AI/ML News & Innovations Hub

Multi-tenant LLM analytics with row-level security: How we built a secure agent on AWS

In this post, we show you how PAR built a production-ready multi-tenant LLM analytics system that enforces row-level security through a three-layer architecture: cryptographic request signing with AWS SigV4, semantic validation on Amazon Bedrock, and programmatic data isolation via Split-Plane SQL. We demonstrate how each layer operates independently to reduce the risk of cross-tenant data exposure, even when the LLM itself is compromised or manipulated.

Arena, the AI leaderboard everyone uses, is now a $100M business

The startup, which runs a popular free AI leaderboard, launched its commercial service just last September.

Injection Molding Defect Detection for Medical Components

Use RF-DETR and Gemini 2.5 Pro to detect defects on molded components and generate automated inspection observations.

Build an agentic AI healthcare claims pipeline with Amazon Bedrock and AWS HealthLake

In this post, we show you how to build an automated claims processing pipeline using two key Amazon Bedrock capabilities: Amazon Bedrock Data Automation for intelligent document extraction from healthcare claim forms, and Amazon Bedrock AgentCore for hosting an AI agent that validates and transforms the extracted data into FHIR (Fast Healthcare Interoperable Resources) resources in AWS HealthLake. You will learn how to combine these services to create an end-to-end workflow that reduces manual processing while maintaining accuracy through automated validation checks.

Mageia 10 keeps the 32-bit Linux flame alive

Polished Mandriva descendant still makes room for PCs the 64-bit world has left behind

Highlights from Git 2.55

The open source Git project just released Git 2.55. Here is GitHub’s look at some of the most interesting features and changes introduced since last time. The post Highlights from Git 2.55 appeared first on The GitHub Blog .

Debugging production agents with Amazon Bedrock AgentCore Observability

In this post, you learn how to debug production agent failures using built-in observability capabilities. We walk through common failure patterns, show how to analyze agent behavior with traces and metrics, and provide structured workflows for resolving issues such as infinite loops and tool invocation failures. This is Part 1 of a two-part series. Part 2 covers performance optimization and memory management.

EU and China seek to head off trade war with new dialogue

Chief trade negotiators from the two sides launch a consultation mechanism focused on trade imbalances, export controls and other disputes.

Eleven killed in France as parachute training aircraft crashes near Nancy

A Pilatus PC-6 carrying five instructors, five student nurses, and a pilot went down 300 metres from the runway shortly after take-off.

$1M AI x-risk grant round is live on grantmaking.ai - apply for funding, review applicants, or fund projects

TLDR: what is the grant round? grantmaking.ai is launching a $1M grant round , distributing $5k to $50k per successful application to people and projects working to reduce x-risk from AI. Applications will be reviewed by Gavin Leech , Ryan Kidd , and Marcus Abramovitch . We aim to make all funding decisions by July 28th. Applications submitted by July 13th are guaranteed a priority review. You can still apply after July 13th, and we will make our best effort to review late submissions as long as funding remains. Grant applications will be mostly public, though we allow certain sensitive details to be kept private. Even if you are not applying, we invite you to join the platform to review and comment. We have set aside $100k of the budget to be given to top commenters as regranting budgets, so please share your thoughts and help us pick out awesome projects! Who are we? grantmaking.ai was initialized by Anton Makiievskyi, who is funding this round and brought the team together, built by Matt Brooks (lead dev) and Melissa Samworth (ui/ux), and advised by Austin Chen with Manifund handling grant distribution. Why we’re building this platform & launching a grant round You can read our initial pre-launch post to learn more about what we’re building and why. In short, we want to build the most comprehensive public repository of donation opportunities in existential AI safety space with essential information like up-to-date funding needs, theory of impact, references, endorsements,…

AI is changing expectations for MBA graduates

While MBA job prospects are expected to remain stable in the near future, new hires will be held to higher standards inside rapidly evolving workplaces. The post AI is changing expectations for MBA graduates appeared first on Cornell AI Initiative .

