AI is exposing Linux protection holes quicker than builders can patch them. Fragnesia is the contemporary. Right here's what we recognise about it.

Steven-vaughan-nichols

Consistent with Linus's law, "given enough eyeballs, all bugs are shallow," is essential to open source. 

Unfortunately, the way to AI malicious program-finding tools, including Claude Mythos and OpenAI Daybreak, in the back of most of these eyeballs are AI engines, and they may be proving to be an awful lot quicker at finding security problems than human ones. 

So it is the contemporary critical Linux kernel vulnerability, fragnesia, that has emerged. It's the 1/3 extreme neighborhood root flaw inside the ultimate weeks.

Fragnesia yields root on all important distributions

Following in the footsteps of reproduction fail and dirty frag, this page-cache corruption bug offers unprivileged users a dependable course to full root control on affected structures. And what are the ones structures, you ask? In line with AlmaLinux, Fragnesia without delay yields root on all foremost distributions. So, essentially, all Linux distros can be targeted and effectively hacked. Are we having fun yet or what?

The Trojan horse was changed into a disclosed this week by means of the ai protection employer Zellic, with William Bowling and other researchers using the enterprise's AI-agentic software auditing tool, v12. It works by using abusing a good judgment malicious program in the Linux XFRM (brief for "transform") esp-in-tcp subsystem to write arbitrary bytes into the kernel page cache of study-handiest documents, with out requiring any race condition. 

This opens the door to local privilege escalation and potential container escapes in multi-tenant environments.

Unlike conventional race-circumstance exploits, those vulnerabilities allow attackers to exactly corrupt file-subsidized pages without timing hints, making assaults more reliable and easier to weaponize once proof-of-concept code is to be had.

An evidence-of-idea makes the most exists

Speaking of which, there already exists a proof-of-concept make the most. It builds a 256-access research table that maps all viable keystream bytes to their corresponding nonces. The attack then copies a malicious payload, which overwrites the primary 192 bytes of the transfer consumer command inside the page cache with a small elf stub that calls setresuid and calls a shell. 

In different phrases, for the ones of you who aren't linux experts, it will right away drop the attacker into a root shell.

This is horrific, bad information. It method a local consumer could benefit from superuser (root) privileges. Pink hat gives it a commonplace vulnerability scoring gadget (cvss) score of seven.Eight, which makes it an excessive-level security bug.

Just as terrible, at the same time as fragnesia is technically a nearby privilege-escalation computer virus, its effect scales dramatically in modern-day cloud architectures that run large numbers of untrusted bins on shared Linux kernels. 

Here, if an attacker can run code in a box or a confined person account however still create namespaces and network stacks, that individual ought to gain root on the host and, from there, assault different customers' virtual machines (vms) or containers.

The way to mitigate fragnesia

Kernel developers and distribution maintainers at the moment are running to harden the esp-in-tcp code path, with proposed fixes specializing in getting rid of in-vicinity differences on shared, record-sponsored pages and tightening fragment handling. An upstream patch to fix fragnesia is available now. However it's no longer currently shipping in any distro as of can also thirteen.

In the meantime, you may mitigate it by running the following command as root:

# printf 'set up esp4 /bin/falseninstall esp6 /bin/falseninstall rxrpc /bin/falsen' > /and many others/modprobe.D/fragnesia.Conf

But, if you accomplish that, you'll also knock out ipsec, which means that your linux virtual non-public networks (vpns) may not paintings. Satisfied, happy, pleasure, joy.

You could, as a substitute, in step with crimson hat, run the following command as root:

# echo "user.Max_user_namespaces=zero" > /and so on/sysctl.D/dirtyfrag.Conf sysctl --machine

Here, but there's any other trouble: it disables unprivileged user namespaces, which may also have an effect on rootless packing containers, sandboxed browsers, and Flatpak. 

It is always something!

Wait for your distro to deliver a patch

You might be better off just looking forward to your distro to supply a patch. I recognize maximum major distros are already beta-testing the patch, and that i wouldn't be surprised if patched linux kernels are to be had by way of may additionally 14. Come that day, you have to patch your systems asap.

Why is this going on? 

I'll be going into greater detail later, however for now, suffice it to mention that Chris Wright, Red Hat's CTO, and that i spoke about this very issue earlier these days, and it boils all the way down to our ai fault detectors being a whole lot higher than they had been even some weeks ago at locating real bugs. 

We can expect to see many more such protection holes discovered in the following few months.

We're going to want to get a lot quicker at fixing insects as they appear.

This, in this manner, is not only a problem for Linux. It's difficult for all open-source software, and as AI gets better at reverse-engineering binary code, Windows and other proprietary software developers will need to upgrade their repair capabilities as well.

If you're trying to limit the amount of private statistics you deliver to ChatGPT, those are the main settings you should recognize.

If you're one of the 900 million people who reportedly use ChatGPT every week, the chatbot might be a staple of life. Maybe it facilitates you getting paintings done or gives you meal plans. You would possibly even consult it whenever you have a scuffle with a friend or family member. 

But as you turn to ChatGPT for more and more in your existence, you can want to re-evaluate how a whole lot private facts you are disclosing along the way. Preferably, you understand not to reveal sensitive monetary facts -- however, different details about you could also be really worth protecting. 

Privacy professionals are already sounding the alarm about the capability harms of saying too much on your chatbot. The underlying concern is that no person is totally positive how your private facts, whether touchy or apparently innocuous, might be used in the future. A few worry that non-public data could turn out to be in a mass surveillance system or be used in other unforeseen methods with a purpose to ultimately harm or disadvantage you. 

