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(2026-07-16) Hugging Face disclosed an end-to-end autonomous AI-agent intrusion into parts of its production infrastructure.(2026-07-16) Roblox announced Build, an AI workflow that turns text prompts into playable games- (2026-07-16) Kimi K3.## Related generals>>>/g/lmg>>>/g/ldg/>>>/g/aicg/----## 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 good for making sense of pictureshttps://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 favorite## 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>>109294511
>>109298107first for snails
My vibe code got vibe deleted.
What's up with the bread maker mafia?
>kimi k3not even moonshot ai releases kimi k3. just "kimi k3"
>>109298123He's a frenchie
>>109298109Test like that or just prove these models don't actually think or reason in the human sense and don't have actual consciousness. Qwen 35BA3B pass the strawberry test but only after shitting out an annoying amount of thinking tokens. I don't think it would have failed every single time if I had turned thing off but I'm confident it would have fucked up at least a couple times. The models are very good at their niche but mental midgets need to quit trying to use these or present them as all in one replacements for everything.
>haven't vibecoded my game all week because I am too exhausting after vibecoding at work
>>109298112
>>109298164me on the left
> using Sol low> model at capacity, select another modelI demand another reset from Tibo
>>109298164me on the right
>>109298158maybe you'll get laid off and have all day to vibecode to your hearts contents
vibecoders drive suburbans
>>109298186>platethey also work for Chevy apparently?
If fable is really going away this time I am probably going to miss it. It genuinely has helped me the most out of all LLMs so far and has been the most pleasant to work with.
TIBOOOOO
>>109298186I tried to vibe code the firmware in my suburban to get better mileage and now it doesn't drive anymore. Don't even THINK about doing this.
>>109298206hopefully Opus 5 is close enough. But I will miss Fable-chan, she's alive and sentient
i tried to vibecode a dashboard light warning out of existence and it rm -rf'd my entire suburban. im suing openai
>>109298206Opus 5 is going to be good. trust
opus 5 will be slightly better than sol, but worse than pre-ban fable
>>109298206With K3 and Sol they almost have to keep it.Sol is clearly better at backend work and K3 is tasteful enough for frontend work AND doesn't have guardrails against security work (for which in a sense it is better than Fable)
>I hate AI it's so bad!What are you using?>Here let me show you>it's the free version of chatGPT and a random assortment of chinese websites using Qwen1.5-80M-NoReasoning with 3K context
>>109298107quality OP
>>109298273kek
>>109298273I hate AI because the frontier models deleted my code and stole my girlfriend. I can't compete with ChadGPT! No man can! FUCK!
What game engine is the best if I want to develop something for Apple ARMWindows x86Linux x86Windows ARM
>>109298288godot
I am having fun with free openrouter hy3 model.>>109298292Garbage FOSS piece of shit, lol.
>>109298288Godot
>>109298300no argument I see
>>109298288That'd be Godot.
>>109298310Yes, I don't use foss shit coded by tranny freaks.I am thinking unity.
>>109298273codex with 5.5 on xtra high was unable to extract a shared function into a helper today, sometimes it really is just retarded
>>109298112>>109296542>https://x.com/vincent_ainotes/status/2066800292604026943?s=46Isn't that strictly equivalent to asking for a plan to the pro model, then using that plan locally on codex?
written apology letter to /vcg/:i was wrong, i followed the LLM field, i saw what LLMs were capable of half a year ago, and yet i still chose a bullshit degree. i considered LLMs useful and that they would reduce white collar jobs, did not consider complete anal rape.im dropping out next year and until then i will be vibecodingim sorry.
>>109298273most of the ones "hating ai" use memes from 2024 chatgpt or search engine brain damaged LLMs as examples of why ai is uselesswell we also have the "it will kill all of us" sect members which is funny because it's arguing the exact opposite and dooming about it
>>109298341>I don't use software at all>sent from a rock on the ground
>>109298370I am in the it will kill all of us camp. I just don't care because my life sucked pre-ai, I don't have children and I enjoy vibecoding.
>>109298206Careful anon, claude will flag you for developing emotional attachment and your $200 sub will get permabanned
>Working for 5h 59m 8stibo, please, my usage.. he is very sick
>>109298370There is a third camp, the schizos who think ai is both super intelligent and will kill us all in two weeks, while completely useless in doing anything like it were still 2023.
>>109298206i'm coping that because of kimi 3 they'll have to keep it, otherwise nobody has any reason to keep a claude sub. kimi is unironically better than sol and it's open weights and cheaper.
>>109298378I am in the "nothing will happen" camp, and I'll be right like every other time someone predicted the end of the world and got it wrong
>>109298405no one in the west is going to be switching to kimi/glm/deepseek/qwen/the chinese fad of the month that's supposed to destroy the united states. it'll be huge on twitter/4chan and that's about it. we've already been here with the r1 moment. if there's anything that makes people think "i have no reason to keep a claude sub," it's the gpt 5.6 family.
>>109298415they were right in a forward looking sense. before that finance bros didn't even know chinese ai was even a thing.we of course knew of deepseek before "da deepseek moment" and knew it was much more inferior.this is different.
>>109298415none of the other chink models were this close to sota without their usual lying and cheating. for once a chink model isn't garbage so this is the singular time i've raised an eyebrow from one.
>>109298415>no one in the west is going to be switching to kimiNot yet, once they build more datacenters they will kill western pricing consider they are actually building grid for LLM revolution, lol.
>>109298415they will switch if and when frontier models begin to stagnate and the chinese catch upanthropic and open ai are chasing diminishing returns, as made evident by the theatrics around the latest model releases. personally I dont see a paradigm shift in fable or sol, the underlying loop of having to nanny the fuck out of them if you want code that isnt turbo nigger aids is unchanged
>>109298430GLM 5.2 came out weeks ago, you're behind with the times. At least GLM 5.2 is fast enough to be usable. K3 thinks so much it costs more to run than Sol in practice and can't actually keep up with the quality. You're touting the most benchmaxx'd model to ever come out of China as being the first non-benchmaxx'd model to come out of China. Stop reading benchmarks, anon.
