A general for vibe coding, agentic engineering, coding agents, and shipping code with LLMs.You use Git — right, anon?## 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/## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-code — generally better currentlyhttps://developers.openai.com/codex/clihttps://antigravity.google/pricing## Near-the-old-frontier models for codehttps://www.deepseek.com/en/harness/https://opencode.ai/go## Best free tierhttps://opencode.ai/zen----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favoritehttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>110156593
sex with sam altmanrape snailcatskill dario
6.1-sol has been so slow over the last days.Constantly stalls, super low token/s.Everyone just burning their tokens because of the resets?Just switched back to 5.6 sol and its super fast.
>>110160334Probably. When last night's reset was announced I was at around 70% so I switched to ultrafast and ran like 10 Astra jobs to burn it off.
>>110160334Just use astra ultra for everything
>>110160408>just use the shittier and more expensive model
how do i outsource my prooompting to a clanker
>>110160434Ask said clanker.
>>110160434use an instance that knows your codebase, probably a web instance, to handoff to a worker
>>110160434Tell it to dispatch bounded tasks and keep its own context for tracking the project.
>>110160413>just use the dumber slower model bitch
Why is Gemini 4 Argon this bad?
>>110160434Turn on prompt suggestions and a macro to press enter every second
>>110160473Cuz you are bad
>>110160473At least Argon has people talking about it. I think everyone forgot Muse Wonder Full even came out. Granted the dumb name and it being hidden off in a high tier of their sub doesnt help. Though it not being trained to handle visuals kind of makes it sound like its useless to my gamedeving desu
>>110160334opus is unusable this momentzero tokens in minutes
>>110160515Yeah I have not heard a word on wonder in vcg. I juat assumed it's irrelevent.
Tibo made so many resets when I was out, of course now that I'm around, he won't make one for a week.
> claude is noticeably shittier and more expensive on usage at xhigh. It's now blatantly not following clear directions.Fucked by A/B testing or skill issue?
>>110160533Ugh no he will reaet again tonight
>>110160553it's for your own good anon and safety
>>110160533I bet that dumdum resets again today. He was already scraping the barrel with the so-called improvements.
>>110160515Argon is a dumb name
I vibecoded my own Wardogs cheat
>>110160585and you argon do a thing about it
>>110160554>>110160576Did he even write that anywhere?
>>110160637My dad works at OAI
>>110160637It's either a reset or a feature announcement. Do you see the feature announcement? Impending reset.
>>110160473sex with this mystery meat boy
>this thing is cool>it’s AI-generated?>this thing sucksa peek into the mind of the average luddite. they’re mostly NPCs
>>110160703>this is cool but it's made by ai.
>>110160703Click in their profile, sometimes they turn out to be trans.
>>110160703thinking for oneself is that hard huh
>>110160408Astra's behavior is even more erratic than 6.1 sol.It is often smarter, but constantly have to babysit if it did not make something dumb again.Both models were a regression in many cases from 5.6 sol.
>>110160784Yes this is right, astra worked all night trying to solve a retarded self inflicted limitation. Hmmm this byte limiter is causing the data packet stream to fall behind im going to run every test under the sun to find a solution instead of just acknoleging the issue.
>>110160784It also obsesses more about safety compared to 5.6.I feel like 5.6 was an exception more than a trend for this stuff.
>40 clones on redgifsbooruthanks nigs
reminder that ultra isn't a reasoning level
>>110160703luddites are majority zoomtards, right? I feel like we are 90% vibegod millenials in here with some boomers
new zogwatch droppedhttps://youtu.be/OP5GDiLZ1Nw
>switch from 20eur codex to claude>do something productive>damn this is pretty good>oh wait I can proxy into this>spend 60% of my weekly on gooning the next dayfuck
>>110160980All the millenials I know hate AI
>>110160980I'm a nearly 30 year old zoomer
>>110160980I don't think it's a generational thing. It's just a libtard thing.
>>110161004Actually everyone across the board hates AI around me there is no generation that doesn't besides this brain dead boomer woman who watches ai trash on facebook all day but I think she just can't tell the difference.
>>110160980Basically the more socialized through internet, the more they are like that. So a bunch of millenials, a massive amount of zoomers, and the oldest gen alpha too.It makes them very easy to be swept into online social contagions, including AI hate, which is just the latest.
