waning gibbous sampler
@waningibbous.bsky.social
economist in tech. opinions my own.
this is one of those bell-curve cases where "how did they do relative to the top of the ticket" is a reasonable-enough metric, if you're interested in that sort of thing, and infinitely more robust than a bespoke "WAR" model that doesn't even know it's trying to estimate something counterfactual
December 21, 2025 at 2:18 PM
this is one of those bell-curve cases where "how did they do relative to the top of the ticket" is a reasonable-enough metric, if you're interested in that sort of thing, and infinitely more robust than a bespoke "WAR" model that doesn't even know it's trying to estimate something counterfactual
I mean, I would argue that, whether or not it exists as a platonic quality, if you can't show that it helps with *some* kind of prediction, you can't really claim to have measured it.
December 21, 2025 at 3:19 AM
I mean, I would argue that, whether or not it exists as a platonic quality, if you can't show that it helps with *some* kind of prediction, you can't really claim to have measured it.
yeah, I mean, causal inference is hard! and it's totally plausible that there's just not a well-identified solution here... but controlling for fundraising amounts, and tbh even opponent quality, strikes me as conditioning on the dependent variable.
December 21, 2025 at 3:10 AM
yeah, I mean, causal inference is hard! and it's totally plausible that there's just not a well-identified solution here... but controlling for fundraising amounts, and tbh even opponent quality, strikes me as conditioning on the dependent variable.
"Wins above replacement" is literally defined as a counterfactual difference... I don't think there's any way to meaningfully define "candidate quality" without appealing to counterfactuals, and I don't know how you can exclude fundraising ability from a measure of quality
December 21, 2025 at 2:54 AM
"Wins above replacement" is literally defined as a counterfactual difference... I don't think there's any way to meaningfully define "candidate quality" without appealing to counterfactuals, and I don't know how you can exclude fundraising ability from a measure of quality
I'm sure that it helps predict outcome... but how do you know it helps accurately estimate the counterfactual of "drop in a randomly selected replacement and see how they do by comparison"?
December 21, 2025 at 2:46 AM
I'm sure that it helps predict outcome... but how do you know it helps accurately estimate the counterfactual of "drop in a randomly selected replacement and see how they do by comparison"?
yeah, but ability to raise money (and deter opponents capable of raising their own money) is a core component of candidate quality, obviously not random / exogenous
December 21, 2025 at 2:37 AM
yeah, but ability to raise money (and deter opponents capable of raising their own money) is a core component of candidate quality, obviously not random / exogenous
I am a professional data scientist, and conditioning out fundraising when trying to estimate "candidate quality" is completely fucking insane
December 21, 2025 at 2:13 AM
I am a professional data scientist, and conditioning out fundraising when trying to estimate "candidate quality" is completely fucking insane
yeah, I think "developer" is just too big a tent. If you're used to thinking about the emergent behavior of automated, but non-deterministic, systems you'll be fine. if you majored in leetcode, and think programming is about memorizing the react api, you'll probably have a bad time.
December 18, 2025 at 2:45 PM
yeah, I think "developer" is just too big a tent. If you're used to thinking about the emergent behavior of automated, but non-deterministic, systems you'll be fine. if you majored in leetcode, and think programming is about memorizing the react api, you'll probably have a bad time.
IMO successfully operating an LLM is at least programming adjacent, bc it relies on the similar skill of being able to role-play how the algorithm / model sees + interacts with the world.
December 18, 2025 at 2:21 PM
IMO successfully operating an LLM is at least programming adjacent, bc it relies on the similar skill of being able to role-play how the algorithm / model sees + interacts with the world.
"the American people are fundamentally misogynist, and female presidential candidates will always be at a disadvantage" is both highly actionable and amply supported by the evidence, but no one wants to admit it because it's way too depressing.
December 17, 2025 at 4:26 PM
"the American people are fundamentally misogynist, and female presidential candidates will always be at a disadvantage" is both highly actionable and amply supported by the evidence, but no one wants to admit it because it's way too depressing.
The Knives Out series is the art that best captures the contours of Trumpism, and it's not particularly close.
December 16, 2025 at 4:19 AM
The Knives Out series is the art that best captures the contours of Trumpism, and it's not particularly close.
In fact, if people care about having a job that "feels" like it matters, you should *expect* a negative correlation between subjective perceptions of social value and wage!
December 10, 2025 at 6:14 PM
In fact, if people care about having a job that "feels" like it matters, you should *expect* a negative correlation between subjective perceptions of social value and wage!
McCain was, pretty straightforwardly, not a citizen when he was born!
repository.law.umich.edu/cgi/viewcont...
repository.law.umich.edu/cgi/viewcont...
repository.law.umich.edu
December 6, 2025 at 10:58 PM
McCain was, pretty straightforwardly, not a citizen when he was born!
repository.law.umich.edu/cgi/viewcont...
repository.law.umich.edu/cgi/viewcont...
