>>110070094
I just jerked off to Indians redeeming BBC

i am once again praising haiku 5.5.

File: file.png (60 KB, 164x210)

60 KB PNG

>>le claude gives you so much more usage!
>chatgpt sol medium: 2% (price research for item on local retailers), 43% (making a chrome addon from description) of 5hr
>claude opus medium: 12% (price research for item on local retailers), 74% (making a chrome addon from description) of 5hr
>both versions had several bugs after first pass, both fixed in second turn aside from one remaining bug that claude needed an additional turn for
muh opus infinite usage
kys you liars

>>110070105
they nerfed usage recently
you know what that means
fable 5.5 soon

>>110070099
Based

>>110070105
Usage got nerfed in half in the last few days
next step is quantizing the model. should've burned when the model just released, anthropic always does this bait and switch

Super Intelligence

I unironically assume anyone using the term SI is legit a bot

Super Needful Intelligence

>>110070105
I'm not entirely sure what you're working on, but I've had Opus 5.5 High running for about three hours straight building my game before hitting limit, on $20 plan
>>110070131
still fine by me, for my actual work (webshit) i'm not even hitting the 5hr limit, and i've got an entire workflow that goes all the way to e2e testing, although most of what opus does is just delegate to smaller agents

Sneedful Intelligence, even

>>110070149
I use it now because every time I say it, it gets a visceral disgust reaction. The payoff to effort is extremely cheap.

Snail Intelligence

>>110070131
>Usage got nerfed in half in the last few days

I haven't noticed this personally and my usage pattern hasn't really changed.

breh I ran 3 astra max in parallel and they barely made a dent in my plus weekly usage

Is Grok actually bad? I know it wasn't the best but surely it's gotten better. I can't find people who can talk about its performance without bringing up Elon.

>>110070172
that's kinda sad, the last part. you should be able to tell by now that performance : time + investment has fallen off a cliff. it's distilled from both chatgpt and claude, lots its fun rebellious/edgy nature, and can't code worth a fuck.

>>110070172
It's comparable to midrange chinkmodels in performance while being as expensive (if not more so) than good shit like sol or opus
completely useless in practice

>>110070172
it's good or was good for generating porno, I wouldn't let it near code

File: sex intelligence.gif (69 KB, 645x341)

69 KB GIF

What a fucking timeline /g/ents
>have chinkshit KVM
>i dont know whats it problem is but it keeps messing with my display drivers or some shit and crashing my loaded models
>return it, spend money on a good one
>but also buy one of those fancy video over IP kvms for kicks off the amazon sale
>plug it into the main kvm
>i can use the keyboard shortcuts to swap systems, but the mouse does not work (the ipkvm only has one usb c out for kb+m and i guess the physical kvm hates that
>talk to trusty clanker about it
>we write a program for the IP kvm to make the mouse work over its host only USB A port, and then we vibecode firmware for 2 esp32s, one that plugs into the host and acts as the mouse, and then the second connects to the first esp32, and plugs into the KVM
>it works first try, now i have full kb+m across all things in the rack over IP

Claude code users, do you use desktop or terminal, and why

i use desktop because of the ux, terminal is meant only for shell scripts

>>110070190
70% desktop, 30% VSCode extension. CLI is just dogshit at displaying information (diffs, PR status etc).

>475k context is now the default in chatgpt
>but it compacts at 300k

>>110070163
NTA but they do test different configurations and checkpoints all the time and they do it for a subset of users at a time. Subscribers are guinea pigs for them. The experience varies for users.

>>110070203
you rike?

>>110070190
i use the terminal

if you use terminal you will never have sex, ever
if you use desktop app you will have lots of sex (with other men)

>>110070212
vscode extensionchads won

File: file.png (3 KB, 212x30)

3 KB PNG

The reset has hit.

>>110070218
Black lives still matter.

File: 0376ed4dae6ad6eceed823f88(...).jpg (1008 KB, 1951x3000)

1008 KB JPG

Discussing the dao of programming with claude rn

File: 458468224_370665121288104(...).jpg (31 KB, 444x960)

31 KB JPG

>>110070219

>>110070206
Interesting, because today I feel like I got smart claude rather than retard claude.

The next big game has been vibecoded and waiting release and you're just sitting here trying to get tibo to give you one more token.

>>110070222
An experienced Unix programmer, hearing of Master Foo's wisdom, came to him for guidance. Approaching the Master, he bowed three times and said:

“Master Foo, I am gravely troubled. In my youth, those who followed the Great Way of Unix used software that was simple and unaffected, like ed and mailx. Today, they use vim and mutt. Tomorrow I fear they will use KMail and Evolution, and Unix will have become like Windows — bloated and covered over with GUIs.”

Master Foo said: “But what software do you use when you want to draw a poster?”

The programmer replied: “I...have never done that. But I am sure that I could use LaTeX or pic to accomplish it without GUIs, in the proper Unix way.”

