ThoughtStorms Wiki

People in KnowledgeGraph / PersonalKnowledgeManagement space are discussing whether LanguageModels / VectorDatabase can replace explicit links.

The issue is that vector similarity can discover similar pages. But most algorithmic similarity is unsurprising, therefore low information (and "low value"?)

OTOH, more surprising links carry more information and value. By this principle BonusFreakyConnectionPoints and WarpLinks are the highest value of all. And of course these really depend on the idiosyncrasies of the individual author making them explicitly.

RobHaisfield noted that maybe "long jumps" for third party viewers represent closer connections in the model of the author which sounds plausible to me.

I replied

OK. This sounds plausible.

Long jumps are visible to me, the author, but not a stereotypical reader or algorithm because in my personal idiosyncratic world-model things are already nearer.

This idiosyncratic folding of idea space is the value each individual brings.

No Backlinks