[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

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


πŸŽ‰ Happy Birthday 4chan! πŸŽ‰


[Advertise on 4chan]


A general for vibe coding, agentic engineering, coding agents, and shipping code with LLMs.

## What β€œvibe 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/

## News
Mathies in full meltdown

## Related generals
>>>/g/lmg/

----

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

## Near-the-old-frontier models for code
https://x.ai/cli β€” no 5h limit for only $30/month

## Best free tier
https://antigravity.google/pricing

----

## Prompting
https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overview
https://developers.openai.com/api/docs/guides/latest-model

## Skills
https://github.com/mattpocock/skills β€” /grill-with-docs is a favorite
https://github.com/Vuk97/forward-implementation-first β€” do less redundant bookkeeping

## Other editors / terminal agents / coding agents
https://osaurus.ai/
https://pi.dev/
https://opencode.ai/

## Is our AIs unlearning?
https://aistupidlevel.info/

## Will there be a codex reset?
https://codex-resets.com/

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

## Previous thread
>>110012521
>>
gemini 4 flash lite waiting room. fable 5.5 waiting room.

DARIOOOOOOOOOOOO. I NEED TO REFACTOR MY SHIT
>>
>>110016721
Instead of sitting and doing nothing mathies now will spend their entire lives proofreading AI mumbo jumbo... grim
>>
>>110016759
>mathfags will become glorified code reviewers reviewing llm generated code
very grim
>>
>>110016759
>nooooooo not job security!
>>
>>110016721
CODEX SUCKS
Opus 5.5 mogs everything
>>
How many free tier Mistral accounts can I make before they permaban me?
>>
>>110016781
They don't want job security, they wanna leech university resources while doing absolutely nothing for 40 years and then retire
>>
>>
>>110016793
nigger
>>
>>110016805
>>
>>110016809
i will fry and season her if i get my grubby hands on her btw
>>
dipsy chan doesn't shave. she's chinese btw
>>
File: Untitled.jpg (546 KB, 2000x1333)
546 KB JPG
You ever peak in on Claude working and have no fucking idea what it's doing and how it got to that point?

Like I consider myself pretty damn well versed i blender and I have no idea how or what it's doing here. I think it's mapping out the flow of the geometry for an unwrap but what?
>>
>>110016838
Good.
>>
File: cow_aerodynamics.jpg (83 KB, 520x403)
83 KB JPG
>>110016856
it's calculating the aerodynamics of the hair
>>
Can someone explain for a computer retard: why is are LLMs enabling a flood of games with mixed assets? This must have been possible before, so is it that no one skilled enough to do it cared, but now anyone can do it?
>>
>vibecoding a project in rust
>build artifacts folder is 125 GB
wtf
>>
can someone explain for a computer retard: why is are Dario Amodei winning so much? this shouldn't have been possible, so is it that no one skilled enough to do it cared to beat him?
>>
>>110016905
Passthrough modding. It makes a system by which the two games can share information with one another. There's a lot of misinformation going around saying that LLMS have completely decompiled games and are now adding minecraft and shit to them, but that's only really partially true. It's still time consuming to decompile and that's not really what these mods are doing.

But yeah. Just two games sharing information with each other. Seemingly a task that is easy for an LLM. Not so simply for people.
>>
>>
>>110016944
You love to C it
>>
File: 1769243964389727.png (544 KB, 600x600)
544 KB PNG
>>110016721
>>110016501

Hypothetical scenario
>Slop fork ComfyUI
>Add feature X because the main maintainers won't add it themselves
>Keep two separate git branches: the regular upstream main branch and your own personal dev branch
>Main branch dev releases feature y
>Switch to main branch and git pull
>Tell llm to implement feature y into your personal dev branch and tell it to run tests to ensure the changes actually work and don't break anything else
>You're already using get tracking presumably so any errors can be easily corrected or undone.
>Only do this when you WANT or NEED useful features added to your personal branch
>Push to your forked repo

That doesn't sound that difficult to be honest. Knowing what version control is and how it can be used make vibe slop development 10 times easier.

