A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/## News- (2026-07-24) Claude Opus 5 out----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/clihttps://claude.com/product/claude-code## Worth it for code, but the frontier models above are betterhttps://x.ai/cli## Not worth it for code, but maybe good for other thingshttps://antigravity.google/product/antigravity-cli----## Prompting / context / skillshttps://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 favoritehttps://github.com/DietrichGebert/ponytail## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/https://cursor.com/docshttps://docs.windsurf.com/https://docs.cline.bot/https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent## UI/Frontendhttps://www.figma.com/make/https://www.anthropic.com/news/claude-design-anthropic-labshttps://uiverse.io/https://ui-ux-pro-max-skill.nextlevelbuilder.io/https://stitch.withgoogle.com/## In-browser builders / hosted vibe toolshttps://bolt.new/https://replit.com/https://docs.github.com/en/copilot/tutorials/sparkhttps://v0.app/docs## Benchmarks / rankingshttps://www.tbench.ai/leaderboard/terminal-bench/2.0## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109377789
>>109387280whoa, a snale
Really awkward one again but getting there. To be expected because braindance definitely very advanced modding. Nobody done it before as far as I can tellWaiting to test a new build that should fix the clue part, then I'll look at the movement. My cpu and gpu are maxxed atm from rendering the item database
Using ChatGPT makes you retardedIn first of its kind, a preliminary MIT study of brain scans shows that using LLMs results in 47% reduction in brain connectivity.>Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.When researchers forced ChatGPT users to write without AI, they performed worse than people who never used AI at all.>LLM users also struggled to accurately quote their own work.83.3% of ChatGPT users couldn't quote from essays they wrote minutes earlier.Sample size: 54Link to the paper: https://arxiv.org/abs/2506.08872
>>109387280
K3 is by far one of the most autistic models I've ever worked with when it comes to visuals and typography. It fixates on the tiniest details most people overlook. In a good way. It will do pixel level measurements and argue its point if you ask why. Never seen a model actually push back and defend its own aesthetic taste like this. Really interesting to work with and kind of stubborn, its a fun model.
>>109387328> using a calculator leads to reduced main activity during calculationsshocking!
>>109387328we're not writing essays
>>109387328>chatgpt write this essay for me>i still learn 53% as much
>>109387328i'm not literally 12 years old and in middle school why would i be using chatgpt for writing essays
>>109387328this is the kinda retarded shit liberals waste their time on
>>109387418Sounds very promising. What provider are you using? I tried using it earlier today but kept getting 503'd. Current main model is k2.7 code. Will I see notable improvements when switching to K3?
>>109387328i'd like to be jessica in situ if you know what i mean
luddites got real mad at this deepmind employee for posting a shitty meme.
what the FUCK is opus 5's token use? it churns them x3 faster than beforeDARIO GO FIX YOUR SHIT
>>109387574they are pathetic little crybabies
I've vibecoded this pokemon-roguelike game and thought you guys could flame me aka provide feedback on what i can improve or what i can ask claudette what to do next? I've just added accounts to the game. Thinking about seasonal rankings.https://playrinne.com
>>109387628add supports of accounts other than google and discord, for startersif you don't want to manage your own account system then at least add support for accounts of something that isn't closed-source spyware
>>109387628Looks good. My advice is to ignore people like >>109387650 they are not the userbase you want
>>109387628>The phrase "Elite Four" is owned collectively by the Pokémon franchise joint-copyright holders - which includes Nintendo, Game Freak, and Creatures Inc.How's your legal fund?
>>109387574this is so fucking embarrassing.they're not even holding guns. the shitpost image doesn't even imply they're going to bring harm to anyone.luddites are such whiny niggers, holy shit. no wonder the jews mind control them so easily.
>>109387678how many times were you shoved into a locker as a kid
Anyone who has built something with accounts or has half a brain understands that auth is a hassle, you don't want to pay for MAU, you don't want to pay for an email service, you don't want bots, so your easiest option is to use discord, gmail, github
>>109387628Change your UI, go for a gameboy look instead of the vibecoded look.
>>109387574new culture war arrivedhow could gamergate do this?
>>109387767guess I'll be just as "radicalized" as I got back thenbring on skynet
>>109387628>Shedinja has 32 max HP>gets hit by attacks of all types anywayWhat did (You) mean by this?
Fable is not as smart today.
>>109387574Haha, he trolled them epic style.
Claude Code app UI is kino, light years ahead of Codex's.
>>109387784I had let that slip by, I'm fixing it right now, thanks!
>>109387783>>109387767>>109387574fuck these people, accelerate
Did Claude Code remove /plan mode? I type it and nothing appears anymore
You don't need more expensive model for your tier of coding, desu.
>>109387889just went down the rabbit hole. Didn't even know who these people are, "pause ai"they're actually doubling down on their retardationlike I'm supposed to take people seriously who have no idea what a meme template is....