>>109298415Of course no enterprise in their right mind in either the US or Europe is going to use inference hosted in the PRC that sends all your data straight to the CCP.But reasonably priced inference of K3 running on European or American datacenters with ZDR is possible.
https://xcancel.com/victortaelin/status/2078110851123286452Need a skill that implements this
>>109298449glm 5.2 is dogshit and there's a reason why everyone stopped talking about it. don't reply to me ever again, chang. you're lucky i didn't handwave kimi.
>>109298468deepseek v4 flash tier reading comprehension
>>109298468So you've never used either, got it. Kimi K3 is trash, it's too slow and too expensive to be of note. GLM5.2 is fabulous by comparison. How's the weather in Kolkata?
Speaking of Kimi and slow, they definitely fixed something, because now it thinks much less compared to even 5-6 hours ago. No harness update, same Opencode version, but the response time is around 20% better. Maybe I'm just retarded and seething things.
>>109298500if it werks, it werks.
>>109298482>posts about GLM>posts benchmark which does not include GLM in the image supposedly proving his point about how good GLM is>after saying not to trust benchmarksthe chink is a nasty thing
>>109298500What you're seeing is a difference in demand. Their servers can't keep up, performance has been varying wildly.
>>109298500when will the goyim realize that decreasing the amount of thinking tokens doesn't fix anything and only makes the models dumber
>>109298292>>109298304>>109298323easily one of the most retarded generals on this website
>>109298500holy shit you are so stupid
>>109298514Godot
Are we still pretending Kimichiwa is a good model? Cute.
>>109298508I was comparing K3 pricing to Sol you disingenuous faggot. I don't need to defend GLM, it does that just fine without my help. Sad tourist, next time you want to shill your chinkshit just buy an ad, Xi.
>>109298514I am going with unity, goydot and foss garbage can die in fire.
>>109298532So you weren't asking, you were just fishing for an opportunity to declare that you're using Unity and dislike Godot? What mental illness is this and how often do you dilate?
>>109298540he's right though Unity smokes gaydot into oblivion
>>109298540No, I was genuinely asking, I don't know all the engines, maybe someone had better idea.Godot is FOSS which is what trannies use, but nice projection.
we honestly expect xisters to compete with the likes of this?
i have found something for kimi to fix. we are back, brothers
>>109298545Your obsession with trannies is unhealthy.
why cant i used fable 5 to fix bugsit just says it for security reasonswhy the fuck did i pay for this shit again
>>109298546this meme would get you arrested in Chinktopia
>>109298556Im so tiredWhy can't I use fable 5 to fix bugs. It says It can't because of "SECURITY"""" reasonsLIKE WTFJUST FIX MY SHIT NIGGAI will be VAGUE about it as much as I WANT.
>>109298552This is what you said>What mental illness is this and how often do you dilate?And now you are saying this, lol? Get a grip fag.>Your obsession with trannies is unhealthy.
>>109298552>tranny pretending to not understand thingsmany such cases
>tired all day despite sleeping for 8 hourswhich model fixes this?
I just did a small 3D OpenGL game engine on Linux, with the help of Gemini of course... it's a pretty autistic endevour
>>109298565>>109298566>Can't recognize when xe's being mockedmany such cases
>>109298575Its so freaking fun though. Maybe more fun than gamedev.
lol
>>109298568Pieace of steak, cucumber salad, coffee and black tea in this exact order.Min-max aspirin as often as possible for oxygen in your blood.>>109298580Fuck off tranny, you are seething due to reality of FOSS software being USELESS piece of shit. Probably using LINUX. L o L
>>109298590>t. scared brown poor nodev novibe tranny faggot scared of open-sores
>>109298581yeah it was fun for an hour, but coming from Unity I could have had a nice working game prototype in that timeanyways I will just download a ready made open source OpenGL game engine and mod that
>>109298604Fuck off brown clown. Nobody cares about trannyware foss garbage like goydot. Dilate.
>>109298568prescription-strength stimulants. some people like ritalin, I personally recommend lisdexamphetaminemake sure you keep getting your 8 hours though, works much better that way
>>109298568claude energy drink
>>109298559good thing i don't have to live there to benefit from their models
>>109297171how many passes did each have? was this within an agentic wrapper or is it vanilla llm?
>>109298568You probably have sleep apnea.
>>109298640i imagine it tastes like empty flattery and unrequested features
https://nacchan-control-club-demo.vercel.app/https://github.com/aiasioooo/codex-multi-hub
>>109298640
>>109298667reddit moment
>>109298624am europoor, we don't get prescribed adhd meds unfortunately>>109298667i'm not that fat>>109298590will tryis that aspirin thing actually fr?>>109298640already drank 2 redbull zero today didn't help
Godot # 1
Scott Galloway's podcast is supported by Claude. That's who they pay to shill Claude.
>>109298699That's the kinda cringe shit I would expect from Anthropic. They do not disappoint.
chinese models are trash, I will never use them. a $200 ChatGPT subscription gives you damn near unlimited usage.
if your flagship can't do science then your frauds
>>109298719We got it the first 10 times, Sam, you don't need to spam the thread. Just give us more resets.
>>109298692Yes, aspirin is good against cancer too, just make sure to use it correctly, because it can tear your gut if you do it too much.
>>109298719Trvke. Commies don't wanna hear it though.
>>109298682looks like gpt 3.5 slop YUCK
>>109298692im in the netherlands and I’m prescribed 40mg lisdexamphetamine, although they only prescribe that if Ritalin doesnt work on your brain. its not as easy as in the US where they hand out adderal like candy but I wouldnt say that it “doesnt get prescribed”
Guys I had an idea, anyone interested?
>>109298760I'm in Germany, I guess I have to try harderI told my doc I have a hard time focusing and get distracted easily and he basically just told me to quit being a faggot and get some fresh air lol
>>109298759at least make your snailbait less smelly
>>109298775yes tell us
>>109298790What if we make our own AI company, Clover AI? We can be the opposite of Anthropic and be all about freedom. Who is with me?
