1 comment

[ 2.8 ms ] story [ 22.7 ms ] thread
Quite good to understand the theory behind LLMs. Latest version always seems to be at arxiv.