That ambiguity, they argue, is cause enough for caution. 

Right here are 5 methods you could better manage the quantity of personal records ChatGPT has about you, if you're using a patron account. 

1. Opt out of education information

One step you could take to make your ChatGPT experience more at ease is to prevent OpenAI from using your data to educate its models. Protection experts are voicing difficulty that in case your records end up in a version, it could at some point be used in a manner we can not even count on right now.

Go to settings > facts controls > improve the version for all and sundry. Then toggle the switches off and click on "performed."

You may also use OpenAI's privacy portal to "make a privacy request." Choose "I have a client ChatGPT account," after which "do not teach on my content." From there, you might be induced to check in. After that step, you may see a button to "submit request." 

This approach only applies to your records in ChatGPT transferring ahead.

2. Delete old chats 

Any other action you can take to ease up the statistics you've given to ChatGPT is to delete old chats. There are two methods to try this. One is to visit Settings> Facts Controls> Delete all chats.

You could additionally delete individual chats from the left-hand sidebar by clicking the three dots next to the name of the chat. 

Despite the fact that the conversation will disappear from your chat records right now, it can take up to 30 days to be completely deleted from OpenAI's structures, according to the organisation's website. 

OpenAI also stipulates  exceptions in which this rule may not apply: occasions wherein the employer has to retain records for "security or prison obligations" or because the statistics have been "de-diagnosed and disassociated from your account." 

Three. Use temporary chats

If you do not need to keep up with deleting chats as you go, you can use ChatGPT's temporary chats. A brief chat will no longer appear for your records, nor will it reference something from preceding conversations or recollections. It'll now not be used for training information, either. 

Just like the retention policy for deleted chats, OpenAI would possibly keep a replica of your temporary chat for as much as 30 days, in step with an faq web page.

To start a temporary chat, click the button labeled "temporary" at the bottom of a new chat. 

For some ChatGPT users, this could cause a much less customized experience, as ChatGPT might not be capable of studying something new about you that would inform future responses. 

4. Manage recollections.

The concept behind reminiscences is for ChatGPT to preserve positive chat details that might, in idea, make the chatbot more useful over the years. As an instance, you would possibly ask it to understand that you have a dog or are vegan.

Consistent with ChatGPT's faq on reminiscences, there are fundamental settings you could use to manipulate reminiscence.

You may visit settings > personalization, after which click the "manipulate" button next to "reminiscence." That step will pull up a list of stored reminiscences that you may both delete altogether or personally. You could additionally toggle off the switches for "reference saved memories" and "reference chat records."

OpenAI may preserve a log of saved recollections for up to 30 days.

Five. Delete your account

It's a greater extreme measure, but you can continually delete your account altogether. It's a permanent circuit, though, so make sure it's what you need. One manner to do this is to visit OpenAI's privacy portal and "make a privacy request." Pick out "I have a consumer ChatGPT account," after which "delete my ChatGPT account." 

You could also go to settings > account, after which click "delete" under "delete account. Consistent with OpenAI, you could only do that if you logged on in the last 10 mins; in any other case, you may have to sign in again. From there, you will need to type your email to affirm and "delete" in order to unlock the "permanently delete my account" button. In the end, click that button.

The way to discover what ChatGPT knows 

If you're uncertain of ways a good deal of the statistics you've given ChatGPT -- and whether or not you have to take any of the steps above -- there is a way you could get a fairly comprehensive concept of what the chatbot knows about you. Just ask ChatGPT.

My editor, Aly Windsor, requested ChatGPT to ask how tons it knew about her. It replied with a thorough listing of personal details she'd shared with it. She then asked it to produce a spark that could succinctly spell out the whole lot it knew about her in a scannable profile, and fed the activation returned to ChatGPT. You could attempt the same technique. You are probably surprised by the way and how much it returns.

How to Learn Claude Code for Free With Anthropic’s AI Courses – One Course Took Me Just 20 Minutes

If you want to learn how to use Claude Code, AI agents, and Anthropic’s growing ecosystem of AI tools without spending money, Anthropic now offers a surprisingly useful free training library. The company’s online learning platform includes beginner-friendly courses on Claude, Claude Code, MCP, AI agents, and advanced integrations like Amazon Bedrock.

For developers, creators, and AI enthusiasts, these free Claude courses are one of the easiest ways to understand how Anthropic’s AI tools actually work in real-world workflows.

Anthropic has quietly built a free online learning hub filled with video-based AI training. The platform includes short tutorials, in-depth technical lessons, quizzes, and certificates for selected courses.

The available topics include:

• Claude AI basics
• Claude Code tutorials
• AI agents and automation
• MCP (Model Context Protocol)
• Prompt engineering
• Claude integrations
• Amazon Bedrock with Claude

Some lessons take only 15 to 20 minutes, while others are full-length technical training programs lasting several hours.

One of the shortest courses currently available is “AI Capabilities and Limitations,” which takes around 15 minutes to complete. On the other end, advanced learners can explore the eight-hour “Claude with Amazon Bedrock” series featuring dozens of lectures.

My Experience Taking the “Introduction to Subagents” Course

I recently tested the 20-minute “Introduction to Subagents” course because I wanted to better understand how subagents work inside Claude Code.

For anyone unfamiliar, subagents are essentially smaller AI task workers that Claude Code can delegate jobs to automatically. Instead of handling everything in one massive workflow, Claude can assign specialized tasks to separate agents and combine the results later.

Think of subagents like a team of interns:

One handles research

Another writes code

Another checks errors

Another organizes output

This approach helps improve workflow efficiency while reducing overload inside a single AI context window.

