Hacker News Reader

Comment by the_snooze | Hacker News Reader

>For example, if you ask ChatGPT’s Agent to book a travel, it’ll open Chromium on Linux in an Azure container, search the query, visit different websites, navigate each page and book a flight ticket using your saved credentials. An AI Agent tries to mimic a human, and it can perform tasks on your behalf while you sit back and relax.

Big tech has repeatedly shown that they are not good stewards of end users' privacy and agency. You'd have to have been born yesterday to believe they'd build AI systems that truly serve the user's best interests like this.

View on HN

I think in this case, Microsoft has shown they don't respect the user when they force shutdown for system updates. This has happened during my time working retail and the mom and pops are helpless when this happens.

I would never trust Microsoft to bake ai agents in..

> shown they don't respect the user when they force shutdown for system updates

Are you familiar with the prior state of things that explicitly motivated this change?

Why does that matter? I should be allowed to explicitly chose the risks I want to take. Not microsoft. Especially not for microsoft to decide, no matter what I'm doing, or what I have open and unsaved on my computer, now is the time they think my risk is too great and tuesday has passed, so reboot reboot reboot.

The automatic reboot has made the world a better place, because too many people were incredibly bad at making this risk tradeoff.

It might still be bad thing for taking away agency. But it was also a massive improvement to society.

That's quiet a bit of stretch to equate forced update reboots to massive societal benefits

The point was about security updates. Without forced upgrades the whole ransomware thing would have been so much worse as just a singular example.

I think it wasnt just reboot inconvenience, I feel like there was a period of time where some software updates would break or make your software experience worse.

I have vague memories as a teenager of running older versions of MSN messenger in compatability mode because after a certain version it was full of ads.

Android phone software is also very good at this now, I still hestate to update my pixel because each update somehow makes my phone worse to use.

Well... this is similar to COVID. As long as your computer is disconnected from any network, yes you should be able to do whatever you want and decide. But as soon as your computer can be a danger for others, then your risk taking decisions can harm others, and then what?

Masks during covid were a matter of public health.

Regular updates are also a similar matter.

It's truly absurd to compare "my computer might be hacked and used by baddies" to "I don't want to wear a mask during a pandemic"

It's not a comparison that bears a response.

Masks were necessary to save lives at a stage where risks were unknown and pressure on health systems was high.

Missing Windows updates does not kill anyone.

Plus, installing Windows updates may cause high frustration because "feature" updates are mixed with them and may alter the OS behavior in unexpected and undesired ways. If Microsoft cares so much about security, they should allow people to stay on fixed Windows stable versions that only get security updates without pestering them. Basically, sell LTSC to normal people.

Not really. Maybe I'm jinxing it, but I've never had a problem caused by failure to update my PC.

Servers I understand because they're exposed to the Internet at all times. Not PCs

Lest one remembers Win 9x or even XP w/ no firewall on residential networks.

Behind a NAT.

Can't remember a single problem with the described setup and I've been using the internet since dial-up was the only option available.

Getting hacked when you don't have any open ports (thanks to NAT) is and was pretty unlikely - what was more likely is some kind of drive-by exploit in Flash or IE. The biggest problem I experienced with old Windows was general instability in the form of BSODs and driver compatibility problems.

NAT has nothing to do with security and it was common that people had a single device on DSL or cable plugged directly into the modem; routers were not common place at home.

NAT was for fancy-pants with multiple PCs.

It's interesting how much different the landscape was in that era: single-device residential environments would have no firewall at all (just a PC with a publicly-routable IP address) and dial-up kind of fueled this due to PCI slot modems, but as the outboard nature of DSL and DOCSIS modems made it easier to build multiple-device residential environments by adding a router, suddenly everyone had a firewall (as a byproduct of NAT). Then you've got malware, which was far more prevalent on PCs through that transition relative to today, but now we've got IoT stuff probably not being updated as it ought to be, potentially hosting malware that serves as a proxy to sidestep an in-router firewall.

Yeah, I remember formatting the HD on a PC back then to do a fresh install of Windows XP.