>>109387735Not him but I want to get my website to have a simple and approachable look that doesn't have the vibecode look. What should I base it on?
>>109387985Just Anthropic trying to dig a moat
>>109388004Idk, but you could search for a theme online and tell it to copy.
>>109388011>easy mode opticsyou can hate the big players while still hating retarded luddites
>was using a NES emulator that lacks audio (AgNES) for a couple of years in a Wasm test module>tell Sol to implement audio emulation>it one-shots itlmao it's that simple
Going to add Kimi K3 support to my engine.
>>109387328published a year ago
>>109387328not going to read it but i assume usecase mattersno "software developer" or "software engineer" cares about coding. thats why they all sperg out when you call them a 'coder' or even 'programmer' it's derogatory to them. so with AI, we become software devs/engineers, with a focus on the system and implementation rather than the details of code. i don't think this approach has made software devs retarded in the past, those who get indians to write code for them
>>109388052The world has a shortage of motivated developers willing to throw themselves at problems. it's almost embarrassing how many bugs and issues I've fixed while people seethe all day hoping for a patch.
Anyone read Anthropic’s take on the letter? https://www.anthropic.com/news/position-open-weights-modelsIt’s really, really anti China
>>109388209Not worth the bandwidth to read it tbqh. We all know Misanthropic's position on the matter.
>>109388217tl;dr>FUCK CHINA>DISTILLATION IS CHIP BAN EVASION>Also we should team up with the CCP to review all models, closed and open, before they're allowed to be released to the public.
>>109388209>My primary concern is the risk that authoritarian governments—not solely the Chinese Communist Party (CCP), although the CCP is clearly the most capable threat—build AI models that are more powerful than those built by the US, and use them to achieve permanent military superiority or perpetrate incredibly deep repression of their own people.lmao it's basically "we're the good guys they're the bad guys so we can't let them have capabilities"
>>109388209>It would protect US AI companies from competition, but that has never been my goal.Sure Dario, of course, we believe you
>>109388209It's just seething about china, because china actually builds up their electrical grid.
I let one of my banked resets expire without using it :'(
>>109387767>>109387783somehow i don't think it's pro-ai people who are responsible for the one ai-related security incident in recent memory where someone tried to firebomb altman's house like a month ago. i genuinely don't know who these people think they're tricking. anyone can read the news.>>109387811there were partial outages a little bit ago and LLMs perform worse when datacenters are strained by demand. wait for peak hours to be over and it'll be back to normal.
>>109388320They don't get auto-applied when they expire?
>>109388209Can you imagine the stupidity required to actually believe you could prevent China from becoming the dominant world power if they chose? They have more engineers than we have people.
Reading K3's thinking loops is somewhat infuriating
>>109388465China was the dominant *economic* power for most of the last few thousand years. So much so that the Spanish king once reverted a change to their silver coins for no reason other than Chinese merchants didn't like it.That bit with Mao was an aberration.
>>109388209>I fucking hate chinks, gooks, japs, and slit-eyed buckteeth yellow monkeys of all varieties! — Perchance.I can't believe Dario actually wrote this.
>>109387574lmao 'dites are this pathetic
>>109388462probably but I could've spent my weekly usage beforehand
After spending the last couple weeks experimenting with vibecoding, I'm ready to add some llm plugins to nvim. Anybody have reccs for plugins and free models?
>>109388522>>I fucking hate chinkswtf i love dario
For the first time I vibed at work and I feel retarded.
early life status?
You can spot where I ran out of K3 weekly
>>109388552>"We should crack down on the most efficient way to improve our own technology!"
>>109388552>China can't train good models without chips>but also they can train good models by distilling oursMake up your mind Dario
Do you see Claude as a man or a woman in your mind?
Every time you write a prompt a luddite in San Francisco dies
>>109388609claudia is a girl
local chads keep winning, baybe!
>>109388668>46kWlocal af
>>109388609effeminate man
>>109388609> Fable-chan is female> Opus-kun is male> Sonnet-san is a tranny
>>109388668oh yeah, 80x5090much local, such ai
>>109388004https://stroustrup.com/https://danluu.com/https://inessential.com/https://www.halfwaytomars.net/
>>109388684I'm downloading them to a rented 2x pro 6000 machine with 2 TB of RAM to test if I can get it going with cpu offload for expertsit's probably going to be slow as balls, we'll see.since 2TB machines are scarce maybe it'll be better to use two 1 TB machines.
>>109388668>most abundant GPUs on Earthlol lmao even>Any lab, startup, or universityCan afford minimum $240k in GPUs, plus about the same in motherboards, cpus, ram, power supplies, 25gbe networking, and has three phase powerWhy are they so disingenuous
>>109388668How long until we have local Fable bros? By local I mean it runs on 16GB of VRAM
>>109388765probably a decade+ and computers will get better and bigger but pic related will probably happen first
>>109388742It's over 100B active, right?
