PendingKetchup's recent activity
-
Comment on Skrillex - FUCK U SKRILLEX YOU THINK UR ANDY WARHOL BUT UR NOT!! <3 (2025) in ~music
-
Comment on Artificial incompatibility - a rant (Dell notebook) in ~tech
PendingKetchup Link ParentThis is why Microsoft-branded hardware actually works pretty great with Linux, I think. As an OS company they understand the pain of hardware that feeds the OS garbage and generally misbehaves, so...This is why Microsoft-branded hardware actually works pretty great with Linux, I think. As an OS company they understand the pain of hardware that feeds the OS garbage and generally misbehaves, so when they make hardware they make sure to implement the specs properly.
-
Comment on Please stop externalizing your costs directly into my face in ~tech
PendingKetchup Link ParentWhy are people doing this? And why are they doing it like this? I understand that you want data to train your model. So some company training a big model, like Google or Meta or OpenAI, scrapes...Why are people doing this? And why are they doing it like this?
I understand that you want data to train your model. So some company training a big model, like Google or Meta or OpenAI, scrapes everything. But these are big operations, they're not supposed to be likely to run around evading IP bans and ignoring robots.txt.
Who is in the intersection of "owns a jillion GPUs to train models" and "acts like an asshole script kiddie with a botnet"? Who is going to buy a crawl dataset from a cybercriminal and use it as a core part of their business proposition? Why don't they freakin' cooperate with each other like Common Crawl and not fetch the same data over and over again?
Or is this just cyber-colonialism coming home, where the software has run out of world to eat, and starts creating economic incentives to eat other software with the same rule-breaking disruptiveness?
-
Comment on In email, Microsoft suggests Windows 10 users trade in or recycle their PC in ~tech
PendingKetchup Link ParentI wouldn't recommend using Windows 11 without joining it to a domain. And even then I've had it encourage me to associate a Microsoft account with my apparently corporate-provided domain account....I wouldn't recommend using Windows 11 without joining it to a domain. And even then I've had it encourage me to associate a Microsoft account with my apparently corporate-provided domain account. Even when Microsoft account sign-in was disabled by policy, so clicking the angry red dot brought up a screen saying I wasn't allowed to do the thing.
So also I guess I wouldn't recommend using Windows 11.
-
Comment on In email, Microsoft suggests Windows 10 users trade in or recycle their PC in ~tech
PendingKetchup Link ParentI wanted to make the joke that letting kids play Bedrock is bad parenting. But it's what MS pushes and I think they only keep Java around to avoid someone reimplementing it. I'm hearing great...I wanted to make the joke that letting kids play Bedrock is bad parenting. But it's what MS pushes and I think they only keep Java around to avoid someone reimplementing it.
I'm hearing great things about ClassiCube, the Minecraft Classic reimplementation that runs on a potato.
Probably what you really want to do is give the old PCs to the kids with a couple screwdrivers. Taking down, rebuilding, and futzing with the critical system settings of a computer that does not have to work should be a mandatory educational experience.
-
Comment on Shorthorn Project - Run modern programs on Windows XP/2003 in ~comp
PendingKetchup Link ParentThose are spelling errors you make by misspelling words, not by using translation. Usually a native English speaker will not make so many by the time they get to be able to program, and a spell...Those are spelling errors you make by misspelling words, not by using translation. Usually a native English speaker will not make so many by the time they get to be able to program, and a spell checker might catch them. But installing foreign language spell check dictionaries is not something I know how to do. It's clear the writer was too lazy to do it, but that might not be very lazy.
-
Comment on Firefox's new Terms of Use grants Mozilla complete data "processing" rights of all user interactions in ~tech
PendingKetchup Link ParentIt sounds like they're writing as if Mozilla is the legal entity operating Firefox on your PC, rather than you. So the process of typing a URL involves you telling Mozilla, in the person of...It sounds like they're writing as if Mozilla is the legal entity operating Firefox on your PC, rather than you. So the process of typing a URL involves you telling Mozilla, in the person of Firefox, the URL, and then Mozilla sending the URL to the destination server to get the content.