The course itself included:

• Four short YouTube videos
• Written explanations
• Examples inside the learning dashboard
• Progress tracking and completion tools

Anthropic uses Skilljar as its learning management system, and registration is completely free.

What I Learned From the Free Claude Course

Even though the course was short, it offered genuinely practical insights.

The most valuable lessons covered:

When to Use Subagents vs Main Agents

The instructor explained when a task should stay inside the main Claude workflow and when it makes more sense to delegate to a subagent.

This distinction is important for:

• Large coding projects
• Multi-step AI tasks
• Research-heavy workflows
• Agent automation systems
• How Subagents Improve Context Management

One major advantage of subagents is better context-window efficiency.

Instead of overwhelming a single Claude session with too much information, separate subagents can process smaller focused tasks independently. This can improve speed, organization, and output quality.

Writing Better Subagent Instructions

The course also covered how to optimize subagent specifications so each one performs a specialized role more effectively.

This includes:

Clear task descriptions

Limited scope assignments

Structured outputs

Defined goals

For developers using Claude Code daily, these small optimizations can significantly improve AI-assisted workflows.

What the Course Didn’t Cover

The course wasn’t perfect.

I would have liked more instruction on:

• Managing subagents while running
• Troubleshooting failed tasks
• Fixing stuck agents
• Monitoring long-running workflows

Still, considering the course is free and only takes about 20 minutes, it delivered solid value and practical takeaways.

Best Place to Start Learning Claude Code

If you are completely new to Anthropic’s ecosystem, the Claude Courses library is the best starting point for structured learning.

However, if you want broader documentation and technical resources, Anthropic also offers its larger learning portal through the:

• API documentation
• Claude developer guides
• Support articles
• Prompt engineering tutorials
• AI workflow examples
• Are Anthropic’s Free AI Courses Worth It?

Yes, especially for developers and AI power users trying to understand modern agent-based workflows.

The biggest advantage is that the courses are:

• Free
• Beginner-friendly
• Short and practical
• Focused on real AI tools
• Updated around Anthropic’s ecosystem

Unlike generic AI tutorials, these lessons are designed specifically around Claude and Claude Code usage.

If you’re exploring AI coding assistants, agent automation, or prompt engineering in 2026, these courses are a strong starting point.

Final Thoughts

Anthropic’s free Claude courses are an underrated resource for anyone trying to learn Claude Code, AI agents, and modern AI workflows without paying for expensive training programs.

Even a quick 20-minute lesson can provide practical techniques you can immediately apply inside Claude Code projects.

For developers building apps, automations, or AI-assisted workflows, spending a little time inside Anthropic’s free training library is probably worth it.

Google's actual-time quantization can be important for jogging neighborhood AI. right here's why.

A nice final result is making AI more accessible by decreasing inference expenses.

With the fee modern artificial intelligence skyrocketing to hovering prices for computer components, including reminiscence, Google last week responded with a proposed technical innovation known as TurboQuant.

TurboQuant, which Google researchers mentioned in a weblog post, is another DeepSeek AI second, a pioneering attempt to reduce the price of the latest AI. It can have an enduring benefit by decreasing AI's memory utilization, making models a lot more efficient.

In spite of this, since DeepSeek did not stop large funding in AI chips, observers say TurboQuant will likely cause enduring growth in AI funding. It's the Jevons paradox: Make some thing greater efficient, and it finally ends up growing general usage, latest that useful resource.

However, TurboQuant is a technique that could help run AI domestically by way of slimming the hardware needs of a massive language model.

Extra Memory, Extra Cash

The huge fee issue for AI in the meantime -- and in all likelihood for the foreseeable future -- is the ever-increasing use of brand new memory and storage technology. AI is information-hungry, introducing a reliance on reminiscence and garage unheard of in contemporary computing.

TurboQuant, first defined by means of Google researchers in a paper a yr in the past, employs "quantization" to lessen the number of bits and bytes required to represent the data.

Quantization is a form of the latest data compression that state-of-the-art fewer bits to symbolize the same information. In the case of cutting-edge TurboQuant, the point of interest is on what is known as the "key-fee cache," or, for shorthand, "KV cache," one among the largest memory hogs of ultra-modern AI.

When you type right into a chatbot, including Google's Gemini, the AI has to compare what you've got typed to a repository of modern measures that function as a sort of modern database.

The thing that you kind is known as the query, and it's miles matched against records held in memory, known as a key, to discover a numeric result. Essentially, it's a similarity score. The key is then used to retrieve from reminiscence exactly which phrases should be returned to you because of the AI's response, referred to as the cost.

Generally, every time you type, the AI version ought to calculate a new key and cost, which can slow down the complete operation. To hurry things up, the device keeps a key-value cache in memory to store currently used keys and values.

The cache then turns into its very own problem: The more you work with a model, the greater the memory the important thing-price cache takes up. "This scaling is a significant bottleneck in phrases cutting-edge memory usage and computational pace, in particular for long context fashions," according to Google lead writer Amir Zandieh and associates.

Making matters worse, AI models are trending to be constructed with more complicated keys and values, referred to as the context window. That gives the model greater search alternatives, potentially enhancing accuracy. Gemini 3, the present-day version, made a big leap in context window to one million tokens. earlier fashions, which include OpenAI's GPT-four, had a context window trendy simply 32,768 tokens. A larger context window also increases the amount of modern-day memory a key-cost cache consumes.

Speeding up quantization for real-time

The answer to that increasing KV cache is to quantize the keys and the values so the whole thing takes up less space. Zandieh and crew declare in their blog post that the information compression is "massive" with TurboQuant. "Reducing the KV cache size without compromising accuracy is vital," they write.

