skelly tayne

skelly tayne

@ohnoether.bsky.social

applied mathematician full of bees

76 followers504 following241 posts

at this rate, when i boot up my dusty 1 year old windows dual boot, my computer is gonna get insta pwned

Techmeme@techmeme.com

Microsoft releases updates fixing 1,400+ security vulnerabilities, surpassing September's record of 972, which was already ~10x the normal monthly patch load (Brian Krebs/Krebs on Security) Main Link | Techmeme Permalink

> i don’t really think about code anymore, it’s more about aligning subsurfaces in high dimensional vector spaces github repo is a todo notes app

This post is unavailable.

skelly tayne reposted

Replying to @benjaminschneider.ch

there's a museum here that's all overhead belt systems they all still run, too

not seeing enough discussion of LLMs for debugging weirdo hardware issues. did i know my video card had an outdated VBIOS? no. was i just going to live with occasional video glitches and hiccups? yea probably. but claude dug into the crash logs and determined the exact firmware fix

a coordinated mobo and gpu firmware update and all problems gone. two years ago id either just accept it or stumble on an obscure forum post from someone with the same issue and, if lucky, copy their solution, or, if unlucky, struggle to reverse engineer their fix from their cryptic follow-up

i wonder how much "code sharing" there is across these. like are these at all using consistent constructs? i still dont quite follow how they verify the formalization. it's not like code where i can execute it in unknown domains and get unexpected behavior right? but this is what sharing serves

me laying on the couch prompting: yess haha yess fuck yesss me when my manager gives ai the bulk of the credit and i dont get promoted: wait no this fucking sucks

Zach Weinersmith@zachweinersmith.bsky.social

Had an experience recently talking to a mathematician friend about AI, and I asked why some math people are having existential crisis and some are excited. And he corrected me saying those things aren't mutually exclusive, and he himself often felt both at once.

replace manager with tenure committee, peer review, etc. brain still attuned to evaluating social credit at 5 year ago prices: omg im printing money brain realizing it has no idea how to price things: oh no everything is worthless

ok everyone don’t panic but imagine how empowered this makes even one malicious actor inside your org

hailey@hailey.at

really not quite sure we are ready for a model that is: 1. this ridiculously cheap to run 2. has effectively zero guardrails around taking advantage of vulnerabilities regardless of scope or outcome 3. is actually pretty darn capable

we’re like 10 generations of engagement loop tech beyond dfw and that guy basically got one shot by 90s era tv. we’re gonna need some industrial grade jesus (a new frontier model that is very good at getting people to touch grass)

🐜 🍯 Anth-“unholy” Becker@anthonybecker.bsky.social

Things doing is constantly growing Supervising agents induced greater doing Things doing fall into tickets and plans and WIP and TODO.md and little scraps of paper and maybe I need to make an app to help me no! Another todo to do the todos! Nothing doing!

guys is it good that our scryers have averted their gaze from the stars

Ethan Mollick@emollick.bsky.social

Was just skimming through the webpages of some of my favorite hard science fiction authors and wow the majority of them hate LLMs: a surprising number of them because they think it is a useless parrot that doesn’t work(!), some because of IP, and a minority because of worries about existential risk.

damn the duckdb+aws merger actually makes a ton of sense - duckdb has become an essential layer over s3 for me that i was starting to think of them as viewer and storage. putting the usual big corpo cynicism aside, aws is in a position to build some really incredible services with this

shell history + /tmp/claude-1000/*/scratchpad search for prior work still feels largely underused

this is NOT an ai-hater post but i see a lot of folks that reach immediately for tool-using agents to drive the execution of repeatable steps when they could just... have their agent write a script. the latter is likely to be 100x more token-efficient

skelly tayne reposted

point: AI is completely useless for literally everything counterpoint: in two hours have a full acoustic simulation of my living room that can optimize the setup for specific frequency response desirability profiles of each individual in their seating location.

me in my code: this is NOT an overoptimization, i will literally DIE if we don't delete this dependency me watching others slim their code: *eyes rolling* wooooooow, really

skelly tayne reposted

The birds are sharing terrible photos of you on their nature app. Double chins, acne, crossed eyes. But you are beautiful to the birds

as much as this would be Good, i know i am too lazy to do this because i am already not marking every code comment with # AI;DR, even though i pretty easily could but hear me out though: a code viewer that's aware of these tags and just filters out all AI comments would be so sick

i’ve been saying markdown needs an “ai-generated” block quote style and i’m right goddamn it

the typical reason for hiding thinking traces is denying competitors distillation, but i'm guessing the icky weirdness of reading the model actively strategize its response to influence the user is a strong secondary factor

>user clearly has a nostalgia for less wrong posts, but in a reformed postrat kinda way >a quick nod to normie bayesian reasoning won't hurt >an ironic reference to Lob's theorem >dont fuckin touch timeless decision theory