[johnfn]

I think this is fairly easily debunked by o1, which is basically just 4o in a thinking for loop, and performs better on difficult tasks. Not a LOT better, mind you, but better enough to be measurable.