If you use the internet, you've probably had at least some personal information go missing. It's just the nature of the web. But this latest discovery, as reported by Wired, is something different. Security researcher Jeremiah Fowler found a public online database housing over 180 million records (184,162,718 to be exact) which amounted to more than 47GB of data. There were no indications about who owned the data or who placed it there, which Fowler says is atypical for these types of online databases. Fowler saw emails, usernames, passwords, and URLs linking to the sites where those credentials belonged. These accounts included major platforms like Microsoft, Facebook, Instagram, Snapchat, Roblox, Apple, Discord, Nintendo, Spotify, Twitter, WordPress, Yahoo, and Amazon, as well as bank and financial accounts, health companies, and government accounts from at least 29 countries. That includes the U.S., Australia, Canada, China, India, Israel, New Zealand, Saudi Arabia, and the UK. Fowler sent a responsible disclosure notice to the hosting provider of the database, World Host Group. Fowler was able to detect signs that the credentials here were stolen with infostealer malware, which bad actors use to harvest sensitive information from a variety of platforms—think web browsers, email services, and chat apps. Following Fowler's notice, World Host Group restricted the database from public access. The provider told Wired that the database was operated by a customer, a "fraudulent user" who uploaded illegal information to the server. In order to ensure these credentials were real, and not just a bunch of bogus data, Fowler actually contacted some of the email addresses he found in the database. He got some bites, and those users were able to confirm the records that he found associated with their emails. That's no guarantee that all 184,162,718 records are accurate, but it's a good sign that most are. As such, it's entirely possible you and I both had credentials exposed in this database. What's worse, Fowler says there's no telling how long the database was open to the public before his notice shut it down.There's a lot bad actors and hackers can do with this type of information. If they know the username and password combo to one of your accounts, they'll not only see if they can use it to break into that account, but they'll use it on other accounts of yours as well. If you reuse passwords, as many do, you could be facing a mass breach. It's bad enough when that concerns Facebook and Roblox accounts, but seeing as there were financial, health, and even government accounts here, the implications are huge.How to protect yourself If you don't have access to the database, you can't say for sure whether your credentials are listed there, or which credentials they have.Still, if you haven't changed the passwords for your accounts in some time, now might be a good time to do so. You don't need to change your passwords as frequently as traditional security advice has taught us, but it certainly wouldn't hurt to give your accounts a quick security audit. Make sure you're using a strong and unique password for each and every one of your accounts. If you repeat passwords, you run the risk of credential stuffing (hackers trying the same stolen password on multiple accounts). In order to keep tabs on those passwords, use a secure password manager. Make sure you're using two-factor authentication (2FA) on all of the accounts that allow it. That way, even if a password is exposed, hackers won't be able to break into your account without the device containing the 2FA code. To boost your security, avoid SMS-based 2FA when possible, and opt for more secure 2FA options, like an authenticator app or physical security key. If your account offers it, try a passkey to combine the convenience of a password with the security of 2FA.

Introduction AlphaEvolve [1] is a promising new coding agent by Google’s DeepMind. Let’s look at what it is and why it is generating hype. Much of the Google paper is on the claim that AlphaEvolve is facilitating novel research through its ability to improve code until it solves a problem in a really good way. […] The post Google’s AlphaEvolve: Getting Started with Evolutionary Coding Agents appeared first on Towards Data Science.