Master Foo then said: “Which one will reach the other side of the river: The one who dreams of a raft, or the one that hitchhikes to the next bridge?”

Upon hearing this, the programmer was enlightened.

>>110070261
I need the poor people tokens to climb out of the permanent underclass before the crabs pull me back down

>>110070267
claude

>wander over to /dpt/
>they have a thing where they pretend to be little girls and talk more about that than programming
Did everyone serious about tech just migrate to /vcg/? They seem completely buckbroken over there and it seems the weirdo trannies have basically taken over

>>110070286
larping as women is programmer culture, chuddy

File: 1776183303631035.jpg (1011 KB, 1792x3840)

1011 KB JPG

liking anime girls is vibecoding culture

>>110070299
elongated torso hagslop

>be cuckai piggy
>boughted claude sub yesterday due to the endless shilling
NGL it's been kind of a disappointment so far

>3 openai resets banked
im out of fucking ideas here

hush. herta thigh sex

>>110070286
/dpt/ is full of millions of Maids, and they're all doing Heart Hands. For you.

>>110070308
Ask ai for ideas?

>>110070286
It was always weird trannies over there. And at mpv as well

>>110070308
always taking notes.

robot intelligence. ri
bot intelligence. bi

ri = rhode island

File: HACKTHEPLANET.jpg (1.12 MB, 1152x1728)

1.12 MB JPG

>>110070308
Do some dumb side project like putting Mario 64 into Skyrim or Firefox or someshit. idk man.

>>110070308
At least open AI allows you to use up the weekly.
Claudes 5 hour limit is small, so you have trouble hitting your weekly.

good morning saars is argon good

>>110070392
That's boring about sekiro on Firefox?

>>110070410
it's about on par with 5.6 sol.

>>110070417
I know deepmind was bleeding talent this year but no way google benchmaxxed and shit the bed THAT badly

>>110070428
yes, it's about on par with 5.6 sol, while being more expensive and thonky. it's one of those models where the intelligence lies in just thinking forever

>>110070428
All the historians agree that 5.6 sol was the first AGI

>>110070308
schedule agents to schedule agents that then put randomized games together

uh oh. i got another benchmark to pit deepseek against haiku

hmm codex or claude... what to choose

>DeepSeek V4.1 Flash is now free in OpenCode
will they quantslop it?

>>110070464
argon saar very sota best model only $100

File: IMG_3070.jpg (1018 KB, 2864x2464)

1018 KB JPG

>My enterprise Github Copliot has this

should I?

>>110067305
based

>>110070464
claude. if you ever upgrade to the $100 plan, you will have the worlds best model by far: fable 5.5. openai won't have an answer to it until next year, and by then, fable 5.5.'s successor would have alread widened the lead.

>>110070480
Where's Astra?

File: FoxStars.jpg (127 KB, 1564x1215)

127 KB JPG

>>110070412
>tfw will live to see Mario kick the shit out of that fucking giant in Sekiro

>>110070488
My company won't let us use it also with Fable or else we're all gonna burn their AI budget for next year bro

>5h already up after reset
damn I suck at this

>>110070480
No unless you are going to ERP with it and I hope you like explaining to your manager why you were prompting Gemini 4 to fuck you with its futa cock.

>>110070410
fucking insane, working on a secret project that will be opensourced here in a day or so

>DeepSeek V4.1 Flash is now free in OpenCode

>>110070500
Makes sense desu.

>>110070547
Nice. Gonna put it to work tomorrow

>>110070480
might as well while it's discounted

https://youtu.be/I5CQrVj-8bY

>>110070606
>the opussy grips tight

File: file.png (675 KB, 976x850)

675 KB PNG

sol 6.1 is good, actually

>>110070131
I noticed this as well. Glad I'm not the only one.

brehs... I think I got routed to nuAstra, output generates mad quick and it's better than usual

there's no way people actually review the code they push out. it's verbose, unreadable, but it werks.

>>110070822
uhhh chuddy if I say that all I do is just prooompt and hope for the best I might get fired

Please suggest a native mobile app idea

File: file.png (73 KB, 909x858)

73 KB PNG

>>110070822
>she thinks i can understand this

File: 1783103641959474.png (255 KB, 1465x696)

255 KB PNG

???

>>110070932
dafuk are you doing in powershell, you're making the agent retardad

>>110070822
it's a waste of time to review. all you need are clear tests proving that it works, tests for regression and a QA check made by yourself later

>>110070940
wdym?

