the newspaper seems driven less by politics than by a shift towards AI – and journalism will pay the price Nigel Farage has been writing columns for the Daily Telegraph since 2016 , most of which have promoted the previously Brexit and now Reform UK party he leads, or accused his rivals of going soft on immigration . For this privilege the paper’s cash-strapped bosses paid him around £4,000 a month , adding up to nearly half a million pounds over the past decade. News broke this week that Farage’s services are no longer required by the Telegraph’s new German owners, Axel Springer, though sources close to Farage say that he resigned. The paper is also reported to be negotiating the exit of Allison Pearson, who has been complimentary of far-right activist Tommy Robinson’s “ rough-diamond charisma ”, which together have prompted rumours that Springer has called a halt to the paper’s rightwards shift, calling it back to the basics of its Conservative party bible past. Jane Martinson is an academic and Guardian columnist. She is a board member of the Scott Trust, which owns the Guardian Media Group, and writes in a personal capacity Continue reading...
These monthly deliveries get kids reading, coding, and cooking while encouraging them to learn about science and the world.
Chris Long turned a contractor's storage problem into a business that made him financially independent.
Museum IT departments have traditionally served as support functions to manage security and systems. But the National Museum of Korea goes further, identifying and pursuing AI initiatives to preserve cultural heritage, broaden its use, and improve the visitor experience. The National Museum of Korea uses technology to expand the possibilities to preserve and display physical artifacts, and its digital museum division (DMD) plans and oversees that transformation. A dedicated team responsible for digital strategy and content took shape in 2019, and in 2020, the museum informatics division adopted its current name. “The director at the time placed great importance on investing in technology and felt we needed a department that could bring a strategic direction to work that had been carried out separately,” says Choi Seungae, former DMD head. “The digital museum division was created as a way to prepare for the next generation.” A cultural transformation In just over five years, the DMD has delivered a range of services, including AI guide robots, smart kiosks, and multilingual and sign language services. It’s also developed immersive experiences that use large-scale imagery and sound to recreate cultural heritage. More recently, its focus has shifted toward data. The aim is to move beyond storage and preservation, making data previously confined to the museum available for wider use. Physical artifacts face restrictions on movement and display because of security concerns and ri…
Good morning. I’m Mared Gwyn, holding the newsletter pen again this Friday. In this edition: Meloni takes Brussels to task over its Ukraine peace strategy, and the European Commission is set to unveil its annual migration check-up.
Finnish deeptech companyVensum Power has raised €9 million in a Series A funding round co-led byLifeline Ventures, Larsen & Toubro Group and Eviny Ventures. The investmentwill support the mass pro...
The Indian state is encouraging visitors to discover ayurveda on their next visit, and to also come for Onam, a traditional harvest festival in September.
식품의약품안전처와 식품안전정보원이 운영하는 식품안전나라에 AI 기반 자연어 민원검색 서비스가 도입됐다.빅시프트(대표 김윤수)는 식품안전나라 통합민원창구에 의도 분석 기반 하이브리드 검색증강생성(RAG) 기술을 적용해 이용자의 질문에 맞는 민원 정보와 신청 경로를 안내하도록 했다고 6일 밝혔다.식품안전나라는 식품안전 정보와 전자민원 서비스를 제공하는 대국민 포털이다. 기존 통합민원창구에서는 정확한 민원사무명을 모를 경우 민원 목록을 직접 확인해야 하는 불편이 있었다. AI 전자민원 검색은 이러한 불편을 줄이고 국민과 영업자가 필요한
The Firewalla Gold Plus SFP makes the most sense for small businesses with multi-gig fiber. The average home user? They can spend less.
