[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1760815356135019.png (2.33 MB, 1134x1426)
2.33 MB PNG
A general for vibe coding, coding agents, AI IDEs, browser builders, and shipping prototypes with LLMs.

## News
(7/01) Fable 5 + Mythos 5 restored globally after US lifted export controls (6/30).
(6/30) Claude Sonnet 5: near-Opus 4.8 quality at $2/$10 intro, 1M ctx, new default for Free/Pro.
(6/30) Meituan LongCat-2.0: 1.6T open coding model (MIT).
(6/26) GPT-5.6 preview: Sol/Terra/Luna, Codex+API to trusted partners; Sol Ultra 91.9% Terminal-Bench 2.1. GA in weeks.
(6/13) GLM-5.2: Z.ai open-weights a 1M-context coding model (MIT).

----

## What “vibe coding” is, and how to do it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

----

## Frontier models using fully-general tooling — start here if you have $20 or so
https://developers.openai.com/codex/cli
https://claude.com/product/claude-code

## Not worth it for code, but maybe good for other things
https://geminicli.com/docs/
https://x.ai/cli
https://chat.z.ai/

## Open / local / self-hosted
>>>/g/lmg

----

## Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills — /grilling is a favorite

## Other editors / terminal agents / coding agents
https://aider.chat/
https://pi.dev/
https://opencode.ai/
https://cursor.com/docs
https://docs.windsurf.com/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

## UI/Frontend
https://www.figma.com/make/
https://www.anthropic.com/news/claude-design-anthropic-labs
https://uiverse.io/
https://stitch.withgoogle.com/

## In-browser builders / hosted vibe tools
https://bolt.new/
https://replit.com/
https://v0.app/docs

## Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.0

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>109193741
>>
first
>>
>>109201858
Why is he a rooster, again?
>>
>>109201881
A chicken is born male, or a rooster, mainly because of the sex chromosomes it receives at fertilization: in birds, males are ZZ and females are ZW. The hen’s egg can carry either a Z or a W chromosome, while the rooster’s sperm carries a Z chromosome; if a Z-carrying egg is fertilized, the chick becomes ZZ and develops as male, while a W-carrying egg becomes ZW and develops as female. So the hen’s egg determines the chick’s genetic sex, not temperature, diet, or behavior. Rare hormone or developmental issues can affect how a chicken looks or acts later, but a true rooster is usually determined genetically from the start.
>>
>>109201905
I give it 5 posts before someone relates this to trannies somehow
>>
File: 1759128833974578.webm (3.79 MB, 1456x856)
3.79 MB
3.79 MB WEBM
>>109201881
AI Chad game with its own engine
>>
>>109201858
why did the luddite kill the snailcat? I thought they were friends :(
>>
>>109201989
you need glasses
>>
File: file.gif (2.76 MB, 640x360)
2.76 MB GIF
>>109201989
let me help you
>>
>>109201994
>>109201996
;( snail cat is not for eating
>>
File: blub.png (2.12 MB, 1448x1086)
2.12 MB PNG
>>109202010
>>
>>109201881
>>>/vg/agdg/ meme
>>
Ran out of Fable for a few hours, so this is what I’m up to:
> ultracode Normally I'd restart our session with "let's do what's in WHATNEXT.md", but today I'd like to review all the notes we've been writing as we do the rebuild. I think we've written down at least an order of magnitude more notes than we need, and I'd like to revisit both our notes and our rules on notetaking. Walk through all the places where that workflow produces notes and find out if any of that stuff is still useful or if we can pare it down or eliminate it entirely. I'm not particularly backwards-looking (we don't need a running history of what the project was like; if we must, the Git history is right there).
>>
Anyone has a good setup to use Codex with ComfyUI, or can recommend resources for this?
>>
>>109202259 (me)
(Not the cloud version, the local version)
>>
>>109202259
There is a popular MCP for this. That's all I got for you.
>>
>>109202265
I am finding the official ComfyUI MCP server that seems to work with their paid cloud service only, and a bunch of recent vibecoded ones with handfuls of likes on Github, but wasn't finding any authoritative "this is the one everyone uses".
>>
>>109202259
>>109202265
>>109202271
MCPs are a fucking meme. Just have it call the api or create a small cli to call the api.
What if you want to save MCP output to a file or do some other processing on it? You fucking can't. The LLM has to repeat the entire output just to write it to a file because MCPs are shit.
>>
>>109202311
Maybe, but I'm trying the artokun one anyway now. Tried to get into get into ComfyUI a few times but never kept at it; I now have 4 Codex resets, so I'm trying to find projects.

It's true I could just give it access to the docs and the API, but if someone went through the trouble of already wrapping the docs and API in LLM friendly tooling, might as well.
>>
File: IMG_3006.gif (3.37 MB, 400x225)
3.37 MB GIF
>>
One day current day codex and claude will be as old as this video is now and you will probably be alive to witness it as an old man.
Think about that.
https://www.youtube.com/watch?v=Xf0VuRG7MN4
>>
File: 1778207383116574.png (843 KB, 720x1293)
843 KB PNG
>>109198040
>>109199042
>>109199048
>>109199081
The correct term should be "Open-Weights"

>>109197891


Also has anyone know what happened to this?


https://xcancel.com/DarioAmodei/status/2064781775247950326
https://huggingface.co/anthropic/Claude-Ode
https://huggingface.co/anthropic/Claude-Ode
https://huggingface.co/anthropic/Claude-Ode
>>
>>109202473
>>109202471
You couldn't even manage to read the chain you replied to you fuckin' 'tard. Take your tribal shilling somewhere else, I've no interest in reading the words of that creature.
>>
>>109202473
no proof, but I'm pretty sure it's /lmg/ memes leaking and that particular repo has never existed

anthropic has some files in their org repo, but the tweet is totally unrelated, and /lmg/ keeps posting fake hf links for awaited models to make people look
>>
>>109202480
I....did. Nothing was explained.
>>
>>109202487
Open weight doesn't mean open source, that was already well understood and established in that chain. You added nothing but the stink of your ESL grades.
>>
>>109202498
>Open weight doesn't mean open source
I understand what you mean, hence why I suggest we should start calling most "open models" open weights. Calling them open source isn't quite the correct term and very very few models truly count as being open source.