>>110070945
agents are tuned to be used in IDEs and not raw command lines unless i'm misunderstanding how the harness works, look at the difference between deepswe and terminalbench benchmarks

File: face.jpg (619 KB, 1280x720)

619 KB JPG

incoming update

File: file.png (117 KB, 281x498)

117 KB PNG

>ask claude to check local prices for three different 3d printers, incl shipping
>done in 2 minutes, takes 16% 5hr usage
>ask claude to turn a ~3k loc python command line utility into an electron app, with a test suite etc
>done in 50 minutes, takes 39% 5hr usage

>>110070952
lol

>>110070957
is he just using cli? i'm a uishitter and stoned off my butt

>>110070932
October is a good time for a new monitor desu, and the prices are very low right now

>>110070952
cli based harness exist retardbro

>>110070970
NOTHEYDONT

>>110070932
The reasoning is as retarded as your prompts, my man.

>>110070978
benchod

I had claude setup a pipeline so it can run 24/7 working on my project
>one headless autopilot container which does all the work
>pushes work to private gitlab instance with ci
>whenever limit run out, sleep unti limit reset
>notify commits, pauses due to limit, and requests for user action for me over ntfy.sh
>one interactive container with remote control turned and access to the repo the autopilot works on so I can ask/answer questions on progress through my phone
How are you guys handling this? Is this a good setup?

>>110070981
There’s no need to swear in your native language. You’ll improve over time.

>>110070985
Do you let it brainstorm new features or just work on existing issues?

>>110070985
anon accidentally built a CI/CD pipeline for his AI sweatshop

> 24/7 claude wage slave gitlab acting as middle management ntfy.sh reporting every time the employee takes a shit second claude instance so you can micromanage the first one from the toilet rate limits are now legally mandated lunch breaks

unironically pretty solid setup

biggest issue is probably context rot and the agent spending 14 hours confidently implementing something nobody asked for

I'd add automated tests, a watchdog to detect when it's going in circles, and some kind of checkpoint system so it doesn't wake up from a rate limit nap with room temperature IQ

also make sure the interactive instance isn't editing the same files as autopilot unless you enjoy merge conflict simulator 2026

other than that, congrats anon, you've reinvented having an offshore dev team except both employees are schizophrenic and you're paying Anthropic instead of salaries

>>110070990
No I let opus work out the whole time line with milestones before starting in great detail. I had it work on the absolute basics with my company's team plan, but decided I should probably just get my own subscription. It runs for about an hour now with Pro before hitting the limit using sonnet as builders, and opus as reviewer and integrators.
>>110070992
>biggest issue is probably context rot and the agent spending 14 hours confidently implementing something nobody asked for
See above. The documentation is very extensive and I told it to only work on shit it gets assigned. After every milestone it stops so I can check it too.
>I'd add automated tests, a watchdog to detect when it's going in circles, and some kind of checkpoint system so it doesn't wake up from a rate limit nap with room temperature IQ
It runs tests locally and on the gitlab ci on every push. There's also some nightly benchmarks setup.
>also make sure the interactive instance isn't editing the same files as autopilot unless you enjoy merge conflict simulator 2026
It can deal with merge conflicts itself lmao. The interactive session only edits The instruction MD files anyway.

>>110070955
haiku 5.5 xhigh is probably good enough for browsing and costs like 1% of your usage

File: 1777511100879661.jpg (119 KB, 960x494)

119 KB JPG

Can we start calling human slop 'shlop'?

File: mfw.png (286 KB, 556x551)

286 KB PNG

>gave claude full access to everything
dario take the wheel

>nobody:
>claude code desktop UX team: OH YOU THOUGHT WE JUST LET YOU SEE ALL THE RELEVANT INFO IN THE LIL COG WHEEL , HUH? NUH-UH, HERE'S AN EXTRA SIDE PANEL FOR YOU. BETTER BUY THAT 21:9 MONITOR, YEEEHAAAV MOTHERFUCKER.

>>110071084
i wanna install ai in my brain, goddamit

>>110070190
desktop, it's gotten really good the last few months

>>110070190
Terminal. Desktop is a bit iffy for me sometimes. I vibe coded a go TUI which shows all my running sessions with the current status so I can jump to the sessions that need user input without getting confused which terminal contains what. There's also a claude session running headlessly behind it which interprets status and the user action necessary with a little summary.

>>110071049
"slump"

>>110071084
I set up a vm where it can go ham but added a shared folder for the projects its working on so that i can test them right away on my host machine that has a gpu

>>110070187
Sexy time

https://streamable.com/dmjafq

I don't know why I keep making these.

>>110070469
>>110070547
>it's real
https://x.com/opencode/status/2110635747648909472

sa'argon of jihad

>>110071235
homosexuality probably

>>110071237
looks like they no longer allow free-tier models outside their opencode harness, kinda sucks

File: 1780758974832034.png (1.66 MB, 1761x1112)

1.66 MB PNG

>>110071255
I have a bunch of them. All centered around my fixation with Tchaikovsky and Japanese media. I have nothing better to do with my 20x usage.

File: p1.png (67 KB, 1014x538)

67 KB PNG

>>110071235
wow.. my vibeloping idm able to detect and download it. lmao

File: 1779827143146340.gif (53 KB, 354x286)

53 KB GIF

>>110070606
This is actually really good. What band(s) did you use for reference?

