Love.
Throw a football at a wide receiver running in chaotic patterns while avoiding oncoming attackers.
Use chopsticks to pick up a single grain of rice.
Olympic level gymnastics via neural-control of muscle fibers in an extremely precise signaling orchestration.
You may say the above are robotics and not “the model can scale to this” domain but I’d counter an animal can do amazing things exactly because of its generalizable and plastic neural architecture and that is what’s in scope.
Shoot I can’t do those things either!
I’ll grant you love, and I’ll also grant you 2030 on these, especially the last one. I suspect the football and chopsticks ones are very achievable by 2030 if anyone cared enough to focus on them.
Have you seen the progress in robotics lately? They aren't yet at a level you described, but quite close - so I'd say it would be possible in 3 years even without intelligence explosion. (E.g. if all frontier AI research stops we'll still have improvements in robotics.)
There are already Unitree robots which are much closer to Olympic level than an ordinary dude: https://www.youtube.com/watch?v=Y3zb4o47yqA You might argue but it's kinda at a level you described - I think if Unitree wanted to make a football-thrower they could just do it now.