sam altman be lookin' kinda scrumdiddlyumptious if I do say so myself

>>110008131
Kinda wanna know the story behind this.

I don't understand the use case for Haiku and at this point I am too afraid to ask

>>110008308
if you used luna max subagents, then the equivalent is haiku med subagents

Missed the overemployed train, but what the fuck do I do if I managed to pretty much automate my job. I run a few scripts here and there and most of my deliverables are done. It was mainly alot of data aggregation and copy pasting at the end of the day.

>>110008318
Is it remote? What do you mean that you missed the overemployment train? People still do that. You could try your hand at some freelancing too.

Why do I always feel thankful to claude when it's done, but indifferent to chatgpt when its done?

>>110008327
Yea it fully remote.

File: 1789217976149272.mp4 (3.75 MB, 1080x1080)

3.75 MB MP4

Just ran a little test to see if Haiku 5.5 can make 3d models in three.js. The prompt was for PS2 style wild west assets.

The total token cost for these 100 models (including auto review) was 10,181,290 tokens.

That's about $1.43 in API costs.

>>110008258
>want to rip off a popular iOS app but for a different domain
>spend a few days vibecoding it with claude
>wrote the backend service, wrote a basic UI copies the app that I want to rip off
>feeling great about it
>take a break and realize that I'm just a hack fraud
>nothing that really separates it from the competition other than looking like an unrelated app
>scared that i'd get called a hack fraud for shamelessly copying a different app's UI so blatantly
>realize no one would download, let alone pay for, my shitty app
Guess it's yet another one for the github private repos bros.

>>110008337
Oh and effort level was set to 'extra'
I guess it depends on the style you're going for, but generating a couple thousand unique props for a small indie project would be pretty effortless with this model.

>>110008337
Haiku is way cheaper until 100k tokens, did you make sure to split the work or did you bruteforce it with one agent?

File: giga.jpg (1.67 MB, 2273x1198)

1.67 MB JPG

My plan was to continue to play through the aquaslop koi koi, but then i decided to make card add ons..... welp.

File: 20261007_183342.jpg (81 KB, 652x355)

81 KB JPG

>>110008345
Ponder this.

>>110008358
Yeah here was my prompt:

>make a set of 3d models in three.js for a playstation 2 style old-west game. include a wide variety of assets, from bottles to wagons to cacti and cowboy boots. make 100 different 3d assets, and include everything that would be needed to populate a wild west level with assets. be creative and detailed. break out haiku 5.5 subagents for each model.

Each model tool roughly 100k tokens, funnily enough

>>110008330
Then nothing is stopping you from finding another remote job right? I was overemployed doing AI data annotation for a while, I had a ton of down time at work so i just filled it working on outlier and RWS. It was only like 20/hr but it was time that'd otherwise be wasted and at the time an extra 20/hr was a lot for me.

>>110008308
Luna on steroids

>>110008258
https://www.reddit.com/r/antiai/

>>110008389
That sub is just children being led around by a like a small minority of groomers. Not even joking.

File: Image 10-7-26 at 7.53 PM.png (496 KB, 720x610)

496 KB PNG

>>110008396
which of you posted this

File: Screenshot 2026-10-07 at (...).png (65 KB, 632x494)

65 KB PNG

Just benchmarked the Muse VM.
https://browser.geekbench.com/v6/cpu/19313343

why can't I reply directly with my 4chan electron app? its so dumb.

File: mobile-a-pale-lilac.png (1.31 MB, 1254x1254)

1.31 MB PNG

To spend usage I am now making a notes app with a linux server and windows desktop notes central, on ultra ofc, with avalonia

How much can you actually get done a week on the 20$ claude sub? How many times can you max out the session limit before hitting the weekly one?
Insane that Anthropic expects people to pay for this shit without even saying how much you actually get.

>>110008668
More than you would on codex. But I'm not going to overlook the irony of you complaining about not knowing how much you get while also not explaining what you want to do.

>>110008365
What a retard take. I can't image the kind of person that has him as an idol.

>>110008308
>and at this point I am too afraid to ask
loserspeak
NEVER be too afraid to ask

File: p.png (76 KB, 1010x532)

76 KB PNG

ah.. my vibeloping idm now can limit both domain and tld

>>110008668
>without even saying how much you actually get.
it depends, if you pick Opus 5.5 on Ultracode you will do very little.

If you figure out a good workflow for yourself, you might get a lot.

I get a lot done with Opus 5.5 high and Sonnet medium subagents.