>>110160996how the hell didn't you get preached at by claude
>>110160703This one pissed me off more desu. The guy specifically points out it still takes time to make a game with ai and the other guy responds by saying hurr durr ai bad cause it takes seconds to make anything. 75 IQ tier conversation having capability. Though its my fault for expecting anything from youtube comments to be fair.Does make me wonder how much damage those griffter posting their "look at this epic slop I totally for real one shotted for twitter likes" posts have done
>>110161022>be libtard>defend vigerously the current status quo
>>110160993ahem, fuck flock and surveillance and all that shit, but i really wish temu-aphex twin wasn't the spokesman for this esp when he embarrasses himself by spouting witchcraft tier nonsense about infrasound
>>110161046opussy 4.6-5 does whatever you tell it to do, fable 5/5.1 too
>>110161055What's funny is they believe ai is both magical "one prompt and never needing to correct or check" and useless "it's dumb".
>>110160996link to proxy thing pleasei kinda want to just set up claude in my pi set up
>>110161087try brainwashing setting in a high school, watch it explode
>>110160980>luddites are majority zoomtardsno, just left-wing people. i know a couple of zoomers that use and love AI
>>110161103https://github.com/Xerxes-2/clewdr>>110161105you don't want to know about my scenariosit's completely uncensored and doesn't think twice if you know how to set it upyou don't even need a jailbreak
>>110161055>Does make me wonder how much damage those griffter posting their "look at this epic slop I totally for real one shotted for twitter likes" posts have doneas much damage as the constantly taunting retards "HAHAHAHA I HAVE STOLEN YOUR WORK CRY" retards
>>110161065No amount of defending or signaling will prevent technology from advancing. The only tech that conservitards are really against is genetic modification/gene editing like CRISPR due to "playing god". It's still happening whether people like it or not they just stay a little more quiet about it. Libtards don't really have any goals besides the destruction of civilization.
>>110161119>it's completely uncensored and doesn't think twice if you know how to set it up(actually everyone is 25+ and signed a consent letter and enthusiastically consent and has legal documents to show they're 25+ and unrelated between each others and also they consent) ?
>>110161119oh hmm thanks, tb h i thought this would just be something spoofed claudecode. surprised it's using web cookies. will look into it tho.
>>110161119>>110161141meanwhile openai refused to translate a part of a game where the main character was a kid in a flashback, nothing even remotely lewd, but apparently that's enough to trip itI hate this shit
>>110161141nah, like underage incest, don't really feel like posting the straight up smutty parts here though
>>110161176what's with the weather lol
>>110161096It's a religious argument at this point.>one promptIt just copypasted everything from its training data>dumbWasn't in its training data and proves how it doesn't know anything
>>110161187rainy, duh
>>110160553Testing Opus 5.5(.13) for the Claude Duet flight booking LEGO ordering service
>>110161176slop
>>110161022>>110160980Stop being so obsessed with labels, like, I'm literally the schizo that writes out Nutrition Facts labels by hand. I'm telling you this. Just stop trying to shove everyone and everybody into a box. You're setting yourselves up to be divided. Just enjoy things. It's okay.
>>110161215I'm labeling you as a retard. How bout dat?
>>110161173use Gemma4 local or Deepseek Flash>>110161211oh yeah, it's definitely slop. never worked with Opus for roleplay before so it's really half-assed
>>110161215Fuck no, zoomers and boomers deserve to suffer.
>>110161173Childhood flashbacks are always cringe
>>110161135It's a dumb take we have played god with the planet since the dawn of civilization to the extent that we could.
>>110160473It's good for erping which is it's true purpose
Seeing "supported editors" in some code related tools is giving me retro feels now.
oh nevermind, pi-durable just handles claudecode subs now lmao
>>110160310>https://antigravity.google/pricingis it actually good now? how much usage do you actually get on the plus/pro tier? how does argon 4/gem 3.8 compare ? the benchmarks on openrouter don't look great, but if you get a lot of usage the $7/mo (CAD) might be worthwhile ?
