Favorites: Minecraft, Medicine, and Minerals
October 19, 2025 — GLP-1s hit mainstream adoption, cabinet small talk reveals big regrets, AI revenue expectations
This is from February, but it’s new to me
Probably reduces evaporation in a way that’s useful if you cover a reservoir like this.
I sorta wonder why they need physical sim cards for this, but still, wild. Here’s the story from abc.
Here’s the original Simpsons bit in question if you’re unfamiliar.
Despite the whole thing nominally being about healthcare spending, the end to the filibuster is the real thing being fought over. Democratic leadership isn’t super clear that this is their goal (cached), but I’m not sure they’d say so if it was.
Longer Reads
• This thread summarizes a Google project to get an AI to play minecraft without it needing to do training runs of actual Minecraft. Instead it learns what to do in training by imagining gameplay, and then shows the best performance on minecraft yet when run ‘for real’. This has significant implications for robotics use cases where it’s not possible to collect tons of real-world training examples. paper, website (src)(cached)
• Great thread on the mechanical physics of screws. (src)(cached)
• Interesting thread about how the Trump admin’s policies around primary care will likely separate the AMA from decisions around how medicare funds are attributed to doctors. There’s a shortage of primary care physicians, and this may help with the shortage. (src)(cached)
• A paper about labor economics and substitutability of skills between non-college and college grads. College isn’t the only way people improve their skills. Here’s more discussion (cached) (src)(cached)
• Thread from an AI safety researcher who is opening up about how OpenAI has been harassing him. (src)(cached)
Flotsam and Jetsam
– Something like a quarter of Americans have at least tried a GLP-1 drug. (src)(cached)
– Someone got an over-the-shoulder shot of conversation between the Secretaries of Treasury and Agriculture. They pretty clearly can tell privately that they’ve made some extremely stupid mistakes. (src)(cached)
– Current math is that AI-related revenue needs to go up by about 20x in the next few years to justify the level of investment we’re seeing.Certainly possible, but there are ways in which AI could be a success without this happening. (Open models perhaps?) (src)(cached)
– Was the recent rain in CA a blessing in return for passing SB 79? (src)(cached)
– In the fight over rare earth minerals (which aren’t actually that rare) it’s important that china doesn’t even dominate mining of the materials, it’s refining that they absolutely dominate at. (src)(cached)
– I’m close to subscribing to The Bulwark just to read the full text of this piece dunking on Bari Weiss. (src)(cached)
– Here’s a quote from Andrej Karpathy from his interview on the Dwarkesh Patel podcast explaining why he doesn’t expect a fast-takeoff scenario where AI becomes suddenly self improving. (src)(cached)