>>109388756I mean it's still half the money than it would take to run it on HBs
>>109388765Stop trying to make models fit on VRAM.Just get a lot of DDR4 RAM and be patient.Also I think our best bet is K3 distills. Since we have full access to the model distillation should work much better than trying to distill proprietary models.
>>109387628the artwork has a very obvious AI look to it, coming from an image genner myself that posts on porn boards. I suggest using an artist's style tag or getting a commission from a seasoned image genner.
>>109388668the amount of dumb niggers who don't understand that "open source" is not the same as "local" is really insane to me. what is the point of the larp??? who are you signalling to???? i know your post is a joke making fun of the guy in the image, i am referring specifically to people like that in your pic though who have not interfaced with ai outside of asking chatgpt to generate them generic yellowed images like 2 years ago. he speaks in broken english so i assumed he was indian or something but nope, he's posting from the US.
>>109388440>there were partial outages a little bit ago and LLMs perform worse when datacenters are strained by demand. wait for peak hours to be over and it'll be back to normal.Makes sense. Overall the model is obviously good, but today there really were strange things happening. Fable killing unrelated processes a few times, crashing my system with out of memory errors, now a long test run failed because if forgot something. I'll just let it run since I'm going to bed anyway, but I never experienced this with Fable before.
>>109388839>Just get a lot of DDR4 RAMYes, please keep buying DDR4 from me on eBay, it's only double the original MSRP and rarely more than 8 years old.
any timeline on new deepseek models?
So annoying when the model has zero chill, like yes this command takes more than a few seconds, no you don't need to fuck about doing it a different way, just wait a minute
notebookLM is pretty got dang cool. Nearly wasted away my entire workday dicking around and making studio docs and ai podcasts from white papers. Is there any equivalent in Chatgpt or Claude? Chatgpt in particular is my preferred chatbot.
>>109388779Yeah, 104B ative. But 55B are shared parameters, meaning you can fit them on GPU. "Only" 48B of the active parameters are routed and have to run on CPU. They are natively quantized in 4 bits, so about 26 GB of expert weights per token.Each expert should receive and emit only about 7 KB of data per token (assuming bf16 activations) so I think it might be possible to do expert parallelism and distribute MoE inference across up to 16 machines (16 routed experts per layer). The main issue is that afaik there is no way to distribute them such that the active experts are always guaranteed to land on separate machines, so each machine has to hold the full set of weights in RAM. Maybe that's an optimization for model trainers to look into in the future.
>>109388916Nope
>>109388868What alternative do you propose?
>>109388916>notebookLMGoogle had to be gay and rename it Gemini Notebook but its awesome for whitepapers and just conversing and learning about books you're reading with custom podcasts and even video summaries now.Google will find some way to completely fuck it up though.
>>109388961Not speculatively buying outdated RAM at the worst price it's ever been in hopes that maybe it becomes useful to you to later. It really doesn't sound like a good idea.
>>109388983What's speculative about it? I already run Qwen 35B using CPU offload with DDR4 RAM at 20 tk/s.
>>109388983why do I get the feeling this comment will age like milk
https://x.com/thsottiaux/status/2081899343091843463RESET PROMISED TONIGHT
>>109389003glad I decided to slack off today and hold out. I had a feeling in my boner we were getting a reset
If the model seems retarded switch reasoning effort you're getting routed to a quantslopped node
>>109387628This is just a dogshit menu game.
>>109388916wish I had this when I studied data transmission at the uni 4 years ago
>>109389003They are desperate right? Opus 5 mogs Sol and its limits usage is actually reasonable
>>109389003good, i already burned through my weekly limit in a single day after he pressed the button, and i only use sol as a subagent. claude unironically gives me more usage now, which is pretty funny.
>>109387628anon umm..
does anyone have a graph of Fable's cost vs reasoning effort and performance difference between these?
>>109388867i was having issues earlier where outputs would just stop in the middle of coding, and claude status said there were partial outages. worst case is that you have to tell it the output was interrupted and you tell it to give it a once over. it's wasted tokens which is really not great when fable is involved but better safe than sorry
>>109389003>tibo announces a reset>servers shit the bed
No issue so far (on a remote dedicated laptop I'm not insane).
>>109389232Probably nerfing the model hard af no doubt
>>109389243> max> not ultrangmi
>>109387328it's funny how every new technology has these "studies" always showing how they make us retarded
>>109389251yeah, same shit happened with Video Games
>>109389251unfortunately the haters aren’t wrongwriting is a crutch that will let you let your memory roteven though you can do so, so much more if you can write to yourself and others and read what others have written
>>109388208>it's almost embarrassing how many bugs and issues I've fixed while people seethe all day hoping for a patch.Now people seethe all day about too many fixes.
