taste, trust, and rare data
- taste: your unique definition of good - trust: when you don't have time to specify your taste... who you trust to specify taste for you - rare data: stuff that a model can't know based on public knowledge
(taste and trust are also special cases of rare data)
Users dismissed Andrew Trask defining taste, trust, and rare data as AI moats as an unnecessary overcomplication of basic recommendation algorithms.
TLDR: taste is not the only moat... rare data is
taste, trust, and rare data
- taste: your unique definition of good - trust: when you don't have time to specify your taste... who you trust to specify taste for you - rare data: stuff that a model can't know based on public knowledge
(taste and trust are also special cases of rare data)
@iamtrask Generalizing a bit – the 3 elements above = context (which at "low" levels and "narrow" scopes we call "local") – as generator/owner of meaning...
@iamtrask Another day, another attempt to overcomplicate basic recommendation algorithms.
@iamtrask rare data is the only one that doesnt rely on another person
unless ur model has a favorite curator
@iamtrask so rare data is basically paying for signal decay that algorithms skip? feels like thats the hard part
@iamtrask the rare data part always gets me
who actually has it vs who says they do is the real filter?
@iamtrask seeing it spelled out like this makes me wonder if rare data is just taste you cant explain yet
taste, trust, and rare data
- taste: your unique definition of good - trust: when you don't have time to specify your taste... who you trust to specify taste for you - rare data: stuff that a model can't know based on public knowledge
(taste and trust are also special cases of rare data)