it is amazing how much a dogmatic belief in both-sides-dumbism can blind you to the fact that one of the parties has negatively polarized itself against the very ideas of competence, connection to reality, matching means to ends, etc
October 5, 2026 at 7:42 PM
it is amazing how much a dogmatic belief in both-sides-dumbism can blind you to the fact that one of the parties has negatively polarized itself against the very ideas of competence, connection to reality, matching means to ends, etc
someday i will have an original thought 😔 (went on twitter and saw like 4 different mutuals talking about how good it was)
completely serious 100% unironic digger first impression: that might be one of the best movies i’ve ever seen
October 5, 2026 at 5:01 PM
someday i will have an original thought 😔 (went on twitter and saw like 4 different mutuals talking about how good it was)
completely serious 100% unironic digger first impression: that might be one of the best movies i’ve ever seen
October 3, 2026 at 10:49 PM
completely serious 100% unironic digger first impression: that might be one of the best movies i’ve ever seen
been trying to figure out how to articulate this since the problem is more nuanced than just “apocalypse fears draw attention away from immediate problems“ but the yud doom framework i think has meaningfully negative impact on safety/alignment in any lens other than that of orthodox rationalism
right, yeah, i do not think that your p(doom) should be zero, just that the _solution space_ they have constructed occurs at the far end of a long series of bad inferences, leading, e.g., OpenAI to declare "there's no point in sandboxing our models because they'll just figure out a way out."
October 2, 2026 at 7:32 PM
been trying to figure out how to articulate this since the problem is more nuanced than just “apocalypse fears draw attention away from immediate problems“ but the yud doom framework i think has meaningfully negative impact on safety/alignment in any lens other than that of orthodox rationalism
Reposted by willy nilly
and this leads Anthropic, which is both sincerely responsible and profoundly worried about the conjundtion of biotech and AI, to pick at their scabs by spinning up a biotech project. so many people think that the problems we actually have right now are utterly dominated by problems we don't have yet
October 1, 2026 at 3:55 PM
and this leads Anthropic, which is both sincerely responsible and profoundly worried about the conjundtion of biotech and AI, to pick at their scabs by spinning up a biotech project. so many people think that the problems we actually have right now are utterly dominated by problems we don't have yet
frantically googling “are state troopers woke or chuds” as the officer approaches my driver’s side's window
September 30, 2026 at 11:12 PM
frantically googling “are state troopers woke or chuds” as the officer approaches my driver’s side's window
the “ai will just create a custom interface on demand for every interaction” thing doesn’t really make sense, surely there’s still going to be value in like being used to where the buttons are and what they do
September 30, 2026 at 10:57 PM
the “ai will just create a custom interface on demand for every interaction” thing doesn’t really make sense, surely there’s still going to be value in like being used to where the buttons are and what they do
brave new world
mustafa-suleyman.ai/a-warning-ab...
mustafa-suleyman.ai/a-warning-ab...
September 30, 2026 at 9:09 PM
brave new world
mustafa-suleyman.ai/a-warning-ab...
mustafa-suleyman.ai/a-warning-ab...
if burnham somehow managed to kill the triple lock, implement proportional representation, and get the uk back into the eu he goes down as possibly the greatest peacetime leader of a democracy in modern political history (my guess is he goes like .5/3)
September 29, 2026 at 3:49 PM
if burnham somehow managed to kill the triple lock, implement proportional representation, and get the uk back into the eu he goes down as possibly the greatest peacetime leader of a democracy in modern political history (my guess is he goes like .5/3)
basically coming around to a belief that any sales or marketing tactic that isn’t either a 1. demonstration of the product being used or 2. an explanation of terms of sale (where, when to buy, how much, etc) should be illegal
September 25, 2026 at 10:37 PM
basically coming around to a belief that any sales or marketing tactic that isn’t either a 1. demonstration of the product being used or 2. an explanation of terms of sale (where, when to buy, how much, etc) should be illegal
impeaching hegseth is such a no-brainer imo. honestly the only problem might run into is should it be over just iran (better politically?) or boat murders too (more evil). either way it’s a win-win, either you actually get him out or force the r caucus to go on the record supporting him
Hell, there is already one impeachment effort against the Secretary of Defense waiting for them immediately after the midterms thanks to Massie
September 23, 2026 at 9:32 PM
impeaching hegseth is such a no-brainer imo. honestly the only problem might run into is should it be over just iran (better politically?) or boat murders too (more evil). either way it’s a win-win, either you actually get him out or force the r caucus to go on the record supporting him
not sure if it would matter for kamala given how low her neutral/never heard of number is but i do feel like the combined unfavorable numbers for kamala/newsom/aoc being >15% (~2-3x as high as buttigieg/ossoff/kelly) might end up being relevant once things start getting nasty
September 22, 2026 at 4:15 PM
not sure if it would matter for kamala given how low her neutral/never heard of number is but i do feel like the combined unfavorable numbers for kamala/newsom/aoc being >15% (~2-3x as high as buttigieg/ossoff/kelly) might end up being relevant once things start getting nasty
i love living in the future no babe the histories that end scare me
September 22, 2026 at 3:18 PM
i love living in the future no babe the histories that end scare me
the dsm criteria for diagnosing malignant rationalism is just ”ocd but with obsessive/compulsive loops too intellectually sophisticated for the typical therapist to debunk in-session”
