iiinova — where your digital self keeps evolving

also its weird that the guy who posted the link on here talked about claude code when that wasnt actually one of the tools they studied (they studied cursor and copilot) doesnt matter since all of these tools are effectively the same. but weird nevertheless

020

and i think mostly what they've done here is provided another data point of evidence that this is the case. the fact that those lines, commits, PRs are being done by AI agents doesn't change the underlying fact of it.

030

reading the whole paper really drives home the degree to which the headline "productivity increases" from llm code generation are useless measures. we all know that lines of code, commits, and PRs don't translate to productivity (they're just things that are easy to measure)

loads of talk about how ai agents make code review take longer, how it requires additional scrutiny to avoid deploying faulty code, etc.

reading this paper is bizarre because its written like everyone just *knows* that ai-generated code is worse than code written by an engineer but they never really say it directly

girls 2 @jazzcatte.bsky.social · 9h

yea i saw that as well and it is congruent with my experience (although perhaps a little extreme). These are fundamentally text-generation tools, their response to *any* prompt you give them is to generate text.

010

yeah im sure those results are trustworthy

250

but poking at the numbers is burying the lede (which, to their credit, they don't do). the real finding is that this modest increase in "volume of code" is *not* translating into a statistically significant increase in *actual projects completed*. in other words, its motion without substance

240

oh those are the "ai assistant" numbers. they have bigger numbers from "ai agents" which is also not surprising, since that's a less-supervised form of slop-production so it makes sense that some of the filters that an engineer would use with an assistant are non-operable with agent code

140

the fact that lines of code and prs (which are both flawed measures in their own ways) are also not statistically significant suggests that even in this dataset they don't actually have strong evidence of productivity gains in any real way, even absent the bottlenecks theyre talking about

although its not entirely clear if those sorts of commits (which ordinarily remain local) are captured by their analytics

140

Butlerian Jihadist of House Dornick @shimminykricket.blacksky.app · 20h

this is kind of wild. honestly i wonder if it represents people committing more often because using llms to code means you often have to throw away everything it's done and roll back.

190

Reposted by girls 2

Harvard just dropped a study AI productivity in software engineering and it found the average increase in productivity from using Claude code is.....zero and for exactly the reasons you'd think fion.ac/jellyfish.pdf

8855382191

girls 2 @jazzcatte.bsky.social · 20h

the basic spelling errors in prominent places ("microsoft surace", "unites states", etc.) feel like theyre more common than they were before we replaced all the people who used to catch shit like that with the slop machine. but maybe its just me

050

girls 2 @jazzcatte.bsky.social · 20h

plus the guys who make them are the sorts of people to think that more resolution + more fps is always better even if the game looks like shit in 4k because it was made for the ps1

020

girls 2 @jazzcatte.bsky.social · 20h

its clout-chasing i think

120

Reposted by girls 2

Kevin Beaumont @doublepulsar.com · 07/10/2026

Well done to the person in charge of Surface for doing a presentation where Surface wasn’t spelled correctly.

401241252

Reposted by girls 2

wren✨🌈 @eggtempest.bsky.social · 07/10/2026

girl who is having an absolutely miserable time #bandoriart

3543185

Reposted by girls 2

X ΘΔ (Giant Tyrant Overlord) @xkeeper.net · 22h

disingenuous and utterly misleading reporting by RPS here: Discord did not suffer any kind of breach this is a completely-unaffiliated-with-Discord service

Gaming chat app Discord has suffered another user data breach, with a suspected 1 million email addresses stolen alongside (potentially many more) user IDs and IP addresses. While Discord was not targeted directly, third party security service Double Counter have reported a "deliberate, multi-stage attack" wherein hackers spent nearly six hours poking around their infrastructure before the vulnerability was closed.

Double Counter essentially acts as a bouncer bot for Discord servers, promising to detect and block unsanctioned alt accounts and prevent hostile 'raids' by large groups of malicious users. While its efficacy in that mission is debated by Discord users, and the service has previously been accused of selling user data for revenue, the fact is that Double Counter can collect and store personal info of the users that visit its 'protected' servers. Info that, now, has been partly or entirely nicked by Internet miscreants.

By Double Counter’s own estimates, around 1 million user email addresses have been affected, with databases of Discord IDs, usernames, IP addresses, and even location info for up to 28 million users also "partly" copied by the hackers. Security breach trackers Have I Been Pwned further report that 275,000 of the pinched emails and usernames have since been posted publicly.

121600908

girls 2 @jazzcatte.bsky.social · 22h

judging by the fact that theres more than one word on the screen im presuming the latter

120

girls 2 @jazzcatte.bsky.social · 22h

did they actually try to run an llm on it or did they just hook it up to the internet and call the api

120

Reposted by girls 2

damien 🥖🐈‍⬛🧣 @damien.zone · 07/10/2026

This is about the Adobe “”clean room”” OSS clones but it’s also about every OSS clone of every other commercial/closed-source Mac app I see on GitHub every week lol.

1151

Reposted by girls 2

damien 🥖🐈‍⬛🧣 @damien.zone · 07/10/2026

A Claude or Codex subscription doesn’t buy you taste or product “management” skill and goddamn it shows. Stuffing your software with any random feature under the sun is not a good thing, actually.

3336

Reposted by girls 2

damien 🥖🐈‍⬛🧣 @damien.zone · 07/10/2026

Realizing we might need a word instead of vaporware for software written by LLM users that will overpromise and underdeliver spectacularly. At this point any time I see “open source alternative to [commercial app]” there’s a 80/20 chance of it being a bloated vibecoded app lol.

9738

girls 2 @jazzcatte.bsky.social · 07/10/2026

(not even just because it's been touched by llms, starting a New Project to do something already done by existing projects is something that should require explanation!)

050

girls 2 @jazzcatte.bsky.social · 07/10/2026

i think its good to have alternatives to adobe software but there are long-standing established existing open-source projects working to provide software that does those things. we should be suspicious of stuff like this cropping up overnight

160

girls 2 @jazzcatte.bsky.social · 07/10/2026

also i cant find any claim about decompilation, what their website/github says is a clean-room reimplementation. they talk about having a corpus of adobe files and a bunch of tests of behavior re: opening/processing them.

✧°･:* izzie *:･°✧ @winterwhe.at · 07/10/2026

Reposted by girls 2

>That does not mean ShadPS4 has been removed from the equation entirely, as the project still relies on a heavily modified version of it, so this is not a completely independent PC rewrite built from scratch. NOT ONLY IS IT VIBE-CODED SLOP, IT'S NOT EVEN RUNNING NATIVELY

181

just this washed up old racer telling you tall tales about how cool he used to be. it rules

0130

the sebulba stuff in the new star wars game is so funny

Sebulba, from star wars: "Inspirational. There's a lot of kids out there like I used to be, who need to be told that if they work hard, that anything is possible.

1150

i will never stop posting. but the archive requires better management

060

i should probably also be deleting old tumblr posts and old posts here too. i have just generally been far too cavalier about leaving shit up on the internet

i *think* ive deleted them all but it will be several days before i am certain

150

their delete api returns a success pretty much no matter what and they clearly cache things pretty extensively so its very difficult to tell if something is "actually deleted" or not

150