Microsoft is now testing a new feature in Notepad that can generate text for you using AI. It’s part of a Windows 11 update being released to Windows Insiders in the Canary and Dev channels with Copilot Plus PCs. The update also includes new AI-powered features for Paint like a sticker generator, and improvements to the Snipping Tool that can automatically crop and resize screenshots based on what you’re selecting.Notepad’s new Write feature can be used to “quickly draft text based on your prompt,” according to the Windows Insider Blog, “or build upon existing content with AI-powered assistance.” You can either right-click where you want new text to be inserted in a document or after selecting existing text you want to use as a reference.After selecting the new Write feature in Windows’ Copilot menu, a prompt will appear where instructions can be entered. The AI-generated output will appear in the document but you’ll have the option to keep it, discard it if it’s not what you’re looking for, or refine the output by entering follow-up prompts. The feature was first discovered earlier this year in code included in test builds of Windows 11.Write follows two other AI-powered tools Microsoft has been testing with Notepad. Summarize, first introduced last March, can generate a summary of the text in a document while Rewrite, which began testing last November, can adjust the tone of text, shorten or lengthen it, and rephrase sentences.Paint’s new Sticker generator creates a small selection of stickers based on your description that can also be used in other apps. Image: MicrosoftJoining AI features already being tested in Microsoft Paint like Generative Fill, Generative Erase, and the Cocreator tool that can create images based on text prompts and reference sketches, is a new Sticker generator. A new button in the Copilot menu opens a prompt where you can describe the sticker you want to create. Paint will then generate a small set of stickers based on your description that can be added to your canvas or copied over to other applications.Paint is also getting a new Object select tool that relies on AI to create smart selections that takes most of the grunt work out of isolating and manipulating specific elements in a photo or painting.Windows 11’s Snipping tool will use AI to automatically adjust the size and cropping of a screenshot based on what you’re selecting on screen. Image: MicrosoftTo potentially make it easier to screenshot exactly what you need and immediately share it without edits, Windows 11’s Snipping Tool is gaining a new Perfect screenshot button on its toolbar. When framing a screenshot using the rectangle tool Perfect screenshot will use AI to “intelligently resize based on the content in your selection,” but you’ll still have the option to resize or reposition the selected region of your screen before capturing it.It’s being joined by a new Color picker in the Snipping tool that lets you see the HEX, RGB or HSL color values below an eyedropper cursor, and it can be zoomed for more precision by either scrolling or using Ctrl +/- keyboard shortcuts.You’ll need to be signed into your Microsoft account to use Notepad’s new Write feature which will use the same credits system as other AI-powered Windows 11 features do. Microsoft hasn’t yet announced what pricing for these credits will be if it eventually starts charging for them.See More:

Start SlideshowStart SlideshowImage: Games Workshop / KotakuIf you were worried that Warhammer’s crossover into gaming may have peaked, I’ve got good news. The Warhammer Skulls 2025 video game showcase was full of teases, trailers, and announcements for tons of new stuff. The tabletop miniatures franchise and its deep, bloody universe of sci-fi and fantasy storytelling continues to thrive across every genre imaginable, from real-time strategy to, uh, retro typing boomer shooters? Also the original 40K: Space Marine is back with a 4K glow-up. Here’s every big Warhammer announcement we got today.Previous SlideNext Slide2 / 13List slidesTotal War: Warhammer III’s next DLC is Tides of TormentList slidesTotal War: Warhammer III’s next DLC is Tides of TormentThe latest Lords Pack arrives in summer 2025 alongside new units, gameplay mechanics, and additional free content. Previous SlideNext Slide3 / 13List slidesWarhammer 40,000: Boltgun gets a sequelList slidesWarhammer 40,000: Boltgun gets a sequelThe excellent pixel art boomer shooter returns with Boltgun 2. It’s coming to PS5, Xbox Series X/S, and PC in 