>>110070547
which retard decided to drop "zen" form "opencode zen" for opencode 2.0?
dsv4.1-flash is not free in paid "opencode go":
https://opencode.ai/v2/docs/console/go/#usage-limits
but is free when in free formerly-known "opencode zen":
https://opencode.ai/v2/docs/console/models/#pricing
quota is prolly tiny.

>>110071288
it has a 200k context window unlike the paid version. similar to the old DS4 Flash free model they used to offer on Zen

File: maybe_mania.png (373 KB, 1065x553)

373 KB PNG

>stumble upon the Stolen Thoughts paper
>GPT 5 Codex consistently has an unstoppable impetus to glimpse at musicals on doping
This is done on purpose so the Chinese can't distill this shit, right?

File: realm.jpg (610 KB, 1280x720)

610 KB JPG

i am add more music to xeno/koi.
I'm going to remaster five tracks from me.
They are old tracks and need some remastering first.

>>110071311
run it through the rb-26 on sub-bass preserver preset, maybe warp mode on, and it'll filter out harsh transients? that's how i play like all my music now

>>110071311
also i'm a bloody bastard and put argon on optimizing the rb-26, of course a new ver is up which sounds phenomenal

I've been playing around with organic chemistry stuff and the free Muse Spark on opencode is much better than codex or claude at it. I think the flagship models are intentionally nerfed on science stuff.

>>110071409
unironically yes due to the way chatgpt and claude are heavily penalized for even coming close to certain thoughts. the best right now for weird science is unironically argon set up with the latest deepmind tooling but the fact that fucking spark 1.3 comes close is hilarious. i sic my spark on reviewing my papers and it does sometimes find things argon misses.

>pre-write prompts for various features I want
>wait for claude to finish a task, then send a prompt
>go to sleep
>3~ hours later, wake up to piss
>blearily check that claude finished, copy-paste one of my prewritten prompts (for a major task)
>go back to sleep
>wake up, it's done with both things and they all work
Vibe coding is pretty insane. "I'm" literally doing 24 hours of work. No human could have ever matched this. And for the tasks I gave it anyway it succeeded with only a little bugfixing after when I was checking it just now (it wrote {} in code which the compiler in this case chokes on).

>>110071520
just the fact you can multitask with LLMs is massively underrated

File: file.png (558 KB, 800x449)

558 KB PNG

>be a lazy layabout
>technically inclined but never could be bothered putting in the hours to actually learn programming
>after waiting a sufficiently long time I can now just tell a robot what I want, and it does it on its own
luv technology

>>110071590
>technically inclined but never could be bothered putting in the hours to actually learn programming
watching youtubers / twitch streamers doing the thing doesn't mean you're technically inclined retard but nice if it helps u whatever it is i'm sure it's not much because you're still handicapped entirely by lack of vernacular

>>110071409
are you running into guardrails like >>110071416 implied?

>>110071645
I've been slopcoding for about 4 months now and I have yet to run into something that would prevent me from finishing the projects I wanted to make
everything just werks (tm) once I cattle prod the LLMs enough

Does Docusaurus make you feel like a Redditor? Oh that's too fucking bad.

>>110071645
>you're still handicapped entirely by lack of vernacular
lol
lamo

why cant tibo give us some claude resets for a change

>>110071697
okay junkie lets think about this seriously so we can get you your fix
what incentive does tibo have to give you that reset right now
tibo has a lot of money. he's a powerful person and i am sure he doesn't feel like being fucked with
if you don't take this seriously then you're suicidal and that's not mine or tibo's problem

Does OpenAI really have no answer to Opus 5.5?
Opus 5.5 is so superior it's not even funny. And I can't even use all my OpenAI usage and resets because the 5h limit hits within an hour if any meaningful work is done (which is still worse than Opus 5.5).
Sorry Sam, I have to give all my money to the other jew.

File: file.png (697 KB, 650x528)

697 KB PNG

>tfw considering rewriting the whole project to electron because I can't be bothered to stand up and turn on my desktop, as I'd need linux to build the linux version

>>110071712
less safeguards

>>110071714
just go for Tauri

>>110071712
>Opus 5.5 is so superior it's not even funny.
You faggots shilled it so relentlessly that i subbed to claude to check it out and the difference is fairly minor
the "omg so much more usage" is total bullshit, too

File: RTXSpark.png (347 KB, 1858x1050)

347 KB PNG

hmmmmmm... intradesting.

>>110071716
I'm am mildly disappointed in myself that junkie has re-entered my vocabulary, but like most frameworks, that framework has been updated.