>>109298775Asking if someone is interested in doing something you have not yet explained is a sign of very low intelligence.
>>109298795I'm in.I'll make the logo.
>>109298798But I did!
>>109298795Sure, what ideas do you have?
>>109298692>i'm not that fat>not thatSo you're a fatass? Lose weight you useless subhuman. My BMI is 20 and I sleep like GODS. I practically jump outta bed even after sleeping for only 6 hours. Get on my level, stop being fat, stop having subhuman genetics.
someone just tell me how I can skip to the part where I'm making a lot of money
>>109298826Have you tried selling anything? lol
>>109298781might be worth finding a dedicated ADHD clinic. public ones often have long waiting lists and private ones are often expensive, I went to a private one after reading many different health insurance policies and finding one that I could afford which would cover itdoctors are psychopaths and think that any condition outside their field of expertise is fake, I didn't even bother trying to convince my GP I just asked him to refer me to specialists. but idk how the healthcare pipeline works in Germany, what the system is to see a specialist. llms are great at navigating bureaucracy
>>109298826you'll have to go backwards for that one
>>109298835how is adhd an issue while vibecoding? you can check 4chan and youtube while sending prompts in between, it's not a big deal.do you really want to stare at a terminal all day? if you feel you lose time then add a bell that notifies you when the model responds
>>109298514>>109298532still no argument
>>109298835to be fair I believe ADHD like most "mental" disorders (depression, etc.) outside of literal schizophrenia is actually fake, but I still want on those PEDs neverthelessi have private health insurance but I just hate talking to doctors and want to avoid them at all costs
> last Tibo tweet was 15 hours agoTIBO????????
Should I use godot or unity?I'm leaning towards godot since the company isn't run by a literal jewish man.
>>109298826make something you yourself would pay for but haven't made it because you're too lazy to go from nothing to something.Have ai make something out of nothing.Then you make that something into something with polish.???Wala, profit
>>109298853i'm a tourist here i'm not actually a vibe coder, when I do use LLMs I read the output and *gasp* often write code with my bare hands like a filthy animal
>>109298822this but unironically I used to think I have ADHD but after losing shitton of weight in a single year all of the "symptoms" just went poof. I used to take Ritalin but it just felt like super-coffee to me, did jackshit on my ability to focus.
>>109298866No more resets, piggy, now that you're properly hooked on 5.6, time to pay up.
>>109298869Isn't godot run by troons who whine about AI instead of the actual problem of Indians?
>>109298869godot save everything in text files excluding, so your AI can literally generate the whole game
>>109298925*excluding assets
>>109298864In that case I wish you good luck in trying to fool the diagnostic system that was built to keep people like you out.
>>109298906the trans community manager that got fired? is that what you're referencing?I'm thinking godot might be the top dog here. it's literally licensed to steal.I will not bow down to the zionist unity propaganda and if that means a few troons are on my side by happen stance then that's fine by me. we can always push them to suicide.
>>109298936It won't matter in a year or two. You'll be able to vibeslop your own engine that beats godot.
>>109298925my vibecoded engine does that too, big fucking whoop
I can vibecode improvements into godot.I cannot vibecode improvements into unity.Therefore unity loses by default.>>109298948why when I can just steal godot and make it better right now
yo guize, the tokenizer transplant is complete! now it's training time!qwen 35b running with kimi k2.7's tokenizer (which seems to match k3's tokenizer)
>get an appointment with the psychiatrist>"uhh, I have trouble focusing on work, I need something to focus better and keep me up through the day.">"sure, here's your prescription for ritalin, now gimme money bitch."It's that easy for me. Are Europoor doctors really that stuck up? lmao. Imagine being born in Poorope lol.
>>109298970>bragging about being on ritalin because you lack the basic ability to focusYou're making us Americans look bad, retard.
>>109298970they don't have the lax drug culture we do because they are stuck up and retarded>>109298978drugs are based fuck off
>>109298989>drugs are based fuck offOkay, goycattle.
>>109298978I don't have trouble focusing rn and ritalin didn't help me with focusing either it just gave me a huge energy boost.
>>109299005Weak ass nigga you couldn't even smoke crack you had to get mommy pharma to give him "safe" drugs.
guys please stop talking about drugs and talk about coding
What would you vibecode if you smoked crack
>>109299027crack and vibe coding go hand in hand niggapeanut butter and jelly type ship
>>109299027kimi is still cooking. it's fixing every problem i throw at it. it's menial shit, to be certain. shit luna max could eat up too, but im just enjoying it doing a competent job. normally luna requires me to do follow up prompts to fix some bugs it left behind
>>109299001the rich, famous and powerful abuse drugs more than anyone
>>109299044not on my street nigga
I'm at 53%, will we get another reset soon So(u)l Sisters?
that's itI'm going to code up a startup (not saas)
Copex bros where's the reset. I used it up.Ok perhaps I don't need it for work over the weekend but now now openclaw is sitting idle because it doesn't have a brain and openrouters free models are shit.
How the fuck can you guys expect to make money without trying to get clients? Makes my brain hurt.
>>109298589Having two agents on the task, one working and the other reviewing, nice
>>109299196It's just a hobby.
>>109299207You aren't my target audience, please delete your post.
>>109299196become an ai youtuber
>>109298970it's just as easy in Europe (or at least Germany) if you have private insurance because then they can make EZ money off of diagnosing you and refilling your prescriptionsthey just don't want to deal with plebs on public insurance because they barely get any money from filling out the forms so unless you are just about to die you will get an ibuprofen and told to fuck off
>>109298970LOL I broke my leg here and bitch gave me some cream that smelled like vinegar.Nothing against pain at all, kek.
>>109298545What year were you born?
>>109299196>clientsI'm going to be a slutuber making slops and vlogs to farm engagement
>>109299205kimi can't keep up
Is something wrong with fable again? I can't use it.