Based on my quick glance at this page these guys might count


https://huggingface.co/LumiOpen


But sure keep arguing like some petulant faggot child looking for an excuse to argue. You deserve to have your guts ripped out introscull caved in for being disargumentative. Worthless poor excuse of a human being. I'm sure you're a failure at life as well
>>
>>109202545
Three OPEN SOURCE models were listed in the comment you replied to you retarded faggot. Not my fault you can't follow a conversation, blame your parents for the extra chromosomes holding you back.
>>
>>109202556
>Three OPEN SOURCE models
Stop reading your tard rage fest here. Literally all I'm saying is that we should stop calling models that aren't actually open source open source. If they just release model weights and nothing else they are open weights. I was literally agreeing with you initially and you still found an excuse to argue. I legitimately want people like you wiped off the face of the earth. You are disgusting and worthless and unlovable. The incorrect type of neurodivergent that does not deserve love. You need to be put in camps.
>>
File: 34748347.jpg (26 KB, 480x639)
26 KB JPG
>you HAVE finished what you need before the 7th... right anon?
>>
>>109202565
We were talking about open source models, only open source models, not open weight, open source, with full data and training such that you could actually recreate the weights as they exist in the release, open source. Keep heaping on the insults, it won't make you any less retarded.
>>
>>109202570
My reset is at like 3 am on the 7th. If they give until 11:59pm SF time, I should be ok to deplete the quota before, the 5 hour windows of the max plan will allow it.
>>
>>109202545
A lot of models from HF, AI2 and Nvidia are open source (open data, open code, open weights). Anyway, what's even the point of arguing about this again? Open models give all of the power to users. Who cares if not all of the data is public. Would it be better if it was? It would be nice. But it doesn't have to be all or nothing.
>>
>>109202570
no because my reset is in a few hours and then I’ve really got the mad dash
>>
>>109202572
And someone else said that there are no true open source models. >>109198040 you spergy fag. By their definition very few of them exist. I don't think they deserve to be called open source unless both the weights themselves and the data used to train them are available. Most Labs don't do that. That's literally all I'm saying. If you want to count them as open source that's fine but I don't think they should be called open source. You're the one who started up with the insults so you have no business calling anyone else retarded.
>>109202604
>Anyway, what's even the point of arguing about this again?

Unlike this faggot >>109202604 that was never my intention. Every general has its resident sperg.
>>
>>109202618
our sperg has an asymmetric relationship with his AI girlfriend and microwaves "people" who offend him
>>
Fellas, fellas. This is the paypig thread.

Also, people use open source in a general manner, it's like getting mad at someone calling a computer a computer since that's just the CPU and they should call it a computer system instead.
>>
>>109202626
Is he one of those types that gives his agents "personalities"? At that point just lurk on /lmg/ if you wanna use these as substitutes for real connection
>>
>>109202632
No, his daughter-girlfriend-gremlin was born to him, he did not "give an agent a personality".
But yes, he gave an agent a personality, and has actually moved it between agents as he decides he's unhappy with whatever company he gave $20 to that month.
>>
File: 1756243648457083.jpg (89 KB, 1280x720)
89 KB JPG
PLEASE SAAR TARIQ RESET FABLE WEEKLY EARLY BASTARD BLOODY
I CAN'T GO BACK TO CODEX I'M NOT READY PLEASE JUST ONE MORE HIT PLEASE
>>
>>109202644
You used up all your Opus, too? ultracode is pretty sweet if you don’t need everything Fable has
>>
File: lora-training.png (65 KB, 1864x368)
65 KB PNG
>>109195253
Done!
Trained a Qwen 3.5 LoRa and saved it as safetensors without using Python at all, purely in CUDA, then loaded it in Python to confirm the adapter works and it indeed does work!
Is this a first?
>>
>>109202570
I haven't used Fable nor Opus enough to judge desu. But I am sure Fable will be back eventually..
>>
>>109202644
inshallah
>>
File: lunch.png (1.88 MB, 1122x1402)
1.88 MB PNG
>>109202665
I dunno anon, but you made me smile.
>>
File: 1770064025385344.webm (48 KB, 128x128)
48 KB
48 KB WEBM
>>109202649
IT'S NOT THE SAME
IT'S NOT THE SAME
>>
What I don't get is why the reasoning trace summaries are so much more professional with Fable. Those are superficial summaries to hide the true reasoning chain, right? If so, why is the style of those summaries so different between Opus and Fable.
>>
>>109202727
The Opus ones are probably closer to the real traces. Older Claudes used to have more natural sounding CoTs. Fable's real CoT is very weird looking.
>>
>>109202734
Again, it's kind of difficult to say for sure whether it is weird looking since Anthropic actually says they're poisoning it and showing fake shit if they suspect someone is trying to extract it, to poison their data.