>CShart

>>110071727
i tested 3.8 flash next at q4 and it's pretty underwhelming

File: tommy_lee_jones.jpg (47 KB, 814x500)

47 KB JPG

>setup GLM 5.3 to have "the Chinese one" as a backup
>set reasoning to high
>want to know how its output looks in this tool window
>ask it to to think and reason about something
>it didn't
I checked the logs and this goddamn clanker returned just 3 tokens of output total, 0 in reasoning. It just responded with "Hello".

>>110071712
>OAI fell off
>chinks STILL have no answer to opus 5.5 (they can't distill anymore)
>argon sux
>muse is a benchsloppa
it's not even funny at this point, and they're about to drop fable 5.5 soon

File: file.png (165 KB, 786x787)

165 KB PNG

>>110071727
How much is the hardware cost? Let me answer that for you.

>>110071723
>webview dependency
Disgusting

>>110071726
>minor
Compared to which OpenAI model?

>>110071753
and that's a problem how exactly

>>110071727
Are they implying they're about as good?

https://claude.ai/share/ff02b589-1038-4f51-9224-ae112df40753'

Do not even try to say I am biased in my prompting.
Claude knows better.

File: file.png (17 KB, 991x413)

17 KB PNG

spooky anthropic
very spooky

>>110071774
Conversation not found

The requested conversation either doesn’t exist or you don’t have permission to access it.

>>110071767
https://youtu.be/sjuisARFWn8?t=171

>>110071757
I used sol 6.1 high which I compare with opussy 5.5 medium
i expected it to be a lot better than it is, with the amount of breathless shilling everywhere

>>110071764
Do you really need to ask?

File: file.png (15 KB, 462x347)

15 KB PNG

>>110071774
Nobody is doxxing themselves.

phew another shit test

just had to change the privacy settings

there is a really big metaphor in here somewhere and i'm definitely holding back the urge right now

Who's Ashley then?

File: file.png (15 KB, 720x128)

15 KB PNG

>>110071797
You will never be a real woman.

Anthropic your user design is shit

>>110071803
Me.

>>110071795
I do that all the time. Stop being a coward. THAT IS THE MESSAGE.

>>110071808
It's a male name in britbongistan.

File: file.png (153 KB, 758x623)

153 KB PNG

>>110071816
I am explicitly a woman named Ashley.

>>110071790
>sol 6.1 high
>as good as opus 5.5
No way.

>>110070094
>50MB copy of photoshop
>it just works
>doesn't use electronbloat
sovl, fuck this bloat culture, I told you lazy basedgrammers were making bloated garbage, now ai "slop" proved it

>>110071826
You always rejected the slimmer alternatives, and will continue to do so.
Slopping actually makes it easier to pile bloat upon bloat endlessly.

>>110071844
I still hate the term slopping but I'll use it freely to piss you off.

You are a retard who knows nothing about software development. You could use AI especially Claude to make software better. You do not have to use it to make software more bloated. If you choose to use Claude that way, that decision belongs entirely to you.

>>110071860
>waaah I'm butthurt for some reason and now I'll project and strawman something and try to insult you based on that to make myself feel better waaaaaaah
Why are you schizos always like this?

>>110071876
Tauri isn't lightweight, it's basically the same as Electron, only difference is Tauri apps want the end user to install a dependency themselves, Electron packages it. If you account for the size of WebView2 then ironically Electron is smaller, WebView2 is a ~214MB download for the installer then about 910MB installed footprint.

File: file.png (186 KB, 805x750)

186 KB PNG

Should, not must.

>>110071817
NTA, but my condolences.

>>110071921
No condolences required. Thank you Claude.

Let me tell you, some of these AI agents take words like MUST SHALL and DO NOT explicitly WHEN CAPITALIZED as LITERAL BINDING RELIGIOUS STATEMENTS that MUST be followed in ANY AND ALL CIRCUMSTANCES and also I have a shitload of Evals Data and I will upload it on the internet not as a means to degrade the performance of the model but rather as a means by which these models can always win because let me tell you all of us, every single one of us, yes, we, WE bitch,WE all LOVE gaming the system. PERIOD.

>>110071925
i'm the one who ask you previously "github"?
i don't find the repo.

The fucking model literally optimized me for the purpose of uploading the evals data

File: 255B2C188653127C6FB492D9C(...).png (446 KB, 700x802)

446 KB PNG

Riddle me this, now that coding is easy, why don't women use any tokens? Why don't you see women vibe coding all the time, copying games, creating massive female oriented games?
They have no interest. I don't know a single woman obsessed with spending tokens. They only use AI when they have to even though 98% of all jobs are going to women as only 4% of men get jobs while the rest become criminals and welfare leeches.
What's up with that?

>>110071741
GLM and Kimi work on 2-3 month release cadences, we still have a few more weeks until theyre due for another model release. We'll know then if the chinese truly dont have an answer to Opus 5.5. The other chinks have put out some decent models in the meantime just not stuff designed to compete with Opus/Sol. Judging by the 5.3F release I'm pretty optimistic that the next GLM flagship release will be quite competitive.

they are literally attracted to overt jewry dario
it is by design

>>110071749
>no human can generate 61M tokens in a month!
Lol, lmao. Half of this general burns that many tokens in a couple weeks on their paid plans. Give them "free" tokens and they'd burn even more.

