Value plus discounted

Logging off isn't a trembling hand equilibrium, unfortunately

socially optimal solution to the posting problem is trivial

my only disagreement is in calling this parochialism. if you'll spend most of your life in a country with more states than Europe has countries and more people in some states than in "big" European countries, it's An Ask to expect you to follow what's happening in your own country + others

Value plus discounted

it's also silly because humans solve for fixed points of stochastic dynamic programming problems every day

one could say he's right in Thierry but wrong in practice

Britpop Damon Hunters

Value plus discounted

unambiguously good and based

Recently accepted by #QJE, “Marginal Returns to Public Universities,” by Jack Mountjoy: doi.org/10.1093/qje/...

So I want to say I am biased, I know the author, and even tried to get him to fund some TRL1 tech I had developed, and his thesis advisor is mine so we are academic siblings, but I think discussions of learning rates missies something that Eric does a good job of proofing.

my current perhaps-crank theory is that learning curves are mostly about the people in the industry, and so their variability is largely about relevant labor market conditions #linklog

I will always love this blog post. You, yes you, have a responsibility to the world not to respond to bad incentives

every claim that "the incentives" support or deter certain kinds of behavior is also a statement about what kinds of external signals the claimant views as rewards or penalties #linklog

It’s easy to point fingers now, but those were different times, and if that *had* been a missile, well, things would look more different still.

Time to update my feed reader. Who else/what other blogs/sites/etc should I add and why?

I don't know that anyone has ever made a very good UX for sampling from high dimensional joint distributions, but etiquettes can work well in their native equilibria #linklog

Value plus discounted

Value plus discounted

between this and a potential IPO: space data centers will probably be a thing over the next 5 years. the open question is whether they'll still be a thing over the next 10 years

AI is a social science, exhibit __

existence of luke combs implies the existence of luke combover

existence of luke combs and luke unkempt implies existence of harry styles

existence of luke combs implies existence of luke unkempt

The spite must flow

hades 2 is pretty fun so far. melinoe actually plays like a different character than zagreus

Value plus discounted

Value plus discounted

From @harper.lol ? harper.blog/2025/04/10/w...

could this be the start of a regular posting pattern? maybe, maybe not, who's to say

[new post] some people love to think "at the margins" (hi)

"Across the highest adult performance, peak performance is negatively correlated with early performance" Is uite literally what you'd see if coliderbias:

does anyone actually prefer the agent editor? am i just old?

Value plus discounted

there are so many good points in this thread but this is probably the best and most important to internalize *cats

I firmly believe that effective LLM leveraging is a skill and we don't even know what being good at it—what being "presumed able to solve solvable problems"—looks like. being definitive is dangerous. I'm bummed I can't find this article on the internet, but I'm gonna keep looking

"a reduced form correlation should hold up in all future periods if it can be plotted in a timeseries" is, when you think about it, an odd kind of structural model

Recurring donors not only give more now, they give more later. As far as ten years into the future, an extra dollar of recurring giving is predictive of substantially more giving than non-recurring gifts. The present value of a recurring donor is ~2.5x a traditional donor per dollar. 9/

I should really get better about posting these regularly. But they also tell me variable interval reinforcement is more effective, so who's to say

Value plus discounted

Value plus discounted

And here it is.

existence of causal inference implies existence of causal training

Value plus discounted

i see the appeal of yolo mode

the real engagement algorithm was the parasocial relationships we formed along the way

influence rly isnt abt how many people your content directly reaches its abt how successful ur brainworms are at surviving long enough in new hosts to reproduce and keep the ball rolling

lol @jamditis.bsky.social with a plea "I’ve already counted at least nine Nieman predictions this year that talk about AI. I think we’d see a much more substantive transformation in journalism if we started by simply learning how to use our computers." www.niemanlab.org/2025/12/plea...