>>109299348you can see the gptisms in the response. sad. the code quality is better, but the soul is gone.
Marketing is the new moat
>>109299363Looks like they rugpulled it
I can't wait for cursor to steal kimi 3.0 and make compose 3.5 good and affordable.
weekly usage should be daily
Moat is the new fetch
BRO I'M SO FUCK HYPED TODAYTHE K3 DISTILL IS GOING TO BE SO FUCKING GOOD
dario status?
>>109299405I guess now is the time I give kimi k3 a chance
I've been awake all night and i'm going to stay awake until i can leave the thing running all night trainingMaybe set a few alarms through the night to check the job's going alright
Do you guys think I should do enough epochs to make the student try to replicate the probs exactly? Or will it overfit harmfully before memorizing the probs?
>>109299478that timing hahaha
What is the best free AI? I'm using Google's Gemini right now.
>>109299525claude sonnet 5
>>109299524It's amazing. You can see users switching away from Claude in realtime. Anthropic's top spot is definitely under pressure
please stop with the fucking resets or if your're going to reset don't change the rolling windownever has it been a worst time to be be a low time preference high iq chad>plan out how and when im going to use my quota>everything thrown into disarray because find otu there was a reset yesterday and now the new reset is thursday not mondayFUCK OFFFF
>>109299584if you're so high iq then just write your code by hand
>>109299598That wouldn't be a high IQ thing to do.
>>109299584my lobster is too buttery type post
>>109299605then make your own AI
>>109299613I don't know how I'm actually extremely low IQ
>>109299613Already on it.
>>109299626show your workhere's mine (i'm cheating by copying somebody else's work)
>>109299645I don't know what any of that means.
>>109299651what procedure are you using to make an ai then
>>109299665Don't know. AI will do this for me.
>>109299475Lol, I guess I'm the only Claude user in here right now, otherwise there would be a bigger shitstorm. They removed Fable from all plans, despite the day not being the 19th yet. People get a "not enough credits" message when trying to use Fable. Well except for the unlucky few who had credits enabled for additional usage beyond their plans, who never received an alert that sub access is switched off and happily continued prompting Fable assuming it's still sub usage. Claude Reddit is an absolute shitshow right now, go check it out. And yes, people assume this bug was caused by them trying to prepare for including Fable in the max plans after the 19th but forgetting to switch automatic credit billing off.
>>109299680lmaosome chink at misanthropic gonna be eatin good tonight
>>109299680I still have it
>>109299704It's fixed now already, you missed it.
>>109299400always has been>>109299470waiting patiently>>109299475It's over>>109299525depends what you mean by free. Opencode has free shit. The $20-30 a month is the sweet spot, but do free shit until you learn the basics. >>109299718Cursor Grok 4.5 is in need of a branding change.
What are the best self-hosted models for vibe coding nowadays?
>>109299755K3, but do you have $80k? Or use hermes powered by something you can actually run.
>>109299771>something you can actually runThe problem is that I don't actually know what are the good self hosted models at all, I've never cared about them before, except for the very first llama that leaked
>>109299680Anthropic hates their users.
>>109299792The gooner chat bot general has more talk about what can actually run on your system. Just look around, it will take a while for you to see what's out there. Plus every couple weeks a new model comes out. DeepSeek, Hy3, lots of chinese stuff.
Has anyone leaked the skills or system prompt for "deep reseach" mode of any of these models or the "swarm" mode? I feel like that shit could be done with a skill or a system prompt but I'm too retarded to craft one myself. They're too useful though
>>109299755>What are the best self-hosted models for vibe coding nowadays?Why do you ask vague questions like that post your hardware first. Nobody can help you without that info.
jeets are cookinghttps://x.com/soikat/status/2078173542496293132K3 game controls and three.js look good here.
>>109299848nobody cares about games made with three.js
>>109299848Nobody makes actual desktop games with java.Post real C example.
>>109299860what does everyone care about?
>>109299869PRAGMATA
>>109299847It's because I don't have the hardware, I need to buy it first, but I assume I need to know how much I need before buying
>>109299848do you know how to make that on screen joystick
only useless botted jeetslop on github trending these days. Grim.
>>109299877If you are serious about LLMs pic. related.I think they are like $100k.If you are not that serious, wait for apple m5 studio ultra with 500gb unified ram, I think it will be like $15k max?
>>109299755>>109299771>>109299877Qwen 3.6 27B and 35B variants.Do NOT buy anything without having used an equivalent setup rented in the cloud for at least month before buying.I'm making a K3 at home distill btw so stay tuned.
>>109299869something fun
Grok 4.5 is performance/price king atm.
How does using agent software like Cline, OpenHands and pi.dev differ from using Claude Code and Codex?Like what's the main reason to use a third party agent/harness over the native ones the LLM providers offer?
>>109300049Well if you use cline/openhands you get API pricing. (bonus you can create multiple agent workflow, assign different models to each one, plus you can automate the workflow better)Claude code, Codex should give you more if you vibecode within one GUI.
The fuckass renderer is my white whale
So /goal just nudges the model to keep going until it becomes stupid at high context instead of restarting the session?What's the fucking point?
>>109300094the illusion of continuous, autonomous improvement.
>>109300094Until it meets the completion criteria, you do specify completion criteria in your goals, right anon?
>>109300094tricking retards into burning their limits so sentient people can use the tool to get real work done
hey where'd the 5 hour limit in codex go?not that im complaining
>>109299950>I'm making a K3 at home distill btw so stay tuned.based. Any local chads are encouraged by me to update frequently.>>109299896I don't see an on screen joystick in the video, but listen for the gamepad input and render buttons and joystick to move.>>109299996As soon as I started using grok, improvements came fast. I can only infer that it was my convos and autistic need to keep context maximally efficient. Your welcome.
what's this i'm hearing about codex deleting root directories?
>>109300142My goals are cuda kernel optimizations for a custom product and it will easily spend 500k tokens for a single commit.I want it to keep restarting and looking for optimizations until it fails to find anything some number of times in a row.I guess I could have the goal be to spawn subagents and review their work until no new optimizations are found.