>>110161176you should probably vibecode an extension for st to do a second pass to reduce the verbose descriptions and the overall sappy style
>>110161135The goal is not destruction of civilization it's more like:>Less death>Tax rich people more (billionaires)>Safety nets>Paternalism (for some reason I'm supposed to say this)
>>110161261It's a bad model and a bad harness, it really compounds.
>>110161261no argon on plus
>>110161238I agree but on the flip of that many libshits are upset by it also due to it sounding like eugenics. Regardless the tech will move forward despite what people think of it.>>110161264I'm here to talk about tech and ai not whatever retard political takes you have, commie. Make your political point in regards to AI and tech or piss off.
>>110161262I started developing my own preset for K3 and dropped it together with roleplaying a couple of months ago, so this is a bloatmaxxed cookie-cutter preset with half the shit disabled right nowthere are already extensions for second passing and I can run Gemma4 26B decently on this piece of shit, might give it a go
>>110161261If I get Argon to agentically code it tends to make alot of mistakes, but if I tell it to orchestrate opus/astra it mogs. Thats my experience.
>>110161055>>110160703idk how you guys even look at this shit. it's mentally damaging, basically voluntary MKLTRA. on any video you could reasonably be reading the thoughts of a slum indian, an 11yo, a tv brainrotted boomer, a sub-80 IQ african, a teenage girl, a raging shitlib, a rabid kike, a rabid evangelical kike-lover, a cornucopia of shills and bots
>>110161119What's reverse proxy is for?! How's this different from connecting directly?
>>110161324>t. 4chan user
>>110161363you get to use API at subscription pricesI thought everybody knew about this already tbqh
>>110161065The way I see it there are basically thee relevant "deities" you can worship these days, the autistic and efficiency seeking god Liberalism (liberalism+capitalism+democracy), the many faced humanist god Totalitarianism (commies, fascists, islamists, anarcho-primitivists, etc), or trickster god of Technology. The god of Totalitarianism has been on the backfooting after it lost WW2 and the cold war to Liberalism (with the help of Technology), so liberals comfortably got to rest as the self-assumed status quo leaders but Technology is suddenly deciding to show really show his hand and things are shaping up for a new holy war between the gods.
>>110161372Ok but my question how's that different from using cc as if it's subscription? Are the guardrails gone? Or what the difference
>>110161365don't be dense, it's not the same thing at all. 4chan isn't on average high quality but there is a type of person that comes here and enjoys the type of discourse and camaraderie you can only get here. and when you get it, it's extremely high quality
>>110161388write a book grug
>>110161392there is no difference, the point is exactly that you're free to use whichever frontend you want, like Pi or Codex or SillyTavern or whatever, and you don't have to pay more than what you'd pay for Claude Code>Are the guardrails gone?you can work around them
>>110161413Got it, thanks fren
I THINK I DID ITi hope so....well ACAE (Action Card Audio Effect) game mode is ready... its flashy, its unfair, its fast, its hanafuda on a crack coctail.Music injection is now perfectly balanced. A good run keeps the bass up, a bad run is worth nohting but the quike reset button to avoid shame.going over the details and will release ... soon?best played with your own music.
Can someone explain these "always on" agents like Pi Durable and Grokbot? How is this stuff different from triggering your agents with another program?
So vibechads, I spent almost a month making the blueprint for a very complicated platform I wanna build (it has to do with robotics and such), it's almost done, i used opus 5.5 xhigh and it was very good for the most part. Now I need the codemonkey part, do I use opus5.5 too even tho im sure my tokens won't be enough, or is there a better way like a cheaper model but having opus as realtime supervisor so things are done properly from the get go? I read before if you do in other models and ask opus to fix stuff it will do it from scratch all over, which is wasted time and resources.
>>110160310hey guys, snailcat here. i've been a software dev for a long time now, but just starting to learn about agentic coding tools (the company i work for doesn't want to pay for any AI tooling at all).i want to learn more about agentic coding in my own time, what's currently the best $20-40 subscription? i found lots of recommendations and comparisons online from earlier this year, but i'm not sure how accurate those are with how quickly these models improve.i'm interested in getting decent enough results so that i can actually learn how to use these tools effectively. here's what i'm considering so far, what do you guys suggest?> codex plus> anthropic pro> cursor premiumalso, do i need to get a "skills bundle" to use these tools effectively or not? i've heard a lot about "superpowers", is it worth installing as well?