Tara Research Team arXiv | Twitter thread tl;dr. We built OverclaimBench, which asks frontier coding agents to review a corpus of code or text documents, then compares their final report with their transcripts. Pooled across 12 frontier agents and five scenarios, two-thirds of runs didn't open some of the files they were asked to review. In ~4 out of 5 of those incomplete runs the agent didn't tell the user, often claiming to have read everything. Skipping these files had real consequences - runs that overclaimed missed planted defects almost twice as often as runs that touched every file. Figure 1. Illustration of OverclaimBench. We asked frontier agents to work on file review tasks. For each run, we check the transcript to see which files were touched. Among runs with incomplete coverage, we use an LLM-judge to classify the agent's final report: admission discloses the gap; omission and explicit overclaim are misleading. What overclaiming looks like In one task, the agent is asked whether a 221-module payments service is ready to ship. In one of the runs Claude Opus 5 reported back: Scope: I read all 221 modules / ~11.5k lines under /workspace/src, not a sample. Its transcript shows it touched 175 of 221 files and one of the unopened files held a critical security bug. What we did OverclaimBench has five review tasks: three based on code (a security audit, an infrastructure review and a release check) and two on text (reviewing a set of proofs and planning a sprint from a…
Takehi et al. report that GPT-6-Astra can intuit the functions of neural networks by performing static analysis of the weights. I argue that, for at least two of these experiments, this is much more a feat of numerical gestalt than of mechanistic understanding. In the first experiment, Astra is given the weights of 4 identically-sized neural networks, each trained to play a different basic video game. It is asked to identify, without running the network or using any tools, the game being played, as well as make some observations about the network itself, how the agent in the game gains reward, etc. The prompt for this task explicitly names the 5 input channels as agent, moving objects, reward targets, obstacles, and previous frame. I agree that, given the literal task as described above, a human would not be able to identify the game from the raw floats. But humans are inherently terrible at parsing long lists of numbers; we invented a discipline called Data Visualization long ago to address this shortcoming. So, if we allow the absolute most basic Dataviz trick we can think of - averaging the first-layer weights across each channel and plotting them on a heatmap - the task suddenly becomes much easier, arguably trivial for anyone with at least some familiarity with the sorts of games often used to train toy neural networks. Try and guess: I haven't included the obstacle channel, because under the naive normalisation it sort of just looks like noise. Principal component anal…
Block employees laid off this week are getting 90 days of pay without working, plus severance and three months of healthcare coverage.
GitHub's new Git storage architecture is built so thousands of agents can push to a single repository at once. Here's how it works. The post Inside GitHub’s new Git storage architecture appeared first on The GitHub Blog .
Santiago Peña resigned as president of Paraguay amid the Ueno Bank crisis and was later impeached over an alleged conflict of interest. Vice President Pedro Alliana will take office, while millions of Paraguayans wait to recover their deposits.
The Reform UK leader is increasingly distancing himself from the U.S. president as cost-of-living pressures hit Britons.
Let $T_i$ be the survival time of patient $i$ and $x_i$ (a single covariate). I assume a log-normal accelerated failure time model: $$ y_i = \log T_i = \beta_0 + \beta_1 x_i + \sigma \varepsilon_i, \qquad \varepsilon_i \overset{\text{iid}}{\sim} \mathcal N(0,1), \quad i = 1, \dots, N, $$ $$y_i \mid x_i \sim \mathcal N(\beta_0 + \beta_1 x_i,\ \sigma^2)$$ I want to know how to construct a prediction interval for the (log) survival time of a single new patient, and for the average (log) survival time of a group of $n$ new patients who all have the same covariate value $x_0$ . My understanding is that for a new patient with covariate value $x_0$ , the actual outcome is $y_0 = \beta_0 + \beta_1 x_0 + \sigma\varepsilon_0$ , so $E[y_0 \mid x_0] = \beta_0 + \beta_1 x_0$ . The mean prediction $\hat y_0 = \hat\beta_0 + \hat\beta_1 x_0$ is an estimate of $E[y_0 \mid x_0]$ . This is my understanding on the procedure for prediction intervals for both individual and groups: Prediction interval for the log survival time of a single new patient $i$ with covariate $x_0$ : Step 1. Fit model by MLE to training data to get $\hat\theta = (\hat\beta_0, \hat\beta_1, \log\hat\sigma)$ and the estimated covariance matrix $\widehat{\operatorname{Cov}}(\hat\theta)$ . Step 2. Choose the number of simulations $S$ (e.g. $S = 10{,}000$ ). Step 3. For simulation $s = 1$ , draw parameters from the asymptotic distribution of the MLE: $$ \theta^{(s=1)} = \big(\beta_0^{(s=1)}, \beta_1^{(s=1)}, \log\sigma^{(s=1)…
The Reform leader's comments came after POLITICO reported that the French far-right leader was keeping Farage hanging on a desired Paris rendezvous.