I think this is an absolute nonsense way to parse the situation, but it seems to be becoming popular in the age of "apps" that go on people's devices and then do stuff there that actively harms those people and advantages their developers.
-
Comment on Funny output from Claude.ai: a codeblock in a question about an early rennaissance theological work in ~comp
PendingKetchup LinkPeople have been training models for contexts where they do have access to code interpreters to do calculations for them. And they have also seen a lot of blog posts and things that say "Now we...People have been training models for contexts where they do have access to code interpreters to do calculations for them. And they have also seen a lot of blog posts and things that say "Now we need to fetch the doobledo" and then break into code to do it.
They also don't have great mechanisms for retrieving things from memory in the way a human would. Was the stuff you were asking about that had been output earlier actually in the same chat? Had it been pushed out of the context? The model might not actually have a way to get that bird to come when called, so to speak, and is (luckily?) imagining that it has a way to recall that information instead of imagining the information from whole cloth.
-
Comment on Shorthorn Project - Run modern programs on Windows XP/2003 in ~comp
PendingKetchup Link ParentI'm inspired. If software is so poorly described that it looks like it was written up by a distracted 10 year old, but I'm somehow hearing about it anyway, it must be interesting software. It's...I'm inspired.
If software is so poorly described that it looks like it was written up by a distracted 10 year old, but I'm somehow hearing about it anyway, it must be interesting software.
It's also probably software I will not be able to get working, but it'll be interesting.
I think usually this is a language barrier thing.
-
Comment on How does one learn how to learn? in ~life
PendingKetchup LinkI've found flashcard programs with multiple choice questions great for memorizing stuff I otherwise don't care about. Instead of needing to force myself to read a boring thing, I just sit and try...I've found flashcard programs with multiple choice questions great for memorizing stuff I otherwise don't care about. Instead of needing to force myself to read a boring thing, I just sit and try to get the questions right until I've memorized all the questions and answers. Getting a right answer is fun. And once I know all the answers, I've memorized the information.
-
Comment on Any VLAN expert here? Will be setting it up on my Mikrotik router and Unifi APs this weekend. in ~tech
PendingKetchup Link ParentGood luck! If you only have a few hosts that need to talk to both trusted and public clients, and they're all things like Linux servers, you might be best off giving each of those hosts interfaces...Good luck!
If you only have a few hosts that need to talk to both trusted and public clients, and they're all things like Linux servers, you might be best off giving each of those hosts interfaces on both the trusted and public VLANs, and doing the firewalling on the host to say what should listen on which address and what ports traffic should be allowed to from each interface. Then you can avoid dealing with setting up good firewall rules at the router for access between subnet.
Though you might need to worry about turning off IP forwarding on those hosts or really writing your firewall rules to filter by receiving interface: I can send a packet addressed to your private-VLAN IP to you over your public-VLAN interface, and you might route it to your own IP and then handle it.
-
Comment on Any VLAN expert here? Will be setting it up on my Mikrotik router and Unifi APs this weekend. in ~tech
PendingKetchup LinkI know one thing about VLANs, which is that they're just quotation marks for packets. On one physical link (Ethernet cable or SSID), you can send untagged (normal) traffic, and also traffic that...I know one thing about VLANs, which is that they're just quotation marks for packets.
On one physical link (Ethernet cable or SSID), you can send untagged (normal) traffic, and also traffic that is tagged with a VLAN number that it is supposed to belong to.
Usually, you use this to put different devices on different networks without having to run physical cabling and install routers and APs for each network. You can take two VLAN-aware AP and tell them to connect SSID X to VLAN Y, and connect them with one Ethernet cable, and devices on both ends on corresponding SSIDs will be on the same network as each other, but not on the same network as other SSIDs.
With a physical router the easy thing to do is assign a VLAN to a port, so traffic tagged with that VLAN internally will go in and out that port without tags. Then whatever you plug into that port is on that network and not on the other networks. The device you plug in does not need to know about VLANs.
