Tailscale Enables Artificial Intelligence
Translations to English for Disinfodrome are coming.
Disinfodrome is more than just the Trump Russia sample server. Regular readers will know of the PressTV success and there are many more leaks out there that involve languages other than English.
Four months ago I pitched for some funding in Artificial Intelligence for Disinfodrome, and I did receive a couple good sized contributions, but I did not do what I described in the post. There have been a number of twists and turns since then:
Loss of the Scraptop.
Death of my workstation.
iPhone8 going dark at the worst possible moments.
Received Pinky & Brain, two Apple M1 laptops.
iPhone XR just magically appeared.
Cheap Dell R430 server found.
What I have only hinted at is that there’s an AI development group in chat. The busiest developer has hardware very similar to mine and there are two integrators who have multiple newer Nvidia GPUs. I am the only one with a Mac thus far and with my desktop’s GTX1060 I’m at the back of the pack Nvidia wise.
I’ve held off on the Dell R730 because there would be both a capital and monthly operating cost. Now thanks to Tailscale Changes Everything it looks like that machine will not be needed. We are finding that for translation the bandwidth needs are small, even the GTX 10x0 GPUs are able to do useful work, and with a tailnet it doesn’t matter where the compute resources live. When people make their GPU equipped machines available translation work will speed along, and the software required is an easy Docker Compose install.
The PressTV work netted a verbal commitment for a donation in the low four digits, and that hasn’t come through, but I think it will, so I’ve been contemplating what to do with it. Both of our older Dell rackmounts are running now, the R430 is getting some service work, and I’m fairly sure a pair of 24TB Exos drives are going to be part of the spend.
The other thing that’s crossed from want to need is a Google Pixel 8 for the sake of GrapheneOS. The iPhone XR is harder than the rest of the Apple phones, but it’s still a consumer OS.
Alternately, that outstanding donation is just enough to put one of these in my workstation.
An incremental, spread out approach reduces risk, and a redundant pair of drives is more useful than a single high end computing. But after the quadruple win in the form of the cheap rackmount, two great laptops, and a fine new mobile phone … maybe the heavy duty GPU is the right thing.
This all rides on customer activity for July. My basic needs are met and if this one big job that’s been stalled should move, the donation covers the GPU and I’ll get the larger drives on my own.
What money can not solve here is the need to get some sort of GraphRAG solution done for our document caches. I think the incremental thing to do here is just get a RAG solution running with Dify.ai so I start getting a feel for what is possible.
There are just four days left in June … can I do it? We’ll see …