2026. In the meantime fans can sink their armored fists into Warhammer 40,000: Boltgun – Words of Vengeance, a free-to-play shadow drop that turns the FPS into an old-school typing game. Previous SlideNext Slide4 / 13List slidesThe original Warhammer 40,000: Space Marine will be remastered List slidesThe original Warhammer 40,000: Space Marine will be remastered A Master Crafted Edition of the Xbox-360-era cult classic will arrive on June 10 (available through Xbox Game Pass) with 4K graphics, modern controls, and improvements to the interface, characters, and audio. First-time fans of last year’s Space Marine 2 can see where it all started while they bide their time for a promised sequel. Previous SlideNext Slide5 / 13List slidesWarhammer 40,000: Rogue Trader’s second story expansion is almost here List slidesWarhammer 40,000: Rogue Trader’s second story expansion is almost here Lex Imperialis is the next DLC and it’s coming out June 24. The 2023 CRPG will have a new 15-hour storyline campaign focused on the Adeptus Arbites faction.Previous SlideNext Slide6 / 13List slidesWarhammer: 40,000: Dark Heresy is a new CRPG in the worksList slidesWarhammer: 40,000: Dark Heresy is a new CRPG in the worksThe studio behind Rogue Trader, Owlcat Games, also revealed Dark Heresy, its next role-playing game in the Warhammer 40K universe, a party-based investigation of grand conspiracies about waging “a secret war against heresy.” There’s no release date yet but it’s coming to PS5, Xbox Series X/S, and PC. Previous SlideNext Slide7 / 13List slidesWarhammer 40,000: Space Marine 2's long-awaited horde mode is out next monthList slidesWarhammer 40,000: Space Marine 2's long-awaited horde mode is out next monthSiege mode pits players against wave after wave of Tyranid and Chaos on Kadaku. Teams of up to three players will be able to earn new rewards in the horde onslaught as well as rank up just like they would through existing Operations. They can also call in Dreadnought mechs for support during matches. Previous SlideNext Slide8 / 13List slidesDid you know there’s a Warhammer racing game?List slidesDid you know there’s a Warhammer racing game?Warhammer 40,000: Speed Freeks just hit its 1.0 release on Steam. The full launch brings player-hosted servers, solo play, 150 new items to unlock and the end of in-game, shop-based microtransactions. The vehicular chaos sim is moving to discrete DLC packs instead. Previous SlideNext Slide9 / 13List slidesClassic RTS Warhammer 40,000: Dawn of War returns List slidesClassic RTS Warhammer 40,000: Dawn of War returns Relic Entertainment has resurrected its beloved 2000s-era real-time strategy game with a Definitive Edition that’s coming to PC later this year. It sounds like there might be some graphical upgrades and usability tweaks, but otherwise it will mostly be the same game operating under the hood. Old-school RTS fans are eating well right now, and hopefully it can serve as inspiration for an eventual Dawn of War 4. Previous SlideNext Slide10 / 13List slidesSupremacy is a new Warhammer 40,000 multiplayer strategy game for mobileList slidesSupremacy is a new Warhammer 40,000 multiplayer strategy game for mobileIt seems like it will be a Warhammer 40K-coded version of Supremacy 1914, a map-based strategy game about high-level army management. It’ll take place on the brutal battlefields of Vigilus and players can pre-register on the App and Google Play stores starting today. Previous SlideNext Slide11 / 13List slidesWarhammer 40,000: Darktide’s first new paid DLC class has arrivedList slidesWarhammer 40,000: Darktide’s first new paid DLC class has arrivedFatshark’s 2022 multiplayer FPS that swaps Vermintide 2's bloody castles for the industrial hellscape of Atoma Prime has finally turned a corner after a rough launch and a long rebuilding phase. With a good foundation now underfoot, the loot shooter is adding the Arbitrator class, a ruthless enforcer with a cyborg hound at its side. It goes live June 23. Previous SlideNext Slide12 / 13List slidesWarhammer 40,000: Mechanicus 2 reveals first gameplay trailerList slidesWarhammer 40,000: Mechanicus 2 reveals first gameplay trailerThe sequel to Bulwark Games’ 2018 XCOM-style turn-based tactics RPG is out on PS5, Xbox Series X/S, and PC later this year, and we now have a better look at what’s new this time around. The trailer shows a glimpse of tactical skirmishes and unit customization. Mechanicus 2 will also be the first Warhammer video game to feature the Leagues of Votann.

