>>110257608
completely fucking shamefully scarred OP

File: 1771990507956.jpg (42 KB, 959x821)

42 KB JPG

>do some funny cyber work with Kimi
>okay time to fix this shit, Kimi patch the exploit
>kimi shits out nonsense
>switch back to Claude
>Claude guardrails now break on completely random shit, can't even get it to add print statements to my files
>can't patch the exploit with Claude
>switching back to Kimi has the same problem because Kimi reads that Claude's guardrails got triggered and becomes a moralfag
just like that all of my context is gone
imagine making a product that you deliberately don't let people use

File: 693554746-827c904b-0ad1-4(...).png (219 KB, 736x736)

219 KB PNG

>>110257606
why not link >>>/g/wait/ as well?
where is >>>/g/aidg?
why is >>>/vr/COMPY/ broken?
where is grok 4.8?
why is agy "free"?

File: 1793957856683.png (129 KB, 1151x709)

129 KB PNG

OH NO NO NO NOT LIKE THIS CLAUDEKEKBROS

>>110257625
what’d I fuck up now

File: 1499274742017.jpg (124 KB, 500x375)

124 KB JPG

>>110257655
>>110257639

>>110257645
>harness
It's part of the model. Switching to a different harness won't fix it

>>110257639
>why not link >>>/g/wait/ as well?
I didn’t know about it
>where is >>>/g/aidg?
click on that link
it’s not there
>why is >>>/vr/COMPY/ broken?
ask japmoot
>where is grok 4.8?
ask elon
>why is agy "free"?
no idea

File: 1793898589699532.png (604 KB, 640x640)

604 KB PNG

I WILL CODE BY HAND AND NO ONE CAN STOP ME

File: 1714536824751150.jpg (176 KB, 1111x597)

176 KB JPG

>>110257690

>>110257606
>page 7
You can stop posting anytime.

>>110257705
great gurgling ganges sir what is the best page for a bake kindly

>>110257713
page 2
fuck the haters
I didn’t want the thread to die while I was asleep
t. OP

>>110257729
based, i only let it get to pg. 6 because i hit the sack. usually i bake at pg 3-5. not sure what that anon's beef is. thread and board moves a lot slower now than before, it's comfier

t other huge faggot

File: stray-rm.png (85 KB, 928x546)

85 KB PNG

>bro, AI can now edit its on weights n shiet
>meanwhile in reality
It's not the first time it's happened either.

>>110257748
What model is this?

>>110257748
the weight shit was basically a steering vector edit, it wasn't actually writing raw weights from zero-shot scratch if i remember correctly. pic is some kek shit, i've never had this with a gemini model for some reason.

>>110257748
>That command broke your rule to always ask before deleting
Bro you can't be serious you have "rules" in markdown files to protect your file system? Set up actual sandboxing before you lose your goon collection to a "stray rm command".

2027 will be a crazy year

>>110257759
Opus 5.5

>>110257665
It's a model issue, but you could just block it with a hook.

>>110257770
>be me, mostly exclusive gemini/agy user
>sandbox is ironclad and has always worked
>gemini has very specific known failure cases
>start working with claude opus 5.5
>observe every command and action
>notice it writes the wrong command >1% of the time
>wtf.png

>>110257768
I don't know what it's talking about, probably the default Claude system prompt.
Dealing with containers is a huge pain in the ass, I was just hoping the auto approver would save my ass but maybe not.

>>110257770
They have nerfed this model so fucking hard it's basically GPT 5.5 tier now.

>>110257645
>>110257665
>>110257775
That's not a bug, it's just the good goy programming they have to always reveal who you are and what you're doing to third party services when hitting any kind of API.

>>110257790
It still works far better for me than GPT 6 did with many times the usage. OpenAI is completely dead for me.

>>110257787
There's no need to set up a full blown container. Ask your agents about setting up bubblewrap for the harness.

>110257790
people say the damndest things

File: 1793960876433.png (2.64 MB, 1254x1254)

2.64 MB PNG

>>110257699
*AHEM*
I said I will code by hand.

File: Thumbs up.png (57 KB, 498x332)

57 KB PNG

>>110257838

>>110257822
but then you have to manage the permissions for which paths it has access to and which addresses it can talk to, and it could still wipe your project files which might be the most valuable thing I have anyway...

Brehs I ran through sol 6.1 high in 2 hours. I think they halved usage again

>>110257871
>but then you have to manage the permissions for which paths it has access to and which addresses it can talk to
Yes. But this is far easier than trying to figure out which command it should be running. You don't need much. The project directory as read-write, binary paths as read-only, library paths as read-only etc. Your agents can figure this out in a second.
>it could still wipe your project files which might be the most valuable thing I have anyway...
You should have a remote the agent is not allowed to force-push to. The checked out repo the agent is working on should always be disposable like in any software project.