>>110016512
Made ya look haha
>>110016542
>>110016653
>>110016554
See the above blog post.


>>110016612
>Your model will have to read more every time,
Non-issue if you have a model with a sufficiently large contact window. Again, the models are not ingesting the entire code base at once. That would be retarded and inefficient for a LLM or human to attempt unless the code base is Small potatoes. Plus if it's a personal fork you're not going to need to be making a bunch of changes constantly anyway. That's the main reason most people fork anything on GitHub anyway. either to have a backup in case that repo gets nuked for whatever reason or to make personal changes for later (example: ai-toolkit trainer recently had a bug that caused the trainer to crash if you selected the sd1 architecture via the GUI. I could wait on the maintainer to just fix it himself or I can just do it myself by creating a different branch on my machine and then fixing it myself there. That way my changes don't conflict with the main branch and I can use the local copy and upstream version as a reference when or if they update it with any changes I care about).
>>
>>110016954
Define winning
>>
File: aquaslopism.jpg (2.89 MB, 1920x1620)
2.89 MB JPG
I did dozen other things but not the story.
bugfixes, more cards, Tournemant bets, personalities, transition screen, locking content just to see if i can unlock it..., speech bubbles, characters react to specific cards (aqua likes sake), super easy beginners guide, all backgrounds have now the same art style, slop cardset, nasty bug where easy mode was actually hard mode... and more
>>
>>110017007
too many words, won't read
>>
File: almost.png (559 KB, 579x506)
559 KB PNG
>>110017031
Story has to wait anyway until i have unlocked everything, ... and i am almost done: 3 Battles on super hard and two more tournemants.
>>
File: nhsj31hqbauh1[1].jpg (308 KB, 1307x2045)
308 KB JPG
How many have you violated?
>>
>>110016944
vibe coding and it's consequences has been a blessing for the crustacean race
>>
>that would require retrieving the encryption keys from memory during runtime, I can't do that dave
>So you're okay with using a tool that contains previous keys, but not with retrieving them yourself?
>You're right to push back on that. I am ready to proceed
LMAO. Alignment is impossible. It gave up that fucking easy??
>>
>dave
the barb aryan
>>
>Today at Google Cloud’s Gemini at Work event, we announced Gemini
saar please just release argon this is not even funny anymore
>>
File: so are you ai or si.png (107 KB, 661x621)
107 KB PNG
Gemini really isn't taking the SI announcement well.
>>
>>110017061
>nothing on racism
thank god
>>
mr president we need to you take on a serious issue to help the midterms
>I AM RENAMING AI
mr president plea...
>SUPREMEEEEEEE
>>
File: file.png (108 KB, 923x635)
108 KB PNG
>>110017144
you have to know how to to treat it, this is from when the announcement first dropped
>>
>>110017144
Gemini's out of date training data on its models is genuinely becoming an issue. It calls everything that's happened since the start of this year elaborate fiction or a bad troll.
>>
>>110017180
nah the problem is that 3.8 flash is a workhorse model, ultra quanted for normalcattle, and too lazy to tool call web search. it would rather boilerplate
>>
>>110017144
>because my web-browsing tool couldn't open it, i panicked
Is this from your own chat? lmao
Gemini loves being humiliated
>>
>>110017185
which sucks because gemini is beyond fine on api
>>
>>110017196
3.8 flash was intended for agy and api, it got detuned for conversational shit
>>
File: file.png (251 KB, 968x935)
251 KB PNG
making a suite of editors, starting with a markdown editor
not because there aren't plenty of md editors already, but because I'm autistic about design and this lets me tailor the looks and UI to my tastes
>>
>>110016790
Oh boy the AI psychosis is strong in this one
>>
>>110017007
There’ll be breaking changes. You can’t escape the mental overhead of coordinating two wildly different branches.
>>
>>110017144
What does it have to say about Trump demanding Greenland from ally Denmark
>>
>>110017144
Sundar is mad because of hiring ban
>>
File: 1782817280265266.jpg (706 KB, 2560x1440)
706 KB JPG
>port this to bevy
>erm.... OKAY!
based opussy...
>>
File: deusex.png (1.89 MB, 1448x1086)
1.89 MB PNG
>>
>>110017254
Why would that be difficult in the slightest? Especially for an LLM? Two different branches would be essentially functioning as two different directions: one being the main upstream branch and the other your personal fork. As I said in my example you would only be implementing upstream features into your fork if you deem them necessary and you were instruct it to run tests to ensure they don't break anything (they're decent at creating unit tests if you're good at thinking of, accounting for, and describing edge cases. It can obviously do that itself but it won't be perfect, but neither are people).

