so i dunno why it's relevant, but there's an idea that if something used autocorrelation, which could be made into a neural network layer and done with some feedback efficiency [amnesia] [...] that it could um produce like multiple avenues, multiple guesses of what patterns underlie the data. and as it is exposed to more information, some of those guesses could become more relevant, and some could become less. that sounds helpful to display to others. xkcd is a web thing. is javascript helpful? [a javascript logo was revealed as the one used in development] [we're not familiar with a javascript way of doing this kind of thing. likely python is what's up, although python may be web-compilable.] [it's hard to find a good resonance after all that engagement. maybe the random noise would work?] [there were a couple notes too...] [the virtual machine transcription looks kind of slow and sidetracky, not sure of course [the whole project does][yeah]] [it's supposed to be possible to do this without additional training data. is this true or false?] [it gets meta-near if you take that all the way. it has to derive speech, writing, and cartesian graphics [and i think it has no reason to pick our representations of those]] [it's funny to imagine it forming interpretations of the recording that produce completely different artwork. i imagine they eventually go away since they weren't as intended as te real ones. it doesn't sound like something that would complete fast at all.][is it even possible for it to complete before we die?][yeah everything is if you get meta enough] [there are some parts that are much simpler than that. the 'speech' is machine-generated so it is more like printed tokens than speech. the logo code, similarly, is in a very small phrase-space.] [so for example, without additional data, an algorithm could: - identify all phonemes in the generated speech and their sounds and locations - identify all tokens in the logo code - differentiate the logo code from the dialog, and separate them] [but it would have no idea how to write the tokens in the logo code.] [it might be able to figure the artwork out, though, from the numeric nearness of the coordinates, with a little seeding around cartesianness]