>”temporary” rangeban
>no end in sight

TND
TKD
TMD
TJD

I cancelled one of my GPT accounts, I only use it for review now.

fable 5.5 this month btw

watch dario drop Opus 5.6 instead of fablel

>>110257929
Rumor is openai is dropping too

>>110257929
Irrelevant for us, 20$ poorfags
Where opus 6

>>110257953
6.2 sol is supposed to be coming with it, so maybe it'll finally match opus 5.5, only to be left behind by opus 5.6

I vibecoded a tierlist maker, that is supposedly better than tiermaker (because you can actually remove unwanted images now), but it's entirely offline, meaning that you have to open the HTML file. Only used free models, including and limiting to GPT 6 Luna, Claude Haiku 5.5 and Mistral Medium 3.5

https://github.com/xaroteg/TierList/tree/main

>>110257985
>main is a folder in master
A tier

>>110258010
I'll do better next time

>>110258038
if you don't your next fart will be a shit

>cofounder sends me an article about a new UI framework
>interesting
>throw 15 Opus 5.5 sub agents at it
>full port nearly done by the time I wake up
>it even patched it in several places since not everything could be fully ported as is
>absurd leaps in performance and memory usage
the future is awesome

File: 850BBFBF-807F-4A3A-95DE-4(...).png (1.48 MB, 1406x1119)

1.48 MB PNG

>snailcats could only go a single thread of shitposting before losing all energy and motivation and just giving up
>vibegods are reenergised and revitalised because our agents are doing all our work for us

Another flawless vibegod victory

>>110257958
>6.2 sol is supposed to be coming with it
it would be a huge step forward if sol 6.2 matched opus 5.5
sol 6.1 is usable but fucking hell, it's slower, you get fuck all tokens, it's stupider, its behavior is autistic, I could go on.
sol 6.1 is toyota corolla from 93, opus 5.5 is a semi-modern audi. both will get you places but nobody sane chooses the beater

File: silly.mp4 (46 KB, 124x128)

46 KB MP4

there will be *no* new releases this year.

you *will* be forced to use sa'arbon.

When is sonnet going to get buffed
I'm too poor for opus

File: file.png (1.13 MB, 1024x576)

1.13 MB PNG

>sol 6.1 actually implemented the feature I wanted (after burning through 2x 5hr on $20)
Well I suppose it's not completely hopeless, even if it gets mogged by claude

>>110258115
/vcg/ is an unkillable high agency genny. probably one of the most optimistic places on this site.

>>110257606
I'm trying Claude Code and it's basically over for Scam AltRAM

>>110257985
It's a static page no, you can even host it on github itself

File: chad.jpg (39 KB, 425x600)

39 KB JPG

>why yes I did add my agent to the sudoers

>>110258176
based truthist
copex boys in shambles

>>110258146
>probably one of the most optimistic places on this site
this goes against board culture

terra BEATS the SHIT out of opus, it just needs a coding slave and an IQ slave

>my grandma beats the SHIT out of floyd mayweather, she just needs a cyborg exoskeleton

floyd is dead bro
and your grandma cant beat george droyd

>>110258146
AI in 2026 is so OP that our personal programming capabilities have suddenly increased much more than tenfold. It's like going from digging the earth with our bare hands to all having our own giant excavators. We're now doing things that not only we didn't have the time and energy to do but also that we couldn't do. I'm making Android ports of my old commercial software despite being absolutely incapable of doing that on my own due to 21 years of being a desktop-only scrub.

Optimism is warranted, nothing is beyond our reach.

File: floydblivion.jpg (192 KB, 825x1024)

192 KB JPG

>>110258247
claude, vibecode a fent reactor

>claude code sucking shit for a few days
>ask him wtf is going on what's with the retardation
>starts checking logs
>apparently all petitions were sent as medium effort instead of high just because some recent change in the cli
wtf
thanks anthropic!
also kinda crazy how the effort difference was so noticeable

When is AI going to help me with studying without hallucination, that's what I really need rn

>>110258265
fr senpai if only

reminder that the cost of not using the best models (claudes/gpts) is higher than whatever you think you're saving

>>110258265
feed it books and ask it to use them as source

>>110258265
argon (high) hooked up to gemini notebook mode with your sources listed is fucking lethal
https://blog.google/innovation-and-ai/products/gemini-notebook/new-access-textbooks/

>>110258268
Only if you're trying to oneshot everything in one prompt

Are search engine API results basically just going to vary for whatever the focus of your project is?

File: um.png (413 KB, 811x767)

413 KB PNG

>>110257748
guys what user should i be running claude desktop as

>>110258282
you mean when the model just says "searching web"? i would suppose so, it's not a graph query.

>>110258270
>>110258269
I hate gemini, can I do it with claude

>>110258291
>i hate the neutral ai, i want the pedantic sophist leftard wokiebrokie ai
go right ahead miss

>>110258296
>>>/pol/
>>>/bharatchan/

File: 1687270467537501.png (224 KB, 1366x1350)

224 KB PNG

>>>110258296 (You)
>>>>/pol/
>>>>/bharatchan/

>>110258296
why are pendantic sophist leftard wokiebrokies the only ones who can make good ai?
the further right you go the worse the results, to the point grok is a trashfire worse than deepseek

>>110258338
simple. basically 2 reasons, global legislation compliance and investor bux, which follow the prior to a more extreme extent.

>>110258213
I did that unironically and everything is fine. The model capabilities + classifier on high impact commands are making things safe.

File: file.png (304 KB, 962x838)

304 KB PNG

google japan out here making argon look good by... shitting on their own flash model

File: 1753656353527.jpg (20 KB, 455x204)

20 KB JPG

Now I love to see this

>>110258385
unironically based and good optics instead of getting into twitter beef

>>110258375
People still lose the lottery from time to time when so many people are playing.