Sora Shimazaki / Pexels In another stark reminder of the constant threats online, cybersecurity researcher Jeremiah Fowler recently uncovered a massive, unsecured database containing over 184 million login credentials from Microsoft, Apple, Facebook, Discord, Google, PayPal and others. The trove amounted to approximately 47.42 GB of data, was discovered on a misconfigured cloud server and is believed to have been amassed using infostealer malware – malicious software designed to extract sensitive information from compromised devices. Recommended Videos A global breach with far-Reaching implications According to Jeremiah, the database also contained over 220 email addresses associated with government domains from at least 29 countries, such as the United Stated, United Kingdom, Australia, and Canada. The breadth underscores the potential national security risks posed by such breaches. Fowlers analysis of a 10,000-record sample revealed that the data included plaintext usernames and passwords, with some entries linked to financial terms like “bank” and “wallet,” indicating a heightened risk of financial fraud. The presence of such sensitive information in an unprotected database amplifies concerns about identity theft, unauthorized access and other malicious activities. Hackread.com has some images from the database provided by Jeremiah. The role of infostealer malware Infostealer malware operates by infiltrating devices through phishing emails, malicious websites, or comes bundled with pirated software. Once installed, it can harvest a variety of data, including login credentials, cookies, autofill information and even cryptocurrency wallet details. The data is then transmitted to command-and-control servers operated by cybercriminals. The discovery of this database suggests a coordinated effort to collect and potentially exploit vast amounts of personal and institutional data. The lack of identifiable ownership or metadata within the database further complicates efforts to trace its origins or intended use. Hosting companies likely do not know that they are fostering these databases to begin with. Immediate actions and recommendations Upon discovering the database, Fowler promptly notified the hosting provider, World Host Group, which subsequently took the server offline. However, the duration for which the data remained exposed and wither it was accessed by unauthorized parties before its removal remains uncertain. I would advise users to: Change your passwords, yet again: Immediately update your passwords for all online accounts, especially if the same passwords are being re-used across multiple platforms Enable two-factor authentication (2FA): This generally requires a text verification code to your phone, or a secondary email address Monitor your accounts: Regularly check your financial accounts and other sensitive accounts for suspicious activity Use reputable security software: Anti-virus and malware software from reputable companies usually help, make sure they are updated. You can check out our antivirus and malware reviews Be cautious with emails and downloads: Avoid clicking on suspicious links or downloading attachments from unknown sources

Large Reasoning Models (LRMs) like OpenAI’s o1 and o3, DeepSeek-R1, Grok 3.5, and Gemini 2.5 Pro have shown strong capabilities in long CoT reasoning, often displaying advanced behaviors such as self-correction, backtracking, and verification—collectively known as “aha moments.” These behaviors have been observed to emerge through outcome-driven RL without the need for supervised fine-tuning. Models like DeepSeek-R1 and its open-source replications (e.g., TinyZero and Logic-RL) have demonstrated that carefully designed RL pipelines—using rule-based rewards, curriculum learning, and structured training—can induce such reflective reasoning abilities. However, these emergent behaviors tend to be unpredictable and inconsistent, limiting their practical reliability and scalability. To address this, researchers have explored structured RL frameworks that target specific reasoning types, such as deduction, abduction, and induction. These approaches involve aligning specialist models, merging them in parameter space, and applying domain-specific continual RL. Tools like Logic-RL use rule-conditioned RL to solve logic puzzles, improving transferability to tasks like math reasoning. Meanwhile, other works propose mechanisms to enhance reasoning robustness, such as training models to reason both forwards and backwards, or iteratively self-critiquing their outputs. Studies analyzing “aha moments” suggest that these behaviors stem from internal shifts in uncertainty, latent representation, and self-assessment, offering new insights into engineering more reliable reasoning models.  