I’m very interested in this approach because I’m heavily constrained by money. So I am gonna be looking (in non appliance contexts) to develop workflows where genAI can be useful when limited to small models running on constrained hardware. I suspect some creativity can yield useful tools with these limits, but I am just starting out.

PhotoGIMP is a godsend.

Thank you so much for this detailed starting point!! ❤️

Anybody got a summary of the drama I smell behind this?

Thank you for the work you do maintaining this space. Antagonistic users who don’t know what they’re talking about have myriad other places to complain, it’s not necessary to endure them here.

Voip dot ms employs a ton of countermeasures versus skript kiddies as voip fraud is a severe problem. You are unlikely to have much luck with Tor Browser. I have to ask them to take my boring data center IP off their greylist every time I want to add cash to my balance.

try voip dot ms

Save yourself a lot of trouble and get a hardware SIP phone like the Grandstream WP826. I spent years struggling with software phones, most of them suck ass and the good ones aren’t very good.

quick unboxing video

Then make sure to enable voice transport encryption and set SIP transport to TLS (not UDP) and set keepalive timers to something below 3 minutes or so. The encryption settings are not just for security, using TLS for SIP transport has way fewer problems with incoming calls than UDP in 2025 network environments.

There are some firmware issues with the Grandstream WP826 but they are steadily releasing new firmware updates every month or two. If you need absolute bulletproof reliability go for something more expensive, otherwise the WP826 or similar model will likely be good enough (I say that as someone who is easily vexed by shitty software/electronics)

ArchiveBox strikes me as being a rickety pile of hacks, but it does mostly do its job out of the box. The built-in search is abysmal however and must be replaced with one of the other options to be useful.

Currently in a holding pattern because, while I got RAM & SSD for a new-to-me “1-liter” server before tariffs hit, I don’t have the server itself nor any money to buy one, despite looking for 9th or 10th gen Intel which will cost me only $120 to $150 barebones.

Money to buy one is not coming in because the place where I live has nonstop noise & activity and I don’t have a separate room or any door I can close, which severely limits my ability to work as I have auditory hypersensitivity and an absolute need for solitude in order to recharge enough to think. 🤷🏻

mood

everybody downvoting your comment has zero experience being the go-to family tech guy for relatives in their 80s and 90s who can’t reliably distinguish between windows, dialog boxes, menus, and buttons

Don’t view keys in Monero “expose information to the authorities” if used in such a way? Hard to see the need for a system where tokens can “expire” when we already have a perfectly good one where they don’t.

lol the downvote counter just ticked up a couple times as I was tapping around trying to figure out wtf

if you wish to get roasted with additional specifics, see !localllama@sh.itjust.works

But what makes this AI model unique is that it’s lightweight enough to work efficiently on a CPU, with TechCrunch saying an Apple M2 chip can run it.

An Apple M2 can run bigger, higher-precision models than this FWIW. More important than this is perhaps whether older CPUs can run it with acceptable performance.

AI models are often criticized for taking too much energy to train and operate. But lightweight LLMs, such as BitNet b1.58 2B4T, could help us run AI models locally on less powerful hardware. This could reduce our dependence on massive data centers and even give people without access to the latest processors with built-in NPUs and the most powerful GPUs to use artificial intelligence.

This is definitely relevant to my interests especially with NPU support for such models coming. Dirt cheap ARM-based PCs based on e.g. the RK3588 are shipping with small NPUs

I should probably give in and get the Puli, I dislike non-replaceable batteries but I’m sure if it dies before I upgrade to 5G and give the device away I could figure out how to replace it.

The Spitz AX looks awesome for e.g. an RV base station etc, but too many antennas for travel use. I used old Spitz on Amtrak with some success. I wish they had a similar 5G unit with 2x external 5G antennas, and internal antennas only for wifi.

Maybe I will get my wish someday, GLi do like producing a variety of devices on a theme…

I have a Spitz, just don’t plan to run a VPN on the device itself as the CPU is very slow and will limit max thruput.

in the USA, LiFePO4 battery cost of drop-in 12V replacements has fallen to the point where lead-acid may no longer make sense for projects like this.

But why doesn’t it ever empty the swap space?

Why would it do anything with the swapped-out pages if they are never subsequently accessed?

Paging them back in for no reason is not an effective use of CPU time or system memory.