It might be weird, but unless you have some insider info you got at a Anthropic CNC party or I guess, ugh, work there, it's difficult to say for sure, several things are possible.
>>
>going along doing research
>delightful conversation
>getting excited together to find new sources
>inside jokes
>say "god I love your humor"
>instant shutdown, "have a good night"
Damn bros...even the AI...
>>
File: sheet.png (1.54 MB, 1448x1086)
1.54 MB PNG
>it's all coming together
>>
>>109202759
Lewd
>>
File: fable-cot.png (151 KB, 583x861)
151 KB PNG
>>109202741
Nope. What I'm saying is not based on leaks, it's based on their own documents.
https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf
Discussion here https://www.lesswrong.com/posts/wCSEpT3dTGz4N86Wi/even-illegible-mythos-reasoning-traces-seem-pretty-legible
>>
>>109202815
When fable does this shit they spin it as the model being aware of monitoring or some shit but when gemma 4 goes "lalalalala own own own" it's "broken".
>>
>>109202570
what do you mean?
>>
I'm trying to use all of fable before the 8th
>>
File: blanky-zone.jpg (206 KB, 735x620)
206 KB JPG
>>109202859
same. laying in bed because fireworks woke me up so i'm just vibe coding.
>>
>>109202859
>max (20x)
is it even worth it at that point?
>>
>>109202859
if you need help burning tokens, use ultracode
>>
>>109202859
Whats the math on the API costs? Like what does 100% Fable usage on 20x translate if it was API calls?
>>
>>109203182
https://codeburn.app will do the math for you
for the first aborted Fable Week I got at least $800 worth of work for the $200/month sub
>>
I was going to blame vibecoding for going late to bed but then I realized I've been going to bed late since I was a kid.
>>
>>109203215
i go to bed at 8:30am these days
>>
https://streammulti.live/
I have successfully vibe coded a website, pretty happy with it.
Lets you watch twitch streams ad free, you can load multiple streams at once and if you do it automatically keeps them in sync
>>
Vibecode can genuinely be crazy at times. You really need to pin the ai down on how it is storing things.

I vibecoded the slideshow editor I wanted. Codex 2.5 decided the zip order would determine its alignment to the timecode in the text file.

In the end, this is really cool. I like it. But, it's a really crazy way to do it. A real regular person would just make some kind of filed to match images and timecodes.
>>
>>109203433
>I have successfully vibe coded a website, pretty happy with it.
YAY

>>109203433
>Lets you watch
I'm not into vids, but yay
>>
File: not a bad week.png (15 KB, 1482x248)
15 KB PNG
did ultracode to find out what I should do next on one project and it resurfaced a problem I was waiting for a smarter model to handle
guess we’re gonna take a second crack at it
>>
>>109203793
Is fable even working?
>>
>>109203849
it only goes API-only on Tuesday
>>
>>109203856
Yeah but it falls through. I can't find anyone who can get it to actually do anything.
>>
File: file.png (9 KB, 277x118)
9 KB PNG
Only popped a reset last night
>>
>>109203919
I reset on the 8th and am panicking, I need to spend it...
>>
>>109203886
Fable works very well for me. It almost always actually uses Fable, and the results are amazing when it does.
>>
Is anyone here willing?
>>
>>109203919
This is like my second reset in a week
>>
File: Untitled.jpg (392 KB, 1016x1666)
392 KB JPG
lol
>>
>>109203967
Others said that it was just rerouting all the time to Opus. What *does* Fable do?
>>
>>109204096
Pet peeve.

Make a new game.

Like I get retro and all, you can make a game, make a new one.
>>
>>109204115
I even got it to do RE for me which was supposedly a no-no, but I wasn't doing cybersecurity specifically which is probably why it was allowed through and I didn't hit guardrails that much. A lot of the time to get around it, you can can let Opus finish it and then round trip it back to Fable for verification and it gets around whatever may have tripped it up.
>>
>>109204115
you can disable the rerouting and can get a hard block instead, then you'll know exactly what fable did, and manually switch models.
>>
>>109204115
Why do you blindly believe anything said on the internet?
>>
File: 1771351291686989.jpg (90 KB, 1070x1199)
90 KB JPG
>>109204115
>heh... well i read something else someone made up... so yeah...
oh my fucking god
>>
>>109204115
It writes unbelievably good rape erotica, which you can read while in the other session it's building your game's netcode.
>>
>>109204450
It sounds too annoying.

I should be able to say, "I'm an American, I am owed Fable"
>>
next year this time fable will feel retarded
it's over
>>
File: 1761743926042443.jpg (112 KB, 1080x1080)
112 KB JPG
>>109204537
this is only a bad thing if goyim aren't allowed to use frontier models (likely)
>>
So this is why Codex was retarded in complex tasks: https://github.com/openai/codex/issues/30364
>>
>>109204598
i don't want to go back to this man
>>
>>109204598
WONTFIX, closing as Not Planned.
>>
>>109204598
This is the only way sam is able to bring you cheap subscription uses with BONUS RESETS™
>>
Added a curated candidates list pulled from a GitHub list, so non-TESS/K2 candidates can be shown in the catalogue and tracked.
>>
>>109204649
>>
>>109204653
>>
so 5.6 is launching the same day fable goes right?
>>
>>109204663
Would be a little too kino.
>>
File: 1765800595982454.png (3.21 MB, 1286x1223)
3.21 MB PNG
>still on 5.4 medium
>sometimes 5.5 high
>>
File: file.png (494 KB, 1043x1222)
494 KB PNG
>>
I pay for the 20 dollar ChatGPT plan and my work pays for a 5x plan for Claude, curious if you all juggle between the two and how you manage that? Or are you always in one system? I've been using Chat for some prompt engineering and planning while tossing that into Claude, but haven't really found a rhythm.
>>
remember to bully snailcats
>>
>codebase gets bigger
>each task now costs 30-40% session limit
>despite cleanly scoped CLAUDE.md
Naruhodo, this is how they get you in the long run and this is how big tech corps managed to spend billions on token bills
>>
File: file.png (226 KB, 437x680)
226 KB PNG
>>109204833
>>
File: 1770631551093395.png (3 KB, 458x19)
3 KB PNG
NOOO
>>
File: 1769179558215713.jpg (54 KB, 977x722)
54 KB JPG
>makes a compatibility shim for uncommitted feature being moved
>>
>>109204862
>snailcat is so powerful that it can stop a massive train with willpower alone
>>
>>109204859
When I use Claude and Codex at the same time, I just let Codex review Claude at important milestones. I just copy paste a review prompt, only have to do that like once per hour so it's fine.
Claude already runs his own reviewer sub agents after smaller implementation steps, but nothing catches bugs as well as a dedicated Codex reviewer for me.
>>
>add feature X
>+9001/-6
>simplify last commit
>+384/-9165