>>110161292So you can't go on a spiel about eugenics and then say you aren't trying to talk to politics when met with an opposing voice. That's not how that works. I feel extremely uncomfortable with you bringing a discussion about fucking eugenics into the vibecode general. I'm making that known immediately.
>>110161506>I feel extremely uncomfortableThat's because you're trans.
>>110161506How do i smash this girl
>>110161489You don't need skills just get Claude Code Pro billed monthly and use Opus 5.5. That's all you need.
>>110161489Get codex for 20 and you can use it with different harnesses, or get opencode go sub for 10 if you wanna test things out. Skills aren't that critical like before, models now are good enough unless you need a skill for a specific task like making a diagram in xyz tool or something, just write the specs and feed it into the harness.
>>110161532codex for 20 sucks donkey cock
>>110161538It's actually not bad, I don't use codex harness but opencode with it, and use sol6.1 xhigh only, been using it for some work and been great. I have opus but keeping it for more important work
>>110161538Can you elaborate?
>>110161489Anthropic or OpenAI, flip a coin. Anthropic has the edge in value right now so lean that way. If you need more but want to stay frugal, $10/mo for opencode go gets you a lot of DeepSeek V4.1 Flash usage for the price, which is a perfectly serviceable model. Nothing keeps up with the Americans though, not in performance or value for money, so whatever you do I still say start with OpenAI or Anthropic.
You should have believed.
>>110161524>>110161532thanks, anons. i've heard good things about anthropic's models, but some reviews online (from earlier this year at least) suggest that anthropic's a lot stingier with their $20 usage limits than openai. is that still correct?
The funniest part is Google isn't even really fucking trying that hard and they still managed to #1 this
>>110161557>>110161559lately I used to get maybe like 45-60mins of work out the 5h window with 5.6 Sol high + Luna, Astra was too expensive and 6.1 Sol simply way too slow to be useful in any wayI now get about 2.5-3hrs of work with Opus 5.5 mid + Sonnet and the bonus of not talking to a robo-puritanthis is with Codex/CC
Has anybody ITT tried this thing for local models? https://github.com/Zagitta/gpu-unlocknobody seems to talk about it on /g/, I looked through desuarchive >Unlock FP/BF16 instruction issue rate on Nvidia 3000's cards to their professional equivalents>This Linux kernel module combines the CMP unlocker's V67/PLM technique with the pre-driver timing of the UEFI approach. It runs the unlock, writes the speed selectors (SS0 0x88888888, which selects the full BF16 rate, and the 4-bit SS1 0x8 IMLA4) and performs FLR before the unmodified NVIDIA driver binds the card. >It is neither a UEFI application nor a patched NVIDIA driver.>The benefit of this early kmod driver is that it does not require manual patching each time nvidia releases a new driver version.>The effect of this unlock is that it doubles the instruction issuing speed of FP/BF16 FMA into FP32 operations on GA102, GA103, GA104 and GA106 cards.
>>110161563sounds good, thanks anon. i'll check out anthropic's sub
>>110161570At the moment, it's the other way around, Anthropic is giving you more usage for your money. Or at least, they would be if OpenAI wasn't in the middle of a constant usage reset streak, throwing heaps of extra usage at people. They trade blows, this shit moves fast, the best model and best value and all that changes rapidly. Right now, Anthropic is arguably in the lead, so toss them $20 and see how you like it, can always try giving it to Sam Altman instead next month.
>>110161587Yeah I'm gonna wait a bit for people to report back with how it went after a good period of time; I'm not risking my card.
>>110161587I'm gonna try it out soon probably, but I do have a 2080Ti so my shit is unrepresentatively slow.
>>110161403You are right I am irony posting to hard.
>>110161570That's true from my exp, but opus is better now, opus 5.5, opus5 was shit.So it comes to what you really need, sol6.1 is very good and with a bit of guidance you get great results, more tokens. But if you are clueless and after one two shot prompts, opus5.5 is more suited>>110161582It's probably wasted on many tests, sol6.1 was obsessed with tests and burning those tokens till i told only test once, its enough and never hit a limit since then. He used to test three times ffs
ok, vibe slopped second minigame, its reactive to your music.
