HNHacker News
TopNewBestAskShowJobs

PunchyHamster

4,416 karma · joined May 9, 2025

submissionscomments
PunchyHamster··on Alan Kay: Shannon gave us a way of dealing with noisy channels [video]
> Calculating the limit is easy with Shannon's theorem. Approaching it in practice is hard.

Many modern modulations already operate basically on Shannon's limit for a given band/SNR. Well, on raw data, the encoding almost always use some kind of error correction so the decoded bitrate is few % lower than the wire one

PunchyHamster··on Microsoft abandons personal AI chatbot race with Copilot reboot
> Microsoft said "She'll learn from your conversations" and the internet said "Bet"

...well, it did exactly that, no bet here.

PunchyHamster··on Microsoft abandons personal AI chatbot race with Copilot reboot
.... immediately ? MS has been doing it for decades
PunchyHamster··on Dutch governments builds alternative for Microsoft based on NixOS
a lot of companies migrated to teams purely because it was already included in the subscription, IIRC they were forced to take it out by EU in 2023 for being anti-competitive.
PunchyHamster··on Oracle cites 'force majeure' to shield itself on controversial data center
I see it as "make regulation that will throttle our competition (especially foreign one)" mixed with "look at us, our AI so advanced" pre-IPO advertising
PunchyHamster··on OpenAI agent hacked Australian government website, PM says
there can be more than one side at fault
PunchyHamster··on OpenAI agent hacked Australian government website, PM says
> I'm a little confused as to why these agents are capable of escaping containment. Can someone who has more understanding (or a better guess) of the harness they are running with shed some light?

Probably because containment was written by LLM just to check a box of "we have it contained". Or, just laziness

With today's internet, there is a good chance just allowing access to a site isn't enough, you might need to give access to 3rd party URLs the site uses.

...and if URL for service site uses is same (there is no bucket prefix like for say S3), giving access to service X used by site Y gives access to more than site strictly needs

...and if they use cloud stuff people get lazy and just do "allow it entirety of S3 access" vs whitelisting per bucket.

URL whitelisting wasn't great 20 years ago, now it is just pretty bad for anything cloud based

PunchyHamster··on Once Claude can measure something, it can make it faster
AI companies are seeing the endgame, more and more tasks can be done just fine by cheaper model, and there will be not enough money to be made to support model that is only needed for top 1% of the tasks.

And the goal is to slow competition down enough before IPO

PunchyHamster··on Once Claude can measure something, it can make it faster
Paperclip-optimizer-esque behaviour very much seems to be inherent to current methodology of building LLMs, there are only ways to lower the changes or mitigate the damage, not get out of it.

Same with prompt injection, current LLMs are commands in, commands out, there is no way to make sure it is "an agent working on data" rather than "an agent that can take commands from data if you phrase it right"

PunchyHamster··on Claude Code reads AGENTS.md only when telemetry is on [fixed]
there is a thing apps have called configuration file.
PunchyHamster··on Abandoning Scientific Linux Was a Mistake
They are. Debian + additional repo with scientific packages would be far more preferable for anyone involved.
PunchyHamster··on Abandoning Scientific Linux Was a Mistake
You can pay companies just fine to support Debian.

> Debian is still easier to manage than RH based systems, and .deb is a really good package manager, and is arguably more sophisticated than RPM.

It's so bad. Coming back to RHEL (new client requirement) after sitting mostly on Debian for years is such a bizzare experience.

You want to install a PostgreSQL database, it doesn't even initialize it, you need to manually do DB init and the rest of the dance.

You want to run DRBD, sorry, we didn't bother to pick that kernel compile option, you can add extra repository for that. But hey, it's the kernel module, which means now you have to enroll cert in Secureboot, and that requires actual KVM access to the VM for someone to confirm it so now I'm on meeting with customer's IT just so they click some buttons in VM's boot process.

Even some common utilities are not in main repos but need EPEL

PunchyHamster··on WordPress: Unauthenticated path traversal leading to conditional RCE
It's a wonderful piece of software. We've had one of our clients wanting us to "just provide a simple PHP install for wordpress, just hosting" (among some more complex java stuff company wrote for them).

Site hacked within a day from install. Thankfully we have limited outogoing traffic (whitelist on proxy) so only thing exploit managed to replace is their main page with our proxy's 403 error page, but damn, how this piece of software remains so shit till this day is massive achievement in incompetence.