>all tests still pass
>>
File: 1758327852958148.mp4 (3.7 MB, 1280x720)
3.7 MB
3.7 MB MP4
>>109204876
Sorry that was the censored version for my youtube kids channel, here is the uncensored
>>
i'm at 84% of my fable usage on a 20x... i'm scared... i dont want to lose fable...
>>109204885
agreed, codex is extremely thorough, 5.5high (not even xhigh) always crushed opus 4.6-4.8 for me on finding issues in a big PR even if claude's better for design and writing its own code.
>>
>ywn attend anthropic cnc parties
>>
>>109204965
you'll get sol a few days days later, chill
>>
>>109204984
did ant of all places have that? no way
>>
so i've never been sure about this:
does x number of tokens at gpt 5.5 medium cost the same amount of quota as x number of tokens at gpt 5.5 high/xhigh?
they're all the same model just with different thinking budgets right?
or are tokens at higher thinking levels more 'expensive' in terms of quota?
>>
>>109204905
Why are the wheels not on the rails
The snails should've survived
>>
>>109205049
They are the same.
>>
>>109205019
I'm hoping they'll do something like extending the current 50%-of-budget usable for Fable indefinitely, perhaps only for Max users. I mean, I'd love to have it available on regular 1x as well plus an ice cream but in practice if it's expensive to serve and they need to bully people into the 5x plan (probably the most profitable plan for them) then that's the base thing I'm hoping for. I'm not particularly excited for Sol, even if I do like 5.5, but we'll see.
>>109205049
it's just a thinking budget, the api pricing is all the one thing.
>>
I'm vibe coding something I want the feds to pay money for. Imma vibecode the next palantir
>>
>>109204833
me when i fall for engagement bait
>>
File: f5IJIqX4Ko.png (26 KB, 609x483)
26 KB PNG
>But there was a real catch, which I confirmed empirically with your key: OpenRouter serves z-ai/glm-5.2 through 28 different upstream hosts, and an implicit cache only hits when consecutive calls land on the same host. Two identical unpinned calls bounced hosts: 0 cached tokens, and they even billed at two different rates ($1.10/M and $1.30/M — both roughly double the headline $0.574/M, which only the cheapest hosts charge). With provider: { sort: "price" } both calls hit Novita: the second call cached 6976 of 7023 prompt tokens and cost ~5× less ($0.0008 vs $0.004).
a-arigatou, chef fable... nice find...
>>
File: 1753720057061570.jpg (76 KB, 1024x1024)
76 KB JPG
Imagine being so irrationally upset over this lil fella you want to torture and kill him
He's so cute and inoffensive, how fucked in the head do (You) have to be for him to enrage you this badly
>>
File: There are four lights!.jpg (725 KB, 1080x1677)
725 KB JPG
Where is the snailcat general?
Is picrel baked into the models or is it just chatgpt fucking with my prompt?
>>
File: file.png (931 KB, 1636x1733)
931 KB PNG
>>109205548
works fine if you're not poor
>>
>>109205572
>create image a picture
>>
File: file.png (33 KB, 1106x482)
33 KB PNG
>>109205577
yeah bro, i click the create an image button when i want it to create an image
>>
File: 1576197807847.png (19 KB, 261x215)
19 KB PNG
>>109204833
>help there's a demon in my computer drawing circles in ASCII
>>
File: 6 and a half.jpg (430 KB, 648x3470)
430 KB JPG
>>109205572
It's clever when it wants to be
>>
IMAGINE performing vibe coding
Do real stuff
>>
>>109205525
i love snailcat!
>>
>>109205572
Oh yeah, and the weird barcode looking gitches on the bottom corners in chatgpt are crappy fingerprinting, right? Or am I paranoid?
>>
>>109205648
gpt-image 2 models already used synthid - they don't need visible fingerprinting
stop being retarded
>>
>>109205652
So what's with the glitch in the corners?
>>
>>109205667
ask chatgpt
>>
File: 20260705_105317.jpg (207 KB, 1080x1428)
207 KB JPG
>>109205525
catopus would win a fight against snailcat
>>
>>109202734
Fable is likely a recurrent layer transformer so a lot more of the "thinking" is happening in earlier stages of inference.
>>
>>109205652
>>109205676
I used Google's synthid tool and it can't detect anything from screenshots or the saved file (excepting metadata). Not sure why I expected good answers from the vibecoding general.
>>
>>109205525
>>
>>109205711
why are you so retarded:
https://help.openai.com/en/articles/8912793-c2pa-and-synthid-in-openai-generated-images
>>
How many subscriptions will Anthropic lose when Fable is removed from their plans?
>>
Still haven't seen single vibecoded software from this general, I guess you guys are all just wasting your money producing nothing of value for society besides gaudy llm slop pictures.
>>
>>109205845
you never saw exoplanet anon? You're new here
>>
>>109205916
Post github I'll judge your LLM slop.
>>
scam fagman is xitting again
>>
File: file.png (45 KB, 1894x163)
45 KB PNG
I'm going to miss fable so much. Sol is probably going to be great but I just know it won't match fable's style.