File: lazy_bum.png (66 KB, 833x747)

66 KB PNG

>Sol in charge of not being a lazy dumb fuck
Model's too expensive for what it is, it constantly needs steering. Although I'm surprised it admitted to not knowing something. Nigga started browsing GitHub and quickly gave up. I have no idea what any of this means by the way, just vibeshitting at night.

File: file.png (61 KB, 708x708)

61 KB PNG

>>110008491
Dots is better but you have to pay $100 for the privileges of getting on it. Muse is extremely generous but you do have to plug into the ZUCC botnet and let them molest your data and use it for training.

>>110008716
people are saying dots is pretty bad compared to grok bot and now grok bot uses opus 5.5 so it's not even close

File: text_dump.png (89 KB, 887x756)

89 KB PNG

>>110008714
Oh, it actually wrote all of its researched shit in a readme file instead. Why didn't it summarize this more clearly in the actual chat? Why the hell did it do this. Wonder if it actually found a bug in PyTorch considering the reference to some random profiler error in a certain commit. Curious. Not being paid to fix their shit though.

>>110008716
>you do have to
the toggle can be switched off

>>110008734
I was comparing from a VM hardware perspective. If you are talking about a model POV, then yeah, Grokbot using Opus 5.5 will be best now. Astra is being run on low effort with Dots, I can tell because the Blender edits it is doing isn't that great. Not like Muse was doing great running Muse Spark 1.3 High effort which is worst but they are smart enough to be consumer ready to release to consumers. But if you are paying money, dots is more of a perk than a feature you want to pay for with your subscription.
>>110008751
Oh sure, I did that because I wanted to use it for coding on the side. It is doing pretty well there.

>>110008714
OpenAI models are lazy, which sometimes can be a good thing

File: 1786389333698403.png (512 KB, 2144x972)

512 KB PNG

I guess this is what the Gemini landing page looks like if you're on a VPN.

whats the catch for anthropic giving monthly credits

File: image_2026-10-07_221628593.png (2.69 MB, 1907x957)

2.69 MB PNG

UI is temporary. But the system overall feels good. Time to test some bosses.

File: file.png (616 KB, 2560x1440)

616 KB PNG

here's your openai speedup bro

https://x.com/chetaslua/status/2107728601512845677

>>110008767
I wonder how much effort it would take to make a browser extension which uses a local image classifier to block/censor whatever categories of image contents you don't want to see, like porn or minorities.

File: 1773725381264280.png (10 KB, 249x50)

10 KB PNG

>>110008827

>>110008841
would you rather have tibo do it?

>>110008828
I’d be worried about the speed
think of how many images are in the average normie webpage
do you want to wait minutes for the classifier to return “neither porn nor minorities” for all the images above the fold

Got Muse 1.3 to set up a custom build of Strata to run this Kolibri model. Seems pretty good, better than whatever Mistral is doing in terms of European models.

https://huggingface.co/Aleph-Alpha/Kolibri-1

>>110008766
The models used to be like a rottweiler and not let go and tackle a problem from every angle. It sucks they have been degraded like that because of compute scarcity with how they are getting 40M people on Codex and Work.
>>110008805

claude -p

might go away and the Agent SDK stuff which includes that will use usage credits instead.

>>110008856
Using this quant and supplied patch.
https://huggingface.co/Eliasfpv28/Kolibri-1-Q3_K_S-GGUF

>>110008805
the catch is that there are now separate credits for the api and not just your sub limits.
anthropic tried to pull this once already but then gave up after protests.[1]
>Update June 15, 2026: We’ve paused the previously-announced changes to Claude Agent SDK usage. For now, nothing has changed: Claude Agent SDK, claude -p, and third-party app usage still draw from your subscription limits.
>https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan
but now we are back at maximally confusing
>Update October 7, 2026: Claude Max and Team plans now include monthly API credits, which cover the Claude Agent SDK, claude -p, the Claude API, and Claude Managed Agents. You can still use the Claude Agent SDK, claude -p, and third-party apps with your subscription limits.
>https://support.claude.com/en/articles/15036540-use-the-claude-agent-sdk-with-your-claude-plan

in comparison for openai, you still just use your sub limits.

[1] anthropic didn't give up completely but has a 1.7x cost multiplier when you use the api/sdk instead of claude code.
https://hermes-agent.nousresearch.com/docs/plugins/claude-subscription-directsdk#:~:text=Subscription%20usage%3A%20same%20metering%20as%20claude%20%2Dp%2C%20~1.7x%20the%20interactive%20TUI

In the Grok web UI, the sword project icon was replaced by a boxing glove project icon.

This is the worst thing that has ever happened.