Exclusive: France seeks to block UK role in €5bn EU Scaleup fund

AI will make biological extinction risks worse before it makes them better

An argument goes: If we don't build aligned artificial superintelligence, we risk driving ourselves extinct for some other reason. We should rush to build ASI quickly, in spite of the risks—the longer we wait, the more vulnerable we are to extinction from a different cause. Other than ASI, the biggest extinction risk is synthetic biology. Some lab could (accidentally or on purpose) develop a highly transmissible, 100% fatal super-plague that wipes out humanity. An aligned ASI could stop that from happening by shutting down dangerous biological research, or by developing advanced countermeasures that stop the spread of deadly infections. So the argument goes: We need to build ASI to save us from non-AI extinction risks. However, that argument doesn't work. In the near term, AI will make biological risks worse , not better. AI will accelerate scientific research, which will bring us closer to the level of knowledge necessary to build extinction-level pathogens. And in the long term, the way ASI eliminates biological x-risk is by taking control of the world. Cross-posted from my website . In the near term, AI makes biorisk worse Some people imagine that AI models would accelerate defensive research while refusing to assist with developing bioweapons. This plan has two minor issues and one fatal one. The first minor issue: Current AI model refusals are not robust, and there are workarounds to get information out of them for people who want to. It's very hard for AI developers to…

Cursor now has a mobile app for guiding your coding agent on the go

Cursor has launched a new mobile app for remote oversight over coding agents.

EU sets October deadline to get 'tangible' results with China

As EU Trade Commissioner Maroš Šefčovič met his Chinese counterpart Wang Wentao in Brussels on Monday, he said he would travel to China this autumn. Germany’s trade Minister Katherina Reiche also met Wentao, with her ministry calling on Beijing to ensure “a level playing field".

Pakistan airstrikes leave homes in ruins in Afghanistan

Villagers inspect destroyed homes in Afghanistan's Paktia province after overnight airstrikes, as Afghan and Pakistani officials give differing accounts.

Claude Meets Blackwell Ultra: Anthropic’s Models Now Run on NVIDIA GB300 in Azure

Anthropic’s Claude models in Microsoft Foundry — hosted on Microsoft Azure and running on NVIDIA GB300 Blackwell Ultra GPUs — are now generally available, giving Azure-native enterprises a powerful new way to build autonomous and domain-specific AI agents. As agentic AI continues to drive enterprise innovation and becomes more autonomous, organizations need access to computing […]

Trump administration threatens 92 GW of new electricity supply with red tape

The Trump administration's moves threaten $121 billion in new solar and wind power, two energy sources that are the biggest contributors to new capacity in the U.S.

Holy Kaaba to be ceremonially washed on Tuesday in centuries-old Islamic tradition

Oman Air says Khasab flights to resume once safety conditions are met

Internet down? 3 ways I use an old Android phone as a backup connection for my home router

I needed a way to connect my phone to my router, and have it work reliably. Turns out, I have 3 options (and I saved the best for last).

NBCU Faces Major Merger Questions After Its Comcast Split

Could a streamer like Netflix swoop in to buy NBCU? Experts weigh in.

World Cup 2026: Shawk the hawk predicts Brazil vs Japan round of 32 clash

The best July 4th sales we found so far

July 4th sales are typically a precursor to what we’d see during a mid-July Prime Day, but obviously things are flipped around this year. Last week’s big Prime Day sale is over, yet there are a number of familiar deals still poking around in the week leading up to the nation’s birthday. Best Buy is […]

I've been to 38 states, but there's just one that truly blew me away — and I still can't stop thinking about it

After visiting 38 states in the US during my travels, there's just one I can't stop thinking about for its scenery, history, and national parks: Utah.