If you want to grant a device access to multiple VLANs, you can have multiple physical NICs and plug them into router ports on different VLANs, or you can plug it into a port on the router that is set to carry traffic from multiple VLANs, with tags attached. (Plus you can decide what VLAN if any should send traffic out that port without a tag.) Then on e.g. Linux you can set up an Ethernet interface for each tag value on the base interface. So to the software it looks just like having multiple NICs, but they're virtual.
If you have multiple routers or VLAN-aware switches or VLAN-aware APs, you want to connect them together via ports that are set up to carry all the traffic for all the VLANs with the tags on. Then ports on VLAN 3 or whatever on any switch or router in the setup will be connected to each other, and everything on all VLANs will ride the shared cabling between switches/routers.
The other way to get a device to be able to talk to devices on other VLANs' networks is by actually doing internetworking. Basically, since these are separate networks, you can route between them. If you set up your router to be the default gateway and DHCP server on VLAN 1, and also the default router and DHCP server but with a different IP range on VLAN 2, then when hosts on VLAN 1 try to dial addresses on VLAN 2, they will see the IP is not on their network and send the traffic to the router, which will then see that it has an interface on the target network and forward it along (unless the firewall is set to block it). You can configure firewall rules on the router to do things like let hosts on one network open connections to and get responses from hosts on another, but not the other way around.
Anything that uses broadcast packets (which e.g. IoT devices might use to find their hubs, or anything that advertises itself to the local network uses) will stay within a single network and won't route through a router separating two networks.
It helps here to think of a router as a computer plus a VLAN-configurable switch. The router has a switch port that talks to its processor, and you can configure VLAN assignments for that just like for all the physical ports. Often this is used to give a router WAN and LAN ports: traffic from each kind of port arrives at the processor tagged with a different VLAN, and the router processor's OS has a virtual interface on each and routes packets between them. So if you do something like turn off connectivity between the router's processor and all the other ports, or set it up so that all the physical ports are on the VLAN that the router thinks is WAN traffic and there are no LAN ports anymore, you can indeed lock yourself out of whatever web UI it has and need to do a hard reset.
-
Comment on TRMNL - Open source e-ink "companion" device in ~tech
PendingKetchup Link ParentThey want to build a deck of cards you swipe through like a slide show. Because they don't have a document about their product, they have a pitch deck. But Instagram-slide-show is not a real thing...They want to build a deck of cards you swipe through like a slide show. Because they don't have a document about their product, they have a pitch deck.
But Instagram-slide-show is not a real thing an HTML document has, and "swipe through" is not a thing a scroll wheel does.
So they build something that assumes you are touching the cards, bolt it to the scroll gestures on the TouchPad on their Mac so they can preview it, and ship it.
EDIT: they do in fact have quite a lot of document about their device. I have no idea why the hot thing in web design is to write a whole Gitbook of documentation and then make a janky landing page that offers no affordances to hide it behind. The crypto and AI people do this too and it's baffling.
-
Comment on Seeking suggestions for Windows virtual desktop (for Photoshop schoolwork) in ~comp
PendingKetchup LinkIf you wanted to avoid working on a bare cloud provider, you could try one of the "cloud gaming" setups people are selling where you can install software on a Windows VM they give you. Maybe...If you wanted to avoid working on a bare cloud provider, you could try one of the "cloud gaming" setups people are selling where you can install software on a Windows VM they give you. Maybe https://shadow.tech/ or something like that might work?
-
Comment on Reddit will lock some content behind a paywall this year, CEO says in ~tech
PendingKetchup Link ParentI am confident that the old Reddit could have pulled off building a Patreon competitor, or an OnlyFans competitor, or even Something Awful. The new one I'm fairly confident is looking to extract...I am confident that the old Reddit could have pulled off building a Patreon competitor, or an OnlyFans competitor, or even Something Awful.
The new one I'm fairly confident is looking to extract money more than they're looking for a way to use costs and transfers to make a community function.
