Wow Boston Dynamics using RL with Atlas! For decades they've favored simple engineering approach. Just last year Gabe Nelson demo'ed bmx riding by (I believe) simple principles, no mention of RL. In fact, no mention of AI at all. 1/ www.youtube.com/watch?v=I44_...

Went down a rabbit hole watching nice videos about myths about wing lift. Best is Holger Babinsky's masterful lecture www.youtube.com/watch?v=QY2p... . Veritassium's 1 min video www.youtube.com/watch?v=aFO4... 12 yrs ago is entertaining but not right; I suspect he'd do it differently now. 1/

Gotta love Mark Rober's LIDAR video. Tesla bit is genius. Make the abstract visceral & entertaining w/ the kid dummy & wil-e-coyote wall. Avoids pontificating, let's you figure out what to think. Interestingly, seems ppl still up in arms about it.

As a sign that aging beats the joy out of life, I saw a big jigsaw puzzle in a shop display window & immediately thought not interesting. I did occasionally enjoy them with family as a kid, my sister still likes them. What happened?

Finally Turing award for Barton & Sutton. Kinda overdue, but amazing to think AlexNet was only 11 yrs ago & Bengio-Hinton-LeCun award was 2018. I would've followed that with a Turing for RL within a couple years, but then again I don't follow who has won since. Recognition well deserved.

I was surprised that deep into 21st century, still had a reviewer complain about p-values. People who "correct" your frequentist stats generally know only recipes & don't understand the purpose of statistics.

Like many, I occasionally digitize data from a published figure to re-plot. Open source web apps work well for analog, but now I can do much better for vector PDFs. Save fig as (readable) SVG, parse manually to see how data is arranged, and explain to LLM helper. Ask for a Python script to...