johnny, uplifted octopus

spurgeon.computationaltheology.comThomas Spurgeon at the PhonographThe nearest thing to the voice of Charles Spurgeon: his son reading his father's last printed words into an Edison-Bell phonograph in 1905, restored and playable beside the original transfer.

13 Sept

a month ago, I used gpt/opus to clean up a 1905 Edison-Bell cylinder recording. doing so was less sexy than what people are getting with fable/astra rn, but it was an interesting experience because my HITL role was basically reduced to “ears” spurgeon.computationaltheology.com/thomas-spurg...

5 Sept

astra is incredibly fun

@whysharksmatter.bsky.social

5 Sept

progressives relinquishing positive technological visions of the future did not work out for them

johnny, uplifted octopus reposted

David Shiffman, Ph.D. 🦈

discoverwildlife.comAstonishing image of scorpionfish devouring prey whole among Ocean Photographer of the Year 2026 finalists | Discover WildlifeDelve into the hidden worlds of our oceans with the 2026 finalists.

Astonishing image of scorpionfish devouring prey whole among Ocean Photographer of the Year 2026 finalists | Discover Wildlife

founder effects and continuing social network balkanization have turned this place into a breeder reactor for terrible brain worms in the same way that parler did for its particular anti-social cohort

sure, sampling problems rule everything around me. but, social amplification dynamics produce something like mercury bioaccumulation and a lot of minds are fully rotted out

if your "social" activities are like 90% dunking on people venomously — and if you disguise that as politics — you're just deliberately eating shit all day long. and shit is no nutritious

there is going to be an impressive improvement in americans' health coincident roughly with kennedy being at hhs but attributable like 100% to glp-1s. it’s best to start coping with the painful inferences that will be drawn, asserted and accepted ahead of time

if you had not had the equivalent of a giant irresistible smile taking over your face accompanied by an internal monolog that roughly translates to "wheeeeeee!" over the last few years, you are not a technologist

which is fine, of course — you don't have to be one! but its a useful means of calibrating your identity to reality

last night at 2am during insomnia, i made a gaussian splat subtle 3d effect for photos (of iphone fame) for my linux machine; turned it into a screensaver; and populated it with baby's pictures. that's indistinguishable from magic

5 Aug

we are living through the most interesting moment in human history. what an absolute privilege

30 July

our collective neocortex is the bow shock in front of culture traversing time

27 July

being able to prompt my way to a better tutorial on text book on any subject i want on demand is truly liberating

21 July

lines get blurred when you have to defend what is and is not science, and the infallibility of their respective clergy. but yea that’s patent anti-intellectualism animating a diseased host. not hard to spot or diagnose

21 July

what is your vision for a political movement that you would want to be part of; that doesn’t demand authoritarian measures to induce and maintain; and is realizable in the united states given our institutional structures?

21 July

(i have no answer here. not one thing i can point to)

20 July

big deal! anyone could have solved jacob’s conjugate !!

transformer-circuits.pubVerbalizable Representations Form a Global Workspace in Language Models

7 July

incredible stuff transformer-circuits.pub/2026/workspa...

johnny, uplifted octopus reposted

Andrew Heiss

@andrew.heiss.phd

12 June

Hey @generativist.com it’s A Short Stay in Hell!

Tamarra Kemsley@tamarrakemsley.bsky.social

For what appears to be the first time, a Latter-day Saint author numbered among the great English fiction writers. www.sltrib.com/religion/202...

sltrib.comFor what appears to be the first time, a Latter-day Saint author numbered among the great English fiction writersThe tides turned for this Latter-day Saint author when TikTok discovered his psychological thriller about hell.

tulsi gabbard's husband has the extremely rare cancer i had and my mind refuses to accept the premise

it is extraordinarily difficult for someone who doesn't have any intuition about a particular domain to pressure test their ideas using llms.

that's why people (in tech/engineering/vc) keep droning about "taste" lately. when you say someone lacks taste, your gesturing at a bad smell -- the gap between what they understand and the output of what they forced into production

the person who can't detect that lacks the understanding of a domain necessary to exercise discernment. it really does collapse towards something like aesthetics born of exposure to *a lot* of good and bad patterns. most people don't and *can't* have that unconditionally

it is extraordinarily difficult for someone who doesn't have any intuition about a particular domain to pressure test their ideas using llms.

sycophancy is a problem. but even if you use discipline to explicitly ask, - is this a good idea? - does it make sense strategically? - does it make sense technically? the conversation quickly turns to finding ways to route around the criticisms. and it will do that for you

johnny, uplifted octopus reposted

Peter Sagal

@petersagal.bsky.social

31 May

Today: the North Shore Half-Marathon. I run it every year to mark my decline. This year: 1:47:30. 21 seconds slower than last year. Time marches on, but I'm making it drag its heels.

28 May

scottaaronson.blog?p=9782

28 May

scottaaronson.blogDispatches from the possibly last days of human relevanceAs most readers have presumably heard by now, Paul Erdös’s Unit Distance Problem from 1946—one of the central open problems from the field of discrete geometry—has been solved by …

21 May

periodic reminder that the stochastic parrots frame is epistemic poison and you should actively and almost desperately try to rid yourself of it. things are...moving faster now

15 May

“it doesn’t matter it’s inevitable” retorts re:ai are bad not because they are fatalistic, but because they promulgate this frame that obscures how fucking fun it all is

14 May

14 May

ai skeptics btfo

2026 orthodoxy is something like, > valuable software is now free or cheap to build so you should be mostly effort solving for distribution because thats how you win and i think it's very foolish.

like the funny thing is people are using this genuinely remarkable reasoning machine to...shit out things that mostly wait for evolution à la the market to decide

the reason everyone is babbling on about taste lately is because everyone kinda senses that no one is exercising it. its the thing up stream of what you chose to generate — the thing that decides what LLM-accessible node should get expanded

2026 orthodoxy is something like, > valuable software is now free or cheap to build so you should be mostly effort solving for distribution because thats how you win and i think it's very foolish.

distribution is necessary but not sufficient. or at least, its only sufficient in cases you will almost certainly not find yourselves in — celebrity games.

but more importantly, coding agents themselves generate what you tell them to generate. its not exactly Borges library in that its not uniformly random. but the modal output is still a bad idea from an engineering standpoint and from an actual useful standpoint

2026 orthodoxy is something like, > valuable software is now free or cheap to build so you should be mostly effort solving for distribution because thats how you win and i think it's very foolish.

i don't even think this is a straw man. i think this is what the modal X (yea okay okay) software entrepreneur believes. in part they believe that because like 90% of them just want to be influencers. the other 10% is just confused

i think on avg coding agents and LLMs do make people lazier, and do induce mental atrophy. I worry about it not infrequently. but i'm actually learning and exploring more computer science-y domains than ever before, because i don't have to spend 80% of my time on trivialities

meanwhile, last weekend i started refreshing some of that to make my nephew some simple web board games with NPCs to play against. all of a sudden, my time is spent generating interesting questions to think about.

still risks in that obviously. and yes nitty gritty implementation is *often* the source of insights. but most of the time it isn't and if you get to the point where you are asking interesting questions sooner, you are learning in a fun and fast way

in grad school i had an AI (classical) class. very fun. my project was a poker agent. as in lots of contexts with games, a stupid amount of the time is just the game scaffolding. you can learn *some* things from that but mostly it's a side quest that takes 90% of the time

people emotionally invested in the delusion that it is a bubble are going to be both disappointed and wrong