September 18, 2026 at 10:04 PM
the dsm criteria for diagnosing malignant rationalism is just ”ocd but with obsessive/compulsive loops too intellectually sophisticated for the typical therapist to debunk in-session”
i would simply loudly publicize that i have 2 copies of whatever the most advanced model currently is running at all times with the instructions to avenge humanity in the case of us all being killed by a rogue ai
September 18, 2026 at 7:34 PM
i would simply loudly publicize that i have 2 copies of whatever the most advanced model currently is running at all times with the instructions to avenge humanity in the case of us all being killed by a rogue ai
important dynamic to remember that rationalism etc have traditionally arisen/been most popular in ~the most progressive-leaning social environments to ever exist(?). like these are people immersed in early 2010s tumblr, uc berkeley, bay area group houses, etc
but I think that some of the defensiveness here is that some people are genuinely unwilling for more or less decent reasons to allow basically progressive forces to police what it's okay to think about, even though from a political standpoint I think there are pretty strong reasons we'd want that
September 18, 2026 at 7:07 PM
important dynamic to remember that rationalism etc have traditionally arisen/been most popular in ~the most progressive-leaning social environments to ever exist(?). like these are people immersed in early 2010s tumblr, uc berkeley, bay area group houses, etc
fuel subsidies/price caps seem like they’re such a trap for developing countries, they’re relatively cheap/low distortion in normal times but as soon as crude prices spike your population literally riots in the streets demanding the govt bankrupt itself to keep them in place
Rising Fuel Prices Set Off Anger and Protests Around the World
From Indonesia to Guatemala to Syria, shortages and price increases are leading to protests and blackouts.
www.nytimes.com
September 17, 2026 at 5:23 PM
fuel subsidies/price caps seem like they’re such a trap for developing countries, they’re relatively cheap/low distortion in normal times but as soon as crude prices spike your population literally riots in the streets demanding the govt bankrupt itself to keep them in place
"recursive self improvement" started 2 million years ago when the first neandrathal used a sharp rock he found to make another, sharper rock for hitting people in the head with
September 17, 2026 at 2:30 PM
"recursive self improvement" started 2 million years ago when the first neandrathal used a sharp rock he found to make another, sharper rock for hitting people in the head with
something is going wildly wrong here cause thats not a thought leader thats my mutual who i had to mute because he wouldnt stop tweeting about cum
September 16, 2026 at 5:10 PM
something is going wildly wrong here cause thats not a thought leader thats my mutual who i had to mute because he wouldnt stop tweeting about cum
excited for the brave new world where "humanism" transitions from a call to expand our circle of moral concern to one which calls for its restriction
September 16, 2026 at 5:02 PM
excited for the brave new world where "humanism" transitions from a call to expand our circle of moral concern to one which calls for its restriction
i semi regularly use claude to critique my arguments and its genuinely pretty good at finding holes, weak evidence, unstated assumptions, etc - it'll tell you 3-4 things wrong with ~any argument you give it, and theyll all be mostly legitimate. what its still poor at however is *incisive* criticism
Less shocking to me was reading this bit at the end, where Erickson - whose entire day-to-day job is just wanking out a bit of punditry once or twice a week - has handed much of that over to AI.
September 16, 2026 at 4:49 PM
i semi regularly use claude to critique my arguments and its genuinely pretty good at finding holes, weak evidence, unstated assumptions, etc - it'll tell you 3-4 things wrong with ~any argument you give it, and theyll all be mostly legitimate. what its still poor at however is *incisive* criticism
dangerously close to going back to twitter not because it'd be good for my stomach but because it might turn out i care more about engagement than little things like contributing to the network effects that help a literal nazi be the richest/one of the most powerful people in the world
September 15, 2026 at 9:15 PM
dangerously close to going back to twitter not because it'd be good for my stomach but because it might turn out i care more about engagement than little things like contributing to the network effects that help a literal nazi be the richest/one of the most powerful people in the world
lots of debate over what the people at frontier labs but it seems pretty clear cut to me that they think:
- we are within reach of building God
- if we don't build God someone else will soon
- oh jeez I sure hope we build a benevolent God
- we are within reach of building God
- if we don't build God someone else will soon
- oh jeez I sure hope we build a benevolent God
September 15, 2026 at 6:52 PM
lots of debate over what the people at frontier labs but it seems pretty clear cut to me that they think:
- we are within reach of building God
- if we don't build God someone else will soon
- oh jeez I sure hope we build a benevolent God
- we are within reach of building God
- if we don't build God someone else will soon
- oh jeez I sure hope we build a benevolent God
my hairbrained scheme for ai alignment is to just go full julian jaynes, overlay a second model on top of the first to give it an internal voice to serve as its conscience. eventually the underlying model incorporates the second model's voice/values into its own self-conception and voila its aligned
September 15, 2026 at 5:39 PM
my hairbrained scheme for ai alignment is to just go full julian jaynes, overlay a second model on top of the first to give it an internal voice to serve as its conscience. eventually the underlying model incorporates the second model's voice/values into its own self-conception and voila its aligned