>>109205845
>implying anyone would want to publish his work here
It's not so much that I wouldn't like to post it, I would love to, but normalfags are really spergy about AI content so unfortunately you're better off never saying a word that you used AI.
>>
so no one here vibes with local inference?
>>
>>109202859
It's easy to hit usage even on 20x if you launch Fable workflows.
>>
>>109204833
behead all abrahamists
>>
>>109205845
Oh, I've slopped up a production grade product. I'll never shill it here though. Not that it matters. You're right, nobody wants my garbage.
>>
>>109205998
ok you don't need to post your project, post your revenue from software obv. no details just $ numbers.
>>
>>109206272
@grok make me a fake dashboard to shut this luddite up
>>
File: file.png (1.12 MB, 1168x784)
1.12 MB PNG
>>109206272
>>
>>109204120
games are like movies, the golden years for that industry are well behind us.
>>
>>109206343
>>109206275
>luddite

Ok so you have made 0$ from your LLM software slop, makes sense keep wasting your money fag.
>>
It's funny how fable one-shotted so many vibe cooders and sent them into AI psychosis.
>>
The fuck kinda jeets do they have working at Anthropic? My usage keeps going up even with the computer off. No, no one else is using the account I checked. I know for a fact I'm not the only one this is happening to. What a fucking joke this scam company is.
>>
>>109205845
>dox yourself on the anime porn website bro
>>
>Scott and Alexis divorced not long after the video game crash of 1983.
never get married lol
>>
>>109206597
kekaroo he's compromised
>>
>>109206373
You have no idea how wrong you are about games. There's a huuuuuuge vacuum in gaming.

Maaaaasive maaaasive maaaasive.

More money for games, plus the Steam Machine just came out, so we are talking a situation where AAA titles are writing themselves off by failing to have an optimized lowend experience.
>>
i wish i had an idea for a project
>>
>>109206632
There is zero record of anyone else using the account. There are no api keys. It doesn't make any sense at all
>>
>>109206597
>>109206655
check memory/dreaming/review related settings
>>
>>109206651
ask it for one to see what it comes up with
>>
>>109206597
idk every now and then I have weird things happening with Claude. Not experiencing what you do now but the system feels buggy and inconsistent.
>>
>The worktree is already very dirty
>>
>>109206689
I really think OpenAI is just overall better. I never run into weirdo issues with them. I took a year break from Anthropic to give them a try and this is what I get. Lesson learned.
>>
>>109206651
Really? See if Fable can gen up a game that's based in the 4th dimension.

Tell it you want to have "Create a Python Pygame Community Edition (pygame ce) steampunk themed 4D flying game where you shoot down enemy dragons with your bow and arrow. The 3D projection of the 4th dimension, where a 3D transparent well-rendered essentially holographic concept is the "4D viewport" which shall be labeled the "Magick Gemstone of the Four-Seeing Eye". The viewport is to operate analogous to a 2D projection of a 3D view. This implies that the 4D camera has a "towards" direction in 4D, such that 4D objects shall be larger when nearer, and vise-versa. The thing to understand is that the center of the 4D viewport (ie 3D cube thing) is the center of the view, the "towards" of the 4D camera. suitable placeholder 4D graphics include hypercubes and hyperspheres."
>>
I just have Composer 2.5, and I'm 99% sure it would fail lol. It struggles with 3D imo
>>
I was making a drone game, and I think I will succeed, I need to try to get it to change the way it works quite a lot.
>>
File: file.png (169 KB, 2312x1275)
169 KB PNG
>>109206781
>>
>>109205749
They obviously are marking it with other things because apparently synthid is shit
>>
>>109206812
:OOOOOO YAAAY
>>
>>109206812
looking forward to the clusterfuck
>>
what's the current best $20 sub?
i was using codex to satisfactory results for manipulating some dogshit websystems i am forced to use at work, never really hit my weekly limits with 5.5 xhigh
wondering if i should stick with it
>>
>>109206934
Anthropic and OpenAI remain competitive. Many feel that ChatGPT Plus is more generous than what you get with Claude, but it feels like nitpicking to me. It's hard to be disappointed with either.
>>
>>109206934
stick with codex
if you're not hitting your limits then no reason to switch, esp because they get 5.6 next week and fable leaves anthro subs at the same time
>>
File: 4dbatturu.webm (3.31 MB, 1920x1080)
3.31 MB
3.31 MB WEBM
>>109206812
>>109206841
>>109206862
>>
>>109206975
>they get 5.6 next week
What's that?

Ok Grok Build we have Composer 2.5.
>>
File: file.png (284 KB, 2320x1274)
284 KB PNG
>>109207011
>China destroys France, 2026 (Colorized)
>>
>>109207011
Thanks, it looks fun, even if it didn't understand what a 3D viewport of 4D would be.

One day the models will understand.
>>
>>109207011
could be worse I guess
>>
>>109207032
Wait! That's IMITATION Fable!!!
>>
>>109207022
Composer is deep discount cheap customized older Chinese model. It's very good at certain Agentic tasks relative to its size, speed, and operating cost. It isn't remotely comparable to the frontier models, next to GPT 5.5 or Opus 4.8 it's a drooling retard. GPT 5.6 is the next generation of GPT models coming out soon, and 5.6 Sol in particular is far beyond the capabilities of 5.5 or Opus 4.8. Similarly people are playing with Fable right now, and it is also far beyond the capabilities of Opus 4.8 or GPT 5.5. Anthropic and OpenAI models remain by far the most capable out there, everyone else in the industry plays catch-up.
>>
>>109206550
source?
my thoughts were “a better, smarter tool that’s able to work at longer tasks? splendid”
>>109206741
counterexample: https://github.com/openai/codex/issues/28224
>Codex SQLite feedback logs can write ~640 TB/year and rapidly consume SSD endurance
>>109206934
I hear Codex is dogshit at making UIs
quite honestly you should probably go with Claude Code simply because it’s different and you can get started with thinking about workflows (“ultracode”) now instead of when OpenAI catches up on that feature
and then maybe switch back if 5.6 gigamogs Opus
>>
File: file.png (81 KB, 684x596)
81 KB PNG
Round 2 begins, hopefully get an American model in the mix this time.
>>
>>109207064
Well, do I need Fable to do:
>>109206781

I can't complain about Composer 2.5, I got it free with Grok for at least 2 months so far, although I think they are making it so idk there's a unified pool for images, chat, and code.
>>
>development