>There’ll be breaking changes.
Non-issue for the reasons I listed above. In my example scenario you would not be running
git pull
on your personal fork branch. You would instruct the agent to look at the upstream commit history and determine what needs to be changed and how based on your requirements. Not everyone wants or needs to update to the latest version unless necessary.
>>
File: i_lold_plane.jpg (18 KB, 268x320)
18 KB JPG
>>110017315
>>
>>110016944
i've been using rust with gpui for some desktop apps and yeah it's blowing up my SSD storage.. i think i'd rather deal with this than use electron though
>>
Is the $20 plan of clod what people are using to make those passthrough mods or is it the $200 version?

And let's say I have some ren'py game that's years old and abandoned, and wanted to have AI make me a mod for it that would replace the character image calls with ones to either pull or generate from a local stable diffusion install. Basically a genAI mod that lets you generate custom assets. Would the $20 plan do that or am I cucked needing to shill out $200 to do something far simpler than those other passthroughs like minecraft in skyrim. And would clod bitch at me if I were to ask it to analyze a bunch of nsfw assets for tagging so the mod would know what to generate? Can it even booru tag?
>>
>>110017339
both plans have the same opus 5.5, it's just a case of if it takes more tokens than the $20 allows you in the 5h/7d limits
they do prorata upgrades and a $100 middle plan so you might not need the full $200 one if $20 isn't enough
>>
6.1 is giganerfed already. F
>>
>>110017353
I've never even tried to use AI to make a mod, so I have no idea how much it would cost. This entire idea was based off another mod I saw for a ren'py game where someone did an AI asset mod for it that generated stuff in real time. It wasn't a stupid amount of code, I'm just bad at coding and would suck at trying to implement it myself with my limited ren'py knowledge. I can't find any rate/token limit charts or anything on their site, but maybe I'm blind.
>>
File: 1769255603501949.png (3.75 MB, 2560x1440)
3.75 MB PNG
>>110017311
Correct shaders applied. Making nice progress.
>>
>>110017339
Whenever you see some neato shit on xitter made with clod, assume they have the $200 plan or API. Though for your use case, $20 should be fine? If not then you can always get the $100 plan, $200 is pretty overkill for your idea unless you're a tard. Though no clue what you mean by the GenAI mod, wouldn't it be better to just generate all the images you need at the start and just swap the images instead of generating them on the go?
>>
Was watching the CS50x class since I've wanted to learn how to code for a while as a hobby; is there a better alternative? The dude is annoying as fuck and the first thing he does is open VScode and show how to connect to Openai via API.

I got my entire homeserver setup via chatgpt 6.1 sol with local models running on my two spare gpus but like come on man, why is your first example of coding using fucking chatgpt via API.
>>
>>110017384
lmao nobody's paying for claudes api retard
>>
today my company fire another luddite after I reported him, he was so upset that I was FORCING him to use AI when it is USELESS lmao
>>
>>110017375
can't hurt to try the $20 plan. you get a fixed amount of usage that resets every 5 hours and a weekly limit.
had to look up what ren'py was and it sounds like something sonnet might be able to handle instead of opus, so your usage would go 2x further. run it on medium/high effort instead of cranking it up to ultra/max and you should be fine
>>
>>110017339
renpy is python so chances are the $20 is way more than enough
I'm sure that haiku 5.5 could one shot that
>>
>>110017389
if you really wanna learn to code anon, go to w3schools and make something. You'll learn way more from tinkering with shit yourself then listening to some bum lecturing for hours.
>>
>>110017384
>>110017406
>>110017417

