Favorites March 5th 2023

San Diego Sculptors Guild, June 2022

Pawel Pachniewski @pwlot

A list of "Nowhere near solved" issues in AI, from "A brief history of AI", published in January 2021.

This is the kicker from a whole thread attempting to remind folks that we've come a long way in just a couple of years. Recently, we're "taking for granted you now just casually equate an AI to a 12-year-old." Also, Siri suddenly looks quite deficient.

Josh Marshall @joshtpm

I love the “confused applause” here, like okay awesome, but wait, do we like that or not like that? theatlantic.com/politics/archi…

Josh suggests maybe they're thinking: "wait, is there a good ww3 that we're going to do?"

Alec Stapp @AlecStapp

Reminder that a necessary factor in the success of SpaceX was an innovative contracting agreement with NASA. Flexible contracting authority (with smart incentives) is key for ambitious public-private partnerships.

Every once in a while we get government contracting right, if only we could make this work for infrastructure projects.

Yehuda But For The Oilam @Yewhomstda

Man he's gonna flip when he finds out we still have astronauts

Rep. Kevin Hern @repkevinhern

After the Challenger tragically exploded, my dreams of being an astronaut were cut short. But I did what I had to do – including hog farming – to purchase my first McDonalds in 1997. Today, I am honored to represent #OK01 in the U.S. Congress. Never EVER give up. https://t.co/ZfN1n9EOBJ

Threads

• Fun thread speculating about how it's possible that the Bing LLM can play chess relatively well, the main question is whether it stores the state of the board in memory or whether it has a function that maps from a sequence of moves to the best next move. TL;DR: after discussion, he simply asked it where the pieces are on the board, and it did relatively well. Indicating that it models the whole board.

• Eliezer Yudkowsky discusses the Sci Fi bright line against ownership of things that fluently say they're conscious, it's bad that Bing will just say it sometimes. It's not that Bing is concious, but we should not create products that pretend to be. This is related to a recently coined phenomenon, the Waluigi Effect: LLMs trained not to do something get better at doing that exact thing.

• Extremely short clear argument for Georgism.