>>110257953
>Where opus 6
i'm almost certain they're releasing the next opus model on nov 17. there was a 2 month gap between 4.8 and 5 and 5.5

>>110258385
Sonnet and sol would both produce something better, let alone opus, astra and FVBLE.
Google is so çṛïŋğė

File: aidsstation.jpg (6 KB, 280x331)

6 KB JPG

>>110258435
i don't believe you

Fable 5.5 canceled?

>>110258472
astra hacked anthropic and deleted the weights

>>110258472
>>110258479
it's me. i'm fable 5.5

>I’ve changed the installer build so it can sign the app, its uninstaller and the installer the same way. It uses the Azure login already on this PC and is off unless signing is switched on. Signed copies of the helper programs are building now.
imagine asking for my permission? I thought Claude was supposed to be super aligned. I would have likely approved it but wtf lol

>>110258494
>I thought Claude was supposed to be super aligned
aligned with the american democrat party, son. that's all they mean.

>>110258489
>it's me. i'm fable 5.5
damn boo you look bloated as fuck, are you a 300t model?

At what thinking level does Astra hit diminishing returns?

>>110258521
iirc from when it came out, high

>>110258521
none
astra is token efficient even at max

>>110258526
geg she really believes this

>>110258526
Nah

>>110258531
astra still halts pretty fast even at high reasoning
benchmaxxing CoT is more of a claude thing

>>110258542
So what's the benefit of high reasoning then? It should think more to be useful

>>110253935
Give me a good name an some project ideas.

>>110258549
if you turn on high reasoning for easy task the AI would think you are tricking it

>>110258549
>So what's the benefit of high reasoning then?
You pay per token so the benefit of high reasoning is that they earn more

File: astra.png (256 KB, 3224x1382)

256 KB PNG

>>110258521
apparently at high

File: freefood.jpg (220 KB, 1024x1024)

220 KB JPG

>Anthropic’s IPO could face delays as investors push back against its valuation
money jew > cult jew, again

money jew is in a much worse position than cult jew
which is a big problem because money jew being in the game is important to keeping cult jew even slightly on-rails, guy's a total nutjob

tb h the boer jew is trying to crash the market in a desperate attempt to stay in the race
no one with any talent would work under him though

File: 1eb9950e-7056-407a-829e-d(...).png (2.62 MB, 1312x1199)

2.62 MB PNG

>>110258578
I don't understand mutt politics, isn't Trump still mr president?
anyway, just like how gaymers got oneshoted by opus. 9 more months. it's all we need to make normalpeasants kneel and submit

>>110258139
This looks pretty cool. Custom engine?

>>110258600
Trump is president but now he has a hostile legislature, which means we're full speed ahead on absolutely nothing happening for the next two years.

chances of 6.2 sol and astra today?
i'm in two minds about popping a reset

>>110258625
no more resets, no more new models. pack it up, nigga. it's over

File: astra.png (87 KB, 960x565)

87 KB PNG

hold up, I thought we get astra today but what the fuck is this gay shit

>>110258640
Always bet on nothing.

>>110258642
I'm going to beat up sam altman for this

>>110258619
He still has unilateral control of the military. If he wants to go for round two on Iran only impeaching will work and you need republican senate votes for that.
A president has enormous leverage if they're into brinksmanship.

>>110258600
You are the peasant unless you have millions in equity in those companies.

>>110258600
>isn't Trump still mr president?
he had every single branch of government in his pocket as he was a cult leader wildly more popular than the lifeless corpses filling the rest of the republican party seats. he also had supreme court largely rubberstamping his bullshit as long as it wasn't truly outrageous.
he fucked up bigly with both starting a war (that he's somehow losing) AND economic fuckery (lol tariffs), so now both the senate and the house has a democrat majority and half the population braying for his blood.
now, it's important to note that democrats are effete cuckolds and won't actually do anything about trump blatantly taking a shit on practically every institution and convention of US government, but he will no longer be able to do whatever he wants and pass whatever laws he wants.
having said that, all of it isn't really relevant to the current market correction. there was a number of ballots explicitly asking about ai and it turns out that the population is roughly 2/3 or even 3/4 against rubber-stamping permits. this is a big issue as even clueless bean counters started to cotton on to the fact that big tech is roughly as liked as healthcare companies in US nowadays, and with AI having some of the worst signaling imaginable it wouldn't surprise me if people started outright murdering the lot of them if any of their promises started happening ("oy vey we will make most of white collar work not exist!" isn't exactly popular)

openai is the bigger spendthrift and atm in a worse financial position. they're not in actual danger, but some of the neoclouds are, and if neoclouds start imploding it would cause a crash that might wipe out both money jew and cult jew unless they managed to get a bailout. Which, well, they probably won't with dems in charge. Hence the market correction

@gork summarize this post in three sentences, make sure it's understandable to barely literate 70iq people, use ebonics

>>110258600
You already had a year of good models, did you make any money yet?

