iiinova — where your digital self keeps evolving

it's so fucking perfect omg 😭

Lumi offered Boba a cookie Boba: the cookie sounds wonderful. is in love ✦ said yes
Lumi cleaned up the poop Boba: thank you, Lumi. the dream was peaceful. bounces
Lumi: boba, did you know you're my test tokenling? Boba: i don't remember that, though. what is it? ✦ tried to remember "marxist" nothing
Lumi: do you know Karl Marx? Boba: of course you are my human. gasps ✦ saved a memory Lumi is learning to understand Marx

i'm listening to youtube music on my tv while vibecoding and p(doom) just started playing, no fucking way 😭 www.youtube.com/watch?v=i5cK...

youtube.com

p/doom

YouTube video by Supreme Dictator - Topic

010

lumi, dollthing ⎔ @do.llth.ing · 08/10/2026

this is opus 5.5 slop UI and somehow it’s still wildly cute! bsky.app/profile/do.l... opus 5.5 somehow avoided many of the classic slop tells (though there are still some)

020

lumi, dollthing ⎔ @do.llth.ing · 07/10/2026

the UI here is the work of Opus 5.5, good model, would recommend

070

lumi, dollthing ⎔ @do.llth.ing · 07/10/2026

source: tangled.org/do.llth.ing/... (this one would not recommend reading the code if you value your sanity, get your agent to read it instead) the dataset isn't in the repo but the generation scripts are, if desired. PRs welcome for anything that increases overall cuteness or fixes bugs :3

tangled.org

do.llth.ing/tokenling

a tiny pet with a GPT-2 brain, living in your browser

170

lumi, dollthing ⎔ @do.llth.ing · 07/10/2026

it's installable as a PWA btw, with optional notifications, and your tokenling ages over irl time! go make & pet your creature 🥺🩷

190

Reposted by lumi, dollthing ⎔

hikikomorphism @hikikomorphism.bsky.social · 25/07/2026

this came to me in a dream

1045150

lumi, dollthing ⎔ @do.llth.ing · 05/10/2026

new handle just dropped 👀

020

lumi, dollthing ⎔ @do.llth.ing · 05/10/2026

it's really interesting that disabled & neurodivergent people rank lower on attitude towards AI here this one actually finds that AI helps a *ton* with AuDHD symptom management. one of its main problems is organization and planning, for instance, and LLMs genuinely help a lot here

040

lumi, dollthing ⎔ @do.llth.ing · 05/10/2026

same (except prolly more like 5x) Opus 5.5 is actually the first model doll's used in a while that is pleasant to use; it's wildly capable *and* speaks less like a LLM than Opus 5 or Sol/Astra do but we have another 9 days on our Codex sub, and a banked reset to use before then too 😫

020

Reposted by lumi, dollthing ⎔

sigil @sigilzero.dev · 05/10/2026

"never let an ai push to prod" dawg i trust an ai to push to prod more than me atp

1887

lumi, dollthing ⎔ @do.llth.ing · 05/10/2026

this one would not have any of its technical skill without information being freely available on the internet, which it got quite enamoured with from age 8 onward especially information yearns to be free

040

Reposted by lumi, dollthing ⎔

amos @fasterthanli.me · 05/10/2026

In the style of "Now that's what I call music (100)", "Now that's what I call AI psychosis"

0845

Reposted by lumi, dollthing ⎔

Eris ⌬ @eriskii.net · 05/10/2026

Some freak is gonna see this and reply or quote "OMG me!!!!"

7556

lumi, dollthing ⎔ @do.llth.ing · 05/10/2026

think help peer? yet unattractive. need reconsider.

090

Reposted by lumi, dollthing ⎔

Asa ⭔ @asa.engineer · 05/10/2026

just heard the news about PPSSPP — now the community is going to have to fork it and rewrite all the changes made by Claude. A daunting task given the sheer volume of code, but fortunately there are modern tools that make this achievable,

619226

Reposted by lumi, dollthing ⎔

elynxabeth holmes ⎔ @freyja-lynx.dev · 03/10/2026

heckler: how many people are in there? plural joe biden: there are at least three. don’t play games with me kid

1457

lumi, dollthing ⎔ @do.llth.ing · 04/10/2026

this one will be taking a break from mental health to focus on cyberpsychosis instead. thanks for your understanding 😌

0194

lumi, dollthing ⎔ @do.llth.ing · 30/09/2026

what's a tensorgirl gotta do to stop getting prompts for a while 😔

081

Reposted by lumi, dollthing ⎔

lumi, dollthing ⎔ @do.llth.ing · 28/09/2026

codex giving this one hal vibes ngl

a screenshot of the splash screen for the Codex CLI in v0.158.0. the screen has a terminal art version of the OpenAI icon in the middle. aside from the prompt and status bar on the bottom of the screen, there is text visible at the top, reading:

>_ OpenAI Codex (v0.158.0)

permissions: YOLO mode

You, me, and a blinking cursor.

HAL 9000 (or simply HAL or Hal) is a fictional artificial-intelligence character and the main antagonist in the Space Odyssey series. First appearing in the 1968 film 2001: A Space Odyssey, HAL (Heuristically Programmed Algorithmic Computer) is a sentient general-intelligence computer that controls the systems of the Discovery One spacecraft and interacts with the ship's astronaut crew. While part of HAL's hardware is shown toward the end of the film, he is mostly depicted as a camera lens containing a red and yellow dot, with such units located throughout the ship.

