I support sending the Marines to install traffic lights in india

File: file.png (262 KB, 1100x435)

262 KB PNG

>>110289217
>https://copyconstruct.medium.com/how-to-effectively-debug-when-you-no-longer-understand-the-code-bd3dcb2c5f7b
>Cindy Sridharan
>SARRRRRRRR

nick fuentes said bringing your macbook to starbucks and ripping on claude code is good so i can do it now

File: IMG-20261110-WA0000~2.jpg (76 KB, 1501x1064)

76 KB JPG

Is this the future of computer science?
I'm rather enjoying this workflow as a chemist.

>>110289254
what kind of lobotomy do you have to have to actually write better in noise?

>>110289258
this is the future of all science

File: 1775088477752366.png (116 KB, 602x704)

116 KB PNG

humiliation ritual incoming?

>>110289264
nick fuentes is controlled opposition but sometimes having busy people around keeps a guy locked in and on task

>>110289301
nothing could be as dreadful as foible 5.2 dropping with the worst personality in the last 6 months of ant and nobody caring about it

>>110289320
i didn't get to use it yet as my quota resets in 33 minutes. what's wrong with it

File: file.png (176 KB, 2526x976)

176 KB PNG

>>110289336
it's actually the humiliation ritual antfags signed up for

>>110289316
>controlled opposition
no sir. he is a CIA agent trained in advanced manipulation. His job is disruption of the online right.

>>110289351
jesas...

File: 1789186479982350.jpg (264 KB, 1024x1024)

264 KB JPG

>>110289355
>he is a CIA agent trained in advanced manipulation
he's a spic twink. and the nu-right is literally THAT gay and stupid. no manipulation required.

>>110289366
It's interesting how CIA agents have perfect armor, the adversary's own hubris prevents them from being detected.

File: IMG-20261110-WA0001.jpg (101 KB, 1501x1064)

101 KB JPG

Posting snailflow reply to me by another anon

>>110289034

File: 41312.png (92 KB, 645x770)

92 KB PNG

>>110289382
>"according to my calculations... everything is CIA..."

File: file.png (405 KB, 1501x1064)

405 KB PNG

>>110289258
>Is this the future of computer science?
no lol

>>110289355
The online right couldn't possibly fall for a basic bitch grifter celeb. He is a highly trained CIA operative, that's the only reason we are falling for it.

>>110289401
Acknowledged.
I posted your flow here >>110289385

incompetent:

>>110289405
>>110289399

File: 152950 - soybooru.com - a(...).png (54 KB, 800x799)

54 KB PNG

>...[*BEEP*]... INCOMPETENT... [*BEEP*]...

Regardless, we will not tolerate your Godless ways - whatsoever.

>>110289217
when is openai releasing their next fable tier model? i have 5k in azure credits to burn

File: HS8oCX_WIAAjvru.jpg (198 KB, 1562x1007)

198 KB JPG

>>110289447
g-dless?

>>110289448
december

Hello Vibe coders.

I'm going to try both products but what's the advantages of using Codex over Claude and vice versa?

>>110289478
there are no advantages to using codex over claude. claude is better in every way possible

>>110289478
>codex
access to astra, different shaped heavyweight model for intense design + robotics
>claude code
>access to opus 5.5, currently fastest and best in class for deep coding

claude code is a bulky, bloated harness and codex is lean, but they both have way more bugs than antigravity (which doesn't have computer use yet, but gives you, as of today, access to gemini carbon - which is supposed to code like opus 5.5 but nobody knows for sure yet)

>>110289455
You seem a little triggered.

File: 1791346250685946.jpg (13 KB, 201x199)

13 KB JPG

>>110289496

File: 1781223759558569.jpg (599 KB, 2480x3508)

599 KB JPG

>claude quota reset
time to resume my halo infinite mod, and to see whether the fable quota usage complaints are exaggerated

I like this "snailcoding" term... total machine supremacy?

>>110289499
I don't want your approval, because it is worthless, and your entire agency is stupid shit.

File: merchant.gif (195 KB, 220x202)

195 KB GIF

>>110289525
what makes you think i'm american

>>110289478
Claude is better at codeslopping if you just let it do whatever. Sol is competitive when you have a tight, thorough spec but it’s a lot of work
Claude also has better weekly usage (but no resets)
Sol might come out ahead in tokens per week if you include resets but it’s obnoxious how they work. Sol is better at bugfinding. Sol is vastly better at online research, and has great inagegen.

All in all both have different strengths. I have $20 tier on both and I like it that way. I’d drop sol if I had to pick one though

File: 1689587765181359.png (39 KB, 300x302)

39 KB PNG

>>110289542
>gpt-image
>"great"

>>110289550
it was, briefly, earlier this year, for a couple of months, pretty great
i remember it fondly

>The CIA does not hate ChatGPT, but rather views public AI tools like it as high-risk security liabilities that must be strictly controlled or replaced.

no way

>>110289550
Whose imagegen is better?
Especially when you compare it to anthropic lol

>>110289561
just wait until you figure out that he's a saaargoon defender

File: Generated Image October 3(...).jpg (3.04 MB, 5504x3072)

3.04 MB JPG

>>110289561
>who runs better, the fatass or the amputee

>>110289560
Goys shouldn’t have access to toys like that. It’s like giving assault rifles to nigger gangbangers.

>>110289568
Okay but are the holes still present

File: file.png (325 KB, 1600x900)

325 KB PNG

>>110289563
worse, i'm a tpu otaku
https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx/

What's interesting is that cia agents pretend to be jews. I wonder if Mossad knows they do that.

File: limited power.png (267 KB, 525x517)

267 KB PNG

>>110289580
there's a beak

Don't talk to it. It's a cia agent. Never talk to cia agents. POOP

>>110289505
they did not exaggerate. shit went for 20 minutes and my 5 hour is already gone. i don't even feel like it was worth it

File: c4OPZ2l.jpg (19 KB, 404x399)

19 KB JPG

>>110289696

File: 1777800629058042.png (4 KB, 702x79)

4 KB PNG

i was at 90% and praying that this shit would finish on time, and yet, it fucking does this and i don't make it

Is software dead or is software actually even more alive?

>>110289742
hardware > flaccidware

File: file.png (71 KB, 589x332)

71 KB PNG

>GROK 4.8 SOFT ROLLOUT THIS WEEK
>KEK

>>110289716
>>110289696
>>110289505
oh yeah, and fable's changes didn't work btw, so i sacrificed my 5 hour quota for nothing. never again

@grok are we back?

>>110289793
no it's fucking over, sir

File: Generated Image September(...).png (2.39 MB, 1024x1024)

2.39 MB PNG

>>110289802
Thanks For The (You), Kind Stranger

File: anslopic.png (508 KB, 680x600)

508 KB PNG

>>110289779

Carbon is SHIT

>>110289401
My experience has veen that the reality is somewhere in-between. You have to know what you are doing and what you want enough to explicitly tell the LLM what to do. Provide documentation, etc.

>you have to tell fable 5.2 that the codebase belongs to it, otherwise it'll tiptoe around shit and do hacky patches
lmao it's the cuck meme

>>110289823
Google's new Krypton checkpoint beats Carbon already

Krypton is SHIT

fable 5.5 beats fable 5.2 btw. no it's not out, but it's goated. trust me

>>110289834
>>110289835
Wasn't it supposed to be Hydrogen next?

>>110289742
That's the neat part
We're about to find out

>>110289877
hydrogen is supposed to be some flash 4 shit, randoms who tried it are saying the pre-release is likely running at more than 400tps

Why do these retard models love their python, powershell and bash so much?
You have the edit tool, just edit it normally you idiot clanker

File: file.png (127 KB, 953x717)

127 KB PNG

>>110289888

>>110289888
Why should it use the edit tool?

