> that building general-purpose compression/intelligence is hard.

Yes, but my point is. that we seem to be doing better at general-purpose intelligence than at compression despite the apparent equivalence of progress.