"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.
"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.