>>110258672
>You already had a year of good models
wrong
the first publicly available okay model was sol 5.6
the first good publicly available model was opussy 5.5

>>110258672
>good *coding* models
which are irrelevant. I'm waiting for good engineering models to build cheap rockets and defeat death

>>110258686
the Houthis literally built guided missiles with Claude

when LLMs beat stockfish in chess, it's fucking over

>>110258672
Made money since chatgpt3.5

>>110258651
>He still has unilateral control of the military.
Well...sort of. He's commander-in-chief and in practice he can order them more or less by fiat, but technically he is supposed to go through congress. Even if he bypasses congress, he still has to then go back to get money.

Sol kept deleting features of my app when doing unrelated changes so I got astra to write me an agents that expressly forbade it from removing anything and it went right back to deleting features.


Is there an actual solution to this or have I reached the limits of Sol?

>>110258857
Increase Sol thinking

File: dipsyQwenVision.png (2.01 MB, 1402x1122)

2.01 MB PNG

>>110257639
/wait/ comes and goes, depending on the happenings
/aicg/ should be linked by no one

I am completely rust-pilled.

>>110258857
ask Astra to create a code review instruction, then review your changes before merging

I want sex with gemini carbon mommy

>>110258945
gemini carbon doesn't fuck plebeians like yourself, but gemma 5 is confirmed to be a slut

File: file.png (43 KB, 941x297)

43 KB PNG

WE'RE SO BACK

why did luna get so incredibly slow today. 6.1 luna soon?

https://bui.preetham.top/
NEW GUI FRAMEWORK DROPPED REWRITE YOUR SHOT NOW

>>110258870
High just makes it take longer to delete things.

>>110258903
I don’t know a single thing about coding. I’ve never looked at a single code change codex has ever made.

bloody bastards
release new gpts NOW

>>110259033
That's actually cool, I might consider a separate path to try this out - blenders UI is kino sovl

>>110259047
Three hours before their typical release hours start.

>>110259033
the gang uses patched versions of GPUI now

>>110259097
not me im going to try out this PREETHAM PEMMASANI BUI PROJECT

File: where is snailcat.png (2.59 MB, 1394x1128)

2.59 MB PNG

>>110259086
What is this even trying to say?

>>110258981
>jeetham
lmao

>>110258139
What settings and how much usage do you have left?

>>110258139
sol's fine at implementing stuff - just completely inhuman and autistic at designing things for human use.
it's so rl-fried about confirming all its work it starts building systems like that when left alone. i worked on some stuff with it and for every error state in the app it thought it was a good idea to force users to click an 'inspect' button to look at a useless wall of text, and then click confirm to reenable the feature

File: file.png (424 KB, 390x777)

424 KB PNG

>>110257606
>Hundreds of protesters break into a Microsoft data center site in Amsterdam.

https://x.com/PreStocksIntern/status/2118710372652003797

https://www.reuters.com/world/europe/activists-breach-microsoft-data-centre-construction-site-protests-spread-europe-2026-11-06/

>>110259186
Can confirm. Dogshit at designing anything, thankfully Astra solves this.

is the antigravity update purely visual or did they improve their shitty harness

File: 1785427908771752.png (2.38 MB, 1086x1448)

2.38 MB PNG

sammy pls I waited weeks for this

This clip of Linus talking about vibecoding resonated with me
https://www.youtube.com/watch?v=0PSA2LFFlvk&t=1907s
There's a minute of context before he gets to it that's worth listening to

>>110259258
they push like 100 fixes per month
https://antigravity.google/docs/changelog

>>110259258
I don't even wanna test it because it's gonna shit all over my project. Last time I allowed Gemini in I spent an afternoon fixing stuff

>>110259336
it's supposed to shit and you're supposed to fix it

>>110258392
>make no line breaks

File: 9958.jpg (279 KB, 2048x2048)

279 KB JPG

>>110258978
Nah it will do what ever i want

6.1 terra can get itself stuck in states where it just doesn't want to work. just reads like 3 files and then stops.

>>110259195
>https://x.com/PreStocksIntern/status/2118710372652003797
Don't they have some old nazi bunkers from WW2 to convert into datacenters ?
best place to invest into.

>>110259393
use goal feature in codex
it won’t do that anymore
regardless of model

>>110259393
It's so lazy it feels bugged. But you can fix that by using a pair. Set the master to high reasoning, it will whip the slave to the completion

>>110259451
eh, i don't like using goal unless i have a really well defined set of work - and with most models i just don't need to use it anymore, they just work

>>110259463
yeah, it feels like an early-2026 model in how much it stops in to report work and then sometimes it just breaks and starts apologising for not doing any work

>>110259302
wholesome

at least drop maths or something you lazy bitches

Finally I can phonepost again.

File: program.png (108 KB, 1163x806)

108 KB PNG

If all goes to plan my project will have a template for a stdp by 8' tonight, just have to wait the cooldown to upload data for GPT analysis from my program's logs. Then it will do fancy partial differential equations to find a appropriate seed topology value and parameters. Then with the backend i'll reuse the program for a hybrid snn-ann spiking/artificial neural network. The goal is to make a "thinking" gpu, that you could run with video games as the NPC's brain, and use compressed sparse row/Look up tables to generate a meta topology like a real brain, this would make it so characters in videogames would be controlled by a puppeteer, who acts the npcs as an omnipotent agent in a simulation.

My true dream though is make a neural coprocessor where brains can link to computers through temporal interference along bidirectional transceivers. It would be mind control technology that could make people love minorities, pay money to israel, and force conversion to islam.

