You Only Launch Once

I still think it's a mistake to assume tasks can be meaningfully assigned generalizable cardinal difficulty levels. the space may be jagged but imo our metric is also not well behaved

switching between data and data just like me fr fr

"You should be confident that the papers you are citing are correct" but this is why I don't like citing myself :(

I think this hits the nail on the head about using current AI for knowledge work with stakes. the explanations generally suck without specific effort to make the things actually try to teach, verification only goes so far, and above all human frustration and overconfidence are killers #linklog

@jamesbrandecon.bsky.social had a great blog post that mentioned how it would be fun to see some other benchmarks that are more about human versions: jamesbrandecon.github.io/blog/posts_h...

it's less weird to me to think "it got frightened by the idea" than it is to think "the correct proof is easiest to find along sequences of {X_t} such that P(Y|{X_t}) contains expressions of fright"

the Colossus 1 flops seem quite expensive

all my findings are counterintuitive (bad intuition)

this is close to my intuition. LLMs sample from P(Y|X), prompts are conditioning info, jailbreaks are local inverses of other conditioning info. the kinds of P(Y|X) LLMs sample from aren't "nice" so local inverses get weird. X as a space is all latent factors mapping to statements, so... extra weird

trillion dollar idea

"is it credible" is how economists say "no" and i think that's lovely isitcredible.com/cases

you won't wanna (0000) eighty six THIS meal wasabi and recursion. while supplies last

this was a very cool project

"Lyman Stone doing whatever he does" ☠️☠️☠️

"interesting long read" is how economists say "your identification strategy, it is totally believable, this is causal" and i think that's lovely

this ONE trick carries like all of congestion games and applied cge calibration

someday somebody's gonna figure out the financial engineering and regulatory play for municipal data centers

I go back and forth on this a lot and tbh it's not obvious to me at all anymore

I'm pretty sure I don't have any fake citations in papers I've written, but if I did I guess by construction I wouldn't know I'd really like to think I don't have any incorrect proofs in papers I've written, but tbh idk. the arguments made sense to me! pretty sure most reviewers don't redo proofs

proposal: "cowening". a straussian read will clarify

back in the day this used to be dlc

steve carrell top chickens 🤝

gooning but it's the culture commentary of elite American publications. need a term for this

vibecoding an app to vibecode the apps that will summarize the reading for my bots (they'll tell me about it)

rooster is deeply uncomfortable

this seems like the kind of place where people will have opinions: all else equal, is a document better with chapters or sections

no. mr bond. the market expects you to collapse

existence of critical thinking implies existence of actoral thinking

evergreen

has any economist active in the last decade had the raw rizz to publish an empirical paper in a top 5 journal and state that it has no relevance to policy

if an agent can be framed as a constrained optimization problem, then there's gonna be something like aggregation theorems lurking

93% weekly usage, 7 minutes left... let's see what i can do

it feels like "composable agents" should be a thing. is it a thing?

existence of theory of mind implies existence of applications of mind

fuck it we're doing digital quintuplets

it's not the token count it's what you do with it

getting negatively polarized into "a June 1 base case not be reasonable if it doesn't really matter for the point you're trying to make"

"the kolmogorov complexity of a LW post" is now a phrase rattling around my brain

we need to move more slowly. in fact we need to really scale up moving slowly. fast

rewatching invincible and man the orbital mechanics make no sense

idk what wet or dry claudes are and at this point idk if I care but I definitely won't be asking

i think it's interesting how "econsky" as a tag seems filled with more newsletter spam than economists talking shop. nice little pooling equilibrium -> unraveling story. what else is like this on here?

Economies of scale and internal rates of return … across all fields … are a calamity (and yes attributable directly to our extractive economic systems).

fwiw i knew folks talking about space data centers in like... 2020. like, actual smart people who know about space and afaict with no money on the line. partly about solar power, but also large part about earth's environment and regulation

now i have the full picture -- the shape of the work IS the work

anti-Discover feed that obliterates your awareness of certain posts and posters

6% gdp growth! > real or nominal? haha it's a good growth print >