<isaac_newton:matrix.org> Interesting never seen shadow network. That runs on kvms? > <@gingeropolous> @isaac_newton:matrix.org: , no its this vibe coded monstrosity: github.com/Fountain5405/monerosim

<isaac_newton:matrix.org> Until it's not > <@jpk68:matrix.org> It's probably fine™ in most cases (this should not be taken as financial/legal/practical advice)

Interesting, I missed that so-called stale data attack (false claim of invalid partial tx). Best mitigation is procedural, e.g. in a 2-of-3 any tx attempt failure is grounds for adjudication.

<gingeropolous> lulz, can't get fable to build the selfish mining toolkit because it gets flagged..... i need to get frontier level on prem somehow

<ofrnxmr:xmr.mx> use kimi or deepseek?

<gingeropolous> yeah. yet another subscription / fee i don't wanna have

<gingeropolous> i mean, opus < 4.8 built most of monerosim .

<ofrnxmr:xmr.mx> cant you run deepseek locally

Opus 5 is good enough for tasks like this

<ofrnxmr:xmr.mx> ive never tried, but i thought it was one of those local models or whatever they are called

GPT 6 Astra is even better, and doesn't seem to be as castrated as Fable 5. But it eats tokens like crazy

Kimi K3 can be run locally, but you need a server with 2 TB RAM and a couple 5090's to run the full model

<gingeropolous> even got flagged on opus 5!

the more 5090's the better

<gingeropolous> apparently anthropic tryna defend the blockchains

<elongated:matrix.org> sech1: They don’t have subscription ?

They do, of course

But it's an open weights model, so you can theoretically run it locally

<elongated:matrix.org> Subscription or pay per token would be cheaper than buying 2tb ram

<ofrnxmr:xmr.mx> 116gb ram is like $200 now lmao

<ofrnxmr:xmr.mx> 16*

<gingeropolous> im too far on the vibe side of vibecode for local llms to work for me I think. i give the models analogies and ramblings and blue sky nonsense

<ofrnxmr:xmr.mx> vik vibe coded his arbiter app using local models, i think

<yokoama:matrix.org> @ofrnxmr:xmr.mx: nope

<ofrnxmr:xmr.mx> nope?

<ofrnxmr:xmr.mx> what did he use

<yokoama:matrix.org> claude afaik

<gingeropolous> ermagerd i need more IPC. these monerosim runs are too slow

<gingeropolous> ok, what should i name this protocol / software: "A peer-discovery protocol with no seed nodes at all: a fresh node finds the network by scanning random IP space for anyone answering on port 7337, and whoever answers, whatever chain they run, hands it peer addresses for the network it wants." 1. Transpeer 2. Sounding 3. Homing 4. Holler 5. other

<gingeropolous> p2p networks unite!

<gingeropolous> end the seed nodes!

Discovery of Service.

<gingeropolous> lulz

Also, it seems like it'd be vulnerable to the majority being able to effectively censor the minority, be redirecting the minority's unwanted service to a fake corralled network.

<gingeropolous> yeah there are various defensive mechanisms in place currently

So. every daemon listening on 7337 would have to speak p2p for every network it can give node info on ?

Never mind, I guess every network would just add a new protocol just for that.

<gingeropolous> yeah, little transpeer sidecar

<isaac_newton:matrix.org> the shadow network has me interested in more ways you can combine multiple low resource instances

<isaac_newton:matrix.org> you could run a full node that way

<isaac_newton:matrix.org> when altcoins were more popular, the bitcoin client could be run with 1 core of cpu and 350 mb of ram - resource requirements are rising. Combining vms could be a way to make better systems