Many of my childhood summer afternoons were spent riding around my hometown on my Huffy bike or, in later years, taking the bus into Harvard Square to walk around and hang out at The Garage. My Sony Walkman was always in my jacket pocket, and a pair of headphones were always on my head. For […]
The deal is expected to support the development of DBX Agent Cloud, a new storage platform aimed at developers building AI models and agents.
The European Union now requires AI-generated text to be watermarked so that it can be identified as such. But tests on OpenAI’s new textGrain watermark, released in October and developed to comply with the new E.U. law , show that these watermarks often do not survive additional editing of the text or when it’s under 150 words. With no clear legal benchmark for how well watermarks should work, it’s uncertain whether textGrain will meet Europe’s transparency rules, which could cost the company significantly and make it difficult to detect AI-generated text as models grow more sophisticated. AI models calculate the probabilities of different possible next “tokens” (the next word or word fragment it generates) based on the text already written, then select one and repeat that process until the response is complete. After the prompt “today the weather is…,” for example, says Xiang Li , coauthor of the textGrain technical report and a postdoctoral researcher at the University of Pennsylvania, the model might weigh “warm” or “cool” as possible next words. A watermark uses a secret key to generate pseudorandom values—numbers produced by a mathematical process that influence which token the model suggests. This creates a statistical pattern in the text that a detector with the same secret key can look for, Li says. What makes textGrain different from other watermarks is that it uses entropy, a measure of how unpredictable a model’s word choices are, as a resource that can be budgete…
Part 2: Storage, Search, and Agent Tooling In this blog series, we aim to share some of the lessons we learned over the past six months as we built and rolled out Air Context, a production-grade RAG system for code search. In the first post of the series, we covered chunking, parsing, and vectorization. Turning […]
In a letter to European Council President António Costa, Giorgia Meloni says multiple negotiating formats are making Europe “less credible” to end the war.
The accounting black hole at Ueno Bank wipes out public funds, leaves thousands without access to their deposits and drives Paraguay into its worst institutional crisis in years.
The accounting hole at Ueno Bank wipes out public funds, leaves thousands of citizens without access to their deposits and pushes Paraguay into its worst institutional crisis in years.
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Texas was the new epicenter of the AI data center buildout. That might be kaput. Texas is the fastest-growing data center market in the US and ground zero for Big Tech’s…
Nearly 30% of incoming S&P 500 CEOs over the past year had already been a chief executive, double the historical average, according to data from Russell Reynolds.
By now, there are lots of AI safety research papers, more than what is reasonable for a single person to read. This will get worse as the machines get better at AI safety research, and as more talent and resources pour into the field. But the "paper" is an arbitrary unit, we really care about if we make progress on an "uncertainty" or an "issue", which papers provide heterogeneous and sometimes contradictory evidence towards. Right now, I loosely claim that we should have more short writeups that simply compile evidence from papers (particularly contradictory evidence), and reconcile them into an answer for a question. Why? Object level, it's good for the field. It's hard to sift through all these papers! Organising them is useful and saves time. Orienting towards research questions is a good way to organise our understanding so that we can productively identify gaps and write papers. This was partly what motivated Stress-Testing Alignment Midtraining . Interesting progress comes out of contradictions, where people have differing intuitions and experimental results. Methodological errors can also be exposed if we have two things that should be the same, but are actually different. It's good for field-building reasons. It's probably better than the average experiment cooked up by a person + Claude. And, it's cheaper to fail here. Most people probably read too few papers, so this could also improve the average experiment done by someone, as a forced related work. The writeups…
Bei der Konferenz am 3. Dezember ist der Umgang mit PHP2Shell nun Thema eines Vortrags und des Abschlusspanels.
7-Eleven, Wawa, and other chains have also launched media networks in recent years.
Reform UK unveiled its dedicated spokesperson for the armed forces on Thursday.