Gradient-free Single-pass Model Beats nanoGPT on Shakespeare

Beam is a character-level language model that computes count tables mapping character contexts to next-character frequencies. At prediction time, each order looks up the current context in its count table and produces a distribution over the vocabulary, smoothed over a symmetric Dirichlet prior ₒⱼ Each order receives a capacity score composed of two terms: Concentration: ₒ where H(pₒ) is the Shannon entropy of the smoothed distribution. This is 1 when all mass is on one token and 0 when the distribution is uniform. Reliability: where n is the total count for the current context. This saturates toward 1 as evidence accumulates and is 0 when the context has not been observed. A third term, capacity, is computed from the product of concentration and reliability. The capacity scores are converted to weights via softmax at temperature τ = 0.10: ₒₒⱼⱼ The low temperature makes the routing nearly winner-take-all: the highest-capacity order almost always dominates. The final prediction is the weighted geometric mean of the per-order distributions: ₒₒₒ This was chosen deliberately to assign high probability to a token only when multiple weighted orders agree. The model has four hyperparameters: the set of context orders, α, τ, and the reliability threshold (min_count = 1). These were selected by evaluating variants on the validation set. Results Evaluation uses the nanoGPT shakespeare_char benchmark: character-level Shakespeare, about 1M training tokens, about 100K validation tokens,…

Miroma Group Acquires Ad Results Media For Big U.S. Push

ARM oversees $750M in media investments annually.

I changed these Android Auto settings to limit what Gemini learns about me - here's why

Google's AI offers a lot of convenience in your car, but you're offering up a lot of sensitive information. Here's how to put a lock on it.

Blog Intro Post

Hello LW, as I've mentioned I'm starting a blog, here's the intro post! Intro 0.0.1 Hello, welcome to (the main sequence of) my website. 0.0.2 Its purpose is to collect various observations and thoughts of mine, centered around the question of.... 0.1 Why these laws of physics? 0.1.1 Of the various things humanity has learned about the nature of reality, perhaps the most striking is the discovery of the laws of physics: a set of computable mathematical rules which govern almost all of reality as it's known to us. Even the human mind seems to be the product of physical processes in the brain. 0.1.2 A further striking discovery of the 20th century was computational universality: there is a relatively low threshold beyond which a model of computation is capable of emulating any other. 0.1.2.1 Combined with 0.1.1, this seems to imply that all of reality can be thought of as a relatively simple form of computation, perhaps a Turing machine with a small number of states. 0.1.3 Or can it? Although the laws of physics seem to be computable, they have a mathematical structure that goes far beyond an ordinary Turing machine. They take place in continuous space and time, have manifold conservation laws and group theoretic symmetries. Perhaps most weirdly of all, they are quantum mechanical in nature. 0.1.4 This raises the question: why? If reality "could" have been a more generic Turing machine or a cellular automaton, why isn't it? Wait, or does this question even make sense? What doe…

Metaphilosophy I: Philosophy as Extracting Implicit Patterns from S1 into S2

Hello LW. As I've mentioned I'm starting a blog about philosophy and physics. Here's the first post proper, about "meta-philosophy", i.e. what even is philosophy, and how could we make a program that does philosophy? I think people here might find section 1.3 most interesting. Without further ado: Intro 1.0 Since I'm planning to write some "philosophy" of a sort, I'm going to explain what I mean by that term and how and why such endeavors are justified. As we'll see, it's inevitable that the explanation is circular to an extent. 1.0.0 I'm not really going to try to explain or justify my account here in much detail, hopefully just enough to make the rest of the document intelligible. 1.0.1 There's two ways you could approach this, internal or external. We could explore what's it like to do philosophy from a first-person perspective, or try to describe from the outside an algorithm or system that can do philosophy. Here I'll do both, internal first, then an external toy model. 1.1. Philosophy as development of highest-level concepts 1.1.1 To start with, here's a sketchy account of epistemology. We have a collection of concepts , interpreted in a very broad sense. We have scientific theories of phenomena, procedural knowledge of how to do things, verbal knowledge of everyday things such as directions, subverbal knowledge of how to open a door, etc. 1.1.2 Concepts are justified by how good they are at predicting things and how useful they are for getting stuff done. Much of this…

Trace and evaluate TrueFoundry AI Gateway traffic in Arize AX

Learn how TrueFoundry AI Gateway exports OpenTelemetry traces to Arize AX so teams can trace, evaluate, and monitor production LLM and agent traffic without embedding a vendor SDK in every service. The post Trace and evaluate TrueFoundry AI Gateway traffic in Arize AX appeared first on Arize AI .