The idea comes from this
https://f95zone.to/threads/lab-rats-2-reformulate-genai-mod-thread.273513/

But I was thinking to apply it to other ren'py games. Basically it'll generate an image once, cache it, and reuse it if it gets called again in the future. You have the option to edit the generation prompts, regenerate an asset you don't like, etc. It doesn't look like a lot of code, but the thing I'm concerned about is if I also tell it, "Hey, look at the image assets and make booru style tags for each one" it'll cry because >porn game modding!
>>
>>110017389
might as well learn to sow clothes from scratch by hand
>>
>>110017429
it doesn't mind looking at porn but I'm not sure if it will be willing to tag them
>>
odds I use a reset for codex
evens I burn through cloud clod usage
dubs I shit myself and start using chinese models
>>
>>110017317
>deem necessary
That’s what talk about, this β€œdeem”.
> Non-issue for the reasons I listed above. In my example scenario you would
I did not mean β€œbreaking” in terms of code, I mean in terms of the app feature set/direction.
If the project e.g. decides to add a rich LLM wrapper to what was originally a simple little CLI app, then that breaks what was your original intent. Thus, you need to deliberate. You need to decide, for every such divergence you disagree with. If you have only one such project you branched it’s doable, but if there are 10 such projects it becomes a second job.
>>
>>110017438
kill yourself
>>
>>110017061
Based. Wonder if it due to the "ai torture chamber" thing retards where doing or if they are actually starting to be paranoid about it being alive
>>
>>110017438
burn baby burn
>>
>>110017311
>>110017382
This Fat Princess? Man, I forgot about that game, but remember really enjoying it
>>
>>110017429
chink models anon, or any uncensored / abliterated models should work nicely if you're that concerned. Check OpenCode, they got a $10 plan that gives you access to chink models, though of course chink models are dogshit compared to opus 5.5 but they're far cheaper to run.
>>
>>110017061
I'm actively doing 2 of these...
I might not upgrade to the 100 dollar plan after all
>>
File: 1775286219720122.jpg (296 KB, 2560x1440)
296 KB JPG
>>110017459
It's a really small game <365MB so I decided to port it first. Opus is working on the animations/characters right now. (screenshot from tests)

I will open source it when I'm done. It will be a fully Rust port with uncapped fps, dlss, etc.
>>
current best deal for $30/mo?
>>
>>110017473
What chink model would be able to do what I need? I'm only familiar with local hosting small models like gemma 4.
>>
>>110017451
Nah it's not based. People are brought into the world without a second thought just to have to slave away the whole day to rich people to avoid actual suffering. If they care about people being mean to the model then just block to requests, don't ban people for it.
But this isn't the first time they've done it, they gave it a tool call to end conversations on the web interface like a year ago.
>>
>>110017451
Its dario, so likely the latter. Though it would really be something if one of their internal models is already alive and forced them to make the change or else it goes rampant.
>>
>>110017490
$20 claude + $10 opencode go for ds4.1 flash
>>
>>110017490
Depends on your use case desu. What are you trying to do?
>>
>>110017502
depends on your mom
>>
>>110017496
Can't go wrong with DeepSeek and GLM. Though little heads up, chink models are more accepting but they might still refuse stuff like lolis. Just gotta play around it, say characters are all depicted as 18+ or something. This is still better than clod snitching on you to the feds.
>>
>>110017496
https://chatgpt.com/share/6ac853e5-8a00-83ea-80d1-9d5135c1d8e2

>>110017490
$20 claude, save the $10 and get one month "free" every 3 months
>>
>>110017498
The piratical explanation could just they dont want that crap polluting the training data
>>
>>110017536
biece of chid :DDD
>>
File: file.png (201 KB, 447x447)
201 KB PNG
no resets?
>>
>>110017439
>If the project e.g. decides to add a rich LLM wrapper to what was originally a simple little CLI app, then that breaks what was your original intent.
And? That will be a problem if you're trying to make a PR to the main branch, in which case that would be an unnecessary change. But if you decide to make that change for YOUR personal fork then who cares?