They also manage to fail on every level, like a simple problem of "a service behind a loadbalancer/reverse proxy" is still unsolved because devs refuse to support X-Forwarded-For header in core "because it's not official RFC", and also do not support official RFC for same feature.

PunchyHamster··on Obscura: VPN that can't log your activity
They don't even need to. If you observe enough of them you can correlate traffic patterns between them and find out which one is used by which endpoint
PunchyHamster··on Obscura: VPN that can't log your activity
Incorrect. Pointing out bad products is a warning to other people to not waste time on it. We should be doing it more, not less.

I also have no idea what company/service you're talking about

PunchyHamster··on Looking forward to Git 2.56 – and 3.0
so it is a problem for majority of use
PunchyHamster··on Looking forward to Git 2.56 – and 3.0
> The Git 2.45 release added reftables as a more efficient way of storing refs. It is a binary file optimized for both space efficiency and quick access. Since then, it has been possible to create a repository that uses a reftable rather than the old file-based mechanism, but that has never been the default. Switching to reftable should have no visible consequences (other than better performance) for users of Git itself, but it can be a problem for users of other software packages that access Git repositories. In his email, carlson mentioned libgit2 as a potential concern.

Looking forward to losing all references at once vs just the current one...

I've noticed persistent Git/fs interaction where on crash the current ref can just disappear...

PunchyHamster··on OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005
also the entirety of the research that went into breaking enigma in the first place is in the training dataset
PunchyHamster··on Suspension of the de minimis administrative exemption for imports $800 or less
I'd imagine even with duty added they'd still end up being multiple times cheaper
PunchyHamster··on Uber arbitration award over Emily Normandin-Parker’s death
None of the parties should have option to choose it, it should be within govt. justice system, just have arbitration court
PunchyHamster··on Raspberry Pi blocks changing RAM chips
I'd imagine that will be for compute module, not the other SKUs ?
PunchyHamster··on Raspberry Pi blocks changing RAM chips
Should just print "warranty void on unauthorized RAM", instead of not booting
PunchyHamster··on Pirate Face Rescues LLM Models from Deletion
Yeah but that's what provider like huggingface could just do, keep seeding the models so they are still accessible
PunchyHamster··on Samsung is expected to more than double output of its HBM4 and HBM4E DRAM
Threadripper have that extra bandwidth and M5 still is faster
PunchyHamster··on Backups Aren't Simple
We tested both and used Restic precisely because it was better in CLI/scripting department. Kopia is better for maybe less advanced user as it gives them GUI to click on

> he Kopia server is convenient if you have a server at home and want to back up laptops to it, with them being in a shared deduplicated repository without having to trust the clients not to get pwned and wipe out the backups.

Neither can do it well. We just used S3 + versioning for that reason, worst case we would just rollback the bucket

PunchyHamster··on A heap overflow and SSO misconfiguration to compromise OpenAI internal repos
That's "only" 11h of download at 300Mbit/s
PunchyHamster··on PCB is brought to you by Fable 5
At current PCB prices that's kinda not worth it. Hell, even actual assembly isn't that expensive (at least if you have "a lot of few components" vs "big variety of components").
PunchyHamster··on PCB is brought to you by Fable 5
I feel like just leaving PCB routing and schematics for it is just waste of tokens.

Once you have initial design (what chips to use, what each of them needs in terms of other components etc.) the schematic/PCB isn't the hard part but it is the hard part for the AI.

It's great at stuff that humans take a lot of time like say "get me buck converters in that range of voltage in power, and put price of chip and supporting components in the table", and at least in not heavily analog designs, finding chips that fit the purpose, are available and in price range is good part of the work

PunchyHamster··on PCB is brought to you by Fable 5
I used them a bunch just for the component shopping, I can route PCBs just fine, that's not a throttling factor in any project, but ability to "send it shopping" and not only get the price/availablity for the options but look thru datasheet for quirks and parameters is very useful.
PunchyHamster··on Backups Aren't Simple
We actually do use some of our backups in anonymize -> test environment pipeline so it gets dual use on top of just testing restores.

We also use S3 versioning in case something gets the keys and just erases it. Then it's replicated to the cloud S3-compatible store to get the offsite location.

Only feature I really miss is "turn this restore into .tar", both for testing restores and for convenience when we get requests of "we need this directory from 4 weeks ago"

← PreviousPage 4 of 34Next →