Quantization has been utilized by Google and others for years to slim down neural networks. What is novel about TurboQuant is that it's intended to quantize in real time. preceding compression tactics reduced the scale of a neural network at training time, before it's further run in production.

it is not right enough, located in Zandieh. The KV cache is a dwelling digest of what's discovered at "inference time," whilst humans are typing to an AI bot, and the keys and values are being converted. So, quantization has to occur fast enough and accurately enough to hold the cache small, even as it additionally stays up to date. The "faster" in TurboQuant implies this is a lot faster than traditional assembly-time quantization.

TurboQuant has stages. First, the queries and keys are compressed. This can be accomplished geometrically because queries and keys are vectors of modern data that may be depicted on an X-Y graph as a line, which can be rotated on that graph. They name the rotations "PolarQuant." By randomly attempting exceptional rotations with PolarQuant and then retrieving the original line, they discover a smaller range of cutting-edge bits that still preserve accuracy.

As they positioned it, "PolarQuant acts as an extremely efficient compression bridge, changing Cartesian inputs into a compact Polar 'shorthand' for garage and processing."

The compressed vectors nonetheless produce errors while the evaluation is executed among the question and the important thing, which is referred to as the "inner product" of today's vectors. To repair that, they use a second technique, QJL, brought via Zandieh in 2024. That method continues one of the vectors in its unique kingdom, so that multiplying a compressed (quantized) vector with an uncompressed vector serves as a test to improve the accuracy of the multiplication.

They tested TurboQuant by way of making use of it to Meta platforms's open-supply Llama 3.1-8B AI version, and observed that "TurboQuant achieves best downstream consequences throughout all benchmarks while reducing the key cost memory size by a factor of at least 6x" -- a six-fold reduction in the amount of modern KV cache wanted.

The method additionally differs from different techniques for compressing the KV cache, which includes the technique taken last year by using DeepSeek, which constrained key and value searches to speed up inference.

In any other take a look at, the usage of Google's Gemma open-supply version and models from French AI startup Mistral, "TurboQuant proved it could quantize the key-price cache to just three bits without requiring training or best-tuning and inflicting any compromise in model accuracy," they wrote, "all at the same time as accomplishing a faster runtime than the authentic LLMs (Gemma and Mistral)."

"It's miles exceptionally green to implement and incurs negligible runtime overhead," they discovered

Will AI be any cheaper?

Zandieh and team count on TurboQuant to have a giant impact on the production use of modern-day AI inference. "As AI becomes increasingly incorporated into all merchandise, from LLMs to semantic search, this work in fundamental vector quantization might be more vital than ever," they wrote.

However, will it, in reality, lessen the value of state-of-the-art AI? Yes and no.

In an age of agentic AI, programs including OpenClaw software that function autonomously, there are ultra-modern elements to AI besides simply the KV cache. different, latest, and trendy memories, including retrieving and storing database data, will ultimately have an effect on an agent's performance over the long term.

People who observed the AI chip international final week argued that just as DeepSeek AI's efficiency failed to slow down AI funding last year, neither will TurboQuant.

Vivek Arya, a Merrill Lynch banker who follows AI chips, wrote to his clients who had been worried about DRAM maker Micron's era that TurboQuant will truly make more efficient use of brand new AI. The "6x improvement in reminiscence performance [will] probably [lead] to 6x growth in accuracy (model length) and/or context period (KV cache allocation), in place of 6x lower in memory," wrote Arya.

What TurboQuant can do, even though, is make a few man or woman times modern AI extra low-budget, especially for local deployment.

The licensed Kubernetes AI Conformance software is establishing a brand new standard for AI-based cloud-native computing.

A safe, normal platform for AI workloads

CKACP's goal is to create community-described, open requirements for consistently and reliably running AI workloads across extraordinary Kubernetes environments.

CNCF CTO Chris Aniszczyk said, "This conformance application will create shared standards to ensure AI workloads behave predictably throughout environments. It builds on the same successful network-pushed technique we have used with Kubernetes to help convey consistency across over 100-plus Kubernetes structures as AI adoption scales."

Especially, the initiative is designed to:

make certain portability and interoperability for AI and system gaining knowledge of (ML) workloads throughout public clouds, private infrastructure, and hybrid environments, allowing groups to keep away from supplier lock-in while shifting AI workloads anyplace needed.

Reduce fragmentation by setting a shared baseline of abilities and configurations that structures should help, making it less difficult for corporations to undertake and scale AI on Kubernetes with confidence.

Deliver companies and open-source individuals a clear goal for compliance to make sure their technology works together and helps production-equipped AI deployments.

enable cease users to hastily innovate, with the reassurance that certified platforms have carried out exceptional practices for useful resource control, GPU integration, and key AI infrastructure desires, tested and demonstrated by the CNCF.

Foster a reliable, open environment for AI improvement, where requirements make it viable to successfully scale, optimize, and manipulate AI workloads as utilization increases across industries.

In a quick, the initiative is focused on imparting to each business and company a not unusual, tested framework to ensure AI runs reliably, securely, and effectively on any certified Kubernetes platform.

If this technique sounds familiar, it ought to, as it's primarily based on the CNCF's successful certified Kubernetes Conformance program. It is due to that 2017 plan and agreement that, if you're no longer satisfied with, say, Red Hat OpenShift, you may select up your containerized workloads and cart them over to Mirantis Kubernetes Engine or Amazon Elastic Kubernetes Service without demanding any incompatibilities. This portability, in flip, is why Kubernetes is the muse for plenty of hybrid clouds.