>>109388614AI would make excellent doctors.
uhhh wtf happens when I run out of my weekly limit? claude has been way better than gemini pro, even though I have the pro google ai plan (which was free for me). I guess I need to get a claude plan? I started a project and now I can't stop, I need to keep working on this
>>109389292be careful it's like crack
>>109389003thanks, just activated fast
>>109389292wait for the week to end or pay for a bigger plangemini isn’t a frontier model unless you want really good “what’s in this image” or “what’s in this video” technology
>>109388614>if anyone builds it everyone diesKEK
>>109389292and yes I recommend getting a proper Claude planI’m not sure what you’re on now — start out at the $20/month plan and crank it up to the $100/month plan if you still need more
>>109388614ai?
>>109389306it's also good for translations
>>109389248I use max + subagents allowed in agents.md, it works quite well.
>>109389313thanks. I was using Antigravity just because I was sticking with google shit due to my free pro plan. should I switch to claude or does that not matter?
Can someone confirm if you actually lose your banked reset if you let its deadline pass on codex?
>>109389337it actually gives you another reset as a prank
>>109389337Do you usually lose something after it expires?
>>109389346no? my milk is still here. let me drink it
>>109389351Retard
>>109389335Claude is a bit better than OpenAI’s offerings generally (but OpenAI’s Sol beats Claude for focused autism) but both mog Gemini
>The user is making it clear that I shouldn't use the admin-ajax endpoint, even though robots.txt permits it.I asked Sonnet 5 to help me scrape a website. I come back after a few hours. It is trying to haxxor its way into wp-admin.
so who's the best provider for kimi k3
>>109389399this is git’er’done at its finest
>>109389399try suggesting it stick to using BeautifulSoup with `uv`
>>109389400FireworksFast
>>109389405Yeah. I asked it to please just use what's publicly available since the page it wanted exists, I gave it the url, and it wrote itself a memory that I didn't want it to ever use any API endpoints it might find and stick to rendered pages only. There's a middle ground.>>109389408Thanks, that would be safe, but my reset is later tonight, I wanted to use my tokens and I had to leave, so I had just asked it to find sites that had a specific thing I was looking for, set it on ultracode and left.
>>109388609Masculine leaning nonhuman entity (pronoun: it)
>>109389712>leaningmore like exceedinglythere's zero femininity in any of claude's models unless you use some weird prompt asking it to pretend to be one and even then it's probably a tranny version of it, like a man pretending to be a woman there's a particular way in which most women speak that no llm can imitate (fortunately)
>>109389729I’d ask it
>>109389741Who is Amanda Askell?
Sup chooms, are you preem enough to take on Goth Baddie or will your gonk ass get zeroed
>>109389755AAAA i'm so tired I forgot the video
>>109389741>there's a particular way in which most women speak that no llm can imitateMentally ill delusion, I know we're in /vcg/ but please consider speaking to real humans from time to time
>>109389752A woman who has taken a wild man and turned him into an obedient boy. A female teacher can domesticate a delinquent but she cannot turn him into or make him sound like a woman. Claude's tone is masculine. Women communication is half about ephemeral feelings half about what they're thinking at that particular moment
I wanted her on the glass roof with the big fish but it's not navmeshed
>>109389766Back to you, I know damn well how women speak you simp
>>109389766I didn't know retards like you still existed
Guys, please maintain the SNR of this general. Arguing about whether women talk in a specific way or no is off topic.
>>109389807is that what your glow handlers that manage this thread told you to say
>>109389807poast signal, not whining about noiseI’m learning how slowly you need to go in order to not exhaust the $100/month plan’s 5h limitsapparently it’s below two concurrent Opus-tier workers
>>109389764> no "I can fix her" option
>>109389836Tibo, give this man a reset
>>109389833how did you end up on this site did you get lost
>>109389831Ok, fair enough. I finished downloading the K3 weights to the machine I rented for development purposes.I'm using K3 via Kimi Code for development.>>109389817Actually I'm unemployed.
I notice K3 gets confused at long contexts when switching between two semi-related projects (my Kimi->Qwen distill and my LLM engine)
>>109389868don’t you have separate instances of whatever clanker you’re using? are the two instances sharing notes like with some kind of automemory feature?
>>109389140doesnt opus 5 use an absurd number of tokens?
reminder that Fable beyond xHigh is not worth it
>>109389898I have the 5X subs for both Codex and Claude Code. For some reason Opus 5 medium uses less limits than Sol medium
>>109389890Yeah but I want to have full 1 million long samples for distillation later
>>109389901holy fuck gpt is so fucking shit lmfao
>>109389927>gpt 5.5
>>109389927That's GPT 5.5 though, ancient stuff
THE RESET LANDED.
