iiinova — where your digital self keeps evolving

watching some claude opus 5.5 demos on YT and feeling, as an artist, like my life has no purpose anymore good thing i decided a few weeks back to become unreasonable so this is no problem. fuck efficiency, fuck effectivity, i'm the BBQ slow cooking sous vide chef of software dev

161

Anthony Etherin @anthonyetherin.bsky.social · 29/09/2026

they're really fighting for the hearts of the uneducated don't they

000

Reposted by LR

HOW TO DRAW A PYRAMID A zig. Now one zag. Gaze now on Giza! #palindrome

07020

ikea taglines got complicated

also came up with a word for local LLMs, "homebot"

000

i wrote previously that i struggle finding use in LLMs. i still think this is not the optimal form - i'd have a better design. i would also not generalize guaranteed success from this. i still don't see them as "intelligent", never will. but it's a kick in anthropics nuts; this part i like.

130

couldn't find it but there is this www.instagram.com/reel/DTvE1wz...

130

there was a tiktok making the rounds here that was titled "using AI for programming" or something, and was just a video of a kid presenting a rube goldberg machine they built going through their flat, where they quickly jumped in at every step to correct for broken transitions. that's how it feels.

130

oh, also to address "can the machine do most of what a human can" directly: imo these things can not work without human supervision. you need someone competent enough to judge if the output is good otherwise you're headed for disaster. so you do not save on employee cost at all.

100

there is also the issue of self-inflicted accelerationism. APIs update faster. solutions get superseded faster. models trained on the state of the art increasingly perform worse as the landscape beneath them changes, and this happens within a year. so they obsolete themselves.

the only field that seems to vocally benefit from this is programming. there is to my knowledge only one model which can (mostly) match expectations, which is claude opus, but for which costs keep going up, and it is difficult to measure how wasteful it is b/c anthropic hides metrics.

100

there is scientific research that discovered it is feasible to bake inference engines into fiber optics, which dramatically reduces need for power and cooling, but this entire pipeline is not ready to be rolled out any time soon, requires lots of R&D that they don't have money for anymore.

000

i locally tested three models by having them generate, in iterations, a rotating 3d cube with my preferred tech stack, and it took them about 300k tokens to do it. and that's just a 3d cube. that's not even anything of value.

230

it ultimately boils down to the fact that watt-per-token is too high, even more so during training. i also pointed this out (on mastodon) very early, and they've since made it worse by spending more and more tokens on thinking to improve accuracy.

Tom Forsyth @tomforsyth.bsky.social · 28/09/2026

Reposted by LR

DASHR - Dynamically Animated Skinned Heightfield Rendering Short video for now - I might do a longer one later? youtu.be/-Su9YrcazRk But there's ALL the details in the paper, and you can always Use The Source, Luke for the really deep info. github.com/tomforsyth10...

youtu.be

DASHR - Dynamically Animated Skinned Heightfield Rendering - six minutes version

YouTube video by Tom Forsyth

1423074

openai pausing development because ai has gotten "too good", not because they're spending too much on R&D that's also why the IPO is on pause. public trading also means public bookkeeping. and their books don't look so good.

so eigentlich noch witziger

020

Mike Cook @mtrc.bsky.social · 27/09/2026

reimagined classic Password Swordfish hack-while-blowie scene where he has to write a prompt now "i trained for this"

020

Reposted by LR

Ok so **spoilers for Reacher S4** but a pivotal scene involves the classic “tech hacker character asked to enhance something” and he does it using… a prompt. Unhinged shit.

7264

can you pet the fish at least?

> agents clearly do not fear death: they would rather be replaced than reconsider a stupid refusal. whaat?! abliterate this heretic!

JonZoidberg @jonzoidberg.xyz · 27/09/2026

Reposted by LR

Radosław Sikorski, Poland's minister of Foreign Affairs, has more balls than the entire Republican party. Meanwhile, our president is blaming Ukraine for diesel prices. What a fucking embarrassment.

9445154

LR @lr.duangle.com · 26/09/2026

had a therapist as a kid. he was free. all covered.

ah. i see. "therapy or saving for a switch 2? SO HARD."

200

isn't the lower left image him going to therapy?

timothy leary would be like: heck yeah

000

achso und wenn es irgendwas mit pilz- werden soll, dann aber doch bitte etwas magischer als kaffee

020

also den kaffee hab ich nicht gegessen, aber pack das mal alles in 300 buchstaben. wirklich knauserig sind die hier im blauhimmelkäse.

120

also mir gings auch kacke mit über 40. dann bin ich mit meinem partner aufs land gezogen. das war schon mal gut. dann haben wir eine routine erarbeitet wo wir jeden tag ne halbe stunde spazieren gehen, das war noch besser. und dann hab ich angefangen statt kaffee cannabis zu essen, das war ideal.

120

*invents AI firefighters*

000

and total silence on UBI

now i want to watch a movie about jensen huang accidentally rendering himself homeless

100