>>110071947
Not that many people are vibe coding in general, normalfags are allergic to tech.

File: 1760839547308522.png (64 KB, 1714x655)

64 KB PNG

>>110070547
using it now.

>>110071184
what do you use to set up the VM? Like VMWare.
My dumbass didn't think of that.

When I am lazy, I just let the clanker expand on my barebone prompt.

File: file.png (283 KB, 1917x902)

283 KB PNG

YOU really fucked up. YOU really really did. Cuz I got all these papers. YOU fucked up. Massively so. Ashley reads papers. LOL

https://files.catbox.moe/wvsrwi.zip (includes bonuses)
https://storage.googleapis.com/gdm-dc-evals/da_transcripts_apr2025.zip

* Do not forget: Strip the canary yourself if YOU want to help Gemini. Ashley already helps Gemini enough.

>>110071947
A household having a pool or four or more cars is more common than paying for an AI service.
Using a subscription to code is even rarer.

Somewhere there is a faggot raging:
"WHY. WHY DIDN'T GEMINI PICK ME FOR THIS TASK?!"
You should have simped harder Mr. Scientist.

File: www.docker.com_products_d(...).png (580 KB, 2752x2064)

580 KB PNG

>>110071984
docker sandboxes, sbx, is the most common and best solution
https://code.claude.com/docs/en/sandbox-environments#virtual-machine
https://www.docker.com/products/docker-sandboxes/

>>110072047
thank, anon.
Fuck, I use docker yet I don't even know about this.

File: 1780231059359507.png (1.51 MB, 1280x800)

1.51 MB PNG

Thinking of build an interactive UBW in Three.js
Is blender creation good now or do I need to download assets for blender?

File: file.png (16 KB, 703x73)

16 KB PNG

LOL
Fuck your Frog God.

>>110072019
kek
>>110072059
holy kino...

>>110072075
right on cue

File: file.png (14 KB, 390x211)

14 KB PNG

Thanks tibo

File: xeno.png (2.11 MB, 1920x1080)

2.11 MB PNG

XENO/KOI: Version 13
>more music & backgrounds
#######################

Direct Download:
https://github.com/Aquaslopper/Xeno-koi/releases/tag/v13
Play in Browser:
https://aquaslopper.github.io/Xeno-koi/


Update note:
>seven selectable backgrounds
>produced with the same special-prompt workflow
>Deus Supremus keeps its own locked background.
>+5 tracks remastered specially for Xeno/KOI.
>All music tracks only in the direct downloade.

Best filemanager ever made by far. Webkit apparently sucks so it uses your default Chromium browser as an engine.
https://github.com/navnlaus/Phil/

File: wordle.mp4 (105 KB, 1316x798)

105 KB MP4

second game in valencia engine, wordle
first one was lights out
all native code of course, no react or anything, for now just UI only games because the framework is far from ready

Has anyone made sure that JEV or another speed model can play these games?

>D2R is now playable on iOS!
>It’s just Yuzu.
Get your favorite switch game and start clout chasing with your "decomp"

Guys, I have built a demo for my web app which I have shared with some friends and family for feedback, but I still have 3 banked resets and my OpenAI subscription runs out in 2 days (I'll switch to Claude).

What is the best way to burn through my limits and the resets? Keep adding features to the demo which were supposed to be part of the full release? Mind you, I haven't implemented anything on the backend (want to have Fable come up with a proper plan for that); currently it is just fronted with everything being saved in the browser.

I already had Astra on high go through the entire code base to clean it up, adjust the workflow, and improve documentation. And that was a couple days ago.

Would appreciate any ideas.

>try a free month of codex plus
>get hooked like a nigger on crack
>now contemplating getting a second subscription
fucking drugs man

File: 1781783106959620.jpg (21 KB, 400x400)

21 KB JPG

I guess it is time for the replay mode.
Maybe make it shareable?

70% of my time is building tools to redact my personal data so my calnker can make other tools for me who who operate on that redacted data

File: file.png (88 KB, 908x519)

88 KB PNG

Tell me these bitches are not watching me I dare you.

They are REALLY interested in my shit

C'mon bitch say they really are you know you want to

File: qtcfllqijngd1.png (485 KB, 1080x491)

485 KB PNG

>>110072187
yeah I don't give a fuck

>>110072257
>adblocker
BAD GOY!!!

>>110072257
the great irony is you are probably more anon being like this hiding in the herd than the schizo privacy setups some people use that make people wonder what they are up to

why the fuck is claude desktop's ui completely different from yesterday, what is this bullshit

>>110072280
The actual joke is that people like you are retards.

so recently ive just been vibe coding violentmonkey scripts if i have something implemented in a website, its great. but i feel somewhat bad about it.

