Good news: Every time I check my GenSys dash, there are new, better features.
This evening I found this deterministic (No Ai) System Analytics panels. Useful
Bad News: Its only Tue and I've burned all the weekly Compute (resets Sat).
I swear it hasn't been this bad, so either #anthropic turned down the screws again, or, they use dynamic pricing to combat sneaky buggers like me who want all the compute.
#sysadmin #vps #vibecode #selfhosting
"Current data loss exposure is total"
Talk dirty to me motherfker !
Uh...I'm going to have to make this (SysGen) tool Model agnostic.
Switching into another LLM when it burns out the compute.
Problem 1. How to modularise the progress logic so we don't burn all the tokens on the model switch.
Problem 2. How to handle variable model competency.
Plus: Having a stack of available models for deterministic launched Ai diagnostics may be very positive. Esp. If we do some adversarial shit.
I'm an old System Admin and in my twilight years I've been drifting back to what I loved doing as a young person. Herding iron.
I'm gathering a slowly growing gaggle of VPSes (Virtual machines)
Some are actually doing useful work.
But because I have very little time to do it well, I often forget where shit is, I lost a server a while back, it's out there abandoned running a terribly slow 1CPU local LLM clawbot.
It's a sad story. Forgot to document it.
I deployed ok monitoring tools like Prometheus/Grafana, but I also have associates to whom grafana dash is ancient runes. So simple English that tells health of a machine is good.
With vibecode you can literally create magic castles, so I'm having useful fun.
It's creating deterministic code, so I will release it for the brave/foolish ones when it's done.
Also, I created a compute governor and I needed a complex long term project...
...that's what I do instead of watching trains 🙄
Thanks for asking.
Wulfy—Speaker to the machines
Aunty