@kaia
And I know what neural networks are for even more malicious reasons — I wanted to make tool to flood a certain forum back in early 2010s. I ended up implementing the part that bypassed the CAPTCHA with ~80% success rate and the part that generated the post text, that was supposed to look like text, not something easily identifiable — it used fragments of earlier posts, this was supposed cleaning up very hard for the moderators as figuring out that it was gibberish required actually reading into it — there were no widely available tools to automate it at the time.
There also supposed to be a part that implemented "personalities" e.g. always using certain user agent when posting from certain IP-address/proxy — this was supposed to make it like real people in server logs, to make automated cleanup of these posts even harder…
But this was the point where I had to deal with too much WebShit™ for me to take and I gave up :marseyemojismilemouthtighteyes: