Using a Markov chain to generate readable nonsense with 20 lines of Python

Posted by signa11 on Sunday November 12, 2023 @12:55AM
from the benhoyt.com dept.

https://benhoyt.com/writings/markov-chain/

( 5 comments | 2 points | on HN )

Re:Using a Markov chain to generate readable nonsense with 2... by jruohonen on Sunday November 12, 2023 @12:59AM (#38237723)

A good reminder that you don't need LLMs for simple nonsense.

[ Parent | on HN ]

Re:Using a Markov chain to generate readable nonsense with 2... by tabitech on Sunday November 12, 2023 @02:03AM (#38238036)

I was just writing about this and possibly using it against Sybil attack, and even using it for banking systems, social media, voting... and so on.

[ Parent | on HN ]

Re:Using a Markov chain to generate readable nonsense with 2... by beardyw on Sunday November 12, 2023 @03:21AM (#38238390)

This is the grandfather of AI is it not?

[ Parent | on HN ]

Re:Using a Markov chain to generate readable nonsense with 2... by andy99 on Sunday November 12, 2023 @06:25AM (#38239299)

How would you calculate or estimate a "parameter count" from the "trained" model? I'd like to be able to compare to small transformer models e.g. some of the very small llama or gpts. Would it just be the number of (w1,w2) pairs that have entries... possible times the total vocabulary size, but I think that would be unfairly sparse. So maybe times the number of unique entries for each pair?

[ Parent | on HN ]

Re:Using a Markov chain to generate readable nonsense with 2... by fluxent on Sunday November 12, 2023 @04:02PM (#38244047)

I built an inside-out Markov Chain to create outward from key/starting phrase. http://webseitz.fluxent.com/wiki/InsideOutMarkovChain

[ Parent | on HN ]