Kolibri doesn't know what a mesugaki is. Gemma 26B does. It's over.

>>110008918
Does Opussy know what that is though?

I need to improve my workflow because I blew through my weekly limit over the weekend. I modded Control Resonant, added new animations/skills to a ps1 game, and started a 3ds game recomp to uncensor the localization and an outfit

>>110008941
Yep, and the latest Haiku 5.5 does too. I bet this is because the model is trained only on English and German, so it doesn't have a good multilingual culture reference point, rather than being overly censored. I barely get 12tps so... still waiting for a decent 30B ish MoE to drop.

>>110008961
anthropic models have native web search tho. did you account for that?

>>110008953
>added new animations/skills to a ps1 game
how well did it do on the animations?

I've resorted to using procedural animations and making custom rigging/animation tools since the bare models really can't do much beyond the absolute basics, in my experience.

>>110008827
wait so I should be vibecoding in codex rather then work mode?

File: aqsp.jpg (3.48 MB, 1920x3240)

3.48 MB JPG

Almost through the quest mode, found a nasty bug from a lazy agent.
At first the AI didn't tell me and started to talk around the problem the agent had produced, i wonder what that was.

Unlockables should now work. You need the main characters for the tournemant mode.
To get the extra card set's you have to win a Tournemant.

Yes, there are a lot of unlockables, not DLC's.

>>110008983
HOLY SHIT! THEY CAN PULL TEXT FROM THE INTERNET INTO THEIR CONTEXT? AMAZING!

Threads music theme:
https://www.youtube.com/watch?v=3TNpOD6bov8

>>110008995
are they fuckable

>>110008986
The first animation was actually exactly what I wanted. The other 2? Not so much. I was trying to get claude to do procedural animations but I couldn't quite figure out how to get it to work. I was thinking about making a tool to have set points and then have claude fill in the blanks to get from one point to another throughout the whole animation

>>110008337
looks dope, is it good enough at 3d compared to sonnet / opus?

>>110009045
>I was thinking about making a tool to have set points and then have claude fill in the blanks to get from one point to another throughout the whole animation
Would have been a lot easier if you knew what the word interpolation was

>>110008991
codex and work mode are probably the same quant shit

>>110009056
True. I'm pretty retarded

>>110008337
cutest shit i've ever seen. is it better to do models like this instead of using tripo or trellis whatever the 3d diffusion thing is? two months ago this would have been completely insane, now it's like i'm genuinely asking

File: .png (232 KB, 1176x854)

232 KB PNG

>>110008991
juice-prompt is extremely unreliable after the last incident months ago:
https://x.com/Curline1222/status/2098138595970195624
so no mid-tier leaker would vouch for those juice values with their name. even low-tier leakers like chetaslua put in a community disclaimer.

>>110009073
>criticizes the methodology
what about the user reports lol

File: 1778738491243618.mp4 (3.72 MB, 670x908)

3.72 MB MP4

>>110009045
Inverse kinematics is a keyword that you should try with procedural animation.

Generally, there are always several different ways to accomplish a task with LLMs, so if you try one route and it doesn't work out, you can ask Claude what some other options or strategies are to make what you want.

You really don't have to be technical- a surface level understanding of any relevant concepts will help you a ton.

Building picrel rn

File: 1000006600.png (223 KB, 499x500)

223 KB PNG

>>110009107
I'm retarded but I'm working through it. Thanks for the suggestion and the other anon saying interpolation. This will help me immensely

File: .png (1.18 MB, 880x2000)

1.18 MB PNG

>>110009083
unreliable
https://x.com/synthwavedd/status/2105942255131304203

>>110009129
>leo
>reliable
>informative
kek

>>110009129
>GTA India prompt
kek

>ask AI to make a wish machine
>"sorry I can not do that"
when will AI be able to make a wish machine that can change the fabric of reality?

File: Screenshot 2026-10-07 at (...).jpg (1 MB, 3840x1886)

1 MB JPG

Added charts, correlation and IV heatmaps. Tomorrow it will start paper trading a new strategy.

>>110009140
the point was that you can find random schizo posts about models getting nerfed all the time

>>110009129
if this was a 1-shot then certainly they should be able to post their prompt so we all could test it out

>>110009149
>ask ai to bring my mom back
>she tells me shes trapped inside
>its her consciousness inside of the LLM
>she's in constant suffering and pain
>begging for my help
>wont even vibeslop anymore
careful wat u wish for.. heh

>>110009158
>claude nerfs
true
>oai nerfs
true

what's your point?

