How hard it would be to create a bot that simulates replies to Jow Forums threads?

How hard it would be to create a bot that simulates replies to Jow Forums threads?

Attached: npc.jpg (971x546, 56K)

Other urls found in this thread:

desuarchive.org/qa/thread/2587940/#q2592632
github.com/typhoid/shitpost-learner
twitter.com/NSFWRedditGif

IDK, you tell me.

Attached: 02.jpg (348x235, 12K)

Not really, look, it takes me 10 seconds


while(true)
System.out.println("kek, based, redpilled, cuck, le s'oy, le frog, le wojak, le xd");

You're not fooling anyone.

That's exactly what i was thinking about.
Nobody would see the difference

thx op i cum on cat she hiss at penis

You mean something like ask god AI

I actually did something like that a while ago. Someone showed me this program called markovified and I fed it a bunch of posts from /qa/ and then set up a tinychan board on a freehost for a bot to dump them to. However, i'm too lazy to set up something like that on here.

The markovifier was a bit crude in my opinion though, as well as what I saw from another anons results. Machine learning would likely go way farther in useful yields.

desuarchive.org/qa/thread/2587940/#q2592632
Link because i cant post images from my phone

I'm actually waiting for the day that Jow Forums would be only bots and just a few real anons.

Just looked at the repo for Markovify and it turns out it was initially developed at Buzzfeed. jsvine (Jeremy Singer-Vine) has BuzzFeed News in his Organizations. Interesting, huh?

>how hard is it to create a bot with zero intelligence
ftfy, op

>He doesn't know he's the only person on this website and these posts are made by bots

Attached: laughing.jpg (600x536, 81K)

Have sex

wrong place

this one is bugged

Easy as shit, scrape the posts and train a LSTM with Keras in Python, there you go.

Already exists
I trained one on ERP threads for fun (erotic roleplay)
Pic related are some of the funniest generated posts it came up with

Attached: ss (2019-04-28 at 04.14.10).png (872x930, 88K)

source: github.com/typhoid/shitpost-learner

that's not how this shit works but telling you how to actually do it would cost you 150 USD/hour

Attached: NPC.png (1200x1083, 945K)

based and redpilled, have a poor man's Jow Forums pass

>shitting on "y'all"
thems fightin words

Attached: gun.jpg (600x900, 60K)

1. Suppose you train GPT-2 to generate realistic comments.

2. Generate 100 candidate comments per batch.

3. Now train a discriminator (maybe conditioned on GPT-2's attention information) to select posts from the batch, but your objective function is the number of (you)s the comment gets.

4. Sit back and watch your robot rake in the (you)s.

Attached: 1555716224601.jpg (808x767, 48K)

>defending degenerate poor American speech