Researchers from the National University of Singapore, Tsinghua University, and Salesforce AI Research address the limitations of relying on spontaneous “aha moments” in large language models by explicitly aligning them with three core reasoning abilities: deduction, induction, and abduction. They introduce a three-stage pipeline—individual meta-ability alignment, parameter-space merging, and domain-specific reinforcement learning—significantly enhancing model performance. Using a programmatically generated, self-verifiable task suite, their approach boosts accuracy over instruction-tuned baselines by over 10%, with further gains from domain-specific RL. This structured alignment framework offers a scalable, generalizable method for improving reasoning across math, coding, and science domains.  The researchers designed tasks aligned with deduction, induction, and abduction by using a structured “given two, infer the third” format based on hypothesis (H), rule (R), and observation (O). Deduction is framed as satisfiability checking, induction as masked-sequence prediction, and abduction as reverse rule-graph inference. These tasks are synthetically generated and automatically verified. The training pipeline includes three stages: (A) independently training models for each reasoning type using REINFORCE++ with structured rewards, (B) merging models through weighted parameter interpolation, and (C) fine-tuning the unified model on domain-specific data via reinforcement learning, isolating the benefit of meta-ability alignment.  The study evaluates models aligned with meta-abilities—deduction, induction, and abduction—using a curriculum learning setup across difficulty levels. Models trained on synthetic tasks strongly generalize to seven unseen math, code, and science benchmarks. At both 7B and 32B scales, meta-ability–aligned and merged models consistently outperform instruction-tuned baselines, with the merged model offering the highest gains. Continued domain-specific RL from these merged checkpoints (Domain-RL-Meta) leads to further improvements over standard RL finetuning (Domain-RL-Ins), especially in math benchmarks. Overall, the alignment strategy enhances reasoning abilities, and its benefits scale with model size, significantly boosting performance ceilings across tasks.  In conclusion, the study shows that large reasoning models can develop advanced problem-solving skills without depending on unpredictable “aha moments.” By aligning models with three core reasoning abilities—deduction, induction, and abduction—using self-verifiable tasks, the authors create specialist agents that can be effectively combined into a single model. This merged model outperforms instruction-tuned baselines by over 10% on diagnostic tasks and up to 2% on real-world benchmarks. When used as a starting point for domain-specific reinforcement learning, it raises performance by another 4%. This modular, systematic training approach offers a scalable and controllable foundation for building reliable, interpretable reasoning systems.  Check out the Paper and GitHub Page. All credit for this research goes to the researchers of this project. Also, feel free to follow us on Twitter and don’t forget to join our 95k+ ML SubReddit and Subscribe to our Newsletter. Sana HassanSana Hassan, a consulting intern at Marktechpost and dual-degree student at IIT Madras, is passionate about applying technology and AI to address real-world challenges. With a keen interest in solving practical problems, he brings a fresh perspective to the intersection of AI and real-life solutions.Sana Hassanhttps://www.marktechpost.com/author/sana-hassan/RXTX: A Machine Learning-Guided Algorithm for Efficient Structured Matrix MultiplicationSana Hassanhttps://www.marktechpost.com/author/sana-hassan/From Protocol to Production: How Model Context Protocol (MCP) Gateways Enable Secure, Scalable, and Seamless AI Integrations Across EnterprisesSana Hassanhttps://www.marktechpost.com/author/sana-hassan/Researchers from Renmin University and Huawei Propose MemEngine: A Unified Modular AI Library for Customizing Memory in LLM-Based AgentsSana Hassanhttps://www.marktechpost.com/author/sana-hassan/Meta Introduces KernelLLM: An 8B LLM that Translates PyTorch Modules into Efficient Triton GPU Kernels