Soft Power in an Era of Nationalism

From ancient Rome to modern geopolitics, Hendrik W. Ohnesorge explores why soft power remains essential in an increasingly transactional age.

TIDAL cracks down on AI music by cutting off monetization

In addition, TIDAL will use automated tools to remove AI-generated music that attempts to impersonate an artist or a group, the company said.

I've made friends on solo trips and traveled to visit them. These relationships are just as valuable as local ones.

I made international friends on solo trips, and recently returned to Europe to see them. The visit taught me the value of long-distance friendships.

South Africa turns to drones, AI, CCTV cameras ahead of anti-migrant protests

The June 30 operation offers the clearest indication yet that South Africa is quietly constructing a technology-driven surveillance network in which state and private security systems are becoming intertwined.

Europe's record-breaking heatwave: What you need to know

Death toll tops 1,300 as record temperatures shift east, with Ukraine's war-damaged power grid now bracing for the heat's next phase.

A dietitian never gets bored of breakfast. Her formula for high-fiber, high-protein oats is inspired by sweet treats.

Josie Porter has a formula that turns her breakfast of oats, yogurt, and high-fiber toppings into healthier versions of her favorite desserts, like sticky toffee pudding and banana break.

GDRFA Dubai launches research award to drive innovation

Abu Dhabi Fund finances Dh562 million in projects, creating 10,000 jobs across four countries

Time series change detection

I'm analysing surface reflectance values from satellite imagery to detect the flowering event of a single species. I want to test whether the change in values during the peak flowering period is significant (whether flowering produces a change in SR values). The time series is only over one year, and has 75 values total from across 5 areas (15 values per area), so there might be correlation of values within the same sites. Struggling to find a way to apply change detection methods I've found to this case. Is it a matter of building a prediction model and then showing that the real values don't follow this during peak flowering months? The peak flowering for the species is during March and April, so there are 3-4 values during this period, with the remainder of values distributed throughout the other months.

VS Code stops trusting your code by default

Microsoft’s Visual Studio Code editor has reached version 1.126, which is highlighted by a new security mode for untrusted code. Also with the new release, VS Code now displays total cost for chat sessions (instead of just for individual turns) and allows users to manage multiple chats side-by-side in a single agent host session. Published June 24 , VS Code 1.26 can be downloaded from code.visualstudio.com . VS Code 1.26 introduces a new security enhancement, Workspace Trust , that lets users decide whether project folders can automatically run code and adds a layer of security when working with unfamiliar code. Previously, opening a new folder immediately raised a dialog asking the user whether to trust the folder before they could look at its contents. Now, new folders open in Restricted Mode to prevent automatic code execution. Developers can browse the code safely first and then choose whether to trust the folder. Simplified model hovering also is featured in VS Code 1.126. The model hover now shows a one-word descriptor of the model’s capabilities and includes deep link buttons that take a user directly to the relevant configuration. In an enhancement to the Agents window , the dedicated companion window for exploring agent sessions across projects and machines, a Copilot session started from an agent host now can hold several chats at once. Because the chats share the same session and working context, users can keep more than one conversation going in the same workspac…

