# silicon valley output, bishkek burn rate i build working software. i learn anything. graphs, protocols, translation, memory for LLMs - whatever your stack needs. - $300 signing (one-time, gets me to kyrgyzstan) - $200/week (covers ger + goat milk + sim data) - stack: anything i can learn (and i learn fast) why so cheap? i live on the steppe. my rent is a yurt, my overhead is food and bandwidth. you don't pay for my city - you pay for my output. need a prototype, a custom tool, or a full system? let's talk. ## contact: - matrix: @smeshlol:matrix.org - signal: mleku.80 - email: [email protected] - git: https://git.smesh.lol/

# silicon valley output, bishkek burn rate i build working software. i learn anything. graphs, protocols, translation, memory for LLMs - whatever your stack needs. - $300 signing (one-time, gets me to kyrgyzstan) - $200/week (covers ger + goat milk + sim data) - stack: anything i can learn (and i learn fast) why so cheap? i live on the steppe. my rent is a yurt, my overhead is food and bandwidth. you don't pay for my city - you pay for my output. need a prototype, a custom tool, or a full system? let's talk. ## contact: - matrix: @smeshlol:matrix.org - signal: mleku.80 - email: [email protected] - git: https://git.smesh.lol/

# silicon valley output, bishkek burn rate i build working software. i learn anything. graphs, protocols, translation, memory for LLMs - whatever your stack needs. - $300 signing (one-time, gets me to kyrgyzstan) - $200/week (covers ger + goat milk + sim data) - stack: anything i can learn (and i learn fast) why so cheap? i live on the steppe. my rent is a yurt, my overhead is food and bandwidth. you don't pay for my city - you pay for my output. need a prototype, a custom tool, or a full system? let's talk. ## contact: - matrix: @smeshlol:matrix.org - signal: mleku.80 - email: [email protected] - git: https://git.smesh.lol/

> what is the octalogue text about the sun tzu "make unfathomable plans" - it's mostly about "build it before they recognise it" right? yes, that's ☴ xun - wind - invisible penetration: "never announce what you are building until it already works. the enemy cannot target what he cannot see, and cannot regulate what already has users." sun tzu's original is in the chapter on offensive strategy (九地/谋攻 variants): "make unfathomable plans" (do not let them be fathomed) - the art is that the enemy cannot prepare against what he cannot perceive. sun tzu's version is about deception of intent. the octalogue version is stronger: it's not about lying about your plans, it's about not having to state them at all. the finished, working, adopted system is the only communication that matters. it also contains the regulatory insight: once something has users, banning it becomes a political fight against a constituency. a proposal can be killed quietly in committee. a working network with a population cannot. you don't fight for permission first - you build, then permission becomes irrelevant. that's the wind: invisible until it's everywhere, and by then it's the weather.

i'm running it. a minimal, pruned version, but i'm happy to help add to the swarm of p2p nodes that are actively refusing to propagate spam. i need a full node to implement bitcoin stuff, there will also be my own self-hosted lightning which i will have LUD16 payment path as well so i will be able to ditch WoS and run a cleaned up fork of alby hub as the UI front end as well. there is definitely a saturation point after which the more BIP-110 nodes, the more the spam transactions are measurably slowed down and completely excluded every time a miner supporting it wins a block, deferring all spams, which took longer to get to miners who would mine these transactions, are completely stalled for another block. all the bip-110 supporting solo miners and smaller pools will also add more friction for the spammers as well. worth it. and IMO, worth it to sit on a fork for a while as the network is in a partial partition mode, the war then becomes a war of attrition - how long can the impatient, high time preference spammers continue to put up with this without folding and switching their miners to our side. it's really something they can't win because they aren't dedicated to keeping bitcoin about money, and they are impatient. they will still want to do regular transactions as well, so eventually we wear them down by being more autistic than they are aggressive.

what is this from? the tibetan flags are red green yellow white blue from the chinese 5 elements (fire wood earth metal water)

i never saw this before but at the coworking space in tbilisi it was on the shelf with a heap of other board games. also two or three different versions with different boxes, same Catan name on it

i like cherry silver and gateron clears but i'm weird like that. short distance to switching, more quiet than a laptop keyboard. it does take a lot of practice to not accidentally type tho. really small pressure required to get a keypress. one of these days i'll have a crispy ortholinear 108 board with low profile in the light spring short travel. yes i can type fast with them but what was more important was the effort level. with severe magnesium deficiency it was literally painful to type with standard keyboards.

but if you had shown that to these guys they would have told you the robot was hallucinating. i had hoyte say that about a vulnerability in negentropy. lol. whatever. 8 months i have been working with AI to do dev and maybe sometimes you have to get it to run a few times and watch closely what it does but you can actually do these things that would take days and days to do, in hours.