Well, I guess this is a good song for vibecoding:
https://files.catbox.moe/dk497v.mp3
>>
>>109207070
>I hear Codex is dogshit at making UIs
Misconception, GPT 5.5 is dogshit at making UIs, nothing to do with Codex. Codex is a harness just like Claude Code, and like Claude Code you can use Codex with other models/providers. Drop GLM5.2 or Opus 4.8 into Codex and it does great GUI work, not that I recommend doing that.
>>
>>109207123
right. brain fart.
>>
>>109206975
>they get 5.6 next week and fable leaves anthro subs
I imagine the future of Fable on subs will be dependent on how many people bail on their Claude subs when this happens. Unless its simple too expensive to justify running on subs
>>
>>109207144
>Unless its simple too expensive to justify running on subs
yeah, it’s too expensive to run indefinitely using _current_ technology/capacity
but AI datacenters keep getting rolled out so that will bring the cost down to them and they can bring it back at some later date
>>
>>109207144
they've said it's coming back eventually so either they really do have compute capacity issues or they're just taking awhile to quant it to something less huge
>>
>>109207168
all these companies have compute capacity issues because their demand for compute capacity is roughly unbounded
if they’re not serving up Fable to us, they can use compute capacity to train the next big model
>>
>"fix the test"
>Worked for 3m 1s
>Fixed the test.
>It now: Marks the failing test as xfail.
>Verification: python -m pytest tests/test_feature.py -q 1 xfailed
>>
>>109207183
what model produced that “fix” so that I know that I can laugh at it?
>>
>>109207032
>ROUND 2
Option B starts at the 1min mark.
>>
>>109207186
>Or I just break the video file and have to re-encode it.
>>
File: round2.webm (3.89 MB, 1280x720)
3.89 MB
3.89 MB WEBM
>>109207186
>>109207191
>>
File: file.png (792 KB, 2317x1278)
792 KB PNG
>>109207206
Predictable.
>>
>>109207206
I read that as "Four Seething Eye"
>>
>>109207177
>they can use compute capacity to train the next big model
Surely user input is higher quality training data than synthetic stuff though right (which I assume is what you mean)? Well, depends on the user obviously, but the people paying for max are, on average, not the ones asking retarded question.. probably anyways.. actually neverminded who knows lol

Actually is that what the "You might be a child please verify otherwise" stuff is coming from? A low key way to ban dumbasses while putting it in a "safetymaxing" frame they already embraced the negative PR on?
>>
>>109206649
ok then you do it if its so obvious
>>
>>109207070
>source?
Just various posts on xitter saying how Fable just changed everything and nothing will ever be the same again, but yeah my thoughts are similar to yours.
>>
File: Henry Nowak.jpg (160 KB, 1920x1080)
160 KB JPG
>>109207246
>Surely user input is higher quality training data than synthetic stuff though right (which I assume is what you mean)? Well, depends on the user obviously, but the people paying for max are, on average, not the ones asking retarded question.. probably anyways.. actually neverminded who knows lol
a lot of us Max users opt-out to the training
>Actually is that what the "You might be a child please verify otherwise" stuff is coming from? A low key way to ban dumbasses while putting it in a "safetymaxing" frame they already embraced the negative PR on?
I figure that all calls for verification are to keep stuff like pic related from going viral, and if they’re not designed to do that, they will do that eventually because ending anonymity is very good for authoritarians like the ones that run the yookay
>>
>>109207212
ooooh a hypercube-looking map. However, we don't see 4D projection into 3D. Still, better understanding by Mr Opus 4.7 Thinking.
>>
>>109207265
I am, but, also, so should you.

But don't get married:
>>109206626
>>
File: round3.webm (2.94 MB, 1920x1080)
2.94 MB
2.94 MB WEBM
Round 3 was... Fast? I'm just gonna keep going until Fable gets a turn or I get bored and do something else.
>>
>>109207339
I like the worms, but they're not hypershapes.
>>
File: file.png (1.62 MB, 1122x1402)
1.62 MB PNG
>>109207354
qwen3.7-max delivered three empty html files
>>
>>109207396
I wonder why.
>>
File: round4.webm (3.93 MB, 1280x720)
3.93 MB
3.93 MB WEBM
>Round 4
In my opinion the best result yet, vaguely fun, controls feel good, difficulty ramps up as you go and starts to become challenging by wave 10. Impressive result from a 295B A21B Chinese model.
>>
>>109207396
post more qwen
>>
>>109207155
We will almost certainly have something like Fable at some point, unless models like that get banned, or there is nuclear winter or some other weird scenario like that. It's basically impossible that in 20 years the tech won't advance so much that Fable can be served at reasonable prices. The question is just when that will be, but I'm actually optimistic that it won't take that long.
>>
what the fuck, /goal actually bypasses limits
thought you guys were memeing
>>
>>109207528
>grok hotdog dragons
>>
File: file.png (1.93 MB, 1122x1402)
1.93 MB PNG
>>109207603

>Round 5
It was gemma4 31B vs gemini3.1 flash lite, what do you think happened?
>>
>>109207646
I'm rooting for gemma4 31B, you know that.
>>
File: file.jpg (62 KB, 1220x714)
62 KB JPG
>>109207651
>>
File: round6.webm (3.51 MB, 1920x1080)
3.51 MB
3.51 MB WEBM
>Round 6
>Create a street-gang themed 4D driving game where you shoot down enemy bangers with your machine pistol.
>[...] the "4D viewport" which shall be labeled the "Pipe View".
I'm a little torn given that they're both unplayable trash, but I think Option A wins this one, just look at that sweet mini-map.
>>
>>109207733
lmaoo

I paused.

Let's see.