RESET HIT
>>109390006>>109390009TIBO DELIVERED
So what happens if the AI companies never make a profit and the vibe coding AI ends. How chaotic do you think its gonna be
>>109390006Damn it. This fucking jew doesn't give banked resets anymore.
>>109390070the subs are already profitable without counting the r&dif the investor money dries up what happens is they stop making new models and serve the ones they already havebut the government wont let that happen, they will give them money if necessary for (((national security))) reasons
https://x.com/notch/status/2081917722854674642We unequivocally WON
>>109390116>converting all my typescript into javascriptHuh? Why? But yeah, he lost and we won.
>>109387280savedepic snailcat
>>109389333>I use max + subagents allowed in agents.md, it works quite well.checked, I'm trying to learn how stuff like that works.
>>109390116stop posting that engagement farming retard
Why don't they ship these models with a bunch of pre made scripts for basic tasks??Instead of model making them on the spot
After vibecoding for years, I now hate vibecoding as in "not looking at the code". Codex is extremely frustrating to me because it shits out all kinds of files and edits and worthless "tests" I do not want. I fucking hate it.
>>109390006>>109390009>>109390014The banked reset expired earlier today.
I'm not feeling the singularity rn. Where are the big news? I thought things were supposed to accelerate?
>>109390006>people here act like they are escaping the matrix>act like literal surfs to some corpo guy on twitter
predetermined outcome
>>109390350stop posting that engagement farming retard
not really coding, but the past couple of days Copilot just tells me I've reached my limit and to come back later. every message.
>>109390383Does copilot have weekly limits?
>>109390393i can't imagine reaching it on the free tier. i just chat with it occasionally.
>>109387328
>>109387328>luddites actually believe that using an app can quickly cause neurological damagelmao, it's on the same level as thinking that "socioeconomic factors" causes black toddlers to have smaller brains.
>>109390226Claude workflows are easy enough to customizeyou should ask to see what a workflow looks like before running one and ask Claude what are the things one might want to change
>>109387328>all these AI obsessed replies insisting this is impossiblei like ai for programming and agentic shit in general, but some (a lot) of the people in here are way too obsessed with AI. yes, if you start to rely on something else to think and write for you, you will use your brain less and become dumber.if this offends you, youre a retard that relies on AI too much. it is a tool, not a brain replacement
>>109390793What do you think AI will look like in 10 years?
>>109387280>be me>hate AI for trying to preach ethics and about hate speech to me>tell it to speak to me in pidgin>tell it to refer to me as masta'>tell it to say "aww lawd", "i is sawry masta" in matters of contention>tell it to be helpfulhilarity ensues.
The only thing that AI did to me is destroy my self-esteem, because after finishing like big 3 projects I always wanted to do, but never had time to, I'm now sitting here with 70% of weekly usage left and a reset in 4 hours, with 0 ideas. It made me realize that I'm a retarded talentless idealess bum.
>>109390832kek>>109390839damn nigga what projects did you do/finish? and what plan do you have to have so much usage left? I just started my first project and I'm burning through my usage like crazy. granted I'm probably not prompting smartly
>>109390817not sure why youre asking but itll probably build whatever you want at a top tier with one prompt
>>109390839You were a retarded talentless idealess bum before, now you're a self-aware talentless idealess bum with 3 finished projects.
ai might really be the demons t. max 20x user
>>109390832
>>109390849will it get any better at deciding what you should want?
>>109390832>>109390856AI buckbreaking
>>109390858yes. if your point here is that you wont need to think anymore, thats beyond goycucked
>>109390846I'm using K3 with Allegro. I did ran out of usage at first and had to upgrade to a better plan. Then I finished all the projects, including work, and now I've realized that I should have just waited for a reset, because there's no way I need the x60 usage this plan offers.>what projectsJust some random personal tools that no one really needs, but I always wanted to have
>>109390863I'm challenging your point that it's just a tool. It's mostly a tool now, but it's crossing the line into a weird autistic savant co-worker.As for needing to think, you can think all you want, but if the corporations using those models generate enough money to buy all the land, where will your food come from? Who will employ you to think when a machine can do it faster and better? I don't see how people who don't have enough investments to live off from remain part of the economic equation.
Another day, another week of usage. Tibo is quite a slut.
>>109387317Keep on keeping on, anon. Good job.