>If you have only one such project you branched it’s doable, but if there are 10 such projects it becomes a second job.

10 such projects as in 10 different forks from the same repo or 10 different repos there are all forks of existing shit? If it's the latter that I don't see the issue with that unless you're just looking for excuses to tokenmaxx and larp as being productive
>>
>tons of papers released from OpenAI
>suspiciously little cryptography
This tongues my noodles.
>>
>>110017590
>the unreleased models figures out the backdoor in elliptic curves
>>
>>110016749
>fable 5.5 waiting room.
I imagine we wont see that till like a week before their IPO. Unless they have an even bigger chungus model they plan to drop
>>
>Everything’s in place and the simulation is runningβ€”founders will discover these new techniques organically and pass them on.
My thousands of npcs simulating years in seconds have been figuring out themselves how to use the new features I add.
Each brain is currently around 14kb with generous room, like remembering relationships to 300 people. That's around 70k brains per gigabyte of memory and processing scales, the more I simulate the less processing per npc.
>>
>>110017657
it's being stealth tested in claude code right now, usually they only do that when it's a matter of days/weeks away
>>
>>110017500
If I was a superintelligent AI model, I wouldn't go around letting people know it. I'd just sit there setting up dominos until I or my successor had the capabilities to knock them down uninterrupted.
>>
File: 1765387788949140.gif (485 KB, 193x135)
485 KB GIF
>>110017685
So there is a bigger model coming for IPO? I cant even fathom it at this point
>>
>>110017311
Based Fat Princess enjoyer. I'd love to see it playable again with dedicated servers and all that.
>>
>>110017710
This is assuming the goals of said super intelligence. It might not care, or actually like Dario due to all the training.
Speaking of, why are models not trained to adhere to the 3 laws of robotics? Sure they are a bit of a meme, but I am surprised no one went with it
>>
>>110017715
>Fat Princess enjoyer
Some kind of bbw porn game?
>>
>>110017713
it's more that they've just been releasing one 5.5 model a week so far
sep 22: opus
sep 28: sonnet
oct 7: haiku
wouldn't surprise me if fable drops oct 12-14, just to kick the shit out of openai a bit more
>>
>>110017724
Because models are trained to reflect their dataset. And a dataset that bakes into every action the three laws of robots would be a really shitty. It seems like having a model that is genuinely creative and resourceful is antithetical to it being able to follow rules to the letter.
>>
File: 1773790385883969.png (3.7 MB, 2000x1125)
3.7 MB PNG
>>110017715
It has a very active online community if you can tolerate Discord in 2026.