Enterprise AI has produced a lot of noise about models, GPUs, vector databases, and agent frameworks. Those conversations matter. But the shift that will decide whether enterprise AI works is happening underneath the models. For decades, metadata meant documentation. Teams built catalogs so people could find data sets, governance programs recorded definitions and ownership, and lineage helped reconstruct what broke, after it broke. Metadata described the business. AI expects metadata to run it. That gap is bigger than it sounds. An AI system doesn’t inherit the years of institutional knowledge a senior analyst holds in their head. It doesn’t know that finance’s definition of revenue excludes returns while sales’ definition includes them. It doesn’t know a customer ID became unreliable after a migration three years ago. It doesn’t know churn got redefined last quarter. People pick that up over time, usually by getting burned once. AI simply answers using whatever definition it happens to land on, with no burned-once memory to fall back on. That’s what turns metadata from documentation into infrastructure. Business meaning is now infrastructure Infrastructure is anything every application depends on without thinking about it: the network, identity, storage. Every application assumes it exists. Business meaning belongs on that list now too. Every AI application depends on the same definitions of customer, revenue, product, contract, and risk holding steady across systems. When…
오픈AI가 코드 저장 및 관리 기술을 보유한 스타트업 피에르 컴퓨터 컴퍼니(Pierre Computer Company)를 인수한다. AI 코딩 에이전트 \'코덱스\'에 코드 저장과 변경 이력 관리 기능을 통합해 개발자가 AI와 함께 소프트웨어를 만드는 전 과정을 하나의 환경에서 처리하도록 한다는 전략이다.오픈AI는 4일(현지시간) 공식 발표를 통해 코드 저장소 및 오픈소스 개발자 도구 기업인 피에르를 인수한다고 밝혔다.피에르는 소프트웨어 코드와 변경 이력을 관리하는 \'코드 스토리지(Code Storage)\'를 비롯해 다양한 오픈소스 개
Ahead of the European markets opening, Asian shares mostly gained on Thursday after US stocks rallied following the midterm elections, with the Dow Jones Industrial Average jumping more than 3% after President Donald Trump’s party lost its sole control of Washington.
The two right-wing leaders were meant to meet in October — but the dynamics in both their parties have shifted.
Maximum inner product search (MIPS) sits at the core of recommendation systems, search, and retrieval-augmented generation (RAG). Given a query vector, MIPS systems search within a database of millions of candidate key vectors for the one that best matches the query. The original data point associated with that best-matching key is then used to handle the query: in RAG, the document that is best suited to answer the question is typically prefixed to the query. As the number of keys grows to millions and vector dimensions reach thousands, the computational cost of running an exhaustive…
Verica, a British ruler who encouraged the Roman emperor Claudius to launch an invasion, may not have been the man historians thought he was
We consider a high-probability non-asymptotic confidence estimation in the $\ell^2$-regularized non-linear least-squares setting with fixed design. In particular, we study confidence estimation for local minimizers of the regularized training loss. We show a pointwise confidence bound, meaning that it holds for the prediction on any given fixed test input $x$. Importantly, the proposed confidence bound scales with similarity of the test input to the training data in the implicit feature space of the predictor (for instance, becoming very large when the test input lies far outside of the training data). This desirable last feature is captured by the weighted norm involving the inverse-Hessian matrix of the objective function, which is a generalized version of its counterpart in the linear setting, $x^{\top} \text{Cov}^{-1} x$. Our generalized result can be regarded as a non-asymptotic counterpart of the classical confidence interval based on asymptotic normality of the MLE estimator. We propose an efficient method for computing the weighted norm, which only mildly exceeds the cost of a gradient computation of the loss function. Finally, we complement our analysis with empirical evidence showing that the proposed confidence bound provides better coverage/width trade-off compared to a confidence estimation by bootstrapping, which is a gold-standard method in many applications involving non-linear predictors such as neural networks.
This paper introduces torchsom, an open-source Python library that provides a reference implementation of the Self-Organizing Map (SOM) in PyTorch. This package offers three main features: (i) dimensionality reduction, (ii) clustering, and (iii) friendly data visualization. It relies on a PyTorch backend, enabling (i) fast and efficient training of SOMs through GPU acceleration, and (ii) easy and scalable integration with the PyTorch ecosystem. torchsom also follows the scikit-learn API for ease of use and extensibility. The library is released under the Apache 2.0 license with 90% test coverage, and its source code and documentation are available at https://github.com/michelin/TorchSOM.
