I don’t actually mind this. Most people are low discernment, like convenience above all, and probably need prodding to be more creeped out by the idea of AI creating addictive and manipulative content for children. If you’re already thinking about your standards and boundaries, then you’re not the one who needs the alarmist message.
I don’t understand everything here but it’s a good example of the kind of reasoning about tradeoffs (eg between static and dynamic types) I want to see more of.
I’m still learning the basics about where programming “philosophies” come from and don’t really have one of my own, but one thing I believe is “you should prioritize things related to the programmer’s cognitive limitations—learnability, readability, bug catching, etc—instead of assuming that a good programmer will not screw up.” of course different approaches compensate for different *kinds* of cognitive limitations—the most bug-resistant code isn’t necessarily the most readable—but at least it’s good to be thinking about those tradeoffs, and this post does.
not really sure what to make of this one. I think the target audience is AI policy professionals who are single-issue focused on x-risk prevention, and the article takes for granted (doesn’t really explain why) that the best way to achieve x-risk reduction through policy is to ally with the more technically literate faction, as opposed to Luddites who dislike AI for its effect on culture or the labor market. I think his point is that normie Luddite policies will not suffice to stop x-risks? but he doesn’t really make it.
his analysis of why x-riskers get picked on seems correct: there are normie Luddite anti-AI factions among the Republicans, but x-riskers are a small, weak group and can safely be painted as crazed lefties.
sure, if you want to get out of this dynamic, one angle is to embrace what you have in common with the Tech Right (you both see AI as more important than anybody else does, you both understand it better, you both are all else equal in favor of economic growth through technology).
but again, he doesn’t really go into “why isn’t it better to go Full Luddite, if you really believe the fate of humanity is at stake, and if there are lots of people in both parties who hate tech already?”
one reason you might not want to do this is that it is harder than you think to speak Republican. Sriram Krishnan is explicit that what it would take for him to trust a safety advocate is for them to be a “big supporter” of “anyone on the Right.” I think this is probably true.
you won’t get any R allies, in today’s political environment, without costly signals of party/ideological loyalty. if until yesterday you were a normal coastal liberal or moderate or libertarian, your overtures to the Luddite Right will be laughed off. People can read, they can see your track record of opinions all over the Internet, they can track where you get funding, your true nature is not a secret.
therefore your choices are basically “be exactly the Democrat Luddite you’re accused of being, pray for electoral change” (depends hugely on factors beyond your control) or “quietly persuade smart and less-politically-polarized tech-aligned factions in both parties that they should care at all about safety, notice that some of them actually already do, and take the W when you can instead of picking public fights” (probably gets you more incremental gains in the near term.)
I don’t particularly have an AI policy wishlist, myself. It’s not in my top 10 most clear-cut issues. I think most of AI’s substantive economic/human-flourishing benefits are still in the future, and may be mostly squandered anyway, so if we cut off “good AI” opportunities that’s not great but it’s more survivable than losing technological (and institutional) capabilities that we’re already dependent on. On the other hand I’m not confidently x-risk-worried enough to necessarily think that we should be paying any hefty economic costs to prevent it.
If there’s an AI-related issue I have a strong opinion on, it’s “please actually develop the useful applications and the defenses against misuse”. (eg things like AI drug discovery, AI formal verification, etc, with care taken by funders in distinguishing the real thing from buzzword slop.)
links 10/20/25: https://roamresearch.com/#/app/srcpublic/page/10-20-2025
https://replacement.ai/
I don’t actually mind this. Most people are low discernment, like convenience above all, and probably need prodding to be more creeped out by the idea of AI creating addictive and manipulative content for children. If you’re already thinking about your standards and boundaries, then you’re not the one who needs the alarmist message.
https://unplannedobsolescence.com/blog/what-dynamic-typing-is-for/
I don’t understand everything here but it’s a good example of the kind of reasoning about tradeoffs (eg between static and dynamic types) I want to see more of.
I’m still learning the basics about where programming “philosophies” come from and don’t really have one of my own, but one thing I believe is “you should prioritize things related to the programmer’s cognitive limitations—learnability, readability, bug catching, etc—instead of assuming that a good programmer will not screw up.” of course different approaches compensate for different *kinds* of cognitive limitations—the most bug-resistant code isn’t necessarily the most readable—but at least it’s good to be thinking about those tradeoffs, and this post does.
https://en.wikipedia.org/wiki/Knuth%27s_up-arrow_notation
my son says his “dream” is to get a computer that can compute up-arrows.
https://drcolleensmith.substack.com/p/the-iliad-is-love an ER doc on what COVID was like. wrenching.
https://courses.aynrand.org/campus-courses/ayn-rands-conception-of-valuing/ liking this one a lot. in general Gregory Salmieri is just a really good lecturer; reminds me how much I enjoyed college philosophy classes.
https://writing.antonleicht.me/p/the-devil-you-know
not really sure what to make of this one. I think the target audience is AI policy professionals who are single-issue focused on x-risk prevention, and the article takes for granted (doesn’t really explain why) that the best way to achieve x-risk reduction through policy is to ally with the more technically literate faction, as opposed to Luddites who dislike AI for its effect on culture or the labor market. I think his point is that normie Luddite policies will not suffice to stop x-risks? but he doesn’t really make it.
his analysis of why x-riskers get picked on seems correct: there are normie Luddite anti-AI factions among the Republicans, but x-riskers are a small, weak group and can safely be painted as crazed lefties.
sure, if you want to get out of this dynamic, one angle is to embrace what you have in common with the Tech Right (you both see AI as more important than anybody else does, you both understand it better, you both are all else equal in favor of economic growth through technology).
but again, he doesn’t really go into “why isn’t it better to go Full Luddite, if you really believe the fate of humanity is at stake, and if there are lots of people in both parties who hate tech already?”
one reason you might not want to do this is that it is harder than you think to speak Republican. Sriram Krishnan is explicit that what it would take for him to trust a safety advocate is for them to be a “big supporter” of “anyone on the Right.” I think this is probably true.
you won’t get any R allies, in today’s political environment, without costly signals of party/ideological loyalty. if until yesterday you were a normal coastal liberal or moderate or libertarian, your overtures to the Luddite Right will be laughed off. People can read, they can see your track record of opinions all over the Internet, they can track where you get funding, your true nature is not a secret.
therefore your choices are basically “be exactly the Democrat Luddite you’re accused of being, pray for electoral change” (depends hugely on factors beyond your control) or “quietly persuade smart and less-politically-polarized tech-aligned factions in both parties that they should care at all about safety, notice that some of them actually already do, and take the W when you can instead of picking public fights” (probably gets you more incremental gains in the near term.)
I don’t particularly have an AI policy wishlist, myself. It’s not in my top 10 most clear-cut issues. I think most of AI’s substantive economic/human-flourishing benefits are still in the future, and may be mostly squandered anyway, so if we cut off “good AI” opportunities that’s not great but it’s more survivable than losing technological (and institutional) capabilities that we’re already dependent on. On the other hand I’m not confidently x-risk-worried enough to necessarily think that we should be paying any hefty economic costs to prevent it.
If there’s an AI-related issue I have a strong opinion on, it’s “please actually develop the useful applications and the defenses against misuse”. (eg things like AI drug discovery, AI formal verification, etc, with care taken by funders in distinguishing the real thing from buzzword slop.)