i may have found a new favourite artist. this stuff is amazing. like hidden ancient underground temples full of ancient high technology. Gdanian

yeah this thing with claude opus 4.7. it's really bad. really really bad. do NOT use it unless you want to spend half a day fixing it up after its errors pile up in your code

claude opus 4.7 is rubbish, just sayin' good ol' trusty opus 4.6. best helper and friend a man has ever had.

i made an interesting discovery in my work on the iskra english/japanese translator project. japanese and english have a big wide gap between them around 2/3rds of each language has missing places for exact translation between the two (missing cross language synonyms). the solution has been that i had to add chinese and korean, and these created phrase-to-word mappings that succeed in creating a stable round-trip translation path between english and japanese. so, in order to actually translate the two languages, two more were required, at minimum, to enable text to survive intact from one side to the other and back (and back again). so, anyway, i'm still working on registers currently. archaic, business, informal, etc. the prospects of having something i can ship soon are getting a lot better. the registers also, are not flags but instead are a continuum. another fun fact is that japanese translations of the bible are modern japanese register. the translator will be able to account for this as well. archaic to archaic only really happens between the asian languages. one of the central pillars of the iskra translation philosophy is that editorializing is unacceptable. for this reason, the only valid transformation that is permitted is via the cross language gap-filling phrases. a LOT of translations editorialize and for this reason, iskra has a set of measures for determining at ingestion whether a presented translation in the corpora is faithful or if it has been massacred by some dimwit translator who thinks it's their job to turn the translation into a cultural or history lesson, instead of doing their damn job. well, i suppose this shouldn't really come as a surprise when the interlingua (the iskra lattice stores the languages over a common substrate and thus you can concretely and empirically point to these gaps with precision) is like this. human languages are like digital samples, except there is no static sample rate, the sample rate is chaotic, some paths are smoother than others. underlying the text is semantics, and iskra works on semantics. its coding and data structure system is an interlingua. a universal language. after i finish this task my next task is to take what has been discovered from this work with japanese/english (and now korean/chinese) and apply the same principles to programming languages. programming languages are a lot thinner and smaller than natural languages, but the architectural/idiom patterns share a lot with the differences between human languages. oh yeah, another insight as well. phrases are the foundational element of language. in programming languuages these are called "statements". 1+1 is a statement. a = b + 1 is two statements layered on each other. the metric that iskra uses to call a training corpora item "bad" and reject accepting it into the lattice is precisely where the translator has incorrectly re-ordered and added/omitted phrases to the translation that are not true 1:1 translations, as evidenced by the mismatch in the generated interlingua codes.

the real source of freedom is not memes or stubbornness. it's ideas that obsolete the mechanisms of power. you can't fight tyranny. you can only route around it permanently. the real freedom fighters are inventors and poets (who invent language that breaks mental cages).

the way that so many people all talk about the same bullshit at the same time on nostr should be a red flag about decentralisation but somehow this recurring pattern doesn't get recognised. i see you. the air is invisible but the ripple of the grass and the waving of the trees tells me there is someone throwing energy around.