grand theft eigenvalue 🔆

Anthropic revenue growth is terrifyingly fast www.bloomberg.com/news/article...

valid haha

I don't recall exactly but ballpark maybe 67%?

love that this gives me the option to download my posterior

mine is -0.03 :)

yes. basically, there are places where research is so critical to the organizational production function that it is a primary job responsibility. i am expressing a normative view that "professor" ought to carry with it a primary job responsibility to teach and mentor.

Many are wrapped in consulting functions. The general gist is you have clients who need questions answered or a better understanding of some particular situation. You then do the research necessary to support their needs. Not nearly as blue sky as academic work can be.

fwiw I don't think the role of "professor" ought to be compatible with the notion of "my job is mainly to produce science". if one thinks one's research product is so valuable as to obviate responsibilities to students, go start a research firm or whatever. commit to your values and face the market.

it seems everyone really wants to "do better science", the action is in disagreeing over what each of those words means

"Big Bacta wants to pump you full of MIDICHLORIANS! we have all natural, all HUMAN, Forceful remedies!" etc etc

the quote you posted describes the singularity as an idea about scale. the post you quoted points out the idea does not make sense at a particular scale. you seem to be arguing here that if the idea makes sense at one scale it makes sense, period? that's fine, just lazy

as opposed to "actually nuclear chain reactions exist independently, and you're being dense if you think it's useful to think about how they might relate to war" ?

no, it's like telling you that you should not insist the only way to think about viruses and neutrons is to neglect war as a whole.

"social reasons" meaning "I refuse to think about a concept involving production without integrating ideas from a field that studies production" is nuts. I get that there's a read by which this position makes sense, but you can always rationalize things by selectively not thinking.

"it rubs the Fortnite lotion on the PUBG skin"

when is the numeraire an attribute of the good? hedonic evidence from the cs:go skins market

doing my part to prevent total collapse with artisanal handcrafted blogslop

TIL you were into AI alignment before it was cool

see you get me

choosing to believe anthropic smashed like and subscribe

But the most important rule, the rule you can never forget, no matter how banging the content, no matter how tempting the skeet, never like on For You.

wow that is a very red big butt! ty

pics???

life pro tip: lazy doomerism says things will be bad in the ways you personally dislike most. don't be lazy! find other ways for things to be bad too.

fwiw i don't think bubbles will be designed as such, they'll just be solutions to maximization problems that may themselves be emergent from interacting "designer" objectives. maybe that's how it is now.

tbf we're all NPCs for someone else's playthrough :)

if I'm right it will be very easy to communicate within your public, and very hard to communicate beyond it. some people will mimic into publics thinking they can reshape them; they will generally fail and be reshaped.

counterprediction: people are just not that original, correlations in For You feeds across and within people are structural and persistent. we'll be sorted into separate engageable publics.

i disagree with the upper limit logic in practice because we will always create new tasks that are even harder for the models to do, because we will use the models to get right up to that frontier and still want more. we will ~always want more.

this is also why humans will always have jobs.

imo this upper limit logic is correct for a fixed set of tasks. still, i don't think distillation is a red herring:

1. rents matter
2. personalities generate rents
3. the economic value of a personality is (also) a function of the personality ecosystem

interesting to me that if this works (you can create Grokkish-Claude by distilling Claude), then the argument that evil vectors are correlated with stupid vectors gets weaker. maybe you gotta break bad to get there, but there's an evil orthant in the not-stupid subspace