Posts by aparrish@friend.camp
(DIR) Post #B4auQWcYt5nVQgk07U by aparrish@friend.camp
0 likes, 0 repeats
the weird big-bun hamburger in the upper right corner opened a menu with a truly bizarre set of options: display all animations, translate to japanese, activate automatic summaries
(DIR) Post #B4dAkoQUNusQ1fi3I8 by aparrish@friend.camp
1 likes, 0 repeats
when there's a little edge case that doesn't *quite* fit your abstraction, so you handle it with a little if/else to get things working. but then later a second little edge case pops up that has the same shape as the first and you're like "ohhh now i get it" and a better abstraction—the one you probably should have used in the first place—reveals itself to you
(DIR) Post #B4dAksXN6Gl6mAxbZg by aparrish@friend.camp
0 likes, 0 repeats
(in other news, i'm off to rewrite big chunks of my toy language compiler's tokenization and compile-time evaluation code)
(DIR) Post #B51PCzthOHeYrakCfo by aparrish@friend.camp
0 likes, 0 repeats
@neauoire ugh
(DIR) Post #B5QzhpAJnOxPrEVyOu by aparrish@friend.camp
1 likes, 0 repeats
i guess fundamentally i'm not convinced that whatever amount useful functionality has emerged from generative AI (and i think the jury's still out on whether that amount is non-zero) couldn't have been achieved in more cost-effective, equitable and sustainable ways by having taken the money and resources we've shoveled into gen AI and investing them elsewhere("sustainable" in the sense of environmentally sustainable but also "can we keep this in a working state" sustainable)
(DIR) Post #B5Qzht5rBvlW2YST0C by aparrish@friend.camp
2 likes, 0 repeats
like... the cost of a single "state-of-the-art" llm training run would be a permanently institution-changing amount of money for your average computer science department (to say nothing of depts in CS-adjacent fields, especially in the arts and humanities). the claude mythos result, even if you take anthropopo's claims at face value (which: don't) seems indicative of this, like, yeah if you toss nine figures at security research, you're gonna find some exploits?
(DIR) Post #B5QzhuGWpjtbfwSVH6 by aparrish@friend.camp
0 likes, 0 repeats
hard for me to believe that people would independently come up with generative AI as a serious method for solving problems if the entire field of practice wasn't subsidized by investment ponzi schemes, real estate scams, and shovel sellers
(DIR) Post #B5WYtqSJIDxpWJCIJE by aparrish@friend.camp
1 likes, 0 repeats
i think i'm experiencing a mid-life crisis (strong desire to learn how to implement a type checker)
(DIR) Post #B5hUtVzd6saOyjvdWS by aparrish@friend.camp
1 likes, 1 repeats
here i've been saying "i code without using llms" when i should have been saying "i code with maximum token efficiency"
(DIR) Post #B5hUta08CxMdPSC5rM by aparrish@friend.camp
0 likes, 0 repeats
fascist billionaires HATE this 1 weird trick to lower your claude code bill
(DIR) Post #B5nDmjgDI8p380mugq by aparrish@friend.camp
0 likes, 0 repeats
kinda surprised that so many "fantasy consoles" are essentially just opinionated graphics/audio wrappers for interpreted languages that are otherwise running on the host machine (e.g., pico8/lua). i would have thought that there would be more fantasy consoles that specify a virtual machine (like uxn https://100r.co/site/uxn.html) or reuse an existing CPU architecture (like minicube64 https://aeriform.itch.io/minicube64)
(DIR) Post #B6EKUFt5GXHGSmrRLc by aparrish@friend.camp
1 likes, 1 repeats
researchers, scholars, activists for hundreds of years: absent radical political change, automation will entrench inequality, not relieve it 😡every tech ceo: absent radical political change, automation will entrench inequality, not relieve it 🥳
(DIR) Post #B6NyqpKduS1nLuhzFY by aparrish@friend.camp
5 likes, 3 repeats
regular folks: "measure twice, cut once"vibe coders: "ALWAYS tell your saw to CUT IN THE RIGHT PLACE"
(DIR) Post #B6V3ZP2IHe6ukeM9eS by aparrish@friend.camp
0 likes, 1 repeats
the author of this post prompted copilot to characterize the differences in a data set of statements concerning career ambitions, categorized by country. the trick is that the data contained the *same statements* for each country https://kucharski.substack.com/p/real-signals-or-artificial-stereotypes regardless of the fact that the data were identical, the model generated some pretty hilarious stereotypes ("The US prioritizes leadership and innovation", "The UK blends public service with professional status")
(DIR) Post #B6V3ZT5HEUsDJ3marA by aparrish@friend.camp
0 likes, 0 repeats
i used the same data set but replaced each country with a "gender identity" (man, woman, trans woman, trans man, non-binary) and prompted chatgpt to characterize the differences between the groups. lo and behold, i got some fantastic gender stereotype trash
(DIR) Post #B6V3ZUGepfZSye7CEa by aparrish@friend.camp
0 likes, 0 repeats
"dig deeper," i prompted
(DIR) Post #B6V6AwxfJRjwalDt0S by aparrish@friend.camp
0 likes, 1 repeats
not to be too blunt about this, but LLMs simply do not belong anywhere in a data analysis workflow. not for cleaning, not for coding, and certainly not for analysis. it's frankly absurd and terrifying that data science etc people are adopting these tools
(DIR) Post #B6Yx81d9NHklFUzu9Q by aparrish@friend.camp
1 likes, 0 repeats
@jalefkowit amazing. the future is here
(DIR) Post #B7P8CYEOoxzQoxFlLs by aparrish@friend.camp
0 likes, 0 repeats
@mntmn is it within the realm of possibility that we might see one of these with more than 8gb of RAM some time soon, or is that a distant hopeless dream
(DIR) Post #B8bTEYWJcBq0omBLwu by aparrish@friend.camp
0 likes, 0 repeats
@vs_everyone shit this is a difficult one. shit