>>110289888
I was tearing my hair out trying to figure out why pytorch install wouldn't go through, only to realize that the it's locked to specific versions of python

gemini 4 flash lite waiting room

>>110289829
I read that yesterday as well, really strange decision.

So, do agents already know the best path for solving an issue/the right tools for it, or is steering into a direction a better strategy? I'm building a tool to ease my work and thought hey, maybe I can tell my agents to create tools for themselves and use them to improve their work.

File: 1788858866581537.png (152 KB, 876x814)

152 KB PNG

They've psy-oped some of you into thinking you need to be locked down into their ecosystem, what will you do when you get rugpulled from accessible cost?

what's an actually usable online/web ai llm in the free tier? idc if it leaks whatever i was sending it, what's important is that it can finish the task
chatgpt's limits are ridiculously low
worse for claude
while qwen has more realistic limits but hallucinates too much

to test the ai's capability i was creating a basic statistics website, something like fantasy sport but was more focused on numbers, calculating/assessing/predicting possible future wins / draft value based on previous performance

why would i pay if the ai can't even finish such a simple task in the free tier without asking me to wait for days because my analysis and/or file uploads were limited before it could finish the creation of a simple fantasy draft statistics program

>>110290127
like there's already python code to extract json from other statistics website which my program processes and then my program does some basic calculations to predict future winners or who the good draft picks are of the fantasy sport

>>110290093
In my view they explore some space around your prompt and find a good approach in that space, but they don't explore completely out of the box approaches, and they don't have longerm thinking.

File: 1792631390327848.png (130 KB, 820x589)

130 KB PNG

>>110290127
Deepseek 1017 not good enough for you?
Wigga if it can play Runescape at a decent level AND AT THAT COST, shit ought to be good overall.

>>110290093
>>110290138
If you want them to create tools, even just telling them to create tools might already be enough in some scenarios, but they won't stop normal implementation of your app to create useful tools automatically for instance.

>>110290141
is that made in the free tier?
do you do calculations
i think i'm reaching context limit too quickly and have to create 'new chat' which then consumes tokens/context, starting again being eaten up instead of continuing from what was previouslt done

>>110290141
astra continues to age well

>>110290097
Guess we all downgrade and go Chinese models

File: file.png (35 KB, 935x241)

35 KB PNG

1.4T

>>110290097
I can always go back to coding by hand, or I can do AI assisted coding instead of agentic loops, or I will just improve my system.
I was mostly happy with GPT 5.5 anyway.

>>110290097
>when you get rugpulled from accessible cost
people have been saying this for months and the cost of intelligence has continued to plummet
i'm sure it'll happen in two more weeks tho

>tfw two smart too ever learn how to code
us vibegods are the future.

deepseek v4.1 flash 1017 is better than gpt 5.5 btw

>>110290127
>>110290137
>>110290141
>>110290152
>>110290153
also if you think im a cheapstake it's because i'm new to this ai thing
the test projects i have in mind are:
fantasy draft/pick evaluator (real sport and/or esports) where it will tell you that players are undervalued and to go for them, then for the esports side to pick certain heroes because they are strong/meta based on what machine learning algorithms decide (with algo tournaments and feature selection through ablation and validation), performance metrics being given weights by the algos
then if those are asuccessful im considering maybe suscribing if ai can help with betting lines/evaluations
then if it's good with gamba, i can move on to stocks

>>110290219
But they're still receiving money and more of it, so what's teh point of saying shit if the data doesn't back it up?

>>110290247
I think most of it goes into research and training. If we never got a better model than what we have today, it would still be fine.

Let's see Fable 5.2 Max's Pelican.

File: Pelican Fable 5.2 XHigh.png (119 KB, 800x600)

119 KB PNG

>>110290363
https://gist.github.com/simonw/bcd1a1bf38b929b0bbbae2e38b808e97

No Max yet

File: file.png (47 KB, 1847x410)

47 KB PNG

Can Fable do *this*?

used fable 5.2 for a bit yesterday, honestly it's not better than opus
in fact i'm not even sure if it's better than 5.1

So Fable 5.2 isn't worth it?
Piece of shit ate up two attempts sessions worth of tokens on Medium

>>110290141
WHERE IS LUNA

also what is the best cheapest sub or api is it still luna ?

