Skip to content

Comment on CauseNet: Towards a causality graph extracted from the webparent

Comments

You're losing interpretability and scrutability, but gaining detail and expressiveness. You have no way to estimate the vectors in a causal framework, all known methods are correlational. You have no clean way to map the vectors to human concepts. Vectors are themselves extremely compressed representations, there is no clear threshold beyond which a representation becomes "proper".

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.