Real analysts (rather than hagiographers and stenographers) might be interested in looking at what AI execs promised their *products* would be able to do in 2025, and checking that against what was actually accomplished.
I emphasize products because a model doesn't do anything alone.