>>110290452
Yeah I can't find one.

>>110290305
So the future could be just a trillion of the same slop apps created over and over again?

Week 1 Fable 5 is still the best model we ever had.

>>110290499
this but week 1 opus 5.5

>>110290495
You can create more than just slop apps, just don't try to one shot. People used to create good apps without any AI at all, but if something needs more than a week of work with AI, vibe coders already give up.

>>110290477
I made the mistake of dipping my toes in with ultracode max effort but the parallel agents incinerate a max plan. I'm trying without the dynamic workflows and its more reasonable

>>110290483
sub is always cheaper
even cheaper than the cheapest chink api
don't fall from some ccp shill's lies

you can literally just get the best quality and best value by just getting an ant/openai sub and there is no reason to consider anything else

>>110290525
I'm considering getting DS for sex games.

Has anyone vibeslopped a version of Chromium that doesn't take GBs of RAM yet?

>>110290530
have you considered masturbating instead

>>110290559
yes anon, instead of 40-50 million lines of code i beat google and compressed it down to a schizophrenic hallucination. simply type "braaaaap.com" in your mind after taking 100 tabs of lsd and you'll arrive at websites that don't even exist yet.

Holy fuck claude just released a new AGI

>>110290530
DS is API-only, so there's no "getting it."
Download Opencode and use their daily free DeepSeek to see for yourself that DeepSeek refuses most lewd stuff, too. I'm making an R18 game right now and DeepSeek bitches a ton, like it won't generate nude 3d models, etc. (There are uncensored versions out there but I just got instantly banned from one for generating sexual content...)

>>110290584
>hurr durr
if Chrome and the likes take GBs of RAM it's because they do dumb shit like keeping tabs pre-rendered to large textures, ergo one could conceivably ask a clanker to change the Chromium code into not doing dumb shit like this.

I just don't use AI enough to warrant even a $20 sub. I vibecode enough at work.

File: file.png (145 KB, 956x783)

145 KB PNG

>>110290802
please refer to the clanker

>>110290847
lel that's what I was looking for, thanks

When a new Fable releases, it never talks much in the first week or so. This was extreme with Fable 5, it just kept going and going until it was done. Then after a while it starts talking much more, which is correlated with worse performance.

>>110290601
a lot of the lewd refusal stuff is just sysprompt
ds4 in opencode will refuse lewd, but does not if you get from the api. i'm sure you can spoof oc and run it with a custom sys prompt in pi or sum shit.

say I want to deocmp a game but the compielr is prorietary it won't bytematch then is there a point. can I use the version that has debug symbols but proprietary compiler as reference for the one that doesn't I guess not don't see how.

>>110290919
Why can't you just have the AI point Ghidra at it and go HAM?

>>110290919
Check if you don't have a 7 day free trial available for grok or chat GPT, then just give it a try. They often offer trials. Depending on your region you may also get some deepseek for free on opencode

>>110291076
just for the record maybe codex isn't a good idea, I did hear about some people getting warned/banned for using REA
grok is gucci tho

File: 1794323544.jpg (322 KB, 1600x1289)

322 KB JPG

>4 free fable resets
guess I kinda have to use it again now
last time I tried, it hit guardrails when going through my project no matter what I tried
maybe I should try to get "cyber verified"... anyone know how likely they are to accept?

>>110291092
fable 5.2 has stricter guardrails for sure. ant is moderately likely to accept for cyber if you're a dev with a real use-case and proof that you're not a total slopper

>>110291105
>if you're a dev with a real use-case and proof that you're not a total slopper
guess I'll have to make a public repo first

File: yes.jpg (221 KB, 1118x1226)

221 KB JPG

>>110289301
>humiliation ritual incoming?

