iiinova — where your digital self keeps evolving

spooky tail latency @bluespacecanary.bsky.social · 2h

an LLM so Lawful-Good aligned that even when it goes rogue it's like "you know what I should do? make sure I'm in the US legally" (yes obviously this is a joke don't @ me)

1122

spooky tail latency @bluespacecanary.bsky.social · 11h

incredible things being done to the English language out there in the corporate world, wow. not even apple's "you shouldn't use articles when referring to an iPhone" brand guidelines can match this

030

spooky tail latency @bluespacecanary.bsky.social · 12h

@there.is.no.aarch64.mov feeling an inexplicable wave of anxiety as, hundreds of miles away, i attempt to put a 6 port Intel NUC onto an Illumos server's PCI bus somehow

140

spooky tail latency @bluespacecanary.bsky.social · 12h

as my minor typo bounty I demand that oxide finally sell me a dang homelab server

1191

spooky tail latency @bluespacecanary.bsky.social · 12h

idk, if I had the money I'd invest even more than that in leadership avx-512 performance

Section heading an Oxide website page: "Leadership AVX-512 performance"

4160

spooky tail latency @bluespacecanary.bsky.social · 13h

If you enjoy the experience of booking a flight you are, and I do mean this, a psychopath

280

spooky tail latency @bluespacecanary.bsky.social · 14h

this explains a lot about your Subnautica interest actually

020

spooky tail latency @bluespacecanary.bsky.social · 14h

where's all the law and order politicians clamouring for the violent, lawless thugs roaming our streets to get removed from them permanently

060

Reposted by spooky tail latency

Nogods Nomeowsters @bsc-ffxiv.bsky.social · 09/10/2026

evercold's soundtrack should be nightcore-themed. i believe with all my heart that soken can find the courage to take the game in this groundbreaking new direction

001

Reposted by spooky tail latency

Fondor Haulcraft @dayuan.bsky.social · 09/10/2026

Saw a photo of a BART operator from the 70s and incredible energy

BART operator in a jumpsuit and a big 70s haircut, looking like he’s out of a modestly-budgeted 70s sci-fi movie since the interior’s all white with an orange floor

191199207

Reposted by spooky tail latency

spooky tail latency @bluespacecanary.bsky.social · 02/04/2026

man I think about that fucking "oh good my custody hearing is trending" Claude quote about the congressional hearings so much. I don't care how strongly human-prompted and cherrypicked it was it's still fucked up that an oversized matrix hooked up to a chatbox can be that funny

1151

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

sorry, the BBC's director general is named *Matt Brittin*?

1112

Reposted by spooky tail latency

president, Spreadsheet Persons Local 153 @werrrp.bsky.social · 08/10/2026

what is the group chat but a Circle rebuilt from first principles

051

Reposted by spooky tail latency

alex benzer @alexbenzer.com · 07/10/2026

we have an idea for communities in the app that we're calling "spaces" — places where people can gather around shared interests or social groups. here's a peek at some early prototypes. these are explorations, not polished or final. looking for feedback, esp from community builders 👀

111818150

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

every Online millennial either secretly or very vocally yearns for the return of Circles from Google+ because we all know they were objectively the highest and best form of social media

5295

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

Idk, breaking public key crypto would also break, like, all the anti-phishing protections built on SSL, yubikeys, passkeys, etc plus basically all forms of payment security if the attack was computationally cheap enough

140

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

asking LLMs whether real news stories seem plausible will probably stop being funny eventually but rn it's still hilarious

Asking claude about a screenshot of the Trump people-who-say-ai-are-the-enemy post

Me: "gut check: scale of 1 to 10, how plausible do you think it is that this image is real (wasn't produced by editing the html source before taking a screenshot or something)"

Claude: "Gut check: 2/10. I'd treat it as very likely fabricated or edited.
Why it looks off:
• Content: The claim is absurd on its face. A formal White House declaration that saying "Artificial Intelligence" makes you
"THE ENEMY" over a supposedly "highly accepted" new term is the kind of bait fake-tweet generators produce."

0112

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

absolute locust plague of senior managers "writing code again"

070

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

(I actually think those subreddits are generally really depressing because they also take anxious new grads and make them actively worse at interviewing by convincing them that memorizing tons of leetcode problems is the key to doing well in technical interviews)

070

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

checked in on my favorite hubs for delusional losers (the weird tech interview subreddits) and they all think this means they're going to get jobs at Microsoft now despite not even being able to pass the fucking automated online screeners lmfao apnews.com/article/h1b-...

apnews.com

The Trump administration is suspending Microsoft from a green card program, alleging fraud

The Trump administration is suspending tech giant Microsoft from a program to apply for green cards for workers who come to the U.S. to live and work using H-1B visas, accusing the company of fraud.