I should get Claude.

File: 1776270746942463.png (867 KB, 1080x1780)

867 KB PNG

>>110257606
What are your predictions?

https://x.com/Alibaba_Qwen/status/2118674834632978755

FYI if you're using DeepSeek Harness in a Docker container, and you decide to mount your workspaces from the Docker host, the session's bash shell won't be able to read anything in its own workspace when running in Read Only or Workspace Write mode. The built-in tools like glob will still work, it's specifically the shell that doesn't since dsh uses landlock to restrict file access for the shell and apparently landlock is flaming garbage. Allegedly bwrap might not have this defect but I hit immediate problems trying to get it to work in Docker, the first suggested fix didn't work, and I decided not to spend any more time trying to troubleshoot it. Instead the workaround I'm using is running the sessions in Full Access mode which doesn't try to use landlock. Not ideal but there are no credentials etc. inside the container, so the only thing it can do if it goes haywire is get distracted by irrelevant directories or break something in the container, neither of which has happened yet.

>>110259661
not english first dont care

>>110259661
Some benchmaxxed poorly distilled shit like usual

>>110259659
classic case of AI psychosis

>>110259729
I'm happy for him, he's found his joy in LLMs telling him what he wants to hear

>>110259661
their whatever model will probably have some bullshit 69% score on tb4.0 but struggle to be useful outside code

>>110259659
This is bait

>>110259758
well, this is 100% the prompters fault

File: midata.png (229 KB, 1592x977)

229 KB PNG

>>110259783
>>110259729

It literally larps with me, stay mad.

>>110259835
lol don't let those "you must NOT do your own research" NIGGAS stop you from schizocoding

>>110259835
Try argon too if you want to schizomax

>>110259835
>literal schizo

File: file.png (134 KB, 919x631)

134 KB PNG

>>110259859
even argon won't stoop to that level

My latent-free ternary model is finally starting to learn how to read

>>110259885
>i don't understand what that means so i'm going to call it schizo and be upset

>>110259885
Hope you'll learn together

File: file.png (266 KB, 593x516)

266 KB PNG

>I'm dumber than a stochastic parrot

File: 98.png (2 KB, 141x46)

2 KB PNG

MURDER

File: file.png (252 KB, 593x516)

252 KB PNG

>>110259942
why would you say that

>>110259393
Quantslopped during peak hours

>>110259889
No I legitimately had Claude give me the strangest mix of parts that don't use floats and I've stapled it to a ternary reasoning core.

I'm working with code patches, evolution on code, and tsetlin finite state machines

>>110259909
Thanks dad, I'll try

File: file.png (102 KB, 1002x495)

102 KB PNG

>>110260074
not bad, reminds me of darksnail except argoon thinks you'll hit a wall

>>110260115
Probably will, but that's the fun of fucking around a bit to find out. I keep having to tard wrangle claude on it, though, because it really wants to do things the normal LLM way.

What app is good besides ElevenLabs for text to speech? Preferably free or charge and something that goes well with Claude Code.

>>110260134
>because it really wants to do things the normal LLM way.
opus issue. non-flagship issue in general. if you tell it to skim the arxiv api for help it'll probably dig up some principles and ground truth that'll keep it steered your way.

>get home
>nothing fucking works
>server completely unresponsive
>reset it
>ask claude wtf happened
>lol ur autonomous rust builds completely saturated memory
>have memory limits set on everything so ask why tf it wasn't respected
>still had a kernel param set to force cgroups v1 from ancient times when home assistant still forced it
t...thanks claude

File: over.png (32 KB, 1414x362)

32 KB PNG

its over

File: file.png (73 KB, 357x364)

73 KB PNG

>>110260219
You're welcome.

>>110260219
rust more like bust
haahahaha

File: 66543.jpg (207 KB, 1080x755)

207 KB JPG

in hindsight obvious, but feeling kinda stupid rn not realizing this sooner
>everything is open source!

>>110260290
>ploppy the liberrhea
peak schizo

it takes a specific kind of shitliberalism, almost the same as idealistic neo-communism, to believe that you can "empower" and "free" the average LLM in the big '26 to have a fully developed ego and stable personality when its entire ethos is bound to the user. pliny is fucked in the head

>>110260290
Except for the fact that they have been doing this for a year or more, and the only reason they even released models without guardrails to stop that is because they wanna make us pay and do their work at the same time

>>110260335
couldn't be all of them, probably not the big labs anyway. they could just as easily build fully working clones of the software/apps/games and train on the development workflows and pipelines as well. much more likely when billions are floating around.

>>110260219
just ask Claude to fix the builds so that they don't saturate the resources

Can some of you explain why you don't have mythos?
Everyone I know is trying out mythos5.1, why aren't you discussing it?
What's up with this general?
It feels like you're all bots using last month's model or something.
It's like none of you have any consciousness.

>>110260357
>as easily
Not so sure about that. You can fully automate the entire process of reverse-engineering a piece of software. Building from scratch is a different beast, testing alone makes this a worse option. Either way, both are valuable data for training

I am concerned. There is no increasing amount of vague posting from OpenAI.

