![Twitter avatar for @IDoTheThinking](https://substackcdn.com/image/twitter_name/w_96/IDoTheThinking.jpg)
![Twitter avatar for @1finaleffort](https://substackcdn.com/image/twitter_name/w_40/1finaleffort.jpg)
This is from a summary of effects of Aukland's massive upzoning 6 years ago. Check out the whole thread, it has amazing graphs. They did it all without the complex value capture system that local groups in California insist upon.
![Twitter avatar for @FreckleEars](https://substackcdn.com/image/twitter_name/w_96/FreckleEars.jpg)
![Image](https://substackcdn.com/image/fetch/w_600,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fmedia%2FFooU2iCXEAQOIKL.jpg)
This is a hard graphic to read, but suffice to say, an M1 Abrams tank has better visibility for a small child than a Chevy Silverado. Sorta wild what we consider safe on US streets.
![Twitter avatar for @TheOmniZaddy](https://substackcdn.com/image/twitter_name/w_96/TheOmniZaddy.jpg)
![Twitter avatar for @JustinTrudeau](https://substackcdn.com/image/twitter_name/w_40/JustinTrudeau.jpg)
Two unidentified flying objects shot down in 24 hours. Nate Silver jokes that it's "definitely not aliens, but if you were making a movie where it was aliens having the Chinese weather balloon thing happen a few days earlier would be a great head-fake."
![Twitter avatar for @WarrenJWells](https://substackcdn.com/image/twitter_name/w_96/WarrenJWells.jpg)
![Image](https://substackcdn.com/image/fetch/w_600,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fmedia%2FFouGx1LagAARJ0C.jpg)
![Twitter avatar for @StreetsblogLA](https://substackcdn.com/image/twitter_name/w_40/StreetsblogLA.jpg)
So the easy to find stat of 40k traffic deaths per year doesn't include this 2k deaths per year from storefront crashes. Plus anything else that happens in a driveway is still unaccounted for as well. Oh, and for some reason 7-11 is liable if bystanders are hit?
![Twitter avatar for @erikbryn](https://substackcdn.com/image/twitter_name/w_96/erikbryn.jpg)
So, first of all, apparently there are Theory-of-mind tests, and no LLMs could pass them pre-2022. In January 2022 GPT3 achieved a score on par with a 7-year-old, and by December could perform as well as a 9-year-old. This is without the model being trained for this ability at all. On the subject, here's a thread summarizing Ted Chiang’s New Yorker piece "ChatGPT Is a Blurry JPEG of the Web". Pretty good for a compression artifact.
![Twitter avatar for @jackclarkSF](https://substackcdn.com/image/twitter_name/w_96/jackclarkSF.jpg)
The three examples he gives are bonkers: 1) You can use Large Language Models (LLMs) to generate lots of text, and then use it as additional training data for a new model, and the results *improve*. 2) You can give LLMs access to deterministic tools (like a calculator) to improve their abilities. 3) you can add in a second language model that trains the first model using reinforcement.
![Twitter avatar for @culturaltutor](https://substackcdn.com/image/twitter_name/w_96/culturaltutor.jpg)
![Image](https://substackcdn.com/image/fetch/w_600,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fpbs.substack.com%2Fmedia%2FFoqwWpmWAAEpN1G.jpg)
Threads
• Cute thread of portraits that famous photographers took of their partners. I'm somewhat surprised how many famous photographers have famous spouses.
• A thread by a journalist from The Economist who has come around to the view that America may have turned out to have a better approach than the EU to regulating big corporations.