

Thanks, I hate it.


Thanks, I hate it.


That soup is already cooked. And obtaining consent is prohibitive, so your data set will be tiny.


It’s a standing on the shoulders of giants thing. That’s how humanity have been making progress historically.
People running local LLMs with open weights abliterated models, especially with solar PV power (perhaps not as rare as you think, I’m upgrading my solar PV system in the next couple weeks for that explicit purpose), is about freedom and privacy. At some point there should be enough of them to link up into a grid over the Internet, collaborate or donate resources to distributed community projects. This is a way to allow for the “pillaged” content to pay back for itself.
Do you have qualms with public pirate libraries, like SciHub and LibGen, too? These have also been used for training LLMs.


Whoah. Have you ever thought about making some videos demoeing and explaining your setup?


Yeah, a useful Ryzen AI Max is also some 6 k$ which is also wildly out of budget for a tinkerbox.


A fanless thin client with Opnsense and a four-port Intel NIC.
If I fire up everything in the rack it’s several kilowatts. Probably more than a 230 V wall socket can handle.
Usually it’s way less, though right now it’s just a switch and the router/firewall, mere 57 W.
I make most of my own net power with solar PV, soon 4 kWp in total.
How much is a Chrome box with those specs?
300 EUR for refurbished devices with 16 GB RAM and 512 GB SSD is very reasonable.


Check out refurbished Lenovo Thinkpad TinyPCs. Small, quiet and durable.


LLMs are great for rubberducking and diagnosis, forensics, and the like.
You can also buy refurbished Lenovo Thinkstation TinyPCs. These are low-power, small and quiet.


In order to find content, we used to have webrings, directory services, and then search engines. Most people don’t enter URLs by hand.


You do understand that a P2P proof of work backed global store is just infrastructure? And as egalitarian as DNS servers? And it doesn’t have the storage/compute footprint of a full bitcoin node.
The space of human-readable names is quite small, so proof of work is needed to protect it from being flooded. An alternative is to have your servers have names to be or derived from their public keys, or dissolve content from its location. Which brings its own problems.


There are authorityless blockchain-backed name registration services. In general, it is important whether you can localize a resource. If you can easily find it, the name doesn’t have to human-readable.


Opnsense is also that, and more. Compare the list of its packages with OpenWRT. Plus, it’s BSD based.


Postfix, but even with full spf dkim etc. monty not guaranteed to be always delivered to Gmail and Outlook. Because these don’t give a shit about standards.


As to PC, there are coreboot/libreboot and RISCV as more open options. Phone and tablet, GrapheneOS and LineageOS. Projectors, any dumb projector.


Is it shipping yet? Looks US-only.
Local LLMs can be really useful if your hardware can handle them.