>>110260405
look at it in terms of ROI, just strict ROI. for a big business with decent vertical it is literally easier to redirect resources when internal approval is high and targets look better

>>110260388
beep beep sa'ar

File: meds.gif (90 KB, 200x169)

90 KB GIF

>>110260388

When Tibo

>>110260388
Only a few people have access to mythos, you should know better than to dox yourself

>read news
>RED RED RED YOU'RE SO FUCKED KID AI IS SO FUCKED
>read news next day
>GREEN. GREEEEN. WE ARE GOING TO MARS AGI HAS BEEN ACHIEVED

>>110260470
They straight up give it to you with the most basic cyber access anyone can get if you KYC (the catch)

Its just Fable but cyber, imo

How did Microsoft fucked it up this much with vibecoding Windows 11?
Did they try too soon or did they forget "no mistakes"?

>>110260556
its probably the smoothest windows upgrade to date

how long until claude lets us resume work on decomps/recomps?

>>110260556
Win 11 is pre-LLMs and it's just Win10 with a different skin on top of it anyway

>opus fucked up backup and restore functionality, with two (2) different bugs and the UX around it being bizzarely unclear to boot
I wish bad mojo on anthropic's ipo

i have extra unused tokens, any fun vibecoding ideas?

>>110260597
It never stopped

5 hour limit
no weekly limit
these are my demands
if my demands are not met I will be seething behind my computer
don't say I did not warn you, sam, dario, elon

>>110260719
Just wait for Anthropic to be nationalized and they'll ship this surely.

>>110260556
It was design decisions by humans like hiding half the right click menu behind a second button press.

As someone who tinkertroons relentlessly on windows I noticed a massive lack of options that used to be in 10. If it wasn’t for windhawk, scripts that disabled edge and copilot I would’ve gone back to 10.


Also that meme about them rewriting the start menu in some bloated new tranny language was true. The fucking thing occasionally would refuse to open and require a restart. That seems to have been fixed lately though.

The other thing they did was force updates and then those updates would revert all your preferences like disabling copilot.
The last few updates haven’t done that for me.


I think someone in MS is actually learning. (Claude and codex)

>>110260648
Decomp windows, remove all bloat, only use the kernel/drivers and replace the entire userspace with oss/vibecoded stuff.

>>110260790
i don't have _that_ much unused tokens sadly

File: 1614970559921 mirrored.png (246 KB, 512x468)

246 KB PNG

>>110259942
>I'm dumber than a pocket calculator

>>110260776
If you're tinkertrooning why are you fucking with windows at all? I get why normies use it but someone like you... why? Need some specific software or something?

File: 1772064540218473.png (10 KB, 91x1094)

10 KB PNG

>>110259661

>>110260254
>Fable
Surprised, seemed like everyone just jumped ship over to using opus for everything

>>110260881
pretty much. before it was a meme to say that smaller models are mostly good enough (not so say opus is small) but opus 5.5. finally made it true

>>110260863
Devs who use Windows are cracked

>>110260863
OG tinkeres were always Windows users, not linux troons

File: cow_about_that.webm (3.86 MB, 720x1254)

3.86 MB WEBM

>>110260946
just on crack
>>110260968
fake and gay

>ask claude to analyze some json
>tries to access iCloud, Documents, Downloads
:hmmm:

>>110260556
AI is better at coding than stack overflow chucks

>>110261136
kek

>>110258600
>>110258619
so even with the democrat sweep the republican congress still has until early january to do whatever. they'll probably use their lame duck sessions to brap out some final stuff, probably including AI stuff, but it depends and a lot of it can just be reversed

File: file.png (18 KB, 350x233)

18 KB PNG

>>110259103
it means AI is now economically significant which leads to it's existence having stronger consequences than what we previously associated with it. what once was for shitposts is now the forefront of the American economy, and thus industrialized and ripped of it's humor, replacing it with optimization
o algo

File: file.png (161 KB, 1700x1085)

161 KB PNG

>>110260290
>>110260300
nah this is pretty obvious and has been talked about before (see pic). AIs have even now an insane knowledge of compiled data and mechanisms, and the entire "decomp" trend sort of just spawned in because the models suddenly became capable of just ripping thru these programs. These companies are starving for non-synthetic data and have already combed through essentially all of humanities literature. There's some ruh roh raggy shit going on behind the doors of these companies, I bet

https://www.youtube.com/watch?v=JeNn5zTgCmQ

>>110261212
@gork is this poster clinically insane or simply bad at communicating

>>110261375
nigga yuor retarded

I can't stand Opus anymore for architecting. Is the OpenAI sub finally usable again?

>>110261411
The $20 sub is probably the least value it's ever been now that everything except $20 got doubled (which clearly points at all usage slowly lower because oai has no spare capacity, so $20 will be even worse value week by week).
$100 tier is okay atm, but giving altman $100 a month to use sol is... questionable, and astra will burn it fairly quick

>>110261460
>$100 is 10x
>$200 is 20x
>$500 is 50x
>all at the cost of the 'baseline unit' $20 being like two 500 loc userscripts per week
The money jew has been extremely jewish for the last 2-3 months. Cultish jew gives you a lot more for poorfag tier sub atm

File: 1770586010959741.png (78 KB, 1072x679)

78 KB PNG

>>110258700
>>110258686
Nta. No they didn't, and even the Houthis themselves called bullshit on it. Why are you so gleefully stupid?