The CD-ROM I had was pre-SP2 (so no firewall), and our internet setup was basic modem + switch. No router with “drop invalid state” or fancy things like that.

So, installed Windows and plugged in Ethernet to fetch Windows updates.

2 minutes later, with no user interaction whatsoever, the PC was infected with malware.

> Servers I understand because they're exposed to the Internet at all times. Not PCs

Gates, is that you ? They have telemetry in PCs those days, you know. /s

> Servers I understand because they're exposed to the Internet at all times. Not PCs

And, for reference, updates are not forcefully installed on Windows Server.

Well, forcefully restarting a server without asking its owner does sound like a bad idea. And disrespecting the users in that way when the competitor OS for servers is free, has significant market share and is known for letting the user to what they want and getting out of the way should probably also be avoided from a market perspective.

Security is the catchall excuse for every bad big tech behavior because they know "security" professionals will defend every f-the-user move they pull [1]. Is it improved security when I lost days of work because microsoft (and you apparently) think their patch is more important then my data? Notice, by the way, that security incidents can cost big tech a lot of money but my lost data is no skin off their back.

[1] It reminds me of dermatologists, so hyperfocused on skin cancer that they tell everybody to hide from the sun, completely oblivious to all the harm their advice causes to the rest of our health.

The other angle is that if annoying enough it gets people to make their own workarounds so it works as they want. The real trouble is when it escalates as each side wants to have authority over the other as they each think they know best, and you get things like laptops on standby waking to try and update themselves in a bag. I've been thinking for a while that windows has been going away from a 'personal computer' OS in that it isn't "mine", it's at the mercy of someone else and efforts to fight that aren't worth it long term.

Yes the security of every Windows computer was much better then, any software that automatically updates itself without user consent is obviously a massive security risk because the user is no longer in control of what software they run.

This is why I still prefer to install programs as root, since then they are unable to update themself. (And also other users can't do that.)

The amount of money lost when millions of small restaurants and other retail shops suddenly become unable to accept customer payments for an unknown amount of time because Microsoft thinks Windows should force update during rush hour rather than allowing the computer owner to wait until closing time, would seem to be far greater than the amount of money lost with once-in-10-years WannaCry attacks

Don't you get out of forced updates if you set yourself regural update point ? (e.g. every Sunday night)

Most users, for better or worse, don't want any update ever, unless they wish for a specific feature. We're at a state where there's only once-in-10-years massive attacks exactly because of mandatory security updates that will be forced on the user if they have no intention to install it ever.

Maybe the 3rd largest tech company in the entire world could spend a little time figuring out how to hot patch their OS. Heaven forbid they actually innovate on something.

You can update without locking the computer. You know... like is done in Linux for a very long time. I have a nice memory of doing a full update of Kubuntu to the next version at the same time that I was playing a AAA game without issues or interruptions.

I hadn’t seen Linux do that. How do they fully do it without ever locking or rebooting the system?

When you run apt upgrade or pacman -Syu that's exactly what you're doing. The files are replaced on your drive while everything else continues running. Generally it won't affect execution of existing software, because they're all already loaded into memory, but some software might crash or get weird behavior as they try to access their files on the drive and those files have been updated, and newly launched programs will use different library versions than other programs which may cause weirdness. You still need to reboot in the end to stop running old stuff that's still in memory such as kernel or existing programs but it's a normal reboot without any extra delay. Canonical does provide Livepatch for Ubuntu Pro for servers that want to update the kernel with security updates without rebooting.

Fedora decided this isn't super stable so they actually went and implemented something similar to Windows updates called Offline updates, where updates are performed after a reboot in a special mode where you can't do anything with your computer while it updates for like 10 minutes, but they give you an option to disable this and do instant updates like described above instead.

