You did it!
#2
by BingoBird - opened
I thought of this too and lo and behold, you did it!
We should be able to 'see a shape' to activated HL concepts from different viewpoints.
For example, when considering originary interest rate explanations, one high-level representation might be from a Monetarist perspective, another from a Keynesian etc.
Unfortunately some 'perspectives' are actually incoherent (nonsense). And people believe them. And they're in the training data.
Maybe this is the block to AGI, all the lies we've told each other...