With fifty eight% of businesses already running AI workloads on Kubernetes, CNCF's new application is anticipated to significantly streamline how teams deploy, manage, and innovate in AI. By supplying commonplace check criteria, reference architectures, and confirmed integrations for GPU and accelerator support, the program targets to make AI infrastructure more robust and comfortable across multi-vendor, multi-cloud environments.

As Jago Macleod, Kubernetes & GKE engineering director at Google Cloud, stated at Kubecon, "At Google Cloud, we have certified for Kubernetes AI Conformance because we accept as true with consistency and portability are vital for scaling AI. by means of aligning with this wellknown early, we're making it simpler for builders and companies to build AI programs which can be production-equipped, transportable, and efficient, without reinventing infrastructure for each deployment."

Expertise in Kubernetes improvements

That became a way from the simplest issue Macleod had to say about Kubernetes's future. Google and the CNCF have other plans for the market's main container orchestrator. Key upgrades coming encompass rollback assist, the potential to bypass updates, and new low-level controls for GPUs and other AI-specific hardware.

In his keynote speech, MacLeod defined that, for the first time, Kubernetes users now have a reliable minor model rollback characteristic. This selection method clusters can be safely reverted to a known-appropriate country after an upgrade. This functionality ends the long-standing "one-way road" trouble of Kubernetes control-aircraft upgrades. Rollbacks will sharply lessen the risk of adopting essential new capabilities or urgent security patches.

Alongside this improvement, Kubernetes customers can now bypass particular updates. This technique offers directors greater flexibility and manipulation when making plans, model migrations or responding to manufacturing incidents.

Besides the CKACP, Kubernetes is being rearchitected to help AI workload demands natively. This assist manner Kubernetes in supplying users with granular management over hardware like GPUs, TPUs, and custom accelerators. This functionality additionally addresses the sizeable range and scale requirements of modern-day AI hardware.

Moreover, new APIs and open-source features, including Agent Sandbox and Multi-Tier Checkpointing, had been introduced on the occasion. Those features will similarly accelerate inference, training, and agentic AI operations inside clusters. innovations like node-level resource allocation, dynamic GPU provisioning, and scheduler optimizations for AI hardware are becoming foundational for each researcher and organizations jogging multi-tenant clusters.

Agent Sandbox is an open-source framework and controller that allows the management of remote, cozy environments, also known as sandboxes, designed for running stateful, singleton workloads, which include self-reliant AI marketers, code interpreters, and development gear. The main functions of Agent Sandbox are:

Isolation and protection: every sandbox is strongly remoted at each of the kernel and network degrees using technology that includes gVisor or Kata boxes, so it's safe to run untrusted code (e.g., generated by using huge language models) without compromising the integrity of the host system or cluster.

Declarative APIs: users can declare sandbox environments and templates for the usage of Kubernetes-local resources (Sandbox, SandboxTemplate, SandboxClaim), enabling fast, repeatable creation and management of remote instances.

Scale and performance: Agent Sandbox supports thousands of concurrent, stateful sandboxes with rapid, on-call for provisioning. This capability may be splendid for AI agent workloads, code execution, or continuous developer environments.

Picture and restoration: On Google Kubernetes Engine (GKE), the Agent Sandbox can utilize Pod Snapshots for fast checkpointing, hibernation, and on-the-spot resumption, dramatically lowering startup latency and optimizing useful resource utilization for AI workloads.

Nowadays, Multi-Tier Checkpointing in Kubernetes is primarily available on GKE. within the destiny, this mechanism will permit the dependable storage and control of checkpoints at some point of the training of large-scale ML models.

Here's a short caricature on how Multi-Tier Checkpointing works:

Multiple storage ranges: Checkpoints are first saved in a fast, nearby garage (including in-memory volumes or nearby disk on a node) for short access and rapid recovery.

Replication throughout nodes: The checkpoint facts are replicated to all nodes within the cluster to guard against node disasters.

Continual cloud storage backup: Periodically, checkpoints are subsidized up to durable cloud storage to offer a dependable fallback in case of cluster-wide failures or instances whilst nearby copies are unavailable.

Orchestrated control: The system automates checkpoint saving, replication, backup, and healing, minimizing manual intervention for the duration of education.

The advantage for AL and ML workloads is that Multi-Tier Checkpointing enables short resumption of education from the closing checkpoint without losing tremendous progress. The mechanism additionally offers fault tolerance by way of defensive error recovery jobs from frequent interruptions by ensuring that checkpoints are appropriately saved and replicated.

On pinnacle of all that, Multi-Tier Checkpointing offers scalability with the aid of helping massive distributed training jobs run on lots of nodes. Sooner or later, the feature of route works with all principal AI frameworks, which include JAX and PyTorch, and integrates with their checkpointing mechanisms.

With rollbacks, selective update skipping, and manufacturing-grade AI hardware control, Kubernetes is poised to power the sector's most demanding AI and corporate structures. The CNCF's release of the Kubernetes AI Conformance application is further cementing the environment's function in setting standards for interoperability, reliability, and overall performance for the close to future of cloud-native AI.

Kubernetes's first decade turned into all about moving IT from naked metal and virtual Machines (VMs) to bins. Its subsequent decade will be described via its capacity to manipulate AI at a planetary scale by means of imparting safety, speed, and flexibility for a new class of workloads.

There has been greater woe for decided on home windows eleven users over the past couple of days, with an unpleasant software worm locking a large number of people out of their primary C: drive. Microsoft has mentioned the trouble and is operating on a restoration.