I think the most interesting innovation are immutable distros, which handle updates entirely differently. They will build an updated image while the system continues running and make it ready so that next reboot will just boot into the updated image. It avoids the partially-updated-system instability entirely and it also makes reverting a broken update instant and easy because you can just boot into the old image (there's usually at least two images). This exists in Fedora Silverblue (OSTree) and Vanilla OS (ABRoot) and AFAIK Android also followed this update pattern with A/B partitions (although they now iterated on this slightly to squeeze a few extra gigabytes out of storage).

I honestly don't know why Windows still sticks to their antiquated offline update system when better options exist and everyone always complains about the way they do updates and they have billions of dollars at their disposal, but I guess lack of any real competition to Windows in the PC operating system market has led to such stagnation

The immutable distro doesn't work for Windows most likely due to disk space. As someone who has informally supported a lot of Windows devices in enterprises it was surprising to me how many Windows problems are a result of running out of HD space and how often updates can't happen (the old fashion kind) simply because there isn't enough HD space for the update. I wouldn't be surprised if something like 5% of updates couldn't happen due to this.

Windows does do hotpatching, but there's a lot of things that aren't hotpatchable. Do you really think that Windows is like "naw, we could do zero reboot updates, but prefer not to because we are so dominant in the OS space"? This would be an incredible feature for the enterprise. In fact the enterprise version added a bunch of new hotpatch support just last year, but still requires quarterly updates and only does security updates. You really think that they did all that, but decided to not do the rest because they're comfy?

Again, I haven't seen Linux or Mac solve the problem fully either, nor iPhone or Android. AFAIK even every cloud provider has to do a reboot. Would Google or Amazon or Oracle have figured this out if it was so easy? How is it that there is no actual software engineer in industry that knows how to do this, but everyone on message forums seems to? Why don't these companies just hire people from message forums?

If disk space was the only thing preventing a much better update system, it would seem very trivial to just reserve a few extra gigabytes and do what Android does with compressed deltas that get applied which emulates A/B but with much less disk space usage (similar to non-A/B)

> Do you really think that Windows is like "naw, we could do zero reboot updates, but prefer not to because we are so dominant in the OS space"?

Microsoft has become complacent with Windows and I think there's no denying that. You need to look no further than the new right-click context menu they thought is acceptable to ship to a billion users. It's lacking half the functionality such as extensions, so they just decided to keep the old one behind "Show more options"? Or maybe no software engineer in the world could solve the infamous context menu 2.0 problem...

No operating system has fully solved every problem with updates, but many of them have solved many problems that Windows still continues to have. Zero reboot updates are probably impossible to do reliably but there are other ways to improve that aren't zero reboot updates. I don't claim to know the ins and outs of Windows and exactly how to implement better updates, but they could surely do better than what they're currently doing.

on linux, if a file is open by one or more programs, and the file is deleted (and replaced, usually, during updates) then the original file isn't actually deleted until everyone who is currently uses it closes it. You never get a "file is in use" error.

Are you aware that MS already sells an operating system that can install patches without rebooting? Are you also aware that Linux can do the same? Why can't a supposedly mature 40 year old operating system do the same? Do you have any concept of the number of man-hours it would save globally? The amount of lost work? The impact on patching compliance and security?

My guess is they don't actually believe they have any competition, and therefore don't care to improve anything that doesn't also improve their bottom line.

every week when I login into my Ubuntu with unattended updates enabled I see this: "system restart required".

the hot patch feature you mentioned is paid

On Ubuntu, when this message is shown, most of the updates except the kernel are already applied so you are mostly pretty secure. And you can choose when that will happen. And it’s just a normal reboot.

On Windows, IIRC, you are blocked during the whole update process which can take several minutes.

Ubuntu's stable builds do not upgrade kernel and its close vicinity every week, AFAIK. I have a couple of servers with unattended updates enabled, and they do not greet me with "System Reboot Required" banner every week, and if that's required, the server is back with all services running <30 seconds.

OTOH, I upgraded my parents' PC yesterday, after three months of downtime. It really took at least two hours and four reboots. The machine was screaming and the task manager showed a blue rectangle for CPU load (uninterrupted 100%) and a green one for the disk load (again, uninterrupted 100%) while nothing was usable all the time.