When you purchase through links on our site, we may earn an affiliate commission. Here’s how it works. JetBrains outlines what's coming in WebStorm 2025.2 David Uzondu Neowin · May 22, 2025 15:12 EDT Recently, JetBrains brought its AI Assistant to Visual Studio Code as an extension. Now, the company has published a roadmap for WebStorm 2025.2, its JavaScript and TypeScript IDE, outlining some interesting updates slated for July. This follows the 2025.1 release in April, which developers are already using and providing feedback on. The upcoming 2025.2 release seems focused on enhancing support for contemporary development tools and improving performance. First up, Bun support is getting a significant upgrade. Regarding this, the WebStorm team detailed: If your project includes a bun.lockb file, WebStorm will automatically detect it and suggest using Bun as your package manager. No manual setup needed. You’ll be able to run files using Bun directly from the context menu, just like you would with Node.js. Work also continues on the new service-powered type engine, a behind-the-scenes rework of how WebStorm handles TypeScript types. It's meant to make things faster, especially for projects with large frameworks or deep component trees. The WebStorm team is specifically working on improving tag and attribute completion in React (TypeScript) and Angular templates, along with addressing "edge cases with third party libraries," such as fixing an issue with missing props from libraries like motion when the new engine is active. CSS is not being overlooked. WebStorm 2025.2 plans to incorporate support for more modern CSS features, including recognition of new property values, recently introduced at-rules, and updates from the CSS Color Module, such as parsing newer color formats like color-mix(). Furthermore, a long-standing issue concerning CSS variables not resolving correctly inside language-injected code, common in CSS-in-JS scenarios, is being addressed. Finally, JetBrains is preparing WebStorm for the future evolution of TypeScript. As the TypeScript team at Microsoft is rebuilding its compiler and language service in Go for better performance, JetBrains is laying the necessary groundwork within the IDE. However, you won't see visible changes from this in the 2025.2 version. If you want to test out these features as soon as they become available, consider joining the 2025.2 Early Access Program. However, keep in mind the usual caveats of EAP builds, such as potential instability and features that might change or be removed before the final release. Tags Report a problem with article Follow @NeowinFeed

For years, the URL bar in your browser has done double duty: you can enter a web address, sure, but you can also use it to enter search requests. Now Norton has a new “free” browser, Neo, which adds a third function: AI prompting. Norton—most famous for its antimalware solutions like Norton 360 Deluxe—is entering the browser market with Norton Neo, an AI-first browser that you can sign up for and join the waitlist. Norton already ships a browser, the Norton Private Browser, which is similar but lacks AI. Neo replaces the search/URL bar with what it calls the “magic box,” where you can “search, generate, and take action,” according to Norton. It’s not immediately clear whose AI solution you’ll be able to use or whether it will run in the cloud or locally on your PC. When you do search, however, Norton promises that you’ll be able to “peek” at a search result to get an AI-powered summary—and yes, it appears to be a list of links, rather than Norton’s own take on Google’s AI Mode. The Magic Bar will also enable you to perform AI-powered writing tasks, such as drafting an email, directly from it. There’s another tweak that Norton is making, although I can’t tell how it’s going to work out: tables browsing. “No more tab chaos—Neo organizes everything, so you don’t have to,” Norton promises. Rather than remove tabs entirely from the AI, Neo will just use AI to auto-group them, according to a related FAQ. Opera One’s Tabs Islands already does this, as does Microsoft Edge, using the “Tab groups” feature. And yes, there’s an integrated ad blocker, which more and more browsers are integrating natively. Norton says that it will not sell your browsing data, leaving room to use “minimal, anonymized data” itself; it also will block “intrusive ads” and trackers by default. Norton does explicitly say that it will not use your data to train AI. Norton doesn’t explicitly say that it’s built on Chromium, the open-source underpinnings of Chrome and Edge, though you can use Chrome plugins, Norton says. Most browsers are free, though Norton waffles a bit in this regard as well. “Yes, Neo is free to download and use with an invitation code for Alpha testing,” Norton says. “We may introduce optional premium features later, but the core experience will always remain accessible.” Although most users still use either Google Chrome or Microsoft Edge, there’s a small collection of third-party browsers, including Brave, Opera, Vivaldi, and more, that promise to do more than the mainstream browsers do. Soon, you can add Neo to the list.