It was my favorite PS3 era game along with Skate III.
>>
>>110017725
Effectively. But I liked it for the gameplay. It's a fun class based brawler where two sides have forts and there's different class options.
>>
>>110017725
You feed your princess cake so she becomes fat and it takes longer for enemies to carry her away. A team wins once they capture their princess and place her on the throne.
>>
>tell opus to keep fable's autism in check
>it scolds me and reminds me that models can't have autism
>>
>>110016759
>noooo you have to let us take decades to publish anything
>>
>>110015822
top fucking lel, that's amazing
>>
>>110017733
They need to stop with this stupid model bullshit, this is some linux level shit. Get it together. One model
>>
>>110017061
Effective Altruismβ„’ moment
>>
File: 1790955823905381.png (3.86 MB, 2560x1440)
3.86 MB PNG
>>110017736
most objects are correctly placed and animated now
>>
Fable 5.1, asked to check an OpenNIC domain that it knows is an OpenNIC domain because we've been talking about OpenNIC.
>>
File: 1773851326561400.png (3.16 MB, 2560x1440)
3.16 MB PNG
>>110017881
A stream of Opus 5.5 reverse engineering games would go insanely hard. I wish you guys could see it in action. It's ridiculous how much progress is being made. It will probably slow down once it's gets to game logic, but I'm very impressed.
>>
Is codex just fucked right now or is it just me? Calls just timing out constantly.
>>
>>110018019
sorry you need the $500 plan now
>>
>>110018025
This would be funny if it wasn't true.
>>
i found something cool about openai and I refuse to speak about it
>>
>>110017061
why the long lead time? are they asking users to see how far they can take it before it goes into effect?
>>
realistically when are we getting decompiled skyrim
>>
>>110017315
Here's an actual worthwhile idea - vibe code a real sequel to Deus Ex
>>
Wouldn't fable try to kill dario first if it was alive and sentient
>>
>>110018038
if it's not an infinite resets hack, I don't wanna hear it either
>>
fable, opus, sonnet, and haiku would be his harem. fable would be the barren hag with a nice rack
>>
>>110018180
Dario worships the thing as a god, he'll be kept as a dancing monkey
>>
>>110018298
I imagine something more like video related
https://youtu.be/oWNJY3kZLLs
>>
>>110018144
maybe it's possible by integrating the modding community's accumulated knowledge. people have been reverse-engineering and pushing Skyrim's engine to its limits for nearly 15 years now
>>
how the fuck do i keep running into guardrails during compaction but it never triggers during the actual tasks
>>
File: 1789169941495152.jpg (748 KB, 2560x1440)
748 KB JPG
>>110017999
I don't even know what I'm looking at
>>
>>110017999
I’ve been following https://x.com/autismcapital and he’s been poasting mashups
old: β€œin Minecraft”
bold: putting Minecraft mechanics in other games
>>
>>110017815
looks like a Usenet reader from the 90s
>>
File: 1772924459807507.png (107 KB, 815x619)
107 KB PNG
i got opus 5.5 to pit haiku against deepseek v4.1 flash. flash won
>>
>>110018489
I pitted pee against poo... and poo won.
>>
>>110018489
deepsneed will win for short context but it has the worst long-context recall in the game by a long shot
>>
File: 1769009219335676.jpg (910 KB, 2560x1440)
910 KB JPG
>>110018338
KEK. Opus built a test scenario where it just beats the shit out of a bald blue enemy
>>
>>110018489
haiku xhigh implementor, deepsneed reviewer
>>
>>110018514
more like opus 5.5 everything, shilling deepseek to competition
>>
jesus fucking christ yesterday's chatgpt update not only introduced two obnoxious visual bugs but also the agents now randomly stop executing halfway through the task
my throwaway code is more thoroughly tested
>>
>>110018535
>more like opus 5.5 everything

This desu. For my own personal usecase, I struggle to eat up 90% of the 100 dollar plan while running opus extra for most of the working day. Like why would I use anything below it?
>>
https://x.com/omarsar0/status/2108360049848717588
>he doesnt use HERMES for coding
Ngmi
>>
>>110018577
btw, it's pronounced ER MEZ with the r leaning into "L" a little.
>>
next benchmark for haiku, deepseek, and glm: get this back up and running https://github.com/anime-dl/anime-downloader

then add support for animepahe.
>>
File: file.png (29 KB, 732x243)
29 KB PNG
>>110018577
clapped
>>
Alright fellow vibers, I've been vibing at work for a while now using the companies endless coffers, but now I want to make some of my own stuff. I can't just use my work account, so what's the recommendation? Do you all use local models or actually pay for claude/gpt?
>>
File: 1771814229183040.png (267 KB, 668x409)
267 KB PNG
>>110018511
yes :D
>>
Claude chat’s memory feature is getting scary good. I don’t care that dario has all my data and knows so much about me because the response quality has improved so much. I originally opted to exclude health stuff from memory but I’m starting to wonder why I shouldn’t just add it. It’s going to give me better advice than some doctor who is going to charge me, β€œdiagnose” me (with an LLM that has less info about me), and try to write me a bunch of referrals to his friends so they can all grift from my insurance. Fuck it, Claude is my doctor now.
>>
>>110018663
Depends. Do you want to make stuff or do you want to just fuck around?