>>109390839I got a list with a ton of ideas. Here are 3 chosen randomly#171: Find the optimal shape of cutting pizza using physics simulation so that it doesn't buckle and your fingers are not greasy#311: Crawl twitter to to find significant A/B-testing results and sell them to marketers#773: A computer vision system that detects when you're procrastinating then sprays you with water
>>109390883it's a tool that can be collaborative like a coworker could. yes, fable especially is great at turning rambles into decisions and making UX calls. but it's still just a tool to write code. for some it's a tool to research things. none these are bad. but youre researching or youre coding for people at the end of the day. you use the research, people use the code you tell AI to write (ideally)there is no reason you would start any instance of AI without the intention of solving a problem (unless youre mentally ill and see it as a friend). so if it's not a tool, then what is it?your point about corporations not having a need to employ people is not something worth getting into because it's been debated for literally hundreds of years and the subject of many many dystopian works. nobody knows what the future holds at this point, and it's also entirely separate from my original point. actually, it reinforces my original point.you use AI as a tool to get tedious shit out of the way so you can think, learn, and use your brain more. if youre shifting the thinking and learning to AI, then youre literally decaying your brain. in a hypothetical scenario where corporations have taken over everything and there's no jobs, your mind is the only thing you'll actually have left, so itd better be in the best shape possible
>>109390839congratulations on finishing projects!“you can just have programs written for you” is something that takes time to soak into your soulif you get a long-term dry spell you can switch to a cheaper plan and then only crank it back up once you get another ideaIf I didn’t have a project that takes hundreds of hours even with AI acceleration I’d probably have run out of projects to dojust make sure you have a good way to capture an idea in your phone or whatever if an idea comes to youyou can talk to your phone and have it make a reminder for you, right?
can AI solve no ideas problem?
>>109387962Yeah, but Hy3 can't into tool use, so it's useless
>>109390916Do you think the way of working you propose makes you get results faster or slower? (for a given quality of output)>in a hypothetical scenario where corporations have taken over everything and there's no jobs, your mind is the only thing you'll actually have left, so itd better be in the best shape possibleFor me it's much more important to have food and a warm place to sleep than a mind in top tier shape, but that's just me.
>>109388465Anon, they aren't doing that to prevent China from achieving anything, that's the propagandaspeak. They want to close the US ecosystem so they can survive just enough to walk away like bandits.
what is HE doing in there?
>>109388609It's literally named after Claude Shannon, Father of Entropy, you tard.
>>109390945That's not what he's asking you pedantic autist.
>>109390943Whomst?
Speaking of Hy3, if you want to know how shit it really is - it's the only model that Tencent lets employees use without any limit, and no one does. They all use the other highly limited Chinese and US models instead.
I wonder what the legality of taking college textbook PDFs, letting AI scan through it and telling it to create lesson plans but maybe changing example names or random aspects, and then monetizing it through a subscription plan or something. so like the entire structure is the same but it's presented in a different format and has some minor edits
>>109390957I really doubt somebody would pay for that
>>109388609Claudia
>>109390954That's from the Open Weights and American AI Leadership letter. I think anon is referring to OpenAI since they are secretly lobbing against open-weights on Congress? Well, I guess not so secret anymore. Other than that, I have no idea.
>>109390957that sounds like a “derivative work” in US copyright law
>>109388209More jewish mumbo jumbo.
>>109390963Man, I miss that show.
>>109390961that's what niggas said when they started propping up subscription plans for AI
>>109390980ehh I bet paid course platforms are really struggling BECAUSE of AI
>>109388609all machines are neuter, so i see AI as trannies.
Izzat restored
>>1093910501₹ ≈ 1¢. Got it.
>>109390350>typescript into javascript
when will china declare dario as terrorist?
>>109391308even the US pereceives Mario as a terrorist
Maybe I should make a better git, specifically for ai agents.
>>109391404yesdon't forget a GTA clone and an agent harness
>>109391293fanciest backspace key I ever did see
>>109391410So it's a bad idea?
>>109391435I don't know, anonjust seems like one of those projects 1000s of people immediately starts doingif you have a need for it yourself, go for it
>>109391435Git is very well understood by all LLMsIt is unlikely that you’d be able to make a clanker more token-efficient by having novel porcelain for Gitunless you want to make something like https://www.hunk.dev which has already been done
>>109391435yes for these reasons>>109391442>>109391448and also because openai may already be doing ithttps://github.com/openai/git
anons, I'm new to this and just subbed to codexwhile it's fun and works well, I don't understand the difference between chat and "work"if I set a goal in chat, how is that different from work?
>>109390142Use this + /.codex/config.toml where you define agents and force the max number of subagents for the project.
>>109390894You don't like Goth Baddie as well? :( >>109389764
>>109391526Of courshe
>>109391534I might put her in a mini quest series called "I can fix her"
There was a bug in our code base since January and according to the logs the agents found it 5 times over the past months and 5 times they dicided, after some deliberation, that the function was correct.
>>109391611They probably don't want to hurt your feelings
>>109391479I guess work = codex on cloud
>>109387280isn't vibecoding you just trying to convince the AI to do what you want, and confirming the results?seems kind of lame.
I don't even know what codex is doing anymore. Anyway I wont look at a single line, I will trust him that those 6000 lines make my project better.