Out of curiosity, have you read this? I thought it was a pretty great "hard"-er exploration of the hive mind idea that dovetails nicely with Pluribus.
www.lightspeedmagazine.com/fiction/the-...
www.lightspeedmagazine.com/fiction/the-...
The Twenty-One Second God - Lightspeed Magazine
We lost so many souls today. Reports stream in across five continents; icons bloom on the map like blood spatters. Broken filters, zero latency, bandwidth that somehow blew through the roof when no on...
www.lightspeedmagazine.com
December 6, 2025 at 6:38 PM
Out of curiosity, have you read this? I thought it was a pretty great "hard"-er exploration of the hive mind idea that dovetails nicely with Pluribus.
www.lightspeedmagazine.com/fiction/the-...
www.lightspeedmagazine.com/fiction/the-...
It's so frustrating that the Biden admin bought into the idea that it's immoral to weaponize a 70-30 issue against your political opponents, because if you do some of them might choose to kill themselves
December 6, 2025 at 3:21 PM
It's so frustrating that the Biden admin bought into the idea that it's immoral to weaponize a 70-30 issue against your political opponents, because if you do some of them might choose to kill themselves
are there any realistic contenders other than amzn for end-state owner of anthropic?
November 30, 2025 at 12:35 AM
are there any realistic contenders other than amzn for end-state owner of anthropic?
coding is also uniquely amenable to structured self-play: you can write some tests, mechanically evaluate whether generated code passes those tests, and keep training until it does.
unsurprisingly, llm performance rapidly degrades once you ask it to write code without guardrails / guidance.
unsurprisingly, llm performance rapidly degrades once you ask it to write code without guardrails / guidance.
November 28, 2025 at 2:16 PM
coding is also uniquely amenable to structured self-play: you can write some tests, mechanically evaluate whether generated code passes those tests, and keep training until it does.
unsurprisingly, llm performance rapidly degrades once you ask it to write code without guardrails / guidance.
unsurprisingly, llm performance rapidly degrades once you ask it to write code without guardrails / guidance.
you also wonder if there was more of a lingering influence from whatever cultural package encouraged the clearing-out of all the other h sap
November 28, 2025 at 1:00 AM
you also wonder if there was more of a lingering influence from whatever cultural package encouraged the clearing-out of all the other h sap
as in, the early Neolithic had less cultural fragmentation that could act as a firebreak?
November 28, 2025 at 12:12 AM
as in, the early Neolithic had less cultural fragmentation that could act as a firebreak?
I think the relevant question is whether the National Review would have done a similarly glossy cover piece for Kathleen Sebelius
November 26, 2025 at 12:43 AM
I think the relevant question is whether the National Review would have done a similarly glossy cover piece for Kathleen Sebelius
yeah but you and I are, uh, not teenagers, and it only takes one.
I've been fairly amused that many of the AI -slop Sketchers ads adjacent to the Friend ads have started getting defaced as collateral damage---to me that speaks to genuine annoyance / momentum.
I've been fairly amused that many of the AI -slop Sketchers ads adjacent to the Friend ads have started getting defaced as collateral damage---to me that speaks to genuine annoyance / momentum.
November 24, 2025 at 12:48 AM
yeah but you and I are, uh, not teenagers, and it only takes one.
I've been fairly amused that many of the AI -slop Sketchers ads adjacent to the Friend ads have started getting defaced as collateral damage---to me that speaks to genuine annoyance / momentum.
I've been fairly amused that many of the AI -slop Sketchers ads adjacent to the Friend ads have started getting defaced as collateral damage---to me that speaks to genuine annoyance / momentum.
I'm sure it was part of the strategy, but there's an irl ubiquity to these ads that isn't matched by anything else I've seen---why is there one at the 96th & CPW B/C stop???
November 24, 2025 at 12:40 AM
I'm sure it was part of the strategy, but there's an irl ubiquity to these ads that isn't matched by anything else I've seen---why is there one at the 96th & CPW B/C stop???
ok, yeah, that is all consistent with my baseline prior of not really trusting Hersh---thanks for the color!
November 23, 2025 at 4:08 AM
ok, yeah, that is all consistent with my baseline prior of not really trusting Hersh---thanks for the color!
www.lrb.co.uk/the-paper/v3...
this is the most complete treatment I've seen. I have no idea what to think, would be very interested in your perspective.
this is the most complete treatment I've seen. I have no idea what to think, would be very interested in your perspective.
Seymour M. Hersh · The Killing of Osama bin Laden
Would bin Laden, target of a massive international manhunt, really decide that a resort town forty miles from Islamabad...
www.lrb.co.uk
November 23, 2025 at 2:56 AM
www.lrb.co.uk/the-paper/v3...
this is the most complete treatment I've seen. I have no idea what to think, would be very interested in your perspective.
this is the most complete treatment I've seen. I have no idea what to think, would be very interested in your perspective.
given that this is literally put out by the farm lobby, why would our prior not be that the results have been manipulated to make Trump / Rs look good?
November 21, 2025 at 5:23 PM
given that this is literally put out by the farm lobby, why would our prior not be that the results have been manipulated to make Trump / Rs look good?