e/acc

Are vector spaces in LLMs a new medium for representing information?

Information was first represented on paper, then analog signals, then bits. Each medium let us do fundamentally new things with information

Can we think of AI this way? Has anyone pursued this line of thought?

Founder, The Roots of Progress (rootsofprogress.org). I write about the history of technology and the philosophy of progress. Former tech founder.

It is similar to how we represent in our brain, but yes. 

Neural nets will also come up with much better encryption algos that are more specific to each medium/format