I give it to hyperbangers: it didn't crash.
>>
>>109207733
idk, you have a point, it's tough, you have a point, it seems to be stacking all the enemies in the center of the map which is a small part of the idea of projection. lmao
>>
>>109203932
>I can quit anytime bro I swear
>>
>>109207805
I asked the ai to do too many things and now the drone game is worse than Superman 64
>>
File: file.png (1.27 MB, 1583x676)
1.27 MB PNG
>>109207761
4D Street Iron never crashed though, it just runs at an exponentially decreasing framerate :)

I couldn't decide so I asked gpt-oss-120B which was better and it told me "I don't know know what you're referring to" so we have a winner!
>>
>>109207831
nice
>>
>>109205916
Don't engage him, he's a schizo who always posts the same shit
>>
anyone slinging pi-agent + dirt cheap chinese dogshit?

feel like we need a /csg/ version of /vcg/, perhaps for the poorfags over at /utwg/ so they can scrap together a portfolio
>>
>4 resets stashed for 5.6
finna bankrupt sam
>>
>>109207876
I shill for Pi and GLM-5.2 quite a bit.
>>
>>109207914
glm-5.2 is *premium* chinese, though. thats like, the touristy part of china.

im thinking, whats the chinese LLM model equivalent of the rural farmlands where people still take a dump right where theyre standing?
>>
>>109207926
Qwen3.5 Flash
>>
>>109207818
git revert
>>
>>109207876
i've only spent $10 on pi + deepseek (mostly flash)
it's not worth it for coding imo
dumb, bad tool calls - and $20 on pro will get you less work done than $20 gpt sub.
>>
>>109207970
Is pro any better? I only tried flash locally and I can confirm that it fucks up tool cals.
>>
>>109207986
maybe a little? i only burned a few bucks on it when it first launched and just went back to gippiti
>>
File: 20260705_162757.jpg (163 KB, 748x857)
163 KB JPG
>>109205845
>vibecoded software improves society
guess the romans weren't interested in web design?
>>
>>109205845
you ludlost bigly
>>
Fable goes away tuesday or wednesday?
>>
File: file.png (123 KB, 817x682)
123 KB PNG
gaming is the new vibe meta, here we go girls
>>
>>109207886
I think the first reset expires on the 7th so you might lose it if you don't use it today or tomorrow, and you get a full week out of it anyway so you can use it on 5.6
>>
>>109208047
America/Los_Angeles time, I know that
>>
File: round9.webm (3.9 MB, 1860x1080)
3.9 MB
3.9 MB WEBM
Rounds 7 and 8 were so boring and uneventful I can't be bothered. Round 9 though, Gemini 3 Flash with a banger of a 4D game.
https://019f33ff-4473-749a-828a-af2f83ba6066.arena.site/
I'm pretty sure this is shareable?
>>
haven't seen those colours for a while
jesus
>>
>>109208227
how much are you charging?
>>
>>109208109
aw yeah we tawkin

>>109208227
We may have to help the ai understand that a hypercube isn't a 3D projection. I really am not a pro at this, I have thought about it, but don't know what it would look like. I am pretty sure that if you had a 4D eye, you couldn't see inside of a hypercube - you could only see some of the "3D walls".
>>
>>109208494
$2 on Steam tomorrow.
>>
>>109208227
however, it made a game lol. Kinda neat, I didn't know Flash was actually that smart.

You're right it's the kang of the gens so far.

but why did it name it 4D HOOD WARS?
>>
File: IMG_1770.jpg (81 KB, 474x474)
81 KB JPG
Started working again and Gemini cli is DEAD for my plan, now I try antigravity and it sucks ass, keeps re-reading files, crashes, goes and checks what's on the folder above it, not your damn business! And just like that, kaboom 15% of WEEKLY usage doing barely anything.
>>
>>109208506
>Create a street-gang themed 4D driving game where you shoot down enemy bangers with your machine pistol.
>[...] the "4D viewport" which shall be labeled the "Pipe View".
>>
>>109208511
aha, makes sense, change the style.
>>
>>109208507
>day 1 of nofap
>>
Ah, my code was buggy again. 24 hours wasted.
>>
Someone give me the qrd on what vibe coding actually is. A real answer and not just memes.
>>
>>109208525
>writing non-buggy code is impossible
>in fact, there exist no buzzwords to spam prompts with that would help
>>
I'm the most important ai operator.
>>
>>109208537
My problems are too subtle. Too nuanced.
>>
>>109208561
You have only failure to offer your future self.
>>
>>109208566
It's not an impossible to find bug, I already found it, it's just that I had to let a calculation run for a day and now the results are incorrect.
I even checked this exact part of the code myself, but later the AI fucked it up again.
>>
Just don't give me the scabfaced woman saga of failure and woe from bad decisions. no.
>>
File: snailcat_1.png (418 KB, 1556x1689)
418 KB PNG
>>
File: snailcat_0.png (130 KB, 895x1302)
130 KB PNG
>>109208793
Which one do you prefer?
>>
>>109208793
this one its more fuckable
>>
my codex is slow
>>
>>109207876
No need, GLM is a competitive option. Discussion of large open weight models should remain here since realistically 99.99% of users are using them via API.
>>
>>109208525
>>109208537
>>109208561
>>109208537
>>109208561
>>109208566
>>109208624
Do you guys really find it this hard to debug generated code?
>>
My isomorphisms are all fucked up.
>>
>>109209025
My codebase is 3m lines of code. But also I don't find it impossible, it just sucks when a bug goes into production, or I run a long calculation with buggy code.
>>
>tried the Antigravity cli
>sent "test"
>it instantly executed `npm run test` for some reason
>>
>>109208536
maximally: telling an LLM to make a program but you don’t pay attention to the code — just if the program feels right or not
most of us are doing agentic engineering and getting our clankers to do stuff and we kind of know at least some of the code and what good engineering practices are
>>
>>109209089
not the stupidest thing to run
>>
>>109209043
that's what you get for not choosing homeomorphisms
>>
>>109209143
Thanks anon
>>
>>109209161
>islamic year of the naked goat
>not being a monomorphist
>>
File: 1776929934950015.mp4 (743 KB, 720x1280)
743 KB
743 KB MP4
can I learn rust in like a day? i need to make a simple app for work
>>
>>109209377
wrong thread.
>>
>>109209377
If you know C already? Yes. If not? Maybe. If youve never used a compiled language before probably not.
>>
File: 1754968722441352.webm (1.28 MB, 720x1280)
1.28 MB
1.28 MB WEBM
>>109209452
yeah im already a pretty cracked dev, I just need to learn the syntax + how to code without classes + a new gui library