Same process takes <10m in Linux (specifically Debian), and an optional reboot, without any hardware load drama.

Weird. My windows PC updates like your Linux machine. How often do update vs your parents? Maybe they had some larger “half” releases pending (I.e. closer to a major macOS release, which also take time)

The machine is on standby all the time. So it updates whenever it wants. In this occasion the machine was turned off for a couple of months, but the updates were not the "half release" updates. The list was .NET runtime, intel graphics drivers, some dynamic update support and the like. I was watching the machine all the time.

Funnily, dynamic updates support installation failed after all the kicking and screaming, and I didn't try. Maybe I'll look into it later.

Upgrade that PC's OS drive to a NVMe. Seriously. We manage thousands of PCs at work and ever since we got laptop models with NVMe drives, updates are a breeze with 6 ± 3 minutes of total downtime.

Oh, OK. It's a PEBKAC case, then, my bad.

I'd rather teach my parents to use Linux instead. Updates will be a breeze with 3±2 minutes of total background work without any interruption and 30 seconds of occasional downtime.

I have no idea how you got to "Problem Exists Between Keyboard And Chair" given that I literally blamed the slow OS disk drive inside the computer based on thousands of data points on my end.

Hey, no, I don’t dispute your data points at all. A bog standard NVMe can handle ~1MIOPS these days, and it’s above and beyond what SATA SSDs can provide.

What I’m against is tolerating a bad OS design with more capable hardware and allowing Microsoft to worsen the experience. This is a pattern of Microsoft since forever.

Oh, that particular PC has no NVMe support anyway. I don’t know why that M.2 port is SATA only.

Especially when every other major OS can handle this more gracefully. I can version upgrade a fully loaded Debian installation in less than 6 minutes, reboot included, on a SATA disk, for example.

Also, while tangential, Windows providing the worst update experience, and calling Linux a major, mainstream OS superior in some ways feels unbelievable when I look back a decade.

Not to derail but there are issues with kernel patching. If it does work you start building a very large matrix of various levels of hot patches and then sometimes it just doesn’t.

If my company was worth a trillion dollars and an entire multi-billion dollar industry (cybersecurity) had grown because of my security inadequacies I would figure it out.

In fact, they already figured out hotpatching and will sell it to you for server 2025.

Note, that you can also keep the userspace unchanged by hibernating and then choosing the new kernel on boot. It is not truly live patching, since you have still downtime, but pretty close.

It is also paid for windows. It shouldn't be.

Off topic, but I'm pretty sure that Ubuntu's livepatching is just kpatch under the hood,

https://ubuntu.com/blog/an-overview-of-live-kernel-patching

> Are you aware that MS already sells an operating system that can install patches without rebooting?

No. Which OS is that ? Even to update Office they throw an annoying popup and then another one to start the update and a dark pattern (close button accesible with a hidden scrollbar and no window controls) one to tell you it is finished.

Server 2025. They upsell it as a subscription because they can. Before that it was also available in Azure.

https://www.microsoft.com/en-us/windows-server/blog/2025/04/...

It is security patches only. To take all the other patches you do need to reboot, which is why it still has quarterly reboots. No real OS has solved this problem.

A 90% answer is better than no answer, which is what we've had for 40 years now.

Linux only requires rebooting for kernel updates, and with kpatch not even that.

Note that you can also only reboot the kernel, but keep userspace.

I'd wager further, is they've by this point long since bled out their top talent. Pretty soon that motor is going to run out of oil.

i dont want a device to tell me when i need to restart it, thats my decission.

Same on boot. Usually when I boot a computer I am not ready to wait for it to install several updates, unasked.

Yes. Since 199x.

macOS does the same thing. When I actually sleep, when my laptop's lid is closed. I wake up. My Mac wants a password instead of a fingerprint. It says it has updated the OS when I was snoring. What's the difference?

Every app, every window, everything is the way I left before closing the lid. My computer is updated, rebooted and ready for the day. Like nothing happened.

