That's exactly what a Markov-chain would say!

@fribbledom I used MegaHAL some time ago, and one of the best answers was:

«If it doesn't work, it doesn't work, and so it doesn't work.»

@mehdorn @fribbledom

- if problem: solution 1/1
- if no solution: no problem 1/1

@fribbledom would say! to you: “Do you want some tea? would be nice to your pet the dog goes to heaven-sent to spam and eggs

I have made a Markov Bot out of a friend group chat and added it, Lärry has been our valid, entertaining, artificial friend for about two years now!

@Ayior @fribbledom
Would you mind sharing some details? This sounds pretty awesome and hillarious!

@Gisikon_Root @fribbledom I don't want to share the bot itself because it technically contains private information, but I can share how it works!
Basically, this telegram chat generates ~500 messages per day based on 6 people in it. At first I took the entire 3 year history of the chat, but it proved *too much* input, decreasing the coherence of the output. About 3 months turned out to be just right. In two years of operation Lärry has never felt repetitive. (cont.)

@Gisikon_Root @fribbledom I stripped the chatlog of any unclean data, that is, anything that is not user-readable text. The final formatting is a simple "Name: Message /n." When (at)ed, the bot generates until two /n have been hit, signalling that at least one full message is generated. It's then spliced after Name: and after the first punctuation, including comma. (Yes I know this could easily be more efficient). The result is just sent back through the Telegram python library. (cont)

@Gisikon_Root @fribbledom I splice when a comma happens because after some testing it turned out short sentences are more believable than long ones. And that's it!
The surprising part is that Lärry seems to have a distinct personality, and it's... The worst of all of us. He (and he has stated his pronouns, and Gender as Genderfluid, no one else in our chat is Genderfluid) is considerably more vulgar and... Incredibly sexual, when we aren't. (cont one more time)

@Gisikon_Root @fribbledom
He can generate sentences that take a lot of creativity to interpret, or he can be on a roll where he has a chain of messages that are sensible in context. Despite being vulgar, he often feels like a toddler. The best part is that he knows all our inside jokes and tends to bring them out in the most perfect moments. The pinned message of the chat (a honor badge for the funniest recent post) is Lärry about 70% of the time.
It's hands down the best thing I have created.

@Ayior @fribbledom Wow, thanks for the neat write up! Is the bot part (vs. the telegram integration part) difficult to put together? I'd like to give it a try as a programming learning exercise.

@Gisikon_Root @fribbledom the whole thing was super easy actually! I'm just a hobby python programmer and this was far from the most difficult thing I have done :) I spend about an entire year trying to create Lärry via an LSTM and Machine Learning until I learned about Markov Bots and did it in a couple of hours.

markov chain bots 

@fribbledom someone on one of the pleroma instances a few weeks ago was talking to a markov-chain bot and thought it was a real person.
Sign in to participate in the conversation

Server run by the main developers of the project 🐘 It is not focused on any particular niche interest - everyone is welcome as long as you follow our code of conduct!