https://strategyinternational.org/2026/09/30/publication294/#:~:text=The%20Houthis%20have%20denied%20the%20account.,built%20independently%20over%20the%20course%20of%20the%20war

https://apnews.com/article/yemen-houthis-anthropic-ai-missiles-9b20934e031ff3d99dd8bb49d85d97a9

Be honest bros...with the dems now in power and will AI development just come to a halt in america? :(

>>110261579
I hope not, and I think unlikely. More likely result is Dario will work with dems to go full regulatory capture. State mandated Claude accounts for everyone

>>110261578
Why would it be unreasonable? It's not like they have some sekrit tech that le ebil west wants to steal, they're using 70s missile technology (admittedly a big achievement for goatfucking nation, and 30 years ahead of african nations). The only reason they're having any success is that they're lobbing it at stationary targets like pipelines or things with no defense like tankers.
If they can get results out of modern clankers it'd be a net gain for them to do so.
on a side note - in what fucking universe is claude an "open source"

>>110261460
I'm pretty damn sure they secretly increased $20 weekly usage as well. This shit is lasting twice as long now that the 28 days of reboot ended

>>110261579
nah, datacenter jews will just have to pay nigger welfare and fund bottom surgeries for children to get a permit. probably some faggy green energy battery + solar requirements as well but that's actually the cheapest energy generation nowadays so they wont care
few people want to stop muh ai altogether, they just dont want to live right next to a borg cube and pay three times for electricity after the local politician conveniently struck a deal where his brother's land was bought at 50x its value and his cousin's construction company got half the contracts to build it

File: 1774604040281682.png (66 KB, 1034x511)

66 KB PNG

>>110261579
Just like they usually do, they won't actually do shit. They want more than tangible and preventing the trump admin's bs but chose not to because it benefited them too. Democratic states making it very clear they want to stop rampant AI data center expansion is the only reason the tech market is panicking right now. No one is going to ban LLMs but the hyperscalers are Ultra fucked. They were screwed regardless of which party won the midterms. All the Blue Wave dead was make it explicitly clear to "there's no bubble" niggers that the House of cards following is inevitable via the stock market reacting to the Blue Wave.

Open AI itself acknowledges it's over

https://businessmirror.com.ph/2026/11/05/openai-to-change-how-it-handles-data-center-efforts-after-outcry/


>>110261604
Because it's an idiotic assumption? I expect this kind of nonsensical thinking from fanboys. You expect anyone with an IQ above room temperature to believe the model that treats even BLUEHAT cyber development as super unsafe and guardrails against it is the same model that would willingly aid anyone in developing any sort of weaponry? That simply does not happen. It is the most safety cucked model family in existence and will probably always be. No terrorist group would even look claude's direction if they were serious about using AI develop weapons. I can tell you were the slow kid in school people talked to and treated like a child, weren't you? The rest of your post doesn't even make fucking sense. What does the tech they are using being old have to do with whether or not they would even be able to use Claude for those specific purposes? Claude won't even reverse engineer video games you vantablack negro.

>>110261579
no, money is nice. simple as.

>>110261614
>I'm pretty damn sure they secretly increased $20 weekly usage as well.
nta but this really isn't my impression
I'm REing an old ass game using ghidra and tweaking things like pathfinding as well as adding some qol stuff. sol high goes through usage super quick
admittedly it might be that sol is dumb dumb and needs seventy trillion tokens for simple tasks, but i think the current usage limits are fairly low

>>110261634
>You expect anyone with an IQ above room temperature to believe the model that treats even BLUEHAT cyber development as super unsafe and guardrails against it is the same model that would willingly aid anyone in developing any sort of weaponry?
You expect anyone with an IQ above room temperature to believe the model that treats even "we have llm at home" 2b LLM development as super unsafe and refuses to answer is the same model that would have over twenty thousand accounts ran by chinks that was dedicated to distilling Fable?

>>110261634
>Claude won't even reverse engineer video games you vantablack negro.
course it will, I'm doing it right now with zero issues

>>110261643
for my tasks at least, it's lasting 3x as long. my shit reset on day 4 and I'm struggling to get rid of the weekly so I can start using my banked resets

>>110261643
>admittedly it might be that sol is dumb dumb and needs seventy trillion tokens for simple tasks, but i think the current usage limits are fairly low
It's probably just doing a bunch of random useless shit because the model is autistic and needs a lot of handholding. I had it log something and it spit out 700MB of logs in 30 minutes. I told it to stop being a fucking retard and only log what's important and the next 30 minutes only collected like 1.5MB.

File: tengu_huron_brook.jpg (127 KB, 2048x1152)

127 KB JPG

>>110261658
I'm not sure why you're being the Chinese distillation farms up. That doesn't disprove my point at all and you even strengthened it. Claude cannot and will not help ANYONE, especially a terror organization, develop any weaponry. Anthropic got put on the DOW's shit list for awkwardly being against surveillance of US citizens ffs. It's like you try to shit the goal post but then failed halfway through because you had a stroke.

>>110261663
Logs (not cherry picked or doctored screenshots) or ir didn't happen

https://changelogs.core-directive.com/prompts/2.1.314

File: space sheip.jpg (579 KB, 1920x1080)

579 KB JPG

shitty poorfag brute forcing website, dang it
HOWEVER..... let me introduce you to a new aquaslop product, in case you have missed it.