Linux is the same deal. If the desktop environment is upgraded a logout and login is necessary (and KDE restores session as well as macOS for the last decade, at least), and if I updated the kernel, I reboot. I'm back in 30 seconds, to the exact point that I left.

Only Windows takes 2 hours, 4 reboots, 3 blood sacrifices and countless frustration sounds to upgrade. While saturating the processor and the storage subsystem at the same time, which makes my computer create the same sounds of the said blood sacrifices.

My experience with Mac is iterm prevents Mac from shutting down so instead some days I wake up and everything on my machine has been closed and the update hasn’t been performed. Lovely.

I don't use iTerm, and close everything that I don't use for the night, which is a habit I have since the beginning of time.

I wouldn't trust a big tech AI agent to act in my own best interest. How do I know I'm getting the best deal and that they're not clipping the ticket? Given so many of these companies are really ad-tech/surveillance businesses, how do I know that they're not communicating information about me to the travel site which might affect the price?

> How do I know I'm getting the best deal and that they're not clipping the ticket?

You should actually expect the exact opposite. There's more money in getting large companies to pay you to redirect customers to more expensive products than in consumers paying for this kind of service. Honey[1] should server as a stark reminder here.

[1] https://www.tomsguide.com/computing/software/honey-scandal-e...

> According to Megalag and other content creators, Honey's core promise isn't true. PayPal and Honey say they'll run through a series of coupon codes to find the best deals. However, the firm is accused of using inferior codes to ensure the retailer gets more money from the sale while promising the user that the best code was used.

> Megalag tested this in his video and found instances where better codes were readily available online, but Honey chose to use a code with a lower discount, claiming it was the best deal.

What happened to Pt2 of his video?

Ironically, Microsoft's slogan in the 90s was "where do you want to go today?"

These days, it's more like "where do we want to make you go today?"

How I yearn for when their marketing had everyday people touting how "Windows 7 was my idea!" Every Windows release since then has felt like they are hostile to user input.

Sidenote, why is it always booking a plane ticket that they hype up? It's like the only 2 things any of the marketing can think of is booking plane tickets and replying to emails

because the people driving these products are disconnected and deeply unbalanced people

It's funny, because it's also one of the most "gotcha-filled" things you can do. Click the wrong box, and they'll stick you in a seat with no leg room or make you pay extra for a carry-on bag. I have very little confidence that an AI would be able to make the "correct" choice on an airline ticket consistently without making a rather impactful mistake.

It will work for a while and then the airlines will game their systems against AI agents just like they currently do against consumers.

It's just a temporary solution. A real solution would be for laws to force them to not do this. But airlines are often very intertwined with the state and a prestige thing for a country.

You'll end up with car insurance, a hotel reservation you don't want and pay extra for the middle seat

(Assuming it even gets the right airport/country).

I think it's hilariously tone deaf that travel booking and shopping are the two examples of "agentic" AI that keep popping up.

The main reason I shop online is the joy of hitting that Buy button every now and then for something I want. I don’t want some dumb bot doing that for me (and getting the wrong thing 2/3 of the times)

The real chore is having to go to the store to get groceries, doing laundry, pairing socks etc … but solving any of that would require more than just bullshit LLM capabilities.

> get groceries

Isn't that what grocery delivery apps are for, if you really don't want to go to the store.

> doing laundry, pairing socks etc … but solving any of that would require more than just bullshit LLM capabilities.

Yes, it's a shame robotics (hardware) is harder than software, but that's not really the fault of AI model developers.

You kind of missed the point of my comment but ok

> not really the fault of AI model developers

It’s their fault for pushing all this crap in all the things and misleading their investors that there is actually “intelligence” in what we now call AI.

> grocery delivery apps are for

These are not popular here and for a good reason - you need to enjoy your food and it starts by picking the right ingredients yourself.

“someone packs a bag for me and delivers it to my door” is just moving the problem somewhere else, not actual innovation.