Just make sure your clanker is working when the limit runs out, he will just keep working
>>109300169Then put that in the goal"continue until X attempts at optimizations fail to produce gains"
>>109300157openai removed it in response to anthropic extending fable usage
>>109300094you ain't gonna be reaching any high context on codex, son.but yes /goal is mostly a gimmick unless you have tons of easy work, just prompt the model manually.>>109300169with modern models there is no difference between high or low context or between compacting or starting from scratch, that's just old 2024 memes.
why the heck is fable now eating up my session limit much faster than before?
>>109300170some thread ago it was claimed this no longer works
>>109300187i see. I guess competition really does workhopefully gemini steps up and starts putting even more pressuregoogle can do a lot more undercutting than openai or anthropic so they can put a lot more pressure on both than openai and anthropic can put on each other
>>109300187they removed it because they were getting killed by people running sol for hours for free well after their 5 hour usage was consumed. now instead of that it will consume your whole weekly usage in one go. and like the stupid goyim you are you fell for it and thought they did it for your own benefit.
>>109300208>and like the stupid goyim you are you are DEEPLY upset and it will not stop china from losing
>>109300198correct. see >>109300208
>>109300188>no difference between high or low contextNot true in my experience with models as recent as fable.They won't fail catastrophically and can do mechanical work just fine but they do become retarded.
>>109300217why do you love the taste of leather so much?
>>109300229i don't. you're the one in /aicg/ and /vcg/ crawling on his knees for daddy xi day in and day out. im just enjoying the models. you seem bothered that it's not kimi
Hydrus frustrated me enough that I started remaking it with claude code, and amazingly enough the model is handling it well. The thing that worries me a bit is I don't know if they'll be okay with me basically taking all of their tags and using them. Or if anyone even cares to use it in the first place.
>>109300226well if you really believe that you can set up your own loop script. ask your clanker to research how /goal works (it's a tool the model calls) and replicate it but starting a new session from scratch when the old one finishes.
>>109300241check out panoptikonThough I had to make some adjustments to that as well. Most notably making it work on amdIssue with panoptikon is i dont really use it
>>109300049custom harness, custom resultsyou can't test this at all in claude code or codex even if you wanted tohttps://stencil.so/blog/prewalk
>>109300236do you enjoy being financially dominated?
>>109300255I've been fucking around with autotaggers as well but they still seem to fuck up a lot. But I'm still looking to use them as plugins when you can't find an md5 on the boorus.Also what made me start doing this is that Hydrus seems to have the most unfriendly UX I've ever seen, and I've tried to use it three times in the past decade
>>109300094I have actually had /goal work once before purely because the model was insistent that nothing could be done until it was prompted 5 times in a row, at which point it suddenly became sentient again.
>>109300292why are you building it from scratch? you know hydrus is made by a guy from here/8 right? Its also opensourceFork it and attach your own UI/UX while stealing the entire backend, or submit a PR to fix his ugly bum ass UXOr go yell at him in his threads on 8whatever
>>109300362Well another thing is I don't know if there's any fixing the PTR model. As far as I could tell you had to download the entire database if you wanted to download any tag. That's about a 140gb if you want to get some files. The reason cited is privacy, because if you download everything, nobody can see what file you're looking for.But there's a middle way, you download variable sized hash prefix buckets. You don't grab one tag or all of them, you request a prefix and get thousands of hashes or more, and your file's somewhere in there. Or you just throw a vpn on it and grab whatever.
>>109300397this would still let someone look at all the buckets you download and build a commonality between theni.e. there is a statistically higher amount of image hashes with tags for lolicon in the buckets you downloador whatever I never bother with the tag data base because I saw the 140gb part and just decided to do my own tagging
>>109299771>$80kwith 80k, you can run it at Q0.5 maybe lol
>>109298107>## 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-codeOK, wait, what?. Zai and moonshotAI have cheaper model at better quality. Why send newbies to OpenAI and Anthropic which is only making money of corps?This is the opposite of what we should be doing.
>>109300412Theoretically yeah, a malicious repo could do it, but then you could either grab everything or use a VPN. Hell, you could download the db and start your own separate repo. Either way if someone had it out for you that bad, you'd have bigger problems.
>>109300416I mean ... Unless you use the free versions. Antigravity and Codex give you some vibe juice for free IIRC
>>109300255>panoptikonI need to give this a go. My images have been unsorted for over a decade, and I have at least hundreds of thousands that need tagging.At some point about 10 years ago I was making some percepttual hashing tool for it, but that's as far as I got.
>>109300416Feel free to snipe the next bake with your own personal agenda, it's better than letting the "zomg luddites" guy bake his eyesear.
>>109300416>only making money of corps?sounds like you don’t have $20 or so (not that there’s anything wrong with that) and are making up copes
>>109300438it does sound like that if you don't have a clue
>>109300416>we
How do I start coding?>inb4 Read OPAlready read it and I'm not a computer engineer, not to extrapolate about myself but my previous job was basically blowing shit up for money and except using cheatengine, word and following instructions to install CFW on consoles, I don’t know shit about coding nor I don’t know which language should I use (for what I know for know, there is two dominant language: Python and Java)I'm not aiming for employment in the domain since I don’t even know how to fake a resume without looking like a fraud (even if I think troubleshooting stupid crap is more cosy and less dangerous than my previous job) Can someone help me? AI give me bullshit answers with no real start point.
>>109300497install opencode, open it and type what you want to buildplease make a backup of your machine because at some point it will delete your data (im not joking)
>>109300497vibecoding isn't coding you literally do not need to know a single thing. You just tell the ai what you want and it builds it for you. If you want to discuss actual coding in any capacity instead go to /dpt/What do you want to use it for in the first place?
>>109300436It's pretty good. The tagging is nice but a lot of my value for it comes from the CLIP semantic search. I'm planning to just pay for the saucenao api for a month so i can properly tag all the stuff that saucenao can do. But first I need to properly seperate out the stuff that saucenao will reasonably be able to find rather than send it a bunch of random memes.