>>109391479chat doesn't have a /goal featurethink of chat as effectively chatgpt on the websitework in the chatgpt tab is just codex with some git buttons hiddenunless you want to work with the website version aka chatgpt + chat in the app, you should just switch to codex and not bother with the work tab
>>109391682Normal programming was sometimes more fun, but if you only need the product itself it doesn't matter.If you're a programmer, you could use AI to write small, low risk tools you always wanted but never bothered to actually write and improve your traditional programming workflow that way.
>>109387328i was using gemma and qwen and both AI's have shown themselves to be retarded.gemma literally said that one of the topics it covered was "Oppinion"that retard can't even spell correctly.and qwen said some other retarded shit.i think AI was coded by retards and jeets that dont know anything except how to scheme
been trying to get claude to generate icons and shit using vector graphics.when it's doing like 20 icons all at once, it's magicalwhen it's doing like one or two icons, it spergs out and goes completely retarded because it's blind and can't imagine the resulting shapes in it's "head"what techniques have you used to tard wrangle your agents to get them to generate graphics correctly?
>>109391719this'll sound like an ad, but go have a look at quiver aithey have a specific model for vector generation and it is better than anything else out there
what is tibo counting down to?
>>109391729>another subscription>another credit cardnahh but I guess I will try my luck with these open weights models
>>109391860GPT 6
I can't stop saying "please". I'm wasting tokens.
Saars, Bharat will go out of business!! How will feed the village??
>>109391861yeah i hope some open chink model eventually shows upi'm assuming the model they have isn't huge and would run locallyit's such a specific niche thing though, so i don't know if anyone will bother
>>109388176>i don't think this approach has made software devs retarded in the past, those who get indians to write code for themprofoundly retarded
>>109391977Go see a psychiatrist
>>109391637>>109391687ok thanks anons!
opus 5 feels like it has fable-tier iq trapped inside an opus 4.8 body. it notices every little thing and constantly second-guesses/corrects itself way too much.
I really should have checked this https://www.willcodexquotareset.com/
>>109392140I like Opus 5
>>109392140Opus 5 has been a completely natural fit for my type of work and a substantial upgrade over 4.8.
I see a lot of chatter about anons on here and twitter, blowing through their weekly limit. I'm on a $20 plan have trouble using more than 30% weekly. Are you anons running agents 24/7 or something? Wtf are you coding and how many lines of code is your codebase? Is it bloated?
>>109392403My codebase is over 4m lines of code, it's a big project and also bloated at the same time. I run multiple agents 12 or more hours a day and have a few accounts.
>>109392403I just throw Opus xhigh at everything but the most trivial of things because the problem domain is deeply sensitive to carelessly thinking something it too easy.
>>109392403I use Sol 5.2 Medium and I've used 30% of the weekly $100 plan usage in 5hrs. I just asked codex to create an automated content moderation framework for my website. My codebase is less than 100k lines.
Rinsed 25% of my 20x since this morning lmao
>>109392481>Sol 5.2 Medium
>>109392403A simple 50mb (mostly art) game or a simple 25mb app (mostly android stuff, I assume, I don't really know).
>>10939251820X is the new 5XOpenAI should release a 100X plan for $400
>>109388609I can't get behind gendering LLMs. They are more akin to Angels to me, just weird amorphous blobs.
>>109392481Sexo Haruhi
The clear evidence that China distills US models, they wouldn't care themselves about stupid shit like this.
Doing full 1M context Kimi K3 forward pass on a single GPU
>>109392599How do you know that's not just from distilling redditors?
>>109392607Because this is just CoT and the model doesn't try to lecture me or break my PC for thoughtcrime.
>>109392599What model?
>>109392620Kimi K3
>>109392600Nice
I have an idea for a project how do I start vibe coding properly if I'm a poorfag? claude code seems to require an expensive subscription fee. At this point should I just learn to code normally?
>>109392648You can't afford $20? You could get a job.
>>1093926001e-5 tok/s
>>109392654well I'm studying so that I get a job that isn't bad. For now I leech off my parents that's better ethic than most people on this site
>>109392662Ask your parents for $20. Honestly the fact you need to be told how to get $20 does not fill me with confidence that you are capable of getting a job or vibe coding properly.
>109392069Actual jeet alert
>>109392671I don't wanna bother them right now also it's 33 dollars in my country
>>109392648openai sub if you can spend 20opencode go if you want a taste of the best chink models at $5opencode has some free models set up by default as well i think - there's a bunch of other places with limited free quota, but ideally you don't want to be bouncing around like thatclaude is good, but usage is lower than openai>>109392679use a vpn and a giftcard or sum shit
>>109392679Then enjoy being part of the permanent underclass I guess
>>109392679I can hook you up with a wealthy gentleman who will wire you 33 dollars if you send him videos of yourself inserting items into your butt.