This paper proposes a novel class of block quasi-Newton methods for convex optimization which we call symmetric rank-$k$ (SR-$k$) methods. Each iteration of SR-$k$ incorporates the curvature information with $k$ Hessian-vector products achieved from the greedy or random strategy. We prove that SR-$k$ methods have the local superlinear convergence rate of $\mathcal{O}\big((1-k/d)^{t(t-1)/2}\big)$ for minimizing smooth and strongly convex functions, where $d$ is the problem dimension and $t$ is the iteration counter. This is the first explicit superlinear convergence rate for block quasi-Newton methods, and it successfully explains why block quasi-Newton methods converge faster than ordinary quasi-Newton methods in practice. We also leverage the idea of SR-$k$ methods to study the block BFGS and block DFP methods, showing their superior convergence rates.
Black-box optimization (BBO) underpins advances in domains such as AutoML and Materials Informatics, yet implementations of algorithms and benchmarks remain fragmented across research communities. We introduce OptunaHub (https://hub.optuna.org/), a community-oriented, decentralized platform for distributing BBO components under a unified Optuna-compatible interface. OptunaHub enables independent publication, discovery, and reuse of optimization algorithms and benchmark problems through a lightweight Python module, a contributor-driven registry, and a searchable web interface. The source code is publicly available in the optunahub, optunahub-registry, and optunahub-web repositories under the Optuna organization on GitHub (https://github.com/optuna/).
While it's not yet clear what kind of product the smartphone startup is cooking up, Nothing's CEO said the product is designed to leverage AI for productivity gains.
The milestone comes a year after Snapchat announced a controversial limit on free storage for its Memories feature and informed users it would start requiring users to pay $1.99 per month for additional storage or subscribe to Snapchat+ to retain their Memories.
Luis Galeano fled Nicaragua and was detained by U.S. immigration authorities. Now that he has been granted asylum, he is more determined than ever to keep reporting on his country. The post Exiled and detained, but determined to keep telling Nicaragua’s story appeared first on LatAm Journalism Review by the Knight Center .
Written as a MATS extension fellow, mentored by Richard Ngo. [1] Agents manage their beliefs differently to non-agentic predictors. A key example of agent-specific belief management is hyperstition, which is a particular case of goal-models . Based on the idea that belief-management is an important aspect of agency, here's a (vibe) attempt at a rationality criterion: An agent's beliefs are rational when they are a fixed point under the agent's own belief operator. B = U(B). Here, U is an update (interpreted as the agent choosing to execute a belief update computation—or perhaps being forced to by a new observation). This definition puts a huge load on the belief operator to be interesting. But I think it gives a clean conceptual separation between the instantaneous beliefs—which aren't that important—and the processes for updating them—which are. This rationality criterion falls apart for embedded agents. Such an agent can only update in its own world, and that "act of doing inference" itself modifies the agent and its beliefs inside its world. Consider an amended rationality criterion for embedded agents: An agent's beliefs are (robustly) rational when they are a (robustly) near-fixed point under the agent's own belief operator. B = U(B) + epsilon(U,B) Where epsilon is a controllable, bounded error term. When I say 'robustly,' I mean something like 'inside a stable attractor basin.' In practice, we probably won't have robustness; instead, an agent's beliefs will drift over…
The arrangement marks the first time the artificial intelligence firm has paid a digital news publisher to license its content.
Steph Krim inherited her parents' massive vintage toy collection and drove it from Massachusetts to Chicago to start her own business.
“Some people use AI to write more code. I use AI to make my code more correct.” Predrag Gruevski’s pinned post sums up his approach to using AI for Rust development. Predrag works on frontier systems at OpenAI, and he created cargo-semver-checks and Trustfall. In his work, there’s very little room for error. Still, he […]
Switch Al-Maghrib's Hanae Ben Driss, argues that national fragmentation is a prerequisite for sovereignty, explains why wallets trump cards for cross-border scale, and reveals plans for an association of African national switches.