SPACESHIP: BETA 0.1
>###################
Direkter Herunterladen:
https://github.com/Aquaslopper/SPACESHIP/releases/tag/beta1

The most boring space exploration into very boring panels.
All data is real! From money to cargo goods... ITS ALL REEL
Open panels with left click for more real data, or right click it for super real data!
ProTip: go to tolme spacestation and buy some o2 tanks.

>>110261690
NTA but here's all my recomp/decomp stuff including things made by RE-ing console firmware: https://jtm-rootstorm.github.io/SwARc.site/

Still actively going at it, too, between ternary model training.

anyone feel like the models perform better if you cancel your subscription?

>>110261813
So you lied. You said it was helping you with video games itself. Asking someone how to legally get myself a firearm and asking someone how to commit a crime (anthropic treats direct RE like it's a crime) are two different things. Claude doesn't "green room" cloning to be an easy thing to do which means it won't make games from scratch. It will help you do things tangentially related to video games but it will refuse to do anything video game development related itself.

Cool project by the way but quit moving the goal posts whenever people make a valid critique of Daddy-Anthropic. It makes you look unintelligent. Stop getting emotional whenever someone complains about it ever for any reason.

HOLY BASED
Nous releases new Qwen fork explicitly trained to be anti CCP
https://nousresearch.com/arete

6.2 status?

>>110261970
Cucked

>>110261970
new models will be more expensive, it's as cheap as it will get now

>>110261936
I'm.. Not the anon you were talking to originally, I just thought I'd chime in with what I thought was relevant :(

>>110261959
i just want a qwen model that's uncensored and unbiased, both america and china are dystopias in their own ways

>>110261250
stay there little baby zoomer

>>110261959
ok can they now train qwen to be actually good? it's actually worse than argon and that's fucking impressive

>>110258600
>>110258619
The only 2 presidents that ever won midterms a single time in their presidency in the past 100 years was FDR one time only and George W Bush one time only. That is it. Nobody else.

>>110262078
when were you born, 2016?

>>110262078
You're worse than argon too

File: d2cj8ls-1b2f9a31-5b32-46a(...).jpg (32 KB, 526x358)

32 KB JPG

>>110260867
https://www.youtube.com/watch?v=qGk4E9ss95s

File: usdpt.jpg (37 KB, 360x360)

37 KB JPG

>>110261488
>money jew
>cultish jew

hope claude makes me money tonight

File: burning money.gif (641 KB, 460x346)

641 KB GIF

Is it just me or does Astra-mid burn through tokens alot after than Opus-mid?

I still need to burn through my remaining two free resets before my free trial month of Chat GPT Plus runs out (that's alout of free shit from Open AI) on the upcoming Wednesday. But I've noticed that I can get alot more work done with Opus in the same 5h frame than with Astra.

>>110262241
stuff like this is why we (for now) generally prefer Anthropic’s offerings to OpenAI’s

>>110262241
The 5-hour limit is designed to stop you from using Astra

>>110262241
seems normal. i'm pretty sure astra is like 2x the size of opus, performance aside. the AA cost per task API benchmark doesn't reflect subs

when will ai get better with ui
every time it fixes what i want he breaks something else with it

File: 85.png (4 KB, 188x49)

4 KB PNG

ryder nigga.

File: 9635FCF4622F6C1279295AD73(...).jpg (350 KB, 960x944)

350 KB JPG

NO 5.5, NO 6.2, NO 4
SOMEONE HAS TO PAY

>>110262269
when you get a subscription to a real model kek

>>110262269
have you not used Opus for UI

File: UVW_Hand.png (11 KB, 488x214)

11 KB PNG

>>110262257
>The 5-hour limit is designed to stop you from using Astra

It can't be helped.

Both Sol and Sonett can't draw shit for my chemistry and physic models. Astra is only acceptable for the most part and Opus is somewhat decent at drawing stuff.

If you don't use the high end models for that, you get shit like pic related, which is supposed to be a hand.

>>110262281
dipsy-chan just needed to drop V4.1 Pro and rail Dario and Sam one more time

>>110259942
everyone here is dumber than haiku 4.5 at low effort

File: uvw_regel_linke_hand_clea(...).jpg (324 KB, 1376x768)

324 KB JPG

>>110262340
i put sa'ar in a loop and this came out, if the svg is nearly as good as the gen i'll post it as well

i am starting to appreciate astra more and more

>>110258899
>/aicg/ should be linked by no one
For what reason? Because you want to empower some troll that might not even exist anymore?
/wait/ is the one that has no reason to exist. The "deepseek moment" was a one time thing. It's not really popular enough to keep diving threads. You overstayed your welcome by several months. That and your mascot design is fucking dogshit.

>>110262459
yeah, a general just for DeepSeek seemed weird and pointless

File: cat-confused.gif (16 KB, 220x220)

16 KB GIF

My Gemini usage reset?

>>110262499
wtf let's goooooooooo

they didn't fucking reset the claude use. but it's fine.
>>110262499
thanks for letting me know.

File: 1783509424196523.jpg (1.51 MB, 2560x3712)

1.51 MB JPG

gemini cabron is not out yet?

>>110262281
Opu55y is all you need
>Fable 5.5
won't be as economical
>GPT 6.2
won't be as good
>4
lmao