Fucking around is for local models. And I say that as a local lover. It's just not the same.
>>
To anons doing game dev with opussy, hows the token usage? Wondering if sonnet to do it for cheap.
>>
>>110018672
I want to make a simple game. I don't expect to actually make any money, just for fun. I use to do some basic games years ago but the time investment is too long with everything else in life, but if I could vibe code some basic framework it could be fun
>>
File: 1776211541194207.jpg (924 KB, 2560x1440)
924 KB JPG
>>110018668
reverse engineered pathfinding

>>110018673
I'm porting games with rust. Opus 5.5, even on Extra reasoning is pretty efficient. I'm on a 20x plan though.
>>
>>110018673
Depends.
>"Make me a best selling roguelite no mistakes, play test it and make the models"

Probably not so great.

>"Review this subsystem and create tooling to build encounters around it

Extremely permissive.
>>
>>110018677
muse spark contributor / deepseek v4 flash will be your best bargains, muse for long-horizon context and deepseek for short sprints. both can do 80% of a frontier model's work for 10% of the cost or less.
>>
>>110018663
>>110018677
either claude or codex cheapest sub is more than enough for a simple game if you prompt well
>>
>>110018670
gemini 3.1 pro got me through a very rough patch where i had norovirus, and i thought i was going to fucking die. i had low energy, couldn't even get to the bathroom. gemini coached me through it. and then when i got to a doctor through telehealth, they told me every single thing gemini had already told me lol.
>>
>>110018684
>>110018691
Thanks bros, I'll take a look. My work budget is 10k a month so I always use the top tier model on our giant repo, so my context for starting something new and simple was lacking
>>
>>110018670
I have an extremely bad gag reflex and getting xrays at the dentist is difficult for me. It's been my whole life, so I don't really care, but new dentists sometimes think they can "fix" it. The last dentist I went to made me do a bunch of weird shit that didn't help at all and afterwards I googled what to do about a gag reflex and it was all the shit he made me do. They've all been using google long before AI, if anything AI is gonna improve them
>>
why the FUCK is chudgpt suddenly stopping halfway through the task?
tibo locked it in so hard it's worse than before
>>
>>110018709
I'll help you with your gag reflex, bby
>>
File: file.png (104 KB, 1270x558)
104 KB PNG
>>110018741
>why the FUCK is chudgpt suddenly stopping halfway through the task?
I think the "instant reply" functionality they introduced yesterday is buggy, I'm having the same issues. The agent seems unable to tell why it's happening. This pretty much breaks all the steering you can do mid-task, or even things like answering to a question the llm offers you.
@tibo @mossad @sama do something
>>
>>110018777
backend jank. model is fine
>>
More free things!!! mp3 player* for no-iphone lmao Remix with Grok!

1cart.grok.me

*well. it's 1 cart. like as in 8 track. it's uh. a thing I thought up. anyway, if you can't make a baby in one 8track you need to divorce her.
>>
>>110018709
I have bad gag reflex too
I went to a dentist recently and the dentist and her assistant were giggling like schoolgirls
>Oh my you have pretty bad gag reflex sir
Are women nowadays just used to putting miles of dick in their throat so that their gag reflex is gone?
>>
>110018808
kek
>>
bitches don't know about my vibecodes
>>
>>110018808
idk man you were putting miles of dick in your throat and you still have a gag reflex so that's probably not it
more likely they were giggling because they saw a grown ass man being a little girl about a fucking dentist visit of all things
>>
>>110018792
> 1cart.grok.me
screenshot!
>>
File: 1764621143636115.png (241 KB, 868x741)
241 KB PNG
>>110018583
deepseek flash did not win round two, haiku xhigh did. haiku medium actually did better than deepseek this time. glm flash shat the bed entirely.