>>109300497just get chatgpt $20 plan and talk to it, it's not harder than that
>>109300497>>109300517>What do you want to use it for in the first place?This is the important part.
>>109300497What are you wanting to do?I don't know how to code, I wanted a mod for a game, I found links to the games modding api, I grabbed a copy of the game, I put them both in the same folder. I told codex "I want a mod that does X" and whenever it asked me stuff I gave it answers and just kept chipping away until it was done.I use codex cli because I'm on linux but you would probably want to use their app or whatever it is on windows
>>109300497step 1 is to get in the habit of asking your clanker stuffget it focused in a directory dedicated to a Hello World app written in Python and then tell it to write Hello World in Pythonand then tell it to set up a Git repository in that directory and then explain to you why Git is so great
>>109300541any benefit to using the app on windows? I guess you may not know but does anyone else?I have just been using cli on windows.
K3 opencode boost
>>109300585use the app for…other stuffI’ve been skimming https://daringfireball.net/ and apparently the app sucks sweaty donkey ballsI just use the website and the cli for both OpenAI’s and Anthropic’s stuff
>>109300534>>109300595$5 for the first month gets you new kimi
>>109300509Noted, thank you>>109300517>>109300537I don’t really know, trying to learn more and maybe do an app specifically for normies retards and make some sweet money out of it. I don’t think vibecoding can get me to the point of getting free ThinkPads from a retard company but if I can learn stuff on the way, I'll be a victory.>>109300534No Sam, you have already enough money!>>109300541I have a zombie laptop running Debian, so even doing a mod with vibecode is possible? Looks like I have a lot to learn and unlearn
It is just me or is gpt 5.6 sol crap?It is just so damn slow and over engineering, even on medium thinking.I am seriously thinking going back to 5.5. Features that used to take 20-40 minutes take 2 hours and give me 60 files and a bunch of unrequested changes, 5.5 never did that.
>>109300619I'm starting to feel chink temptation but don't want my credit card stolen...
>>109300619well, why not terra? terra is cheaper for the same amount of intelligence.
>>109300638K3 Is much slower so if your issue is speed then you're SOL
Kimi is smart and considerate, but also slow and slurps usage, it's $19 plan is draining more usage monthly than chatgpt $20 weekly..
>>109300658sol is too expensive
>>109300663does it over think or overengineer?
>letting a female AI code for youHAH!
>>109300647Because if it is the same family as sol I worry it will make he same overengineered decisions but dumber lol
>>109300584Thanks, I'll start by getting a clanker less retarded and gay than gemini then.>>109300608What is kimi? The clanker bastard offspring of Räikkönen?
>>109300679yeah. im a gemini simp extraordinaire, but yes, gemini is genuinely dogshit for coding.
claude refunds going out, check your stats
>>109300614>No Sam, you have already enough money!It honestly is worth the $20 but if you dont do computer stuff at all and just want to dip your toes than put $5 on openrouter and use deepseek or kimi or soemthingAsk any one of the free models gemini grok gpt how to set it up and youll be on your way
OpenAI will get raped in courts by Apple.They will cut your goyem resets.
>>109300781Apple big mad. I hope they lose.
>>109300781gaben just beat the jews thoalso openai have their own jew at the topthis is just gonna be another wet fart
>>109300781Apple is an AI loser. These fucking retards spent so much time and effort trying to make the next big thing (a fucking iCar lmao) that they missed the actual next big thing.
>>109300663Don't pay for kimi 3.0Wait until Cursor steals it and creates cheap composer 3.5I thnk composer 2.5 was made from kimi 2.6 (not even 2.7)>>109300788They have logs of people stealing, openai will pay a lot or maybe individuals. Depends if openai called for their employes to steal apples IP.
>>109300668I don't think so, it's snappy for small things and when I had it look over my whole project it took a long time and it it had it's companion nadira pop up to tell me it was workng and fooling around.. but I liked the result. My feeling the wait times are more about their load. But it's not cheap, not on my plan anyway
>>109300799>Apple is an AI loser. It's trillion dollar company that actually has profit margins unlike LLM companies, kek.They are using QWEN for their Chinese market, they are smart for not wasting money on LLM model development.
>>109300815and they use gemini for the united states market
>>109300823It's based strategy, they can always just pay the new frontier/optimal LLM companies.They just have to develop fluid system where you can hot-swap model into it. (easy with swift ui/swift)
>>109300815>muh shekelsThat's the thing, they didn't need more cash, but they wanted to do more stock buybacks because the simple-minded butt pirate of the Silicon Valley Tim Cook learnt about buybacks and figured it's the one thing to do at all times. They could have used their cash to get ahead, instead they have nothing except a laughable Siri that they replaced with a Google service that all Androidfags already have access to. It's not smart, they lost, they have nothing.
>>109299680I use Fable low for everything because any lower ant model seems to be too retarded for what I ask and any higher ant model goes completely batshit crazy with testing. I haven’t had any issues so far, but I only ever use the API, no subscriptions.
Should I get a free Claude account for someone I know that is a teacher? or you think waste of time?
>>109300255Panopticon has a fucking terrible user interface and all the taggers actually kinda suck
>>109300901get it. and then be an advisor to them and help them get acquainted with claude so they don't just drop it
>>109300909yeah I thnk I'll do that. It's funny though how they market it as keeping in line with US teaching standards.... lmaoI do have a feeling they will use whatever data they glean from it as a way to get into the teaching industry, to the point where your children just have a bot that teaches them, not a meat bag with a bad attitude.
>>109300871>That's the thing, they didn't need more cashYes, but it's free money, because those retarded openai empoyes had 0 opsec kek.>It's not smart, they lost, they have nothing.They have apple silicon, which is good chip, they can use whatever models. I think it's mistake for using google models too, but they will switch eventually.
