A post likens Markov’s 1913 letter counts to training a language model
The post says Andrey Markov tallied 20,000 letters from a novel and calculated how likely vowels and consonants were to follow one another.
TLDR
Calling Markov’s work the “first LLM,” the post likens his 1913 calculations to “training” a bigram by hand. It describes counting 20,000 letters from a novel to calculate four probabilities: vowel after vowel, consonant after vowel, vowel after consonant and consonant after consonant.
Combined views
13.7K
2 Sources, first seen 18d ago