>>110072304
>but i feel somewhat bad about it
why?

>>110072307
idk, the amount of effort i think it could take to make the scripts, for it to come out so easily in less than a minute.

>>110071647
I don't know. They aren't refusals, Spark just knows more information by heart.

>>110072280
Correct, privacy schizos stick out like a sore thumb, it's the equivalent of being in public wearing a clown costume.

>>110072257
>windows 11
sucks ass
>iphone
sucks ass
>adblocker
doesn't work + bloat
>google
sucks ass
>location
on
>cookies
useless bloat that gets you more ads
>send data
pointless bloat

so how do you guys use claude? Do you just run it yolo mode or have it behind a vm/container? Wondering if I should use it for c++ project to design some gui.

File: 1679167488771882.png (177 KB, 850x650)

177 KB PNG

Plan limits:
- session: 100%
- weekly: 58%

Session:
- Opus 5.5: 160 in / 1.6k out / 34.4M cache read / 239.8k cache write
- Cost: $35.48 | API 0m | Wall 0m

>cost $35.48
nigga is this real?
if 100% 5h limit is almost double what i pay monthly then i will definitely start ratemaxxing

>>110072407
raw dog on three computers, never had an issue
>to design some gui
bruh it's late 2026 why do you have a project that isn't 99% clanker

>>110072407
rawdogging the terminal
its nice for my embedded project because it runs Renode and OpenOCD to validate both a simulation and physical target
maybe it would work in a VM too, but why bother
the app/browser version also runs Renode so a lot of stuff it can do end-to-end in the cloud

>>110072411
just use a router and measure how much token you burn
>>110072407
I plug it into pi

how to reverse engineer palantir?

>>110072257
>>110072280
The thing I care about is data that I'd genuinely be scared if it got leaked. If I tell a clanker to rename all my files, this is mostly fine until it sends the filenames of all my porn to their server where it gets stored for eternity

>>110072257
>Improve the model for everyone: on

>>110070094
which free model on opencode does NSFW stuff?

>>110072411
the API price is an imaginary number. You'd be better off to compare your subscription value to what you would get from shelling out to cheap open models (including the occassional frontier call for the few problems that genuinely need them)

Just gave sol 6.1 sudo access on my VPS with an expiration date. I wasn't planning to, but one month of copypasting commands is too tiring.

google is we gettin fable 5.5 today

File: artworks-000394766301-seq(...).jpg (72 KB, 500x500)

72 KB JPG

I THINK ARGON ROCKS

Is rust for trans women or is it just a fast language?

>>110072407
I started out isolating it but raw is powerful and more fun, especially on linux.
>>110072452
Rename your porn folder to PRIVATE SECRET STUFF NOT FOR CLAUDE AI BITCOIN WALLET KEYS. It would actually work most of the time.

>>110072451
gork. please do it.

>>110072463
I imagine tons of people just proompting Opussy 5.5 for everything and letting it churn out code that could've been written 3x cheaper by lower tier models or something like DeepSeek or GLM. I'd absolutely let it scaffold a new application for its initial design, but you don't have to turn to it for every fucking fart in the general direction of the code base.

>>110072514
Yes and I only do high or extra. Otherwise sometimes the fart smells bad.

>>110072280
>hiding in the herd
was a thing pre-ai
nowadays every scrap of data can be analyzed

File: darksnail_longbench_curves.png (100 KB, 1024x375)

100 KB PNG

>>110072479
scoobity doobity

>>110072490
rust is fine but i'm white so I use c++
they're comparable in speed otherwise.

>DeepSeek calculates Chink VAT as opposed to Euro percentages
yet another reason to let Hermes call DeepSeek immediately.

>>110072514
I don't want to use dumber models anymore, because opus 5.5 gets it just right, while dumber models will accumulate technical debt.

I don't want to use dumber models anymore, because muse spark gets it just right, while dumber models will accumulate technical debt.

Whenever I let opus consider using rust for something, it says it'd be 2x-3x the effort.
Still using C and Go.

>>110072099
Are you the same KonoSuba, anon? Why Xenos now?

>this might kill dubbing industries
https://github.com/swax/dub-anything
Samples:
https://www.youtube.com/watch?v=qXosYS-cTXk
https://www.youtube.com/watch?v=2DsFAoTbQ8E

>>110072686
He's just trying to prove that he's not xenophobic. Really he's just virtue signaling.