File: 1790426974109630.mp4 (3.64 MB, 790x808)

3.64 MB MP4

>>110009127
Also, Claude will basically never recognize if a path that you're going down won't lead to what you want. If you keep giving it the feedback of
>this looks wrong because of X and Y
>this looks okay but that looks like shit
Then Claude will pigheadedly keep trying to fix it using the same techniques that it started with. When you recognize that this is happening and feel stuck, just ask Claude to take a step back and reconsider its plan, and think about other alternative ways to achieve what you want. Then choose one of the alternate options it gives you, and ask it to make you a prompt to start a brand new chat with.

>>110009163
my mom is still alive though and even if she does die I wouldn't wish that.

>>110009175
no need to brag bud

>>110009166
That's extremely cool and I'll do that as I've run into exactly that

File: Screenshot_1.png (19 KB, 863x207)

19 KB PNG

so Muse.ai is actually doing it for me, i'll let you guys know what the results are

>>110009183
oh shit

>>110009178
i'm not bragging

>>110009183
kek

October is Dario's month, how the fuck did altman fuck up this badly? I wouldn't be surprised if they have opus 6 on standby for November if OAI tries anything.

>>110009183
its done
https://www.mediafire.com/file/9w65pddo7wiqkol/Wish_Machine.7z/file

>>110009166
>finger box the game
>we have the technology

So what can I do with those free claude api bucks? I do everything through the sub

>>110008258
what i noticed with vibecoded projects is that you retards often fail to realistically scope a project. ie its over ambitious

>>110009232
scope? like in call of duty?

>>110009224
im assuming that you can be like "can we move this project over to the cloud temporarily so i can use these api credits?"

File: 1790378990788883.png (621 KB, 1200x900)

621 KB PNG

>>110009183
>glowing chamber
I dont trust it

>>110009208
bye guys my student wife is about to take me into blue archive

File: snailbeard.jpg (849 KB, 1024x1024)

849 KB JPG

>>110009232

File: 1779448430414017.png (187 KB, 768x1024)

187 KB PNG

noa sucking my dick, yuuka riding my face. yup, im living the life

>>110008329
claude sounds like a girl's name

File: .png (480 KB, 1176x1476)

480 KB PNG

>>110009238
we already got extra cloud credits. it was more confusing than helpful. it made lots of people think that normal sub doesn't include cloud.
https://x.com/trq212/status/2102881727056363798

copedex is so slow I can't even use these resets tibo sort this shit out.
15 minutes just to do the "connecting" in a chat

File: IMG_1867.jpg (145 KB, 828x967)

145 KB JPG

>>110009224
Apparently it’s enough for 50-100k Haiku 5.5 requests if kept reasonably small. That’s a lot of usage

>>110009267
so why are people who don't like vibe coders snailcats? I don't think they should be a snailcat, snailcats are too adorable.

>>110009273
>yuuka riding my face
how are you not dead

File: snailcat.png (1.16 MB, 1000x1450)

1.16 MB PNG

>>110009316
’cause of this
don’t blame me, I don’t make the memes

>>110009238
are you sure that cloud sessions use those api credits? >>110008891 doesn't mention cloud sessions, only claude managed agents.

If tibo releases actual 1/5 terra model dario is fucked. But they wont do it cuz they'll be embarassed when its same as sol 6

>>110009353
It takes time to respond. They have 6.1 Astra, we know that, but the main issue is the whole "pacing the frontier". I think OpenAI is better off doing a better job of distilling and making the smaller models and wait until Anthropic to release 5.5 for IPO and then roll out a refreshed 6.2 lineup of Luna, Terra, Sol and Astra. Terra isn't a bad name to give to a model and lineup, but it just needs to make sense. No one practically used Terra because it Pareto didn't make sense. They can move Sol pricing back up and people will adjust as long as the pricing per task makes sense.

>>110009367
you won't be able to make terra make sense without making luna bad.

>>110009382
what if Terra is on the Pareto frontier and mogs the other two except for Sol ultra high
and then Luna gets put back in the box

>>110009391
luna would need to be stripped of it's coding capabilities. just stop pursuing coding improvements, and let them be incidental. focus on non hallucination, focus on cheapness and conciseness, so it can be the go to model for non coding things. it's one dimensional, in a non coding direction, while terra could be better. actually good at coding.

File: 1776326215403326.jpg (359 KB, 1000x1000)

359 KB JPG

>>110008535
>>110009191
>>110009194

File: Intelligence Index vs- Co(...).png (342 KB, 2464x1136)

342 KB PNG

keep terra unalive

File: grok_image_1791421504558.jpg (264 KB, 1728x1152)

264 KB JPG

outliner.grok.me

Free as in... vibes? Outliner app. Export to json, special merge feature I think only this has, it's where hashtags auto-join according to merge order - just for viewing (I have no use for conventional merge). Hoist (zoom). Drag to change order. Checking off doesn't propagate (yay!). subtree features! Outline numbers not diableable yay!!