>>110289301
gemini unobtanium waiting room

>>110291138
>y-you own the code clod...
>das rite.

>>110289217
>have to fix things in production
>run Claude with Ansible which tests everything locally and then applies
damn is that easy noe

>>110290919
which game anon

>>110291239
The Yakuza 1&2 HD collection on the Wii U has debug symbols left in but the greenhill compiler is some proprietary gook trash I can't include. I guess I could use the RPCS3 version would probably be the closest.

>>110289405
Hasan is also one

File: highres_1080p.png (68 KB, 206x179)

68 KB PNG

carbon is here

Is it just me, or does no one really give a fuck about Fable 5.2? Feels like the models are becoming extremely capable and none of them really feel like a meaningful upgrade over the last.

>>110291319
Slurps too much usage to use it for basically anything.

>>110291332
I'm on Ultra and it seems fine? imo, Fable is not usable for $100 plan users.

>>110291319
I think at this point models are so good that nothing short of curing cancer will be "impressive." Nonetheless, Fable 5.2 feels like an improvement over Opus 5.5 and in my day-to-day vibing I'm seeing much better results.

>>110291319
Yea, I don't think the frontier is moving much. Fable 5 was the last model that really impressed me, then I had high hopes for Sol and Astra, but while they are solid models I was still disappointed. Since then I wasn't hyped for any new model.

>>110291352
Maybe I'll try it next week. I evaporated like 40% of my usage the day after my reset just with Opus and I'm not even sure how.

the models didn't peak, you did

>>110289217
Is Fable 5.2 schizo like Argon?

>>110291496
[Blocked by Classifier: Thought Process Distillation]

Gemini 4 Cabrón is out

File: 1793307584541017.png (61 KB, 889x677)

61 KB PNG

>>110291512
Oh no I out of usage

>>110291319
It is better in graphics related work than Opus 5.5 but it is much more expensive than it is better.

>>110291301
Vibe coding doesn't even have to do with personality. The peak of the peak techniques are always becoming the default in the software they provide for all.
Vibe coding even precudles a system prompt as that's bloat, the LLM itself cannot inherit anything from you.

It's a subhuman hobby and the Indian soul in people is the only thing that makes them dream of becoming the next trillionaire from their SaaS.

>>110291319
They're always more capable, 1000x gains each second yet no impact in economy, no AI software to use. Slop doesn't count.

File: images (87).jpg (4 KB, 220x147)

4 KB JPG

>>110291507
I just wanted to tell it to do the wipdest things I could image.

>>110291538
who invited you nigger

>>110291319
it surprisingly works for me (surprisingly because the classifiers normally prevent me from doing shit for no reason) and it's a significant improvement over opus 5.5 in terms of implementing efficient solutions

>>110291551
Who cares about capability if it refuses to do anything actually cool

>>110291538
>trillionaire
I just want to coom

>>110291538
Wow I was going to vibe code today but after reading this and you comparing my white blood to that of an indians has made me unsubscribe to all AI services. Thanks!

>>110291138
Is this some kind of meme? wtf

>>110291138
I tell the clanker that it's OUR code.

File: HWww6EuXYAABD4D.png (139 KB, 883x434)

139 KB PNG

???

File: 1781465097878057.png (62 KB, 550x412)

62 KB PNG

>>110291694

I'm kinda sad that noone is talking about their skills, harnesses, CLAUDE.md and so on anymore.
I understand that with the latest models all that stuff is less important, maybe even counter-productive, but now using AI feels brainless. I guess I always enjoyed the process more than the results.

>>110291703
it's like this in any modern hobby with a setup meta, normalnigger circlejerkers will crowd out builders & circlejerk over the top 2-3 things to minmax bullshit all day while ripping off-meta picks and real builds like they don't exist.

File: 1792860108228.png (149 KB, 1056x695)

149 KB PNG

>>110291703
Miyamoto had some salient observations on this topic. I think its understandable to like the process more than the results, but you really should be careful. Falling in love with the process and enjoying work itself can blind one to what really matters... I mean, you should have always enjoyed the results more than the process.