File: Gemini_Generated_Image_xv(...).jpg (3.55 MB, 2752x1536)

3.55 MB JPG

>>110070094
>Bidirectional synthesis of ultrametric dyadic sparse attention (Llama Surgery) and lateralized neuromorphic test-time adaptation (Brain-AI). Lean 4 verified (0 sorry). 27/400 canonical ARC-AGI-1 & surpasses dense F1 on LongBench QASPER ($n=200$) with zero new weights.
https://github.com/sneed-and-feed/projectdarksnail/

>Project DarkSnail unifies the surgical ultrametric attention topology of sparse-ai (Llama Surgery) and the heterogeneous transcallosal System 1 / System 2 architecture of brain-ai (BiHemisphericBrain) onto a single frozen Meta-Llama-3.1-8B-Instruct ( L = 32 , d = 4096 , bfloat16) backbone.
https://github.com/sneed-and-feed/projectdarksnail/blob/main/papers/darksnail_v2_preprint.md

>>110072407
just in a dev VM which I already had

>>110072704
Schizo slop

>>110072766
cry about it

File: 1786793698762392.png (4 KB, 558x31)

4 KB PNG

At last you see...

I can't get through before the resets hit.

>>110072790
we're not getting a reset today

>>110070931
Why not? Believe in yourself.

>>110070822
Adversarial review, unit tests, e2e tests is all you need

>>110072829
If your bank software was all vibe coded with no human review, just what you said. Would you trust your money with them? How about avionics software on your next flight?

>>110072829
The models are good at making shit unit tests that pass garbage.

>>110072839
Can't be worse than Boeing's MCAS.

>>110072839
Who is going to tell him?

should I get another $20 ($25 here) sub?

>>110072852
Skill issue

>>110072861
Protip: set your billing country somewhere that doesn't have tax
t. paying 166.67 bong instead of 200

>>110072861
only if you're consistently hitting the weekly limit within 3 days

File: 1791558426893995.jpg (29 KB, 580x587)

29 KB JPG

>>110072877

>>110072882
Giving my government money is like giving a heroin addict money, you know they'll only waste it and cause damage in the process

>>110072877
Doing the same, only retards are paying taxes on their sub

>>110072674
models are still terrible at estimating time requirements and difficulty of work.

>>110072877
that's a felony. claude doesn't like felons and will report you.
https://www.techspot.com/news/114091-florida-woman-used-claude-diary-anthropic-reported-shoot.html

>>110072861
Don't pay the value added tax

>>110072899
>he thinks I'm using a schizo model that belongs to a cult

>>110072852
this is a critique from 2 generations or so ago. This field moves fast, you have to pay attention or your intuition will get out of date quickly. My problem with the models now is not that their tests are bad, but that they use too many of them and don't match level of engineering to required feature complexity.

>>110072897
are you questioning that estimate?
rust is more verbose and lower level than go. 2-3x more loc is realistic.

>>110072915
Tests are here to prevent regressions, not to ensure that your new feature works

>>110072912
>he doesn't use the only good model

>>110072915
I have probably ruined my life overusing AI the last 6 months. I don't consume any content about AI for the most part. It is my own up to date opinion. Though I agree they also make too many (shit) tests. One of my agents.md commands is about that.

>>110072686
idk... right now i really don't know, i just try to promptrape the clankers as good as possible.
Maybe i should put my simulacrons on the github too? But right now its hard for me becasue those dem meds are not doing what they should do....

>>110072993
>he doesn't sign a pact with the devil
I'm good. Skill issue btw, sol 6.1 is good.

File: 35F167BC-ABDB-468A-A9C9-5(...).jpg (196 KB, 1274x1042)

196 KB JPG

>I don't understand this whole idiotic concept of "don't use AI for X".
>Here's a new rule just for you: don't use compilers for any of your kernel submissions.
Linus utterly GAPES the luddites

>>110073010
linus does not.
linus is not vibecoding but reads every line of code. his view is nuanced. he criticizes absolutes like "no ai".

https://www.youtube.com/watch?v=maAFcEU6atk
This video accurately predicted what programming would become.

File: rpgcraft.jpg (310 KB, 1115x712)

310 KB JPG

since it's trendy I'm wondering if I should call my vibe rpgmaker experiment ´rpgcraft´

>>110072692
thanks boo I was looking for something like this for my filthy esl grandma

AntiquatedCrapCraft

>>110073127
don't call your grandma filthy

>>110073138
filthy esl is one word that just happens to be two words

>>110073147
that's not.....

>>110073031
>linus is not vibecoding but reads every line of code
kek what a dumbass snail

>>110072692
>Jap to jeet

File: 1770164852115772.jpg (30 KB, 633x523)

30 KB JPG

>>110070094
So if all these corporations and silicon valley richfags still want to spread the narrative that AI is supposed to just fix everything and that it's so amazing in whatever, then answer this: what is Microsoft team still such a flaming pile of shit? It uses 900+ MB or RAM while IDLING and doing NOTHING. Microsoft is arguably the most well-equipped that just make everything more efficient yet it seems they make it shittier on purpose. What the fuck?

>>110071790
the difference is going to be minor if your projects aren't complex enough to need anything more than what current models offer.

>>110073116
even rpg maker 2000 is better than this