16 September 2026

Dr. Robert Wachter of UCSF wants to slow down AI, but not in healthcare. Slowing down AI is like going on vacation. You may get a respite, but the rest of the world presses on. China will not slow down. Neither will the rest of the world. AI doesn't need to slow down. Humans need to speed up. That includes doctors, because AI engineers will roll over them and end up telling them how medicine will be run.

And Everyone Says Datacenter Moratoriums Are Killing the US Buildout. We disagree. Yeah, they're still being built. Datacenters are far less intrusive than those windmills that blight the Columbia Gorge, ruining the natural scene, and turning it into some Simon Stålenhag feverdream. Once they MNR technology is perfected, the dataceters will be self-sufficient, and the main complaints about them will be moot. 

––– 良くない–––

I agree. Good grief, The Atlantic has yet another math and AI piece: Math Can’t Go On Like This. Yes, mathematicians will have to find out how to adapt. EVen Science published this: OpenAI breakthrough triggers ‘existential crisis’ in math. Mathematicians will need to learn how to use LEAN and use modern AI models. Learn and adapt, or die. This isn't theoretical, like it still is for medicine.

Sam Altman thinks that GPT-5.5 is as good as an average math professor. It just has a good working memory of math. But I don't see any models taking the place of real professors.  Check this out:

––– 良くない–––

Here's a handy table to read off when the latest LLMs' training cutoff dates are. The latest ones are accessing online search before they spit out an answer, so these dates will mean less. But there is research that shows that many times, they will answer questions from memory (using their trained weights) without checking first. And you won't know unless you demand that they check before they answer.

––– 凄い –––

AIX Global and Seed IQ claims to have used quantum computing to solve all six of the remaining Millennium Problems. It doesn't seem that anyone is taking them seriously, though.

––– 良くない–––

Want to know how to get those awesome visuals?  The amazing videos or the impressive images? Here is a collection of prompts that were used to get those. Study them well.

––– 凄い –––

Here is Dario Amodei's handpicked staff that will save the world from AI. Not quite as bad as Biden administration staff but uncomfortably close. And read this:
Since its founding, METR and its affiliates have received funding from EA kingpins like Democratic megadonor Dustin Moskovitz and Skype cofounder Jaan Tallinn — both of whom are Anthropic investors.
Tallinn again! His involvement definitely scares me.

––– 良くない–––

Reddit Is Still Useful. It's Also Becoming Impossible. Yes, there are still some bits of Reddit that are worth following, but it's slowly becoming less convenient to use. And convenience trumps quality.

––– 良くない–––

Blue states with strong teachers’ unions have lower student literacy levels: study shows. Ho, ho! The folks in Oregon know that too well, don't they? This is from a WSJ study. Unions ruin so much.

––– 良くない–––

Calliptic is like a Star Trek universal translator. This reminds me of the Hearing Helper appthat lets you communicate with people who are hard of hearing.

––– 凄い –––

Knight family gives $1.1 billion to Providence St. Vincent. And they could use a boost like this. Phil Knight is an amazing guy.

––– 凄い –––