Databending. Which is applying random noise to a file in order to generate something that still decodes well, but has some weird/cool glitch. I went to a workshop about this in some underground club that no longer operates; still thinking about it and toying around from time to time. This was waaaay before Dall-E et al. were cool ways to generate quasi-crappy images.
Try taking a 1MB compressed text file (with some interesting text), cutting it in half, and replacing the second half with random numbers, then uncompress it.
It tries to decrypt the random data using the corpus built on the original data.
so the end of the file looks like:
"Hesn’ed a tembers, ea ' mo byarrledMaelcallrlynwigr oysTold yo joinnt bnan twas The spwt?" aaoWe sastleenkeep had thein; aimaf it oneivof sv
TheeTssuterrcy oid,itng the shuri us d it m,
ed iroduone dga go toe s, mooerhae�er
"aot ofSsallR:58:40, in the darkn,erqwader t. "Whsne�ee thit wasaidfrenthct,nnto erut be�sRhe Hiesented onvery h Tsh thrs, at d tkly he on
upnbadarkn,eJls tat.iwheitu rs p.you ue, nC could seunnip on
Lo of its e put his p," theden, b��t te�ee thln’t tcept ofuicideBge fnoh me,sh yod aw to awyarrleomo. rare
Hutofases vred ve s"s. "Hicsuty sely’s lher uri and
watcosThe yfiile.cn,
aub, baelcumsemit wChcycleis the ��d taJ hair b"
https://en.wikipedia.org/wiki/Databending