-
Comment on Is it okay to use ChatGPT for proofreading? in ~tech
PendingKetchup Link ParentYeah, one thing ChatGPT has going for it is that it is in fact a product and not a pile of script glue. And you might not have the local power to run a model as big as you need. But it still might...Yeah, one thing ChatGPT has going for it is that it is in fact a product and not a pile of script glue. And you might not have the local power to run a model as big as you need.
But it still might be worth looking for less-powerful options, or options that don't grant OpenAI so many rights to your text.
-
Comment on Google Maps now shows the 'Gulf of America' for US users in ~tech
PendingKetchup Link ParentI heard they pulled in an update to a government database they use for place names. But they seem to be making up facile process-based excuses like this to try and make people ignore that the net...I heard they pulled in an update to a government database they use for place names.
But they seem to be making up facile process-based excuses like this to try and make people ignore that the net result is obviously wrong.
-
Comment on Overwhelmed with the realm of data exploration (datalakes, AI, plus some c-level pressure) in ~tech
PendingKetchup LinkCongratulations on your new research project! If you build this thing, you will be a top-tier AI researcher contributing meaningfully to advancing the state of the art in the field! Now go make...Congratulations on your new research project! If you build this thing, you will be a top-tier AI researcher contributing meaningfully to advancing the state of the art in the field!
Now go make sure your bosses understand that.
By combining things like LangChain and its tool use, DeepSeek and its chain of thought instincts, and RAG as you mentioned, and maybe giving it a page of text that explains how to do business strategy, or a pass of training on explanations of good and bad ideas in your problem domain along, with the database queries that fetch the evidence to support them, you might be able to make something.
It's not clear that this is a better investment of time than if you just built the data analysis tools that would help a human answer these questions themselves. Especially since you mostly have to build those tools anyway to feed the AI. A model can just write SQL queries, but the easier the tools are to use the smarter the system is going to be.
The sampler/framing code/driver logic is going to be where all the real work is, and it's going to be a lot of work. Have your boss read up on Williams syndrome and Broca's aphasia, because really with a language model you have a linguistics research project. Everyone involved needs a good understanding of how language is a special-purpose function of the human mind only loosely connected to the other parts of what one might call intelligence, and how a project like this amounts to "build the rest of the brain".
-
Comment on Is it okay to use ChatGPT for proofreading? in ~tech
PendingKetchup Link ParentIt's a live service provided by what (is? soon will be?) a for-profit (non-B?) corporation, which necessarily means it is designed to put the corporation's profits over the interests of the user....It's a live service provided by what (is? soon will be?) a for-profit (non-B?) corporation, which necessarily means it is designed to put the corporation's profits over the interests of the user.
In this case that might look like an incentive to cultivate a dependence and then jack up the price, a lack of attention (or of drawing the user's attention) to problems with the output that are cheaper to pay damages over than to fix, or a standard enshittification where it finds people like advertisers or ad-targeting companies to sell the user out to.
There's also the problem where people doubt they have clear title to their training data, but they use it anyway. And the way they jealously guard their position and hipocritically get mad every time somebody like maybe DeepSeek trains on data generated from their model.
-
Comment on Is it okay to use ChatGPT for proofreading? in ~tech
PendingKetchup LinkI am going to come in with a rare no: do not use ChatGPT for proofreading. You should be able to do LLM-based proofreading with a much lighter, cheaper, and more efficient tool than the...I am going to come in with a rare no: do not use ChatGPT for proofreading.
You should be able to do LLM-based proofreading with a much lighter, cheaper, and more efficient tool than the umpteen-billion-parameter, needs-more-GPUs-than-you-can-buy ChatGPT. Something that takes a tiny model and runs it over a few dozen tokens of context and lights up red or something when the text is highly improbable. You ought to be able to solve the problem pretty well with a few watts on a laptop, so the extra hardware and energy used by doing it with millitary-grade ChatGPT aren't justified and thus should not be used.
Here's an example of local grammar checking on Mac with a 7B parameter model, with a bunch of Automator stuff to tie in to text boxes in apps, though I think it still uses Q&A and not the model's probabilities directly, which probably would be better.
Here it is on SoundCloud
I liked the part in the middle where it was seized by Atlantic Records and replaced with silence.