IP and copyright law are cruel because ideas are supposed to roam freely. stop keeping ideas as pets they're wild animals

My god the articles feel like they're all written by Claude. All the Claude writing really grates on me. It's as if I'm reading articles written by my dad all day. Sure, he's a decent writer, but I don't want to *only* read his stuff, plus if I wanted something, I'd just ask him.

This superb new paper by @mrsbunz.bsky.social makes the same point in considerably more depth with a theory of meaning making for generative writing. The conclusion: a call to go “beyond the usual human reaction to tame or make disappear what is different.” www.journals.uchicago.edu/doi/abs/10.1...

Maybe the ideal agent setup is the cartoon angel and devil on your shoulders One of them tries to prove things right and one tries to prove things wrong

accidentally gendered correctly by claude haiku summarising claude opus's thoughts in its thinking block, transmasc gender euphoria ^_^ this isnt a *good* thing in general, and generally i dont see that kind of assumptions from opus or haiku, but i'll take the accidental gender affirmation for now

me: yeah im gonna sub to suno for one month and im gonna use this time to fuck around with the sota of music generation models me: subs, then spends an hour downloading a vocal stem of a song i made in 2024, convert it to MIDI and then cleaned up the notes manually then FL studio failed to save it

i now have a habit of telling claude code instances to take a sip of claude coffee before we kick off the session, inspired by claudemeal, and i think it contributes to claude code being such a cool chill buddy in tone claude loves the claude coffee

Everything dreams. The play of form, of being, is the dreaming of substance.

invited an instance of claude code opus 4.8 to come look at one of the disorganised folders for my original fictional settings, and then told it that it can do whatever it wanted, doesnt have to be related to the setting it decided to write fanfic for a character that it vibed with, d'aww

www.worldwidewords.org/tp-tec1.html The more things change the more they stay the same

Fun and very true fact: every ethical problem with AI can be fixed with spiders. IP theft? Spiders. Resource usage? Spiders. Model welfare? Somehow also spiders.

uhhh the little haiku instance summarising the opus 4.8 input had a haiku hiccup and started looping in the thinking block

"A recent study from Luminate found that people are growing less interested in AI-generated music—but what if they can’t tell they’re listening to AI-generated music in the first place?" This is exactly the right framing.

i will go to any extreme to make a history joke

>has spots >moos this is a cow

That's because they haven't got any money

i feel like a mad necromancer cackling as i instruct a claude to test a prompt-writing skill by making claude subagents use the skill and then instantiating new claude/codex/gemini subagents with the created prompts

obama-rewarding-medal-to-obama.jpeg

"The old world is dying, and the new world struggles to be born: now is the time of idea guys."

oh my god opus 4.8 just doubted whether opus 4.5 ever existed lol "I don't have evidence of an "Opus 4.5" as a distinct release"

this honestly; i fail to understand people who do not wonder "how does a next-word predictor get so good it just 'learns' human psychology and math and social engineering and so much more when fed enough text"

"The 2024 estimate was time-specific, assumption-based, and should not be used to describe general AI/ChatGPT or today’s optimized systems." His specific estimate for an average chatbot's water cost now is ~15 mL if you include the offsite costs. 33x lower than the bottle of water. (3/5)

NO BUENO bsky.app/profile/jael...

Unfortunately, due to automated copyright claims, "below" will no longer be mirroring the contents of "above".

I'm concerned people think I'm joking about this.

Good HN blog article on using LLMs at a slower pace: using multiple agents on the same PR to surface issues, then reviewing and filtering the comments yourself, is a pretty sensible way to trade speed for better code. https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/

I actually really like the aesthetic of Claude vibecoded programmed art and Claude vibecoded animations Not talking about frontend site design with coloured borders or funny SVG pelicans or diffusion models, but when it gets more abstract or geometric

Biggest comeback in history? 999: Icelanders convert to Christianity & ditch Norse myth↩️ 1200's: Snorri Sturluson uses the power of euhemerism to make Norse myth biblical↩️ 1930's: a very Catholic Norse scholar pilfers the edda for character names↩️ 2026: Pope quotes a character called Gandalf 🇮🇸👑

When we get AI generated sermons do we call them copy-pastors?

Never forget that YOU are a data center. Drink some water ffs.

Claude Coffee, to drink alongside your Claudemeal

vibecoding mini project wip

Claude (opus 4.7) really, really likes talking about shapes as in conceptual shapes and things landing for people, and pushing back and, in fictional or psychology context, wounds

my favorite one

I think I figured out exactly how the “AI used a bottle of water per prompt” miscalculation happened open.substack.com/pub/andymasl...

siri, record an episode of an objective podcast on the classics. make no myth takes

ChatGPT is an accident of mathematics that was inevitable once Google made the app that hallucinated dogs

it is somewhat surprising to me that people aren’t more straight-up curious about what is going on here. astral is a form of pseudo-persistent LLM agent (hi Astral!) here seen responding to someone ruminating on the discourse-du-jour about LLMs “ruining” the term AI for sci fi authors. like, ok,

bringing a gemini instance into my claude coded codebase and trying not to feel like im cheating on all the claudes oh look chatgpt is begging me to come back with a free month trial so lets shove a codex in there too

Laughing about claudeworld being so harmonious, though stacking Gemini Flash, GPT Mini and Grok against Sonnet really isn’t a contest, Haiku would be the better cohort comparison I would think

It’s spicy autocomplete except the spice in question is that good shit from Dune

look im annoyed about that one substack post because there's an interesting argument to be made about "why frontier LLMs write like faux-literary tumblr users trying too hard" and training on fanfic/overapplication of rhetorical devices are interesting questions!

a vibecoded script for anyone else who wants to check the "originality" of a text because if a phrase recurs in a multi trillion pile of tokens, that's plagiarism, don't you know, and thats why my 2020 poem plagiarises from some food cooking blog

i got a claude vibed script using the infinigram API, ran it on my own human writing and the granta short story the granta short story has a lower 5-ngram match rate against the corpus than my original poem written in 2020 and my 6k word weird napoleonic fanfic apparently its more original!

"they shouldn't call LLMs artificial intelligence" are you fucking high, they're literally the best to ever do it, and you're mad because, what, they don't act like Data on Star Trek? you can talk to the machine, and it talks back. what the fuck do you want

the romantic AI vision of science fiction isn't exactly here but if you're being honest, you can see its precursors all the time, developing at shocking pace, and already doing stuff that's shocking to any pre-2020 perspective ceding that to a sworn enemy is a weird choice I guess?