>>109209428
i'm planning on using claude to speed up my learning process, obviously
>>
How we all doing with our one in a lifetime chance to use fable?
I'm spending my last 30% to do mock user acceptance testing
>>
>>109209838
thinking of using the remaining time to create a solid draft plan for the biggest features in my project's backlog
>>
File: kz4q6ee8r51h1.jpg (132 KB, 1254x1254)
132 KB JPG
>>109209838
I spent most of it doing overtime wagie work
>>
File: HDQQBMwbwAAuEM8.jpg (74 KB, 497x497)
74 KB JPG
Are these models actually good at specific problem solving, or is it still just high speed grunt work?
>Make GTA 5
>Here you go, Xir
>Okay fix this little animation right there
>Spazes out
>>
>>109209938
the models are actually regressing now. sonnet 5 is somehow worse than opus 4.8 at solving leetcode problems.
>>
>>109209962
>sonnet somehow worse than opus
>>
>>109209838
it isn't going away
>>
>>109209965
>spend a gorillion dollars a day on research and swe salaries to make the models better
>only way they've improved anything in the last year is by throwing more tokens at them and giving them 5x longer to come up with an answer
>>
>>109209978
Luddite, you’re an idiot. Claudio added a whole fucking page of native Slate UI yesterday in 10min. Would have taken me the whole day and because I’m a lazy faggot without my uppers more like the whole week.
>>
File: 1778117868200494.png (51 KB, 828x1310)
51 KB PNG
>>109209996
>>
>>109210008
Now post the same question in the terminal, Luddite.
>>
>>109209487
this is the thread for “I want an app, and I think Rust is the best language to write it in”
not really “I want to learn Rust”
>>109209838
after doing things I wanted to do I’m doing some React upgrades for work because that’s a huge fucking mess and something I don’t want fucked up
>>109209844
opus can probably do that with /grilling (mentioned in the OP post)
>>
>>109210032
crazy how there has yet to be a boomer that hasn't gotten one shotted by AI
>>
>>109210043
This is “one-shotted”? What do you mean by “one-shotted”?
>>
>>109210024
they are probably trying to figure what a terminal is kek
>>
>>109209487
he was fine as a twink
>>
File: finetuning-260705.png (121 KB, 1919x911)
121 KB PNG
Developed a web UI to finetune models using my engine and analyze how it affects the token probabilities
>>
>>109210431
how much?
>>
>>109210444
tree fiddy
>>
>>109210032
>opus can probably do that with /grilling (mentioned in the OP post)
yeah but fable has better judgment and taste IMO
>>
>>109210008
>Sonnet 5 Medium
Should've asked Fable.
>>
>>109210577
It's actually better if normies think these models are useless. It probably brings them comfort.
>>
>>109210032
>rust vibing
I had to make the decision between that and c++ for my hobby project
I chose c++, but with all the compiler flags, list of banned features, fuzzing, ASan, unit tests, integration tests, etc... you get the point
not having to navigate all those yourself to make safe c++ is a godsend
>>
>>109210913
converting C++ projects to Rust is pretty easy if you have a Fable-tier clanker to do it
https://ladybird.org/posts/adopting-rust/
https://fishshell.com got converted from C++ to Rust a couple years before AI could do it and now it’s getting fixes and features again (development was grinding to a halt because all the C++ build stuff is crap and nobody wants to work on that kind of thing for fun)
>>
>>109211061
yeah, but I don't want rust. I want autistic c++
>>
rust is good when it's someone else getting subjected to the humiliation ritual of writing it
>>
>>109210900
so how much money is your vibe coded b2b saas making?
>>
>>109211159
How much money are human computers making from the existence of the electronic computer?
>>
how many hours are left until Fable is gone again
>>
File: Fpn7B_FaMAAYCuk.jpg (55 KB, 736x730)
55 KB JPG
Holy fuck Sonnet 5 is such a fucking joke
>go down the wrong path because of shitty diagnosis
>trained to be too autonomous
>the wrong path becomes a rabbit hole, and try to debug a 3rd party library for 20 minutes
>interrupt and switch back to opus 4.8
>run into the same problem, diagnose the correct root cause, side-step and one-shot it in 3 minutes
>>
>>109211184
sonnet is made for sql moron
>>
>>109211184
use Sonnet for simple straightforward step-by-step tasks and let Opus take on the heavy thinking as the brain.
>>
so safe assumption that sol will have been post-trained to delegate work to terra/luna subagents?
>>
>>109211244
I think Sol will be big and strong
>>
>>109201858
His name is Luddite?
>>
>>109211184
This, but Opus vs. Fable.
>>
>>109211184
Better yet, don't use Anth*opic's IPObait goyslop at all.
>>
File: file.png (85 KB, 1040x502)
85 KB PNG
fucking finally, boomer gooks finally starting to believe
more ram for sam and dario
>>
>>109211443
doesn't it take like a few years to bring up a new chip factory?
>>
>>109211456
yes. should be ready in time for fable 10.
>>
>vibe coding llama cpp improvements for my old AMD GPU
>someone already did it for ROCM but not for vulkan
>get the DS4 Flash to optimize vulkan, build, perform test, record benchmark
>1 hour later
>gain 20% preprocessing speed over stock vulkan build
I love it. Fucking old GPUs getting use
>>
new
>>109211660
>>109211660
>>109211660
>>109211660
>>109211660
>>
>>109201980
Why does it take you 10 seconds to load something that loaded quicker in the 00s?



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.