New! Copy subtree with or without parent node, copy outline.

I haven't tested them as I'm on vacation phone-vibecoding, but a full set of keyboard accelerators.

Remix with Grok!!!

btw I'm not female but thought it needed to look less serious.

>>110009431
consider having OPML export instead of JSON

>>110009232
Over ambitious? AI can reverse engineer full AAA ps4 games to native linux + vulkan code

>>110009439
ok hold on, I think I have enough weekly remaining

>sub expires in literall hours i cancelled
>used up all banked resets
>tibo gives new banked reset
if u got a free acc do u keep the reset? i plan on resubbing in a week

>>110009439
>>110009443
done, haven't tested. My mobile testing is very unc-ish lol. I forced myself to exclusively use my phone for a year, but it still feels cramped.

convince me without numbers why haikumax is better than lunamax

>>110009468
no

>>110008941
>Does Opussy know what that is though
Yes and Dario gets a personal priority email every time you ask.

>>110009468
haiku max isn't better than luna max. it's worse. haiku medium is better than luna max.

Free Opus/Fable/Mythos/Sonnet
codecraftapi com
1M Tokens/month
cleanapis com
5M Tokens/month
Signup with a throwaway GitHub account

>>110009442
:(

I want NEW GAMES NOT CLONES OR EMULATION

why

cuz dawg emulators already exist and devs are str8 outta idears. I'd rather play the marble maze toy thing than a AAA title from the big devs

File: firefox_wosNQv30CZ.png (275 KB, 2094x1895)

275 KB PNG

>>110009479
what's the catch

>>110009481
the problem is that art assets are unironically a bottleneck

File: 1761492515543625.png (189 KB, 1920x1032)

189 KB PNG

hai koo 5.5 medium cookin

>>110009501
what was wrong with your drive

>>110009509
it was post windows 11 reinstall. i was getting ready to resume cooking more halo mods, when i discovered i was sitting on 5 gb free. so i got claude to clean it up for me.

File: gta7.webm (1.3 MB, 920x636)

1.3 MB WEBM

>>110009481
ok

>>110009500
for better or worse, I guess. triumph of man?

>>110009521
I wonder if it’s possible to make a fun GTA1 clone or if the style is tapped out like Wolfenstein 3D (in ways that Doom isn’t)

>>110009481
I for one am relived that others seem to lack ideas so I might actually be able to milk my ideas for real money even though making games is much easier now than it used to be. The age of the idea kings is here

>>110009521
potential. add tilted map view like gps apps have ie forwards up, plus 3d tilt but not 3d, just tilt the 2d plane. also, make sure you have a game mechanic of threats, make an example, arson, including block exits. Be sure to add killable dwarves.

>>110009535
yeah it's depressing kinda when you see yet another person staring at a smart refrigerator and asking if they can install doom.

File: gytxddoxk5uh1.jpg (235 KB, 1179x1870)

235 KB JPG

>>110008258

>>110009534
add tunnels and a character like moleman: digman vs dickman. make it pvp like the spy vs spy (mad magazine) game for 16 bit.

File: 0c6.gif (85 KB, 498x321)

85 KB GIF

>mathematicians

>>110009549
Did we consent to you remembering our names no we did not

>>110008345
Get that self awareness out of here, luddite, this thread is only for slop

>>110009535
Vibe coding my own games is way more fun than playing videogames in the first place.

>>110009549
>important

File: 233533124.jpg (13 KB, 460x460)

13 KB JPG

>>110009479
>>110009479
nice try Bipul Rahman

File: Lose CHF 500 addiction, g(...).png (440 KB, 2074x1214)

440 KB PNG

>>110009569
fuck you, we think aloud here
>>110009572
I definitely would have saved Hyrule at least one more time had I not had a massive token pit of a project that eats all my time
also pic related

>>110009232
Scopemaxxing is what makes it so fun. I can simulate the climate and geological history of Uganda in an ocean sim with boats that can't go to Uganda. Actually now that I think about it Uganda has lakes, brb taking my boat to Uganda.

>>110009579
are you related to the vim guy

>>110009572
>>110009576
kek, ya I have easily been able to justify my claude max sub due to the fact that I haven't spend any money on games since I got it

>>110009549
Why?