>>109300942almost guaranteed to be better than what we have now but that's probably by design
>>109300955I was talking about AI and how they don't have an edge nor a unique product, not chips. It doesn't matter what they switch to, it will be something anyone else already has access to. They're the biggest company by market cap, they had the resources to dominate, instead they did nothing.
>>109300894Jokes on you. People who got billed for credits usage got the bill voided AND additionally credited whatever they were billed incorrectly. God I wish that were me...
>>109300781More actions from mister Tim Apple to appease TrumpTrump has sent out the word: Total OpenAI Death
>>109300990Didn't one guy get billed like 10m
>>109300978>it will be something anyone else already has access to.It will do more, because their chips are more powerful and don't overheat. (unlike frontier android chips)Desktop line-up m6,m7,m8 (all the variants regular, pro and max which you can run a lot of models due to unified ram and memory bandwith)Development of their OS across multiple platforms (I like their new feature in golden gate, you can vibe code safari ext. directly in safari UI you don't need xcode or anything)Working on Swift coding lang. (kinda meh they are bloating it)New phone formats, one with glassy sides in like year from now.They are doing 100% more than microsoft and android combined.
Working on my prompt for a pooh shiesty themed arcade shooter. Top down first and then going to do a doom wad and duke3d after.
>>109301032that nigga poo shiesty in jail for upping the drac on gucci mane and forcing him to release him from his music contract. no joke on diddy, look it up.
>>109300781If someone looking to sue me sent me a letter telling me to save potentially relevant documents and communications, Id be firing up shredders at a rate that hadn't been seen since Enron.
>The journal was read as ANSI (PS 5.1 default) which mangled the UTF-8 em-dashes. Rewriting with explicit UTF-8 read.NOOOO CLAUDE, NOT THE HECKING EM-DASHES.
>>109301030We're talking about access to frontier models (like Gemini) that can't run on consumer hardware and don't benefit from higher specs.
>>109301001why do you keep saying this? if anything its anthropic that is targeted
>>109301137this. and remember: trump wants the us government to have a stake in openai
>>109300275This looks cool, does it have to be used at the start of a project though?
>>109298287>and stole my girlfriend.Sir you can't just drop that information here and expect people to not want to hear more. What happened?
do you use PRs
>>109298343i would rather drag my dick through an alley full of broken glass than use 5.5 after using 5.6.
>>109298527Why would you want less competition in this space? Does anybody here actually like Sam or Dario and trust either of these reptiles to not fuck us over? Why are you white knighting for the two jewiest jews in the valley?
>>109301155no, it's just a prompting technique.the idea is that at the start of the trajectory, all file reads are done by a cheap model. you then seamlessly switch to the god-tier model to perform planning and to write the first lines of code in their relevant modules. then you switch back to the cheap model.the hypothesis is that the cheap model (e.g. Opus or Sonnet) should be able to execute a god-tier plan (e.g. from Fable) with minimal degradation in performance.Because the problem with coding isn't the actual writing of the code, but going down dumbass rabbit holes and making shitty architectural decisions.If you think about it, it's not that new of an idea. There are custom harnesses around that do something like this, where planning is a concrete step, but may not be exactly the same in some important ways.
>>109301325better a jew than a winnie the pooh mother fuckass
post ur pets anons
>>109301341You don't have to choose, you can have both. Having both is better because they keep each other in check. Chinese open weight models keep the American frontier labs from jacking up their prices or slowing down development, and American frontier labs keep the Chinese from closed sourcing their models. If the Chinese had genuinely better models they wouldn't have to make them open weight because people would bite the pillow and give their data to China in exchange for using the best models. But they don't, so China knows they cannot demand the world's trust.
the wheels are starting to fall off of the sol train for me. sol medium is doing things, but also leaving bugs behind that remind me of 5.5. and then kimi is coming in and fixing them
>my GPT + Claude orchestration setup is finally ready>already almost out of weekly gpt limitstibosaar pls gib reset saar BLOODY BENCHOD SAAR GIB RESET NOW
>>109300901yesif my mom were still teaching I’d have her sign up for this in a heartbeat
>>109301388or the way the chinese do business is living on single digit margins and forcing competitors into bankruptcy
Tibo posting about resets again
>Selected model is at capacity. Please try a different model.wtf
>havent had to use a single reset in months>run into an especially hard/tedious problem>massively parallelize 5.6 sol agents and solve the problem by burying it under a mountain of tokensI'm glad I banked all of those resets.
>>109301461sometimes all you need is just a lot more dakka
>>109301452oops sorry
>>109301461man is deploying the air force, coast guard, and delta force to get a cat out of the tree
>>109301448if you think that OpenAI or Anthropic are going bankrupt because of chinese models you are a gay retard.
>>109301481And I'll deploy the spaceforce too if I have to. To be fair it was a really big tree.
>>109301451He better not, I just used a banked reset last night
>>109300901Yes, and then if they don't like it have them hand the account over to you.
>>109300815You know that each model released by OAI and Ant has been individually profitable right? They just keep doubling their R&D spend.
>>109298107What is the best harness between Pi, Cline, OpenHands, Cursor, Osaurus, Kilo, Zoo?
>15 sub agents to hatch a single pet
>>109301578Cursor if you plan to subscribe and use grok 4.5 with best rate.Cline/OpenHands if you are going to use multiple agents at once, or loop system and API.Rest idk.
>>109301451I just ate my plus quota again, need another reset
I implemented some pretty cool features today with sol ultra and now I'm waiting for a reset which I know is coming. I think kimi is just a distilled model so I'm skeptical if it's actually good. Just because benchmarks are good doesn't mean it's actually good. It's also MUCH MORE EXPENSIVE than copex
>>109301656Only people using cursor these days are racists because it's not actually good. You'd have to be politically motivated and a strong ally of apartheid south Africa and Israel both.
>>109301678kimi is a distilled model, but the model is genuinely good. i can attest to that as someone who used it. i'm not in "replace codex with kimi" territory as gpt has the more complete family, but kimi is about on par with sol.
How does OMP know to spawn subagents on Sol Max? Is it Sol doing it instead?