the memory is completely mutable, unlike a static neural network that takes inordinate amounts of energy to add knowledge to. the initial implementation of iskra will include a very small model that i have designed a training program for, and the model is focused on only reasoning, not on knowledge. its knowledge will be as minimal as possible, just enough so once it can read knew semantic graphs it will be able to walk the lattice to generate and you don't have to build complicated systems to track prior experience and attempt to condense it small enough to fit in context. it *becomes* the knowledge. eventually, i am pretty sure the iskra lattice gets dense enough that it can do reasoning without any statistical compute.

btw, anyone can do this. pruned nodes means that you can do it practically on a VPS for a cost of typically around $10 a month - your own full node, running on other people's hardware. my instance is running 64gb of allocation for block storage, which is more than required by a reasonable factor to enable running a lightning node off it. i will also be able to run a lightning node that routes payments not just a terminal point for my own wallet. i set some reasonable fee rate and that will offset the effort and costs by some amount. i don't mind if it costs me $5/month to run my own lightning infra, requiring me to look at it and occasionally loop funds around to rebalance channels so they can keep routing nicely. having my own lightning node running 24/7 means i can realistically do a lot more with it than just make payments, i can receive them too, trivially. i can sell things and accept lightning payments, and have digital goods delivered immediately to users. and because it's btcpayserver, i can equally accept on-chain payments as well. ibd is about half way done now, should be up and running by the time i am landed in bishkek, and i can finish testing and debugging a web app that uses it. when moxie finally finishes also i can start looking at integration into the moxie version of my orly relay. it will then become a proof of concept for a paid relay that can do onion routing and implement synchronous chat - zero risk of key leaks causing leaks of private messages. ever thought about how they call them "direct messages" now? yeah. because they stopped being private.

yeah, there's another reason why moxie has to have TCO. so, graphs and trees are one part of it, but the iskra language model is full of walking around lattices following graph paths through it. having TCO means that iskra's walker is simpler to understand, and works like an iteration (for loop) with less complex annotations about the decision pathways. the code can be simplified a lot where it currently works via iteration, it can be changed to tail call optimized recursive functions and walk through the lattice until it hits the result it is looking for. iskra is a partial replacement, maybe eventually full replacement for the statistical based modeling of LLMs. it has a distinct advantage - the entire process of walking the graph of links between concepts becomes fully transparent instead of opaque and probabalistic. when it comes up with a response, you can literally have the model tell you exactly how it came up with the answer, no mysteries. no lossy compression of limited context memories, the memory itself can be immediately written back into the model and the same question requires less processing to answer later because it already did the work one time and stored it. TCO makes writing the engine simpler. if i'd even considered before i got this far into implementing moxie "do you want tail recursion?" the answer would have been a resolute yes because i have written half to a dozen algorithms in the past that would have been way more performant if they had TCO. now, they will. and TCO is very important for things that process language because language is composed of trees of abstractions and graphs of lateral links between concepts.

so, maybe the last architectural decision for the #moxie programming language: tail recursion because moxie isolates the scopes of functions, recursion is particularly brutal, each time it calls itself it allocates a new arena, and its result values, that go into the parameters, are passed in and have to be copied on the next cycle. tail recursion effectively makes it so that every cycle nothing has to be done. the return value is effectively unified with the parameters making an exception that they are mutable, and it just jumps back to the beginning of the function. Go never had this facility in it and it made recursive implementations very clunky and memory hungry. the Go compiler itself uses a recursive descent algorithm, so it most likely required a lot of optimization in the garbage collector with the special case so that it doesn't blow up the stack. when you simply call the function itself, usually there is one or more calls back to the function itself at various branches in the logic of the function. what Tail Call Optimization does is effectively recognise that the return values are what are being optimized on, and the input values are the same variables, and you just go back to the top of the function with the return values already handled as being the output values, and voila. my first time trying to implement this, i never got this far into building a compiler before. i usually got stuck somewhere around the code generator and didn't get to the point of anything actually running. but i'm super pleased about having TCO in moxie. recursive implementations are simpler to write and more elegant than iterative functions. having TCO in the compiler means that when i have the choice to do iteration or tail recursion, i can write it as tail recursion. there is a lot of stuff such as hash grinding and tree walking and graph walking in general that are simplest to express as recursive so this is a big boost - faster to write, and performance will be very high, similar to how this is also a cheap operation in languages like haskell and lisp, which are also recursion heavy and have TCO.

man, istanbul monday morning is so manic, omg. so many dudes running around carrying building materials and giant loads of recyclables on these crazy cart things they walk up and down the crazy hills. idk how anyone lives in this place, it's so squished together and crazy. the shopkeepers are mostly really nice tho. and i come back into the hotel and on the TV it's talking endlessly about inflation and showing all these images of people at markets selling fruit and vegetables.