The conclusion: "One interesting idea is that, at their core, AI models are nothing more than compression models that take the corpus of data humanity has and b

The conclusion: "One interesting idea is that, at their core, AI models are nothing more than compression models that take the corpus of data humanity has and boil it down to a set of weights" is also good, there is a interesting paper about compressing weather date with neuronal networks: https://arxiv.org/abs/2210.12538 https://arxiv.org/abs/2210.12538 What's missing a bit is that the comparison is more for general purpose data, there are some very interesting and super fast compressing algorithms for e.g. numbers (Turbopforc, gorilla, etc...) Daniel Lemires blog is super interesting about the different algorithms and how to make them faster.