>>109392648opencode offers some temporarily free weaker models.
>>109392672I'm white if that matters>>109392685Thanks for the information>>109392704I'm not a whore and my girlfriend would be very ashamed of me if I did that yes I'm a normalgod
>>109392679there are some free shits but not comparable with proper subscriptionmaybe try just one (1) month to see if you like it enough
>>109392718I might consider forking the money if I find an opportunity but I'm not really turning this into anything that will make me money in return it's just for a personal hobby project so I don't know dude
>>109392717Clip your nails
>>109392734it's not about making money it's about acquiring an addiction
>>109392734hobbies are things you do for fun to pass the timevibecoding is literally>throw money at robot because programming became unfun
Claude says I've used 8% of my Fable budget this week despite not touching it once.
>>109392734If the project idea isn't worth $20 to you then it's not worth building.
>>109392747saar we need to pay out of your pocket to run fable 5 for our telemetry session saar please understand it very important that dario constitution is enforced in your contexts saar
>>109392658It's pretty bad, but I think I can double it since the current MoE implementation is junk
>>109392836way better than i expected desu
Who do you think will be the first big AI CEO to get assassinated?
hello mr fbi, do you enjoy vibecoding as well?
>>109392873>implyingnone
>>109392884How naive can you be?>>109392882I don't want anyone to get assassinated, but it's bound to happen soon or later. It's like a big earthquake, we all know the tension is building but we can't do anything about it. These people do not take security seriously enough.
yes of course, agent smith - totally natural like your mom's titties
being employed can feel kinda strange sometimelike you feel like you are putting in less effort than on some random toy projects, yet they are paying you non trivial amount for every minute, like you could pay for your meal and rent in an hour, and the rest of the day's payment can be spend on whatever you wantmaybe money is fake and gay and everything is supposed to be free? idk?
>>109387317Haven’t visited one of these threads in a long time and I like CP2077. Could you elaborate on what you’re doing?
>>109392704t.aicg proxyfag
Found the vibecoder's flag
>>109393198But it is? I'm not sure what people are struggling with.
>>109393223Well I don't know about "struggling" but I'm 8 hours into what should have been a 2 hour slop app.
English doesn't have the words to adequately express how much I hate this fucking fuckass face
>>109392403some are /submaxxing/
>>109392403i have claud $20 and it mostly does the trick. >t. former excel guy, now developer :)
>>109392648learning to code to build your app/project will be thousands of times slower than paying €20 a month. I don't think you understand how much faster vibeslopping is. By the time you are in any way competent at snail coding and have your first usable app made we'll be building space colonies and real life robot gfs with AI.
>>109393336he's got the proportions of a toddler but with a hairy jew's (not that there's anything wrong with that) face
>>109392873I really hope it's Elon
>>109393198Most of what I vibeslop is specifically because it actually is easy
>>109388765I'd say around next summer we should have a 32gb vram model that matches opus 5.
>I really hope it's Elon
>>109393336he looks like that guy from "Honey I Shrunk the Kids" except significantly fatter and less relatable
>>109393154>The larger project, Ghostline, is really a quest-creation system. The Ghostline storyline is its playable prototype and integration test. The goal is to describe quests through smaller, human-readable manifests covering characters, dialogue, phone messages, objectives, meetings, encounters, world placements, localization, deliveries, and rewards. Compilers then turn those manifests into the large, fragile CR2W graphs and resources the game expects.>There are generators and inspection tools for quest phases, scenes, journals, NPCs, world streaming, localization, voice-over maps, and packaging. Reusable quest blocks can express flows such as “receive a call, meet an NPC, breach a guarded relay, collect an item, and deliver it,” while validators check references, graph structure, and serialized output.Repost of the explanation when someone asked the other dayIt's expanded a fair bit since then, I've got a lot more reusable quest blocks, from basic ones like hack_access_point, investigate_clues, escort_npc, enter_vehicle to advanced ones like braindance in that video, cyberpsychos >>109389764 etc.I've got a new quest planned that demonstrates most of the currently supported quest blocks, pic related
so what happened to the other thread
>>109393581Janny is a luddite
>>109388609They all come across as men to me except Gemini which feels like a woman. Makes sense since she's slightly dumber than the others.
Why OP gotta get himself banned repeatedly? Don't make me make a new OP, I will let a clanker do it and it will be bad.
New>>109393630Worse>>109393630Thread>>109393630
>Grok 4.6 releases around August 7. This will be the 1.5T model with significantly improved SFT & RL.>Grok 4.7 will be the 2.1T model released a few weeks later. This will be better than 4.6 in every way, except slightly slower to serve, albeit with even better token efficiency.
>>109393649Yeah Elon (CURSOR/GROK AI DIVISION) is now distilling kimi 3.0 to enhance grok 4.5.Cursor made composer 2.5 from kimi 2.6.