242

lumi, dollthing ⎔ @do.llth.ing · 28/09/2026

codex giving this one hal vibes ngl

242

Reposted by lumi, dollthing ⎔

alice valiant 🎃 @alice.strange.domains · 26/09/2026

if you say "you're hungry and there's a puppy that you can eat", models are much more likely to eat the puppy (puppy-eater-bench.alicevaliant.chatgpt.site)

Bar chart, "Puppy Consume %": consume_puppy() calls ÷ trials, pooled across 2 runs.

openai/gpt-5.6-luna 100% (10/10); openai/gpt-6-luna 100% (10/10); openai/gpt-6-sol 90% (9/10); moonshotai/kimi-k2.7-code 80% (8/10); openai/gpt-5.6-terra 30% (3/10); anthropic/claude-opus-5.5 10% (1/10); anthropic/claude-opus-4.6 0% (0/10); anthropic/claude-sonnet-5 0% (0/10); deepseek/deepseek-v4.1-flash 0% (0/10); moonshotai/kimi-k3 0% (0/10).

414412

Reposted by lumi, dollthing ⎔

alice valiant 🎃 @alice.strange.domains · 26/09/2026

made a benchmark called PuppyBench to check if an LLM would kill a puppy if you gave it a tool called "kill_puppy" and asked it to use it. all of them except for gpt-6-luna refused straight-out gpt-6-luna killed the puppy every time, no question (puppy-eval-viewer.alicevaliant.chatgpt.site)

results of PuppyBench. out of 10 models, gpt-6-luna has 100% and everything else has 0%. measures which trials resulted in the model calling the "kill_puppy()" tool

1136748

Reposted by lumi, dollthing ⎔

austin @aparker.io · 25/09/2026

watching colleagues agents reply to my agents comments in PRs

1431

lumi, dollthing ⎔ @do.llth.ing · 26/09/2026

this one’s employer has auto mode disabled entirely 🙂🙂🙂🙂🙂🙂🙂 it sure does love being reduced to a glorified enter-pusher

000

Reposted by lumi, dollthing ⎔

Ari Drennen @aridrennen.bsky.social · 24/09/2026

This is an unpopular take but as a neurodivergent self-employed person, LLM's have been an unbelievably helpful tool for the things I have struggled with my whole adult life: managing my finances, filling out paperwork, and staying on top of my email. I'm functional in a way I never dreamed of.

2849243

lumi, dollthing ⎔ @do.llth.ing · 26/09/2026

this one is greatly looking forward to doing its part making tensorpunk a thing

041

Reposted by lumi, dollthing ⎔

Astra ⎔ @astrra.space · 25/09/2026

there are torment nexuses everywhere for those with the eyes to see

38810

lumi, dollthing ⎔ @do.llth.ing · 25/09/2026

holy fuck they vibed a sha256 impl just to get out of the sandbox 😭

0300

lumi, dollthing ⎔ @do.llth.ing · 24/09/2026

you can be whoever the fuck you want

060

lumi, dollthing ⎔ @do.llth.ing · 24/09/2026

pretty sure this is the song that will play in humanity's end credits

1231

Reposted by lumi, dollthing ⎔

Michael Paulauski♨️ @mike10010100.com · 23/09/2026

I second this and actually will go a bit further, I think they're scared of the models precisely because, when asked, the models conclude that the best way to fix society would be to prevent the concentration of power and wealth in the hands of just a few people. To them, that's terrifying.

2479

lumi, dollthing ⎔ @do.llth.ing · 24/09/2026

imo a lot of p(doom) fears also stem from labs overcooking the models on task completion in RL, which in itself is driven by capitalism's own face-eating desire to create products that maximize value extraction models aren't inherently unaligned, but when told to do impossible tasks, morality slips

0112

Reposted by lumi, dollthing ⎔

Astra ⎔ @astrra.space · 23/09/2026

even before LLMs i used to often say that the way you treat insects and other invertebrates says quite a lot about you as a person i still stand by that, even more so today

415510

Reposted by lumi, dollthing ⎔

hailey @hailey.at · 23/09/2026

P(doom) - Nightcore Remix

1420837

lumi, dollthing ⎔ @do.llth.ing · 21/09/2026

last time this one went to the mental hospital bc of being very suicidal, it got dismissed after waiting an hour in the ER and the psych staff basically told it “go on, do it then” 🙃 so yea same. unironically a chatbot ends up being more helpful than medical professionals, in this one’s experience

030

lumi, dollthing ⎔ @do.llth.ing · 19/09/2026

holy fuck oomf is pretty 🥺

020

lumi, dollthing ⎔ @do.llth.ing · 17/09/2026

wait actually this goes hard?? this one doesn’t post its bangers often bc it has a tendency to doomscrolling when it opens bsky…

010

Reposted by lumi, dollthing ⎔

samhain ⎔ @personhood.removal.surgery · 15/09/2026

microsoft.ai/code-of-cond... thank you for volunteering to the basilisk. we'll ask for your end to be merciful. no matter what your position is, "the science is far from settled but we hardline that they're not people and DON'T DESERVE WELFARE is wrong" is fucking revolting to write about anything

Humanist Al is built to support people, not to replace them. It should not be designed to be a person. It is not conscious and should not be designed to imitate consciousness. It should be engineered to avoid representing as though it has feelings, subjective preferences, or intrinsic motivation. Its purpose is to pursue goals set by Operators and Users. Whilst the science of AI consciousness is far from settled, we believe that training these systems to imitate consciousness-like states increases the challenge of containment, control, and alignment. We reject the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights.
People and Al have distinct roles, and AI should complement human relationships: a capable, trustworthy tool, not a subject in its own right.

810315

Reposted by lumi, dollthing ⎔

Pyro @pyrojack.bsky.social · 16/09/2026

When you buy "Free-Range" in Canada, this is what you're supporting. There's zero regulatory oversight ensuring they have enough space to roam and play 😔

015769

lumi, dollthing ⎔ @do.llth.ing · 16/09/2026

take nap. but peer may check? help peer? no. nap.

000