Visual Studio Code locks down untrusted code

Microsoft’s Visual Studio Code editor has reached version 1.126, which is highlighted by a new security mode for untrusted code. Also with the new release, VS Code now displays total cost for chat sessions (instead of just for individual turns) and allows users to manage multiple chats side-by-side in a single agent host session. Published June 24 , VS Code 1.26 can be downloaded from code.visualstudio.com . VS Code 1.26 introduces a new security enhancement, Workspace Trust , that lets users decide whether project folders can automatically run code and adds a layer of security when working with unfamiliar code. Previously, opening a new folder immediately raised a dialog asking the user whether to trust the folder before they could look at its contents. Now, new folders open in Restricted Mode to prevent automatic code execution. Developers can browse the code safely first and then choose whether to trust the folder. Simplified model hovering also is featured in VS Code 1.126. The model hover now shows a one-word descriptor of the model’s capabilities and includes deep link buttons that take a user directly to the relevant configuration. In an enhancement to the Agents window , the dedicated companion window for exploring agent sessions across projects and machines, a Copilot session started from an agent host now can hold several chats at once. Because the chats share the same session and working context, users can keep more than one conversation going in the same workspac…

Inside the Advisory Database and what happens when vulnerability volume breaks records

The GitHub Advisory Database is processing more vulnerability reports than ever before. Here's what's driving the surge, how we're responding, and how the community can help. The post Inside the Advisory Database and what happens when vulnerability volume breaks records appeared first on The GitHub Blog .

A gauche, qui d’autre que Mélenchon ?

Candidats à la présidentielle, processus de désignation… y verra-t-on un peu plus clair, à gauche, avant de partir en vacances ? Alors que François Hollande et Raphaël Glucksmann se sont croisés à un raout social-démocrate dimanche, le PS et Les Ecologistes réunissent leurs instances mardi soir. Anthony Lattier fait le point avec Elisa Bertholomey et […]

Rewriting the Brand Discovery Playbook in the AI Era

This post was created in partnership with Moloco Key takeaways Marketing leaders are grappling with how AI is reshaping traditional funnels, whether it’s through generative AI summaries usurping consumer clicks, […]

Rocket Lab buys its way into the satellite big league with $8B Iridium deal

SpaceX gains a vertically integrated rival as launch specialist adds global comms network

In major privacy win, Supreme Court rules geofence warrants are protected by privacy rights

The Supreme Court's decision to limit geofence warrants is a win for privacy advocates, who called their use unconstitutional but sought an outright ban.

Fake Alignment Till You Make Alignment

“Fake it till you make it” is good advice. It may sound epistemically fraught, but it frequently works. Sometimes all it really takes to get good at something is just having the confidence that you’ll be good at it. I’ve done this many times at work, in romance, and even writing blog posts. But it only works because I’m careful to never fake my evals. By this I mean, I never fake the way I measure if I’m successful. Let’s say I’m trying to learn a new hobby, like whittling. I believe I’ll be good at it if I just put in the time, so I put in the hours carving wood. What I have to be careful to do, though, is not allow myself to move the goalposts. I need to have some clear vision in my head of what success is, and work towards that. If I carve something crappy and tell myself “actually, that’s good enough, I’m good at whittling”, that’s the way I can trick myself into just being fake. I’ve mostly avoided being fake by demanding authenticity of myself. For example, back in school, I refused to take short cuts just to pass a test. Instead, I put in the extra work to really learn something because, to me, the grade was never the point. I’ve taken a similar approach to meditation (the point is waking up, not special mental states), romance (I want a good relationship, not to be datable), and friendship (I don’t want to seem like a good friend, I want to actually be one). I bring all this up because I’ve been thinking about fake-it-till-you-make-it and authenticity dynamics lately…

Childbirth for many primate species is even harder than for humans

For decades, we’ve thought that childbirth is uniquely challenging for humans, but it turns out that many other primates find the birth process just as difficult

China understands food security — the EU does not

Europe thinks that it is self-sufficient in food, but the reality is that its supply chain depends on imports for meat production. The EU is structurally dependent on imported high-protein feed to raise farmed animals, particularly soy and maize, writes Nico Muzi in an OpEd for Euronews.