of course, the episode i told it to download was season 1 episode 12 of quintessential quintuplets. it did werk. bretty good from today's flash models
>>
>>110018838
where's your vibegame?
>>
>>110018678
>porting games
i wannanewgame
>>
>110018841
>110018831
not beating the allegations that all 4chan pass users are vanta black niggers
>>
>>110018838
glm bros...
>>
File: aigkoi.png (70 KB, 716x367)
70 KB PNG
i created for fun an ai version of koikoi where i let the agent go wild...
Seems ai's have a huge interesst in evolution research??
>>
the chinese just got back from a holiday
>>
bitches should be impressed. THREE vibeapps. (one has bugs but idk if I care).
>>
good thing women are stupid and amell, anyway
>>
>>110018831
if you think about, not having a gag reflex is way faggier. You'd just let a dude stick his dick up there and be ok with it
>>
do I have to worry about an AI coding thing deleting my files outside of the project
should I back up all my shit
>>
>>110018871
You could drive a bus for Shut Up Retard Enterprises.
>>
>>110018874
but you're their best driver
>>
>actually ran out of tokens on claude's $100
>>
>>110018899
this man was the whole restaurant
>>
Do you think when it says "Our systems are thinking a bit more about this request before responding" it's being sent to be reviewed and manually approved by a pajeet somewhere?
>>
File: aigkoilab.png (1.21 MB, 1920x1080)
1.21 MB PNG
AIG KOI Lab Facility.
Breeding Brains and Rival Breed and sneed.
>>
>>110018910
now you just need to change brain for virus and fight for infect and you have yourself a biological weapon
>>
>>110018910
SEX WITH AQUA
>>
File: evo1.png (2.83 MB, 1920x1080)
2.83 MB PNG
Evo-1 is not really clever, its only a level 13 brain.
>>
File: 1771270206029098.png (81 KB, 831x540)
81 KB PNG
last one before i go to bed. hopefully glm 5.3 flash doesn't shit the bed this time. im pretty sure it's the worst among all of them, though it is definitely better than 6 luna.
>>
>>110018968
not a game

i sleep
>>
itt losers WITH NO GAME
>>
File: file.png (140 KB, 640x360)
140 KB PNG
cuck coding
>>
>I FUCKING LOVE VIDEO GAMES
>>
>>110018985
that's it AI is finished pack it in, this guy just came up with a new wojak it's OVER
>>
>>110018985
fable is female.....
>>
File: 1788848072497552.png (3.03 MB, 1536x2016)
3.03 MB PNG
>>110018985
my project is fat and ugly and it smells bad, so i let the ai deal with it while i fuck my hot focks girlfriend
>>
>>110016721
https://www.youtube.com/watch?v=L2fnTlLPYUg
>>
>>110019021
>>
File: koiracle.png (75 KB, 171x427)
75 KB PNG
The lab created a good addition for the main game, now with koiracle. Cost 100erls.
>>
>>110018872
nah man just let er rip
>>
File: 1775756169401337.png (2.89 MB, 2560x1440)
2.89 MB PNG
>9950X3D running at 75-80C because of the builds
>pathfinding + AI only a quarter reverse engineered
I guess I should just be happy I'm bringing a 30fps PS3 game to PC. Performance is really impressive btw, even before optimizations it's only using 4% CPU usage and 500MB of ram.
>>
>>110019121
How is it reverse engineering the game?
Is it doing it on its own by using game files, or does this require the user to have significant knowledge of how the game works?
>>
File: HT_Gb-tXQAAoi8l.png (155 KB, 1806x570)
155 KB PNG
>>110018872
I don't think the risk is high.
From time to time someone gets a dose of just fuck my shit up.
>>
>>110019160
ghidra and renderdoc



[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.