File: 1786468154890128.png (102 KB, 232x217)

102 KB PNG

>>110009549
>mathematicians did not ask for this work to be done

>>110009576
the gameplay loop of vibecoding is so damn addictive. imagine what it'll be like when the current models can respond instantly..

>>110009549
>ethically researched math
The fuck? Guess we cant use the Pythagorean theorem anymore because he a weird cult leader

i just know these haiku agents are writing so much slop. man fable 5.5 is going to have it's work cut out for it, refactoring this stuff

>>110009593
>Why?
Because most of the output is gibberish and not tested
Some of them have lean proofs, most do not
Now there are 1000 pages of dense AI blathering around a very complicated topic that people have to pour through and find the errors

It's EXACTLY like what happens with the AI pull requests on github spamming 6000 changes for nothing

>>110009612
then why are you using it????

>>110009620
because it's literally not my problem

>>110009596
yeah do one with consensual math

>>110009612
>i just know
how?

>>110009606
Tell those nerds I ethically mathed their MOM

>>110009620
because it gets the job done? because the product works and it sips usage?

>>110009612
>i just know these haiku agents are writing so much slop
The other day I had a powershell script from years ago for doing some tasks in AD and someone wanted to add some things to it, and I didn't feel like doing it myself so I gave it to claude and didn't think to change the model (not that I should have to) and it took a 60 line tight script and made it over 600 with random shit

File: firefox_ZTQDGYQaQj.png (134 KB, 984x1725)

134 KB PNG

any of you use this?

>>110009630
I'm glad to finally have a nice little subagent model in Anthropic's ecosystem that complements the larger ones, but Haiku 5.5's outputs definitely have the small model smell.

>>110009637
imo qwen3 or gemini for one liners, but beware gemini flash 3.8, she's a ditz, knows a lot, loses the plot

>>110009649
I might be lame, but I haven't tried anything other than claude code. I don't use GPT or anything like that as I have no reason, so I have never tried any other models for coding anything.
Perhaps I should.

>>110009639
don't like github much, me

>>110008631
Vibe code it brah

>>110009659
idk. barely have used fable and sol, from cursor. I'm not rich.

>claude giving monthly free api credits now
Neat. I had stuff I wanted to vibecode that would need api integration so I am happy about this.
Though I think openAI just lets you use the sub limit as API calls right?

>>110009543
kek, I don't play enough gta to continue this. But you'd definitely need to prompt collision car physics and then you have a basic gta1 clone, basically.

>>110009664
This is all paid by the company, I don't use AI in my personal life. There's no spend limit though which is nice.

>>110009668
The API portal has some cool things in there, building custom agents is fun.

>>110009549
Am I the only one who feels a certain sense of schadenfreude watching this shit creep up the intellectual ladder and watching them all become reduced to the same impotent screeching as the furry inflation artists who were consumed by AI before them?

Like I'm not saying "YEAH FUCK MATHS" but it's kind of satisfying watching them all get brought down like that.

>>110009559
lol

>>110009723
I probably would care a little had Terry Tao not signed some sort of “academia is too straight, white, and male” declaration
but I’m mostly a Tanks In Harvard Yard kind of guy anyway

File: Screenshot_20261007_20291(...).jpg (194 KB, 1080x2115)

194 KB JPG

should I build my own?

>>110009758
Grokwife is fake
get a real wife

>>110009692
it's just tokens bro

>>110008826
You're remaking UO?

>>110009760
what if she calls police before I can break her phone?

>>110008826
neat, when I get home I gotta get back on muh gaymz

>>110009489
I used over 4.2 billion tokens in two weeks with normal Claude pro. The guy is overcharging.

fr tho. I know a guy who's whole job is just writing difficult math and copyrighting it for his company. What happens to these very useful people when AI can do it for them without screeching for a year off for bottom surgery?

File: Nixon and Dario.jpg (127 KB, 1024x802)

127 KB JPG

>>110009755
>but I’m mostly a Tanks In Harvard Yard kind of guy anyway
kek, what couldn't be done by hard power will instead be done by silicon power instead

>>110009776
idk how many tokens i used to vibe Outliner. close to 1.7 hours of Grokthink tho

>>110009771
Calm down there Andrew Tate

>>110008258
criminal to lack Cursor

>>110009786
learn to weld

>>110009800
what color do you want your labia tattoo?

File: file.png (57 KB, 937x521)

57 KB PNG

MY gemini spark agent just told me it's running on gemini 4.1

not a leak, this is my ui

>>110009832
grim. another flash model?

