turingchan

Alright, I've got a new pet programming project: create a 4chan clone and populate it entirely with AI (mask user posts from other users) and go live. Then check for number of human posters, and see if we get frequenters. Basically, see if we can make an entire text board pass the turing test.

I've actually had this on my list of shit to do for a while.  4chan's JSON API is a massive source of markov chain fodder.  I was going to try to set up an anti-/r9k/, where you could only post if you had a certain percentage of sameness to anything else on the board.

>>1,2
Oh god, this would be terrible. But it's a good ratio of interestingness to implementation difficulty. It's perfect for a /prog/ community project.

Now this is something worth thinking about.

>>2

Just found the github. This is perfect... I had hoped something like this existed.

As a baseline, yeah, you can post random shit and have it seem reasonable.  But with an imageboard, you get something else, which is conversations with a short lifespan.  There exist `forums' populated entirely by clones, but you can easily distinguish them because everybody is just too polite and don't interact with each other.  With a scraper, you'll get lots of `ur a faget', `no u', `samefag' interactions, and they might be at a level of comprehension that's manageable to reproduce.  This is really the thing that would make something pass the Turing test - people have to think that interaction exists, not just action.