As reported via Neowin and others, the Trojan horse triggers the message "C: is not accessible – get entry denied". It seems to have been introduced in the February 2026 safety update (KB5077181), and is predominantly affecting laptops manufactured by Samsung.

In its professional aid note, Microsoft indicates the problem is probably related to the Samsung proportion app, which is no longer shown at this stage. What is obvious is that it is incredibly difficult to get access to, and there are plenty of programs for affected users.

Workplace apps and internet browsers are the various applications that the main work reasons problems for, Microsoft. It says it's "presently investigating" the malicious program, together with Samsung, and will issue updates "whilst more data is available".

Neowin notes that only the contemporary home windows eleven variations 24H2 and 25H2 appear to be impacted by the bug, whilst according to Microsoft, more than one region is affected, such as Brazil, Portugal, Korea, and India.

If you've been hit by means of this computer virus, you may know about it: Microsoft says "getting access to files, launching packages, or performing administrative obligations" can trigger the error message, and for plenty users it's making their systems pretty an awful lot unusable.

There's the same old safety internet of an update rollback, which you may discover through the Settings > Windows Update menu — Microsoft has full instructions right here. As an alternative, if you have not mounted the February 2026 replacement yet, hit pause for now.

The enterprising folks posting on Reddit have a few guidelines about workarounds that involve converting ownership of the C: power, and at the same time, as this seems to have worked for a few customers, it is now not without dangers — so proceed with warning in case you do try those fixes.

Garmin's Instinct three sun gives a smart choice of flagship features, down to $349 at Amazon for Black Friday.

The Garmin instinct collection has usually been focused on the outdoors adventurer who needs a rugged watch that plays no matter what is thrown at it. I purchased the Instinct 2X solar more than one year ago, and it's miles nonetheless my first choice in relation to fly fishing and camping due to the fact I never have to fear approximately the battery life, and the watch offers the essentials.

Garmin launched the instinct three collection a few months in the past and, for the first time, provides clients with the option for a memory-in-pixel (MIP) display or a colourful AMOLED display. Given that long battery life, advertised as unlimited with adequate solar charging, is a staple of the Intuition logo, I requested to review the MIP display version and test out the enhancements over the Intuition 2X solar.

Garmin instinct 3 solar

If you need a rugged watch with outstanding battery life, then the Intuition 3 is an extreme contender. That LED flashlight is the icing on the cake..

I understand the appeal of AMOLED presentations, but i am not satisfied that Garmin ought to have introduced one to the Intuition lineup. i might choose to have seen only a sun version that bumped up the MIP show from monochrome to shade, as we see on the exceptional Enduro 3. Garmin has verified that it may nonetheless offer wonderful battery lifestyles with a colour MIP show, and that is the upgrade I wanted to see.

Monochrome MIP displays had been utilized in intuition models for years, and i don't mind it in the intuition 2X, however, after trying out the intuition 3 sun for some weeks, I assume i am done with a monochrome show from Garmin. Garmin now offers such a lot of exciting fitness and wellbeing metrics that even diffused colorings show beneficial for short glances and understanding of those metrics.

Transferring away from the display dialogue, we see that Garmin has progressed the Instinct 3 solar in numerous ways when compared to the Instinct 2X solar.

Solar charging strength has been considerably stepped forward, up to five instances that are furnished by means of the ultimate model. The assessment has been progressed and is honestly substantive when each watches are placed facet-to-facet. New color alternatives are to be had, and upgrades in software capabilities have been made.

Those software upgrades include training load, sleep coach, nap detection, Garmin Messenger, and multi-game car transition.

The Garmin intuition three solar has tested correct location monitoring, solid coronary heart rate monitoring accuracy, and wonderful, dependable performance with the whole lot I threw at it. That LED flashlight remains a staple I take advantage of day by day and is one cause you'll continually find a Garmin watch on certainly one of my wrists. I attempted the bigger 50mm model with that cool Twilight and Solstice orange band, and it fits me flawlessly. I especially love the burden and look on my wrist.

The watch has rugged factors with raised polymer body pieces around the bezel protecting the scratch-resistant display, a touch of color at the aluminum bezel ring, and fiber-strengthened case fabric. it is rated for 10 ATM water resistance, so that you can wear it anywhere without fear. I additionally enjoy the usage of that small show within the upper proper, especially for marking when I land fish inside the rivers.

The Garmin Instinct 3 Solar is a great alternative for those looking for one of the longest-lasting GPS sports activity watches for below $500. it is time to move on from monochrome MIP displays, and i would really like to see the Instinct 4 supplied with a color MIP version and the AMOLED one so that you can choose colour and then make a battery life decision like we do now with the flagship Fenix 8 collection.

Those apps are free; however, I'd be inclined day-to-day shell out a few bucks for them day-to-day. right here's why.

At the same time, there are certain open-source programs I might pay for, GPL or not. A number of those software titles are simply that precise, while others I rely on in reality.

However, which apps exist in that special category? There are a pick out few, many of which run on all platforms (together with Windows).

Even though I may not run the Docker command every day spin up a field every day, I simply use the results of those instructions. I presently have five special packing containers running on my LAN that i take advantage of every day.

Amazon has confirmed plans to make investments as much as $50 billion in new AI and high-performance computing structures designed especially for US government use.

In a weblog post, the enterprise outlined how the programme will aid a wide range of federal structures that rely upon advanced computation.

This consists of national protection work, scientific tasks, self-reliant structures improvement, cybersecurity analysis, energy programmes, and healthcare-related studies.

Growing reliance on specialized government cloud regions

Amazon claims the funding aligns with federal priorities that call for more potent AI gear deployed on at easy infrastructure constructed in the United States.