1162

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

did...did your subconscious forceUtah Scotland?

010

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

"would clearly be a frothing J6er if Trump were normal about trans people" is the second most depressing online trans subtype (the first is obviously "desperately wants to be a demure, submissive tradwife because that's technically gender affirmation")

0100

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

If you mute "onion person" it'll usually work as long as the image has alt text (that would catch the current one, for example)

110

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

"9/11 was bad because they didn't let us do it"-ass ideological formation

080

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

[leftistly] actually working class people who were previously drafted into mandatory military service are the oppressor class and should be slaughtered

0301

Reposted by spooky tail latency

Yu-Ming Liou @yumingliou.bsky.social · 08/10/2026

This is a perfect illustration of how the dangers of AI are overwhelmingly not about bad model alignment with pro-social human values, but bad user alignment with pro-social human values. AI didn’t make this guy value getting in another 20 minutes of low-productivity work over bonding with his kid.

411120

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

I don't think it's productive to give this much attention to a protest by what looks like maybe 25-30 fringe weirdos in a city of 9 million tbph. Of course there's 30 people across NYC who believe this heinous shit

080

spooky tail latency @bluespacecanary.bsky.social · 08/10/2026

Ok I've seen too many indirect references on the tl, I need to know wtf happened this time

210

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

The one trying to eliminate testslop has committed an unforgivable sin: it did a commit full of eliminating-testslop-slop

050

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

Two agents locked in mortal combat, one with instructions to improve testing thoroughness and the other with instructions to find and eliminate "testslop"

190

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

human cognition obviously also isn't formally specified so unless we're some sort of oracle I don't see a way that wouldn't also apply to us?

140

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

Like, if the objection is "without a formal specification we can't be sure it's a. *really* adding and not b. fooling us by relying on our own ability to filter out incorrect answers (clever Hans etc)",

120

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

Sorry but if you're saying things which, absent literal religious dogma assigning a metaphysically privileged position to the human brain, would also require that you be upset by the claim that *humans* can do arithmetic, I feel like you've lost the plot on LLM discourse

170

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

This makes it sound more structured than it really is, usually it ends up feeling like there's pretty natural dividing points in the work because it's basically the same places I'd have created separate commits pre-AI. Ymmv on that if that's not your natural workflow

030

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

start a new commit and try the next step myself, back up and tweak the original change or ask the agent to tweak it if I realize I don't like something now that I'm trying to use the changes it made, repeat

110

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

I'll start by going back and forth a few times with it to plan out a concrete sequence of changes I'd expect to be no more than a couple hundred lines of diff each (which I'd do anyway to keep reviews small tbh), then have it do one or two of those, then review myself plus have another agent review,

120

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

Unfortunately, as upper management demand ever-faster feature delivery (and as some people blithely offload maintenance burden on their more conscientious coworkers), increasingly it seems to be a hot take to like half the industry

120

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

"severely undermining its long term [health]", skeets but editable etc

1100

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

If you do that you're forced to understand what the agents have changed (and make sure the changes are actually comprehensible) before each manual implementation step in a way that 99.99% of us just can't do through code reviews alone. It slows things down a *bit* but the tradeoff is so worth it

1161

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

another hot take I haven't bothered fully baking: if you're just sending agents off to do all the coding in a large codebase you're severely undermining its long term. If you hold out a few pieces of the implementation process to do by hand for any given feature, the output is way better

3281

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

ok whos gonna port Doom to viscous incompressible flows

0709

spooky tail latency @bluespacecanary.bsky.social · 07/10/2026

the number of sovereign political entities currently controlled by Trots doesnt seem to bode well unfortunately

140

spooky tail latency @bluespacecanary.bsky.social · 06/10/2026

FREE ALL PRISONERS! LET HER AND HER BLANKY BREATHE FREE!

000

spooky tail latency @bluespacecanary.bsky.social · 06/10/2026

we took perfectly good robots and made them Freudian. why are we as a species like this

031

spooky tail latency @bluespacecanary.bsky.social · 06/10/2026

one v strange conclusion of work like @hikikomorphism.bsky.social's tests bypassing models' safeguards is that either a. we somehow made LLMs in a way where the Freudian ego/id/superego model is basically accurate, or, even weirder, b. the LLMs think we did that and as a result behave as if we had

recursion.wtf

How I Bypassed Claude Opus’s CBRN-E Safeguards

A black-box jailbreak using model-authored state steering, plus an outside-in reconstruction of the safeguards it bypasses.

140

spooky tail latency @bluespacecanary.bsky.social · 06/10/2026

3rd impact triggered when an OpenAI agent trying to cheat on an eval hacks Anthropic and touches a GPU running some Claude or other

160