What if the AI capex bubble may not be a bubble after all?

"knowledge is the log of compute" is a weird sentence.

Knowledge is the collection and storage of information. The _what_. Understanding is the deep comprehension of knowledge. the _why_.

So how then can knowledge be the log of compute? It's simply the amount of data you can steal and store on your disk-arrays.

The models should generate _understanding_ and i think that is what people expect.

What they do I think is actually _consilience_. I would expect them to be good at that, since its essentially pattern-matching. The models can also use existing knowledge items and apply them more consistently across a field or to more complete picture, more datasets etc.

Ofc i am not super educated on LLM/AI stuff, but it kind of 'feels' (sorry) like that is how it is.