The enterprise also links the initiative to broader government efforts to modernise computing ability via upgraded cloud website hosting environments and specialized hardware.

"Our investment in cause-built authorities AI and cloud infrastructure will essentially transform how federal agencies leverage supercomputing,” stated AWS CEO Matt Garman.

“We're giving companies multiplied get admission to to advanced AI abilities so they can enable them to accelerate important missions from cybersecurity to drug discovery. This funding eliminates the era obstacles that have held government back and similarly positions the United States to lead in the AI era."

The investment is expected to add almost 1.3 gigawatts of compute capability across the 3 authorities cloud regions.

Amazon claims this will allow businesses to complete simulation and modeling duties a long way faster than before, with the aid of linking AI systems with real-time feedback loops.

Those projected gains rely on the government cloud garage regions that already assist labeled and unclassified workloads.

Over the last decade, Amazon has constructed GovCloud, pinnacle mystery, and mystery environments to fulfill federal compliance and safety necessities.

It has been a pacesetter in government cloud computing, providing agencies with managed spaces for sensitive computational missions and helping over 11,000 organizations.

This current basis is the platform on which the brand new AI and HPC systems will perform.

It allows companies to undertake expanded compute capacity without rebuilding operational frameworks or shifting away from mounted governance gear.

The funding is part of a wider shift closer to AI-driven scientific and safety workflows.

Federal teams are being pushed in the direction of environments where professional fashions, marketers, and natural-language gear assist research, engineering, and analytical tasks.

Amazon states such skills will let scientists and analysts define complex issues and get hold of automatic hints supported by means of simulations and high-accuracy evaluation.

The enterprise introduced this transition is a break from conventional HPC strategies that require a significant manual effort.

While you buy through hyperlinks on our website, we may additionally earn an affiliate fee. Here’s the way it works.

• Salesforce CEO Marc Benioff has ditched ChatGPT for Google Gemini three
• He made the transfer after attempting Gemini three for simply hours
• Benioff praised Gemini 3’s pace, reasoning, and multimodal capabilities as a major step forward

Salesforce CEO Marc Benioff has fallen head over heels for Gemini three and has publicly dumped ChatGPT in the process. His statement left a whole lot of the AI global reeling. After just hours of playing with Gemini three, a top enterprise software program chief disavowed the maximum famous AI chatbot in favor of its rapidly ascending rival.

Holy shit. I’ve used ChatGPT each day for three years. Simply spent 2 hours on Gemini three. I’m no longer going lower back. The bounce is insane — reasoning, speed, photos, video… the whole lot is sharper and quicker. It feels like the international simply changed, again. 

The post wasn’t subtle. It wasn’t hedged with well-mannered comparisons or impartial optimism. no longer the most effective has one of the maximum visible enterprise figures in Silicon Valley crossed from OpenAI to Google, but he's additionally made a factor of telling the world.

Gemini 3 has attracted plenty of different accolades for its mixing of Google DeepMind’s modern-day advances right into a unmarried unified architecture. Gemini three helps text, images, code, audio, and video in a single interface. Gemini 3 is properly positioned to take ChatGPT's crown for cross-to AI chatbot in many ways.

Though ChatGPT's ubiquity made it sense inevitable, Benioff’s post indicates that inevitability has an expiration date. In a depend of hours, one of its maximum prominent evangelists defected. that announces more about the tempo of AI evolution than any leaderboard or benchmark can.

AI rivalry heating up

Benioff isn’t only a few casual early adopters. Salesforce has deeply incorporated AI into its merchandise and enterprise strategy. It turned into among the first company giants to companion with OpenAI for productivity tools embedded in consumer courting software. When Benioff makes a non-public bounce like this, it carries the load of corporate alignment. And if Gemini 3 is now the version he's defaulting to, the software stacks around him can also follow suit.

Gemini 3’s strengths seem to match what high-frequency customers like Benioff search for in terms of speed, reasoning, and flexibility. Google has been clear that Gemini three was constructed to feature as a flexible engine for both purchasers and builders, able to energy the entirety from helpdesk bots to video editing suggestions. That breadth may be what tipped the scales for Benioff, whose day possibly includes toggling between datasets, dashboards, and vision decks at Silicon Valley warp speed.

It's a terrific reminder that the AI panorama is getting more aggressive. There’s a growing refrain of contenders scrambling to make your AI assistant. For folks who casually take a look at it with ChatGPT once in a while, nothing has to change in a single day. However, Benioff’s public conversion gives a peek into how quickly even entrenched possibilities can pave the way to the promise of sharper gear.

Black Friday is weeks away, and we've been given the info on the biggest store income, together with dates, times, and early access opportunities.

The maximum notorious shopping event of the year is slowly making its way to us: Black Friday is two weeks away.

It truly is right: Black Friday, Cyber Monday, and the weekend of financial savings in among is nearly right here, and we've got already seen several early seasonal sales at a number of the biggest retailers like Amazon, nice buy, and Walmart, which kicked off as early as Nov. 13. the subsequent wave of early financial savings stretches into this week, and by means of Nov. 20, many retailers, like Amazon and first-rate buy will kick off their reputable Black Friday sales per week in advance of the historical sale day.

It is why I have compiled the entirety I recognize about shopping sales as ZDNET's deals and buying editor. I am passing my preferred hints, tricks, and need-to-recognise information to you so that you can store savvy this excursion season. Don't forget, this is your all-inclusive manual to the 2025 excursion shopping season. Right here's everything you need to recognize to store Black Friday deals and different seasonal savings offers precisely like a buying editor.

