This helps explain why it's so hard to automate farm labor!

It's not that it's too hard to make a robot pick crops.

It's that humans are really, REALLY good at it. It's hard to make a robot that's BETTER at it than people.

Cursor is, in my experience, much better at creating test code than production code. It gets me wondering if refactoring production code might get easier because we can have higher confidence that the new code works the same as the old.

I wanted to respond to an issue my friend Jared Bernstein
raised about my conclusion in my recent @foreignaffairs.com piece that real wages are below trend.

My conclusion is VERY robust to just about every major wage measure and trend.

I will expand & explain in this 🧵.

I have a new piece in @ForeignAffairs titled, "The Post-Neoliberal Delusion and the Tragedy of Bidenomics". They were generous about giving me a lot of words but were less generous with charts--so this long thread partially rectifies that. (And links to the piece at the end.)

In 2024 I was lucky to collaborate on six research papers exploring topics ranging from the human tendency to fixate on numeric information to why we're better at pursuing goals in tandem with friends. Here's a short 🧵on the papers I had the good fortune to co-author in '24...

@fchollet.bsky.social describes o3 as deep learning guided program search to solve the input problems (finding the rules that create the output from the input).
Could o3 apply the same technique to those NL programs, essentially asking "how could somebody have come up with these NL programs?"

I propose we replace semantic versioning with pride versioning

Really interesting post that summarizes a lot of debate about AI