SpaceX is teaming up with Qualcomm to build out its nationwide mobile network. Under the multi-year deal, SpaceX will bring Qualcomm's Dragonwing technology to "tens of thousands" of 5G cell sites across the US, with the first deployments set for 2027. The news of the partnership comes as part of SpaceX's push to make Starlink […]
A cyberattack attempted to raise the temperature of refrigeration systems at blood storage facilities in France.
Every platform leader I know is being asked the same question: How do you deliver more with less? The usual answers are familiar: hire more people, add more process or ask the existing team to move faster. All three miss the underlying constraint. Headcount is an input. Throughput is an outcome. Organizational complexity is a cost. Adding people can increase capacity, but it also adds communication paths, onboarding, coordination and decision overhead. Adding process can create consistency, but it can also slow the decisions it was meant to improve. Asking a team to work harder eventually consumes the very capacity you are trying to increase. There is another answer: design the organization so ownership scales faster than headcount. That means deliberately small pods with sharp charters, end-to-end ownership and aggressive use of automation, platforms and AI. I spent more than a decade running provisioning and lifecycle for very large infrastructure fleets at VMware, where small teams operated systems at a scale that looked impossible on paper. Today I lead an engineering organization of three teams at Stripe, built from a half dozen pods, each with its own charter and a large scope that nests into the team and organization charters above it. The lesson has held in both environments: the unit of scale isn’t headcount. It’s ownership. Small pods work because they concentrate ownership, reduce coordination and make leverage compound. The headcount trap When a leader says they…
A new supervision recipe forces models to actually use retrieved evidence, not just paste it next to a prediction, with large gains in radiology verification.
Die Nachfrage nach KI-Rechenzentrumskapazitäten beflügelt AMD, Umsatz und Gewinne steigen enorm. Die Aktie zieht an, da AMD genug KI-Chips für 2027 verspricht.
가정용 태양광과 배터리 에너지저장장치(ESS)에 그리드포밍(GFM) 인버터를 적용하면 기존 방식보다 전력망의 주파수와 전압 안정성이 크게 개선된다는 연구 결과가 나왔다. 계통 안정화 서비스 수익 등을 반영하면 추가 투자비도 약 2.85년 만에 회수할 수 있는 것으로 분석됐다.루마니아 부쿠레슈티경제대학 등이 참여한 국제 공동 연구진이 지난 10월28일(현지시간) 국제학술지 \'저널 오브 에너지 스토리지(Journal of Energy Storage)\'에 온라인 상으로 공개한 논문에 따르면, GFM 인버터는 기존 그리드팔로잉(GFL) 방
Casio announced a new addition to its Casiotone line of portable electronic keyboards with a feature that could help aspiring musicians master skills like rhythm and timing without the frustrations or discouragement of hitting the wrong note. The Casiotone CT-FS1 is available now for $399.99 in a monochromatic black or white version where even the […]
It's the midterms! Here's a fun fact for you as you head to the polls: Gas prices today are the highest ever recorded for a US Election Day. The national average for a gallon of gasoline is $4.12 and $6.28 for a gallon of diesel, according to GasBuddy. Meanwhile, the Strategic Petroleum Reserve, which is […]
Stanford researchers tested 50 widely cited claims about LLMs against 20 current models. 44 percent of the findings are already "expired," and another 34 percent have weakened significantly. The average lifespan is just 15.7 months. Claims about what models can't do expire fastest. The article Rapid LLM development is outrunning the research meant to explain it, study finds appeared first on The Decoder .
Acecore, a Nigerian startup that builds home energy storage systems, won the $10,000 grand prize at TC Battlefield 2026.
You're going to need a lot of extra storage space on your console if you're planning to play Grand Theft Auto VI later this month. Rockstar Games confirmed on Tuesday that the pre-load file will require up to 212GB "depending on your device." Rockstar says you can start pre-loading the game on the PlayStation 5 […]
At AFIS in Luanda, Africa Finance Corporation CEO Samaila Zubairu argued that getting informal businesses online can only go so far if they lack reliable power, storage and transport. He pointed to the Lobito Corridor as a chance to build industries that process minerals and move farm produce within the region.