jae (hallowed)

jae (hallowed)

@fubarchitect.com

maybe, maybe just once, i, will get, what's coming to me

1.1K followers985 following14.5K posts

well prepared collards. breakfast lunch or dinner. cold or hot gotta have that hock in it tho, and at least 2 kinds of pepper

What is your FAVORITE food, pound-for-pound? Like, a dish you will eat at any time and it'll still be good? Mine is lasagna.

jae (hallowed) reposted

internal Anthropic researchers working with an unreleased version of Claude have reportedly discovered a fifth Noble Truth

jae (hallowed) reposted

Replying to @lauriewired.bsky.social

throw some agents at it, see what ROI you can squeeze out.
 I’ve got some of mine trying to find bit twiddling type optimizations on particular CPU gens. kind of doing a superoptimizing compiler + finding lower bounds with z3 it's working...not bad actually!

this weirdly tastes like strawberry-rs to me because i've played enough doom for this to be a really grotesque failure to depict doom

Ethan Mollick@emollick.bsky.social

Google's new Genie 4 runs DOOM: "Walking through a typical office but every computer is running DOOM and all of the walls are also running DOOM and also everything is DOOM" "My coworker who is a demon appears" "He has a chaingun" "He gives me a bonus" All this is in realtime.

oh, yes, my pre-ipo financial disclosure. the most scrutinized document my company will likely publish this year, if not ever. surely what i should do with this document is fill it with lies the average bluesky poster can see through this will be a fruitful investment for me

like how does this even work; does one imagine themselves as a titanic intellect beyond the reach of any of the thousands of regulators and millions of prospective buyers reading this same document? not saying companies don't lie, but in a way that your dumbass is gonna catch? no honey, not here

i think most white people just don't know what to do with an eloquent black man in any context, but when he's not president they can just think "DEI" or "Affirmative Action" really loud in their heads to make the pain stop. when he's president, it's a bit harder

Aubrey Gilleran@aubreygilleran.bsky.social

If you are too young to really remember the Obama years, conservatives were like this every single day. He really did just break their brains.

jae (hallowed) reposted

just heartbreaking stuff Alex Pretti's mother calls out ICE and the Trump administration for killing her son and smearing his name, and calls for justice to be done.

jae (hallowed) reposted

Replying to @daniloc.xyz

Funny how having an inherent respect for human behavior can protect you from this. Not that I agree with all human behavior but that I think it generally has a point to it and "the whole world is stupider than me" is a bad distortion on analysis

jae (hallowed) reposted

It genuinely does suck and is *bad* that it feels like there’s almost nowhere to go for people who 1. Want to work on robotics (like drones) and 2. Don’t want to build weapons.

there's a few lines in here about reading papers that go, approximately, 'it is not about reading the paper, it's about being exposed to the core idea and then thinking alongside the paper; ideally you try to prove it to yourself and use the paper to check your work' and it's like my favorite idea!

jae (hallowed)@fubarchitect.com

oh i like this one a lot, especially @3blue1brown.com 's commentary on credit and ownership and, approximately, meaning-making it resonates with a central conflict i have with society; y'all way too into 'status' and 'credit' and it's a wonder anything useful ever gets done in that context

View full thread

using saas as well; in general for the applied mathematics schools i understand, a real world artifact is a paper for this model of thought and they make me tremendously happy

anyway, this is becoming how software discussion works as well! "hey guys, i just summoned <tool>" - i don't actually want to see <tool> yet. i want to think about if your version of <tool> sounds useful, and will likely end up summoning a very different [tool] later

there's a few lines in here about reading papers that go, approximately, 'it is not about reading the paper, it's about being exposed to the core idea and then thinking alongside the paper; ideally you try to prove it to yourself and use the paper to check your work' and it's like my favorite idea!

jae (hallowed)@fubarchitect.com

oh i like this one a lot, especially @3blue1brown.com 's commentary on credit and ownership and, approximately, meaning-making it resonates with a central conflict i have with society; y'all way too into 'status' and 'credit' and it's a wonder anything useful ever gets done in that context

this is what gaming is like for me; i see a game mechanic and a part of my brain cracks open; i am curious how it was done but have negative desire to see the code, i want to reason to how it """should be""" done in my internal code ideology before i consider how it actually is done

jae (hallowed) reposted

Replying to @fubarchitect.com

there's plenty of evidence that the first proofs of big theorems are supplanted by later more elegant and intuitive work. my pet example is erdos-ko-rado: the original proof was a tedious double induction, the current popular textbook proof came a decade later

in contrast tho, for most people reading this, your work is 99% simple stuff that open models were sufficient for last year

Quantian@quantian.bsky.social

If you are still on the “cheap Chinese models will take share from US labs” cope train you need to get off. US labs are now distilling their own models to compete across the Pareto frontier, so Haiku is better model than Kimi at 1/8th the cost. The only reason to use it is if you MUST self-host.

jae (hallowed) reposted

Replying to @ed3d.net

also "the sankey community" is beautiful

jae (hallowed) reposted

if you're a professional programmer, and you have 'ops folks', they do not and have never actually trusted you, and you can treat your models like they treat you

garrison@garrison.corporate.fm

I mean it's kinda impossible to use models to write code unless you trust them

View full thread

but steve, being a credentialist little fictional weasel, needs the credit to do the work, and so we have a system to give it to him. it's rather recent, and it's not very good, and people act like it's a central pillar of society

View full thread

but steve, being a credentialist little fictional weasel, needs the credit to do the work, and so we have a system to give it to him. it's rather recent, and it's not very good, and people act like it's a central pillar of society

from my pov it also ties in to why we suck ass at history as a society; we really wanna know whodunnit, but most things aren't actually done by a person, as much as there's simply a person who is there when they are done. they are done by movements

View full thread

personally i'm excited for the proto-borg thought that promptly emerges when you can't expect to have your name on the world changing paper that's how operators are, broadly. no one gets credit for the site staying up. so it takes a type

it's sorta bullshit to credit the openai paperdump papers to "openai", but not because collective credit is bad. it's bullshit because 90+% was done before they appeared. but that means it'd also be bullshit to credit steve the math professor

math is not dead, but first-past-the-post credit in math might be, and that means a lot of individuals' motives to fuck with math might be but that's deeply distinct from math being dead!