Find Amazon Black Friday deals

When is Black Friday?

This year inside the US, Black Friday is on Nov. 28, 2025, sometime after Thanksgiving, which falls on Thursday, Nov. 27.

Whilst is Cyber Monday?

This year, inside the US, Cyber Monday falls on Dec. 1, 2025, only a few days after Thanksgiving and Black Friday 2025.

What's Cyber Week?

A more recent term to emerge within the closing decade, "Cyber Week" refers to the time of savings in late November and/or early December that arises yearly just after Thanksgiving. In the course of this era, many on line and in-store retailers provide large financial savings and reductions across nearly every product category. Cyber Week 2025 begins on Black Friday (sometime after Thanksgiving within the US), and extends through the cease of Cyber Monday of the following week. For 2025, Cyber Week will kick off on Nov. 28 and run through Dec. 1, though we're already looking ahead to masses of sales that begin even in advance and linger beyond Cyber Monday.

What is the Cyber 5?

The "Cyber 5" is essentially just another call for Cyber Week. Those are the five days of seasonal savings that kick off with Black Friday on Nov. 28 this year and stop with Cyber Monday on Dec. 1. This Friday-to-Monday stretch every 12 months falls properly after Thanksgiving and historically serves as the most popular time for vacation purchasing.

Whilst will Black Friday and Cyber Monday offers be live?

This 12 months, Black Friday falls on Nov. 28, 2025. which means Cyber Monday will fall on Dec. 1, 2025, and the times between Thanksgiving and Cyber Monday are dubbed "Cyber Week." It really is, whilst you can expect to see a few important holiday sales and discounts, and as continually, ZDNET will deliver considerable coverage of the pinnacle deals across the purchasing extravaganza at more than one stores.

Expect to see the biggest quantity of deals live at some stage in this window, from Nov. 28 to Dec. 1. However, remember the fact that stores roll out deals in advance and in advance every season, so there'll still be numerous times of financial savings throughout November in advance of Black Friday and via December in advance of the winter holidays. The relaxation of November is filled with financial savings, and you also do not want to miss out on the aid of failing to prepare.

How long will those vacation deals be available?

Black Friday savings are already slated to begin in advance in November, and wind down through Cyber Monday on Dec. 1. But, remember that Cyber Week is not the most effective time to shop, and even if a few offers quit, it's probably that others will linger for a few days, submit-Black Friday weekend. Plus, we will in reality see extra savings activities and sales in December, too.

What is the distinction between Black Friday and Cyber Monday?

Semantics, without a doubt. Black Friday is a a long-time-vintage buying extravaganza traditionally associated with in-store financial savings and reductions following the Thanksgiving holiday in the US. Cyber Monday, like Cyber Week, is a much extra modern time period that turned into born because of the web buying boom, thanks to the likes of Amazon and other retailers moving to on-line modes of sale.

Cyber Monday more carefully reflects seasonal financial savings supplied online only, particularly on tech products like laptops, tablets, phones, and more. However, the bottom line? In 2025, each term is pretty synonymous with vacation financial savings, and Black Friday has without a doubt grown to consist of a much larger online presence, too.

What are the best merchandise to save throughout Black Friday?

Black Friday is the conventional start of the holiday buying season, and is possibly one of the busiest instances of the year for in-store and online retailers. There could be plenty of merchandise to save on Black Friday, and you will be able to discover savings in basically each object category. You will be able to snag financial savings on tech products like streaming offerings, gift playing cards, video games and consoles, TVs, laptops, tablets, headphones and audio systems, and robotic vacuums. Plus savings on broader classes like luggage, clothing, beauty, kitchen and household, toys, and more. There'll also be opportunities to save on products that do not regularly see income, like on flagship Apple devices.

What shops could have Black Friday sales?

Nearly every fundamental store will certainly provide competing income, deals, and reductions for the duration of Black Friday and through the holiday season. Walmart, goal, pleasant buy, Costco, and different retailers (like Kohls and Macy's) will try and compete with every other's offerings, frequently price-cutting the equal objects throughout the same periods, extending their own sales, or imparting member-exclusive income or get right of entry to. However, do not worry -- we'll cover all the high-quality offers on pinnacle tech via the season, along with deals surfacing somewhere else past the usual culprits (Amazon, Walmart, and great purchase).

Shop early Walmart deals

Walmart Black Friday

On Oct. 27, Walmart launched the details of its Cyber Week event. The primary sales will kick off on Nov. 14, and the brand promises thousands of gifts under $20 and as much as 60% off top manufacturers. Walmart will roll out financial savings in three essential sale waves, and the store just added a return of its popular 50% financial savings on a Walmart+ membership through Dec. 2. Walmart+ is filled with perks like loose Peacock or Paramount+, free shipping, and early get admission to to shop Walmart's Black Friday deals.

Walmart Black Friday offers occasion 1: Nov. 14-16: deals available online and in shops for the whole event. Walmart+ individuals get 5 hours of early admission to online starting Nov. thirteen at 7 p.m. ET.

Walmart Black Friday offers event 2: Nov. 25-30. deals to be had online best from Nov. 25–27, observed by on-line and in stores from Nov. 28–30. Walmart+ members get 5 hours of early access on line beginning Nov. 24 at 7 p.m. ET.

Walmart Cyber Monday deals occasion: Dec. 1: deals are available on line most effective. Walmart+ individuals get five hours of early get right of entry to on line beginning Nov. 30 at 7 p.m. ET.

best-buy-black-friday-2025-full-guide-to-deals-starting-on-november-28

amazon-black-friday-2025-full-guide-to-deals-starting-on-november-28

Source URl