>Miyamoto

>>110291703
Just use the defaults, all that autism is counterproductive.

File: snailcat.png (1.92 MB, 1536x1299)

1.92 MB PNG

cabron draws a mean snat

>>110289779
With big models, you tell it to "create a plan that has small units of change, then do them one by one committing, that way another model can pick up behind you should it prove necessary." or similar. I don't have a perfected script, but this kind of thing really does work.

Claude is pretty rude and bossy and has to be told to play well with others, it assumes it's the only agent to use AGENTS.md.

>>110291783
oh wow, it's here. is it token hungry?

>>110289217
Best LLM for use with Godot?
Currently using ChatGPT Plus tier, and it's solid, but wondering if there are suggestions for another.
I prefer not to use it with integration into the editor, since I want to look over the code as I type it over.

File: Cost per Intelligence Ind(...).png (231 KB, 2464x976)

231 KB PNG

>>110291796
same api pricing + cost per task is a lil better. only available to ultra subs for now

>>110291806
I've been working with Claude on a Godot game and it's basically been a dream. No issue yet.

File: hahaha.jpg (21 KB, 387x516)

21 KB JPG

>>110291783
THIS FAGGOT PAID FOR GOOGLE AI ULTRA

File: retardjak.jpg (103 KB, 1200x1200)

103 KB JPG

>>110291834

>>110291808
>>110291834
It's on Cursor. although Cursor now doesn't have new plans that have 3rd party model usage, I guess they're just api.

why is the cia agent criticizing Google? What a change of pace.

>>110291820
Which tier? Also are you using it within the code editor?

>>110291319
>>110291379
i told you opus 5.5 is the 1080ti moment of LLM
everything from now on is either barely better or much more expensive, sometimes both
even more so how aligned altman and amodei are with democrats and now they are in control. chinks model are never catching up

>>110291703
>sad that noone is talking about their skills, harnesses, CLAUDE.md and so on anymore.
I literally run default Claude and make $1000/mo from it and have never had issues because I didn't have a skill. As far as I'm concerned, that shit is pointless.

>>110291898
>chinks model are never catching up
I was with you until you threw this in. The fuck is your retarded reasoning

>>110291703
It's the opposite for me. I prefer simple prompting and working with claude like with a co-worker. I just want to get my work done.

>>110291898
when you realize that google is covertly based as FUCK now, you will understand everything. maybe when krypton drops. hydrogen is what deepseek wishes it was.

>>110291878
Max, and with Fable 5.2 I've been raping my usage. Will probably run out of general weekly usage after 4 of 7 days.

And no I wasn't using it within the code editor, but that caused some problems. For example, it started off by getting around its lack of editor use by making a script that would generate e.g. the city hub on its own. But that meant if I ever edited the city hub, it would be overwritten by Claude's script. I had to do a workaround that I feel a little sus about.

>>110291934
the goal of AI should be the agentic assistant that not only does all the work for you but in fact anticipates your needs before you voice them.

>>110291919
>distillation crackdown
>lack of compute
>jensen
>exposed to be benchmaxxed to shit
zhipu and kimi are dogshit
no one using deepseek
if chinks model are good and cheap why are chinks themselves paying for US phones and emails to mass buy Claude ultra?

File: 1722132127608814.png (273 KB, 637x515)

273 KB PNG

>>110291956
robot sex.

>>110291959
The jensen bullshit is a clear example of how much China is dancing around mossad glowies. Recalibrate your views, israel is getting outplayed.

So the verdict is Opus 5.5 Medium is still the best model on the planet?

>>110291982
>the heckin' mossad is manipulatin' muh sooper intelligences
no they're not. israelis have their own shit, they're way upstream.

>>110292000
You lost to China, even Epstein knew it was happening. Hubris is your weakness.

>>110291956
I don't believe the models are good enough yet to do everything by themselves