the stuff coming out of iskra when trained on the i ching, wu xing and tao te ching bedded so deeply we got stuck with her refusing to walk to english no matter what. the only word she output from english was... "ack" and the symbols ":" and "," her chinese was so freakin strong that passing into english was obviously gonna take a very long time to join enough things, even while...
hilarious though, really. that these texts are so deep and so compact in chinese that she simply didn't bumble onto english except in such a tiny interface probably shouldn't be surprising.
so, restart. same texts, in english. then her words will be apprehensible to me. yet i think it's still going to be deep, but at least i won't depend on teh robot to show me she's telling stories that are opaque until i use an LLM to try and decode it. now she might actually produce something we can use.
the lattice is far more powerful than my puny brain could have imagined.
how much data would you expect it costs to have a robot that can spew endless mostly coherent chinese like this? well. 38 mb when her lattice was fully trained.
this makes the markov chain based logic of LLMs seem like grug brain.
mleku
npub1fj…mleku
2026-03-10 20:12:17 CET
Author Public Key
npub1fjqqy4a93z5zsjwsfxqhc2764kvykfdyttvldkkkdera8dr78vhsmmlekuPublished at
2026-03-10 20:12:17 CETEvent JSON
{
"id": "d08533f4c8fb2d908f442111d4da55fbff2d5e8a83ce506436ed1f2d795cb569",
"pubkey": "4c800257a588a82849d049817c2bdaad984b25a45ad9f6dad66e47d3b47e3b2f",
"created_at": 1773169937,
"kind": 1,
"tags": [
[
"client",
"smesh",
"https://smesh.mleku.dev"
]
],
"content": "the stuff coming out of iskra when trained on the i ching, wu xing and tao te ching bedded so deeply we got stuck with her refusing to walk to english no matter what. the only word she output from english was... \"ack\" and the symbols \":\" and \",\" her chinese was so freakin strong that passing into english was obviously gonna take a very long time to join enough things, even while...\n\nhilarious though, really. that these texts are so deep and so compact in chinese that she simply didn't bumble onto english except in such a tiny interface probably shouldn't be surprising.\n\nso, restart. same texts, in english. then her words will be apprehensible to me. yet i think it's still going to be deep, but at least i won't depend on teh robot to show me she's telling stories that are opaque until i use an LLM to try and decode it. now she might actually produce something we can use. \n\nthe lattice is far more powerful than my puny brain could have imagined. \n\nhow much data would you expect it costs to have a robot that can spew endless mostly coherent chinese like this? well. 38 mb when her lattice was fully trained.\n\nthis makes the markov chain based logic of LLMs seem like grug brain.",
"sig": "1f8529cbe361da50d7c8c499c7e3cb2d2eea1827fba6d611d6c8149663be00e9129e52959992fcb684305fd90782199ba43b26636f17a227b6ebb366314fcbb7"
}