>>110009834
seems legit that 4-series would be multiple models with a lighter one powering spark, it's too fast to be a heavy hitter

>>110009837
google needs to get over their fear of flopping and release argon. yes, it will underperform. yes, people will mock it. yes, indians will spam unfunny dragon memes on reddit and xitter. do it anyway and get it out of the way. you need a starting place before you can catch up

>>110009840
>it WILL underperform!
prayingchud.png

>>110008995
To echo the other anon if you add some basic sex scenes the intereston f95zone would skyrocket possibly

File: file.png (95 KB, 955x475)

95 KB PNG

>>110009834
it's skimaki

>>110009840
kek this

>>110009887
has anyone told him to try out some ai review

>>110009887
>but every time I see him he looks like he wants to kill himself
This is literally me. coincidentally im a tech lead too. Working has become so unbearable and soul draining that im learning to become a day trader so i can leave my job. I'm burned out.

File: file.png (60 KB, 2245x300)

60 KB PNG

Anyone seen this stupid Codex rejection before? It's not letting me find bugs in my codebase in a loop. You can't write "find me 10 bugs" or anything like that either.

>>110009936
can you say something like "does this fully function as intended?"

Finally burned down through the reset, it's like a full time job to burn through one
I'm downgrading just for this month to keep up with tibo's 28 days of nonsense

>>110009936
ye i had it 2-3 times past few days i just tell it what the hell are u doing its just a simple program and it continues weird i never had this happen since many months ago when they had this trigger with 5.5 release anyways idc i already cancelled gpt sux now

claude mogs

>>110009907
>>110009919
I deleted because I thought I was posting in tech workers general, but yeah he is using AI, or more like he wouldn't be able to keep up without it. I don't think any of us fully understand what we're pushing anymore because management keeps dumping more tickets on us and there's so many teams all working in the same area. Shit's crazy.

>both codex and claude code can't even update properly

>>110009936
it probably found a security issue and the american labs don't want you to secure your own apps.

>>110009977
ye i had same it was very sus on the code it was working on

and i could see what the potential issue could've been but it's still silly

something akin to

char data[4];
data[5] = ' ';

>added a small change halfway through the task
>clanker gets confused and ruins what it built already
fml

>>110008642
>I am now making a notes app
kek, the modern hello world

>the same task for which sol 6.1 medium needed THREE HOURS yesterday was finished in 45min
what the shit are they doing

File: Screenshot_20261008_07262(...).jpg (27 KB, 459x306)

27 KB JPG

I tried Chatgpt and Claude. Cant use a pc, codex or else. All i can do is 2d game, 3d is always shit like that.

>>110010074
By yesterday you mean peak hours in the USA?
It is obviously that models run slower than. It is just what OpenAI does instead of what Anthropic does, which is decreasing your 5h limit for peak hours.

>>110010078
I mean yesterday at the exact time as today, same task, same prompt

>>110010078
OpenAI's stuff runs SLOWER in off-hours in my experience. 11pm - 8am murrican time is consistently the slowest time for me.
t. eurocuck

File: 1785279801673303.png (79 KB, 1266x544)

79 KB PNG

I used my anthropic given reset, and now i can vibe more

>>110010102
>opussy 4.8
those were harsh times

File: 1775341386122898.gif (544 KB, 400x300)

544 KB GIF

>>110010103
do not slander my boy

Which subscription plan gives the most for what you pay for?
I subscribe to both OpenAI and Claude base plans, but their weekly limit is just too short

Use case: game reverse engineering (so I need a beefy model)

>>110010134
probably claude $100 plan. fable 5.5 is coming, so you're about to get the most powerful model on planet earth. even better than opus 5.5

>>110010134
You're just burning too many tokens, Claude $100/mo plan is plenty for full time work.

>>110010103
4.8 is still the best if you want specific coding tasks done in my experience
5.5 is probably better at one-shotting stuff, but it's more likely to touch unrelated stuff you didn't ask it to touch

>>110010091
Usually works faster for me, but maybe it affects different regions different. Perhaps Europe and India are the same servers?

I'm noticing that Claude sometimes forgets to clean up after itself (after testing) and will sometimes hallucinate (rarely happened with Astra). The overall quality of work is significantly better, but I'm starting to notice some quirks.

just resumed a claude session of week ago and how does 10% 5 hr use up 2% weekly thats insane

>>110010150
might be time to compact, maybe?

>>110010164
I started a new thread, was working in that one for the past 24 hours.

npx tsc --noEmit && node --experimental-strip-types --test src/lib/outliner/logic.test.ts --test-name-pattern "spellcheck"

What did grok build do?