They always mess up a few things, make brain dead substitutions, or get low quality produce. I had bags show up smelling strongly of cigarettes. All for a premium price, an app that takes a surprising amount of time finding things on, and the complete loss of discoverability.

My experience with other shopping sites makes me suspect that with all the ads, tracking, captchas, etc bogging things down, it might be faster to just go to the store yourself.

Can you refuse to pay, since you didn't got what you ordered?

Actually, for Robotics hardware is a solved problem. Software is struggling to keep up.

> Actually, for Robotics hardware is a solved problem.

I understand the sentiment but this couldn't be further from the truth. There are no robotic hand models that get close to the fidelity of humans (or even other primates).

The technology just doesn't exist yet, motors are a terrible muscle replacement. Even completely without software, a puppeteered hand model would be revolutionary.

Every time I hit a "buy" button it brings nothing but horrible anxiety over what future bullshit I'll have to deal with, either because the product will be garbage or the seller will be garbage. And that's after doing an hour of more research for every god damn thing.

Getting groceries is practically relaxing at this point

Groceries are hysterical to me. The ultimate first world problem.

It is just too much to go to the store, put what you want to eat in the cart, pay and walk out.

It stresses me out too much and takes time away from wasting time on my phone.

Because for the average person there isn't really that much they get out of todays agentic ai. This is all project managers can think of that applies to the average layperson.

It's just shitware being added to everything at very few people's benefit just so they can score some points on the stock market AI hype leaderboard.

I think there are two factors:

1. "Help customers buy crap" is one of the vaguely plausible use-cases which excite investors who see the ads, even if it isn't so exciting for actual customers.

2. The ideas seem sourced from some brain-trust of idle-rich, rather than from the average US consumer. Regardless of how the characters in the ads are presented, all of them are somehow able to prefer saving 60 seconds even if it means maybe losing $60 on a dumb purchase or a non-refundable reservation at the wrong restaurant, etc.

> The ideas seem sourced from some brain-trust of idle-rich , rather than from the average US consumer

I think it says more about the economy currently. The "average US consumer" is the wealthy right now. Just 10% of the population, the highest earners, drive nearly 50% of consumption currently and that number is growing.

That is the new average US consumer, hence the ads and use cases targeting a more well-off demographic. Everyone else has been left behind.

Adding context: The upper 10% for household income across the US is about $160k/yr.

Limiting the scope to people living in high cost-of-living cities (probably smaller than their ideal customer field) that might be $300-400k/yr.

I think my marketing professor said something interesting about it a decade or so ago. Basically, in the US we are moving towards heavy bifurcation. You can cater to the well-off or not well-off. The class was full of kids, who did not seem to understand the implications of what he was already saying then ( not that it technically is that mindblowing, the signs are there.. ).

searching for a flight and booking it is legitimately one of the most painful online things that exists. it's like the booking industry is feeding on suffering

It’s intentionally obfuscated because the product developers don’t want to share profits with brokers. They also do not want to compete on in the open because that too lowers odors Otherwise, we would have a system where it would be insanely easy to monitor and alert for price breaks. Hidden cities is probably the best example of how it could work and easily presents the price charts over time. Yet they too were cut off from some providers.

Probably high priority because the dev and literally everyone else is sick of microsofts selfservice platform for travel.

The industry has decided that 'agentic' stuff is The Future, and has bet the farm on it. However, actual useful applications are, ah, thin on the ground to say the least. Accordingly, industry obsesses over the few use cases which have shown up, even if they are not necessarily use cases that anyone particularly _wants_.

Travel booking is time consuming and frustrating. In doing it now and hate it. If some YC company wants to fix this I’d be hugely appreciative.

It's like the "store your recipes" to sell home computers 45 years ago. Not the problem we need solved.

Or the "write code more quickly" for LLMs. NOT the problem we need solved.

>Big tech has repeatedly shown that they are not good stewards of end users' privacy and agency.

I can understand Google or Facebook being bad because their whole business model is based around selling your attention and agency. Microsoft shouldn't be as bad because they are selling a product but in many ways they appear worse.

Reply on news.ycombinator.com