a thoughtful web.
Good ideas and conversation. No ads, no tracking.   Login or Take a Tour!
comment by wolf
wolf  ·  3433 days ago  ·  link  ·    ·  parent  ·  post: Neural network generates Magic: the gathering cards

I used to run a MegaHAL IRC bot (actually I think a different incarnation is still around in, of all places, a major IRC channel for the Rock Band series) and it was absolutely hilarious (at first) to see the bizarre stuff it came up with. I liked to feed the bot with ONLY the conversation that was said in the channel, so it would be "relevant"... sort of. I did do some experiments with feeding it other things (articles about the games, as one example) but really the best was to start it blank and let it evolve itself. I ran 3-4 of them at once, so it was interesting to see how completely moronic one bot became while another regularly spouted reasonable (for a bot) conversation.

The latest incarnation uses a different Markov chain-based system, and works much much better. The originals would get very messed up with URLs and other strings of numbers/letters and would inevitably spout spammy nonsense after long enough. But it's been a really fun experiment to run a bunch of different "AI" bots and let them do their thing and see what happens.

Edit: Of course I didn't write these bots, they were just scripts for Eggdrop (or a standalone script in the case of WolfServ, the latest) but I wouldn't mind trying to figure out just how to do something like that. I'm not very deep into coding stuff yet; most of my coding experience comes from hacking up eggdrop scripts to make them do what I want, and then eventually writing TCL scripts from scratch one I figured out enough about how it works. And an EXTREMELY basic roguelike I started in python using the libtcod tutorial. But I've always been interested in trying to learn more about how such bots are constructed and how I might be able to do something like that myself.