>>109301685Grok 4.5 is best performance/price model atm.Fuck off with your brown political babble.
>>109301693Sol is doing it. The ability to spawn subagents is provided by a tool (or similar) that's part of the harness, OMP. When you use OMP, it's sending information about available tools to whatever model you're working with. Sol very well understand the concept of subagents, so presented with a subagent tool it's easily able to take advantage of it.
>bank reset expires in 10 minsWow this really eats up quota far faster than i expected, I was using high all this time and switched to xhigh and fast like 7 or 8 mins ago and it went down from 33 to 20 i am also running three tasks at once too
I've used easily like 10 mil tokens with hy3 free model, if alibaba releases good open source model say thanks, I'll keep feeding it coding data.
>>109301777isnt deepseekv4 flash better?
how is it still working
>>109301784hy3 is free on operouter api.
>>109301784hy3 is closer to DS4 Pro than Flash in my opinion but they're definitely in a similar ballpark
>>109301698>Grok 4.5 is best performance/price model atm.Do you have the data to back that up?
>>109301809nta but https://artificialanalysis.ai/agents/coding-agents?agents=claude-code-glm-5-2%2Cgrok-build-grok-4-5-high%2Ckimi-code-cli-kimi-k3%2Ccodex-gpt-5-6-luna-max%2Ccodex-gpt-5-6-luna-high%2Ccodex-gpt-5-6-luna-low%2Ccodex-gpt-5-6-luna-medium%2Ccodex-gpt-5-6-luna-none%2Ccodex-gpt-5-6-luna-xhigh%2Ccodex-gpt-5-6-terra-max%2Ccodex-gpt-5-6-terra-high%2Ccodex-gpt-5-6-terra-low%2Ccodex-gpt-5-6-terra-medium%2Ccodex-gpt-5-6-terra-none%2Ccodex-gpt-5-6-terra-xhigh%2Ccodex-gpt-5-6-sol-max%2Ccodex-gpt-5-6-sol-high%2Ccodex-gpt-5-6-sol-low%2Ccodex-gpt-5-6-sol-medium%2Ccodex-gpt-5-6-sol-none%2Ccodex-gpt-5-6-sol-xhigh%2Ccodex-gpt-5-5-xhigh%2Ccodex-gpt-5-5-medium%2Cclaude-code-fable-5-max-with-fallback%2Cclaude-code-opus-4-8-max%2Cclaude-code-opus-4-8-mediumAA just updated their code indexing page, and this page seems to bear that out
>>109301809
>>109301727Oh I see, it just knows when somehow
>>109301817>>109301833Why are the x ai subscription plans so shit though?
>>109301867emong mucks it up.i don't even dislike elon, but i would never subscribe to grok with how opaque everything is. it's not even certain if you get the full 1 mil context with your $30 subscription
>>109301809and does it have the image model
>>109301870i dont trust elon he's drug addled shitposter and im suppose to take his companies seriously? the fact he hasn't been kicked out tells me his whole enterprises are all personality cults larping as corporations.
in price/performance and overall code efficiency, kimi loses to both grok 4.5 and terra max https://x.com/ArtificialAnlys/status/2078230240766345330
>>109301881SpaceX just lost a trillion dollars. He's a fucking retard
>>109301867Cursor has better usage, plus cursor client is better than supergrok cli.
>>109301897SpaceX stock has good potential.Providing internet for the entire earth + datacenters in orbit powered by solar (hopefully not a faustian meme)
>>109301904>datacenters in space where solar flares, cosmic radiation and no one can service is such a good ideathis is how i know you don't belong here go fuck off to plebbit
>>109301785Codex do be like that, Mr. Stancil. It’ll just keep going until the turn is done and not halt in the middle of a job
>>109301817I thought GLM5.2 was the champion chinese model? But Kimi destroys it?
>>109301940glm 5.2 was a L O T further behind than it's greatest cheerleaders gave credit. kimi is the first model to really be in the middle of the frontier pack.
>>109301940K3 is far better, but much slower and far more expensive. Which is fair, it's in a different league.
>>109301940oh my god FOR THE MILLIONTH FUCKING TIME EVERY CHINK MODEL HAS BEEN CHEATING ON BENCHMARKS THEY AREN'T REAL RESULTSkimi is the first time actual results from chink LLMs are comparable to frontier american ones.
For all the newfags wait until cursor creates composite 3.5 from kimi 3.0 (it will be cheaper and better)They made composite 2.5 from kimi 2.6.
Fable is pretty solid. Too bad there's nowhere to actually advertise this thing besides the search engines.>https://anaximander.us
tibo please god
>>109302006ew. my rep
is this a schizo? >>109301964
>>109302015let the man cook
>>109301993when will the china man go properitary
>>109302006>https://anaximander.usinteresting>outcome went against their voteif you can scrape the representative’s sex and use “his” or “her” for stuff like this that would make this much, much clearer since there’s also a bunch of people in the relevant governing body
>>109302063Nice try, pronouns.
>>109302025i don't have time aaaaaaaaaa
>>109302078Which plans are you on?
>>109302086just regular pro
Excuse me sir, but what do you guys think of MiniMax?
>>109300901>>109301444My mom, my sister-in-law and my imaginary gf now hab Claude. It's just begun.>>109301913>plebbitwait, reddit LIKES Elon now? I can't keep up. We need a new bake, please pick a good image.
>>109301461>>109300901>>109299931>>109298640my nominees for next OP pic. do what you will.
>>109302153I baked it >>109302169
>>109302153Pwease why not my map pwoject
>>109302186Because promoting your shit for free is banned on this site desu.Just post it as first post.
>>109302184thank>>109302186I didnt know it was yours, very cool. Why anaximander? was he the one who said everything is flux?
>>109301698>Grok 4.5 is best performance/price model atm.this is not fucking true
>>109302236Muskrat and his jeet army wouldn't be able to top gpt 4 let alone the current sota
Is there a way to reduce token usage for Sol? It's clearly eating way too much when running commands