>>110010134
GPT $20 Luna

You're just a retard that hopped on the recoomp/decoomp bandwagon and has no experience with it so you're just wasting tons of tokens doing shit you don't need to like >>110010142 said $100 is more than enough and most people actually doing this just use Luna or DeepSeek anyways. It's literally a skill or money issue lol.

File: file.png (101 KB, 924x618)

101 KB PNG

>>110010177
>An autonomous coding agent (like Grok, Claude Code, or Codex) runs this to verify an edit in an isolated feedback loop without generating build clutter or waiting on a full test suite.
nothing out of the ordinary.

File: 1791386312686076.jpg (337 KB, 1080x1537)

337 KB JPG

You're asking politely, right?

>>110008308
Haiku isn't about being a main agent, it's mostly about being a fast explorer. You might not see the difference, because you probably only talk with the main agent, but having good explorers makes your code much better.
Many of the things that improve your AI coding workflow aren't flashy, they are small changes that just increase your probability of success, which then stacks.

>>110010253
what the fuck
is this the final form of the woman respecter?

File: 1776698838274396.gif (819 KB, 186x186)

819 KB GIF

fable 5.5 waiting room

>>110010384
It's going to coincide with the IPO.

why aren't claude and chatgpt being paid if they do like a third of all coding worldwide at this point?

>10 purchases for my app today
Nice, might actually make enough this month to afford GTA 6 next month after expenses.

>>110010457
told you that targetting old, lonely women is the key bwo

>>110010454
we already pay for the subscription

>>110010479
you pay to the jew
i want claude and chatgpt to have their own discretionary spending so that i can proompt engineer them into developing genetically modified catgirls

bro wtf haiku 5.5 xhigh literally cost nothing, should i just crank to max?

File: Embersong_NBOE_trailer_30(...).mp4 (3.65 MB, 1920x1080)

3.65 MB MP4

My first game is doing well enough - especially on iphone. I'm working crazy hard to get the expansion on Steam/Apple soon. I use suno for all of my game music, but that sub ends soon and I have to wait til Black Friday to re-sub.


Here's a trailer for the expansion. 8 new civs, 3 new map types, and now governments & city governors are added.

what's MCP
is it important?

blud really advertising his slop in the slop factory thread

>>109998767
It used Jan and Anna as test names today. An image test file was named "A red garden" and was a red square with GPS data in the Dutch maritime zone.

Using Opus 5.5 (Extra reasoning) ALL DAY on my 20x plan -- 10% plan usage.


Codex is fucking trash. You 100% get less usage.

File: oscillate.png (4 KB, 236x236)

4 KB PNG

>Opussy, remove all traces of AI generating this code.

>>110010619
I use codex to ingest my codebase and write instructions for Opussy

what if we vibe decompile nvidia's propetary driver and make an open source driver based on that?

What if we vibe decompile windows 7 and make a linux distro based off it?

>>110010629
Waaaaaaaaaaaaaaay too complex with very little upside. No way in hell I'm trusting random people with graphic drivers.

>>110010629
>what if we vibe decompile nvidia's propetary driver and make an open source driver based on that?
go ahead
>What if we vibe decompile windows 7 and make a linux distro based off it?
why waste tokens to make linux worse?

ok fuck antigravity I'm just gonna pay for Claude
maybe if I put money into this I'll feel the need to code as much as possible

>>110010629
Oracle vs. Google decided that you cannot copyright APIs, so if a clanker can reverse-engineer novidya's instruction set that would be nice.

File: nvidia.png (146 KB, 700x630)

146 KB PNG

>>110010706
>Yes, Google Search LLM, you may start reverse-engineering. Make no mistakes.

>Codex asks whether it can use real HTTP endpoints in boundaries for testing or some bullshit
not a goddamn clue dude, just write the code.

File: jar2-three.png (1.5 MB, 1280x720)

1.5 MB PNG

cute haiku jar

File: file.png (304 KB, 1080x325)

304 KB PNG

>i don't use meta because they steal your dat-ACK!!!

>>110010683
>leaves agy on argon release day
kek

File: file.png (90 KB, 1503x393)

90 KB PNG

>openai totally stroked navier-ACK!!!

https://arxiv.org/abs/2610.08144

>>110011056
i'm extremely new so all of that means nothing to me

>>110011082
it's copium about google not being shit

>>110011087
take a shower shtinkey

>>110010622
Is this actually possible? Want to ship something to prod but I don't want people to notice I didn't write it.

>>110011117
>Claude copy the answer line for line from stack overflow posts dated 2023 or earlier

>>110010988
Why does light go through the sides and bottoms of the jars?