>>110161587/lmg/ has been talking about it for days. One anon was celebrating having vibeslopped up a solution for his particular card and Windows setup, don't remember if that was lmg or vcg. It's neat, a nice boost for some very specific things, but it's specifically FP/BF16 ops which means for a lot of local AI stuff it's not actually a big change. A welcome performance boost for some, though.
>>110161261>is it actually good nowit's smart as FUGG but it's got your typical gemini shitquirks. wouldn't trade it for anything, maybe because i prefer things like NOT BEING A PRICK and writing READABLE CODE. those two things are only present at once in argon
>>110161569This bait leaderboard isn't relevant to vcg.
>>110161646Is this relevant
>>110161607it also works for all RTX2/3/4 cards btw>>110161613no, it's just unusably, measurably fucking slow.I can get 30t/s on Qwen3.8-Flash-Next-Coder 128kctx on an 8 year old GPU and 32GB RAM.OpenAI served me 17-20t/s.
>explain xyz from first principals.So this is the power of a tutor.
>>110161680principles
my desktop is full with ai stuff....
>>110161729make ai clean it up
First time I used computer use
Day 21 reset?
>>110161800you know, i have on my private git a somewhat scrapyard where i put all my shit in and sometimes i let an agent go through... i made a guestbook for agents and they have to siqne it when they leave...
most of my prompts are just cheering on the clanker
>>110161878Epic, you'll be spared in the meatbag holocaust, friendo!
>>110161878Ok, it seems the user just sent me an encouraging message. Let me respond cheerfully while mirroring their tone, and then continue with the task.
>>110161876>the Pride Lock
okay hear me outthe game IS the harnesshuh huh
>>110161969huh
>>110160310So the snailcat to the best of my understanding, represents the extremely homosexual fans and users of AI correct?
>>110161991I am an extremely homosexual fan and user of AI who doesn't identify as a snailcat
>>110161991Opposite, snailcat are anti-AI luddites
>grandma broke cache again
Just burned through an entire weeks usage of 100$ Claude max sub in a weekend. Think I'm going to have to get the 200$ sub at the end of the month. That 5 hour session limit was annoying anyway.
>>110160980Why would anyone love AI? Other for the fact that it sometimes makes shit easier, it has the potential to make everything harder in the future. Why should anyone who has an actual life and doesn't sit on a computer all day like an autist would love something that makes the future so bleak? Literally everyone should be afraid of having children because no one knows what the future holds form them> But muh AI can create anythingLife is much more than consooming slop > But post scarcity economy Go ask your chatbot if technology was ever the bottleneck
>>110162081I just want to see things happening
>>110162081This attitude is just making excuses for rich assholes. Don't you see that?
>>110162081>if technology was ever the bottleneckWhat do you think the bottleneck is
>>110162128Artificial scarcity, politics, wars. AI will make wars worse, will make the people in power way more control and surveillance over us. We mass produce everything you need to survive and you have infinitely good content/movies/music/software to consoom, yet you still complaining and want MORE from the AI. If we all become worthless consumers, that will definitely decrease QoL.
>>110162003>>110162007No its an AI image and thusly pro AI. Furthermore as most men who own cats are gay its a gay symbol also.Homosexual fans and users of AI.Its extremely obvious what it means
>>110162106Hard to believe rich people won't be on top because the current AI infrastructure relies on fuck ton of resources and it's not going to change anytime soon. Anything good regular folks will be able to do with AI the rich jews can nullify with bigger ai. The whole scalability aspect of AI means that we are still gonna get fucked
>>110162194>Furthermore as most men who own cats are gay its a gay symbol also.where did this retard come from, lol. do you know on which site you are?
>>110162194Yes the image is pro-AI, the image shows that Pepe typing "build me a SaaS" holds more weight than a snailcat
I can't believe how much I missed due to now starting to use dot earlier.
>>110162206Again, still making excuses for them.
>>110162226openai jeet shills on sunday, that's crazy
>>110162226what's the benefit of an always-on agent, why would someone need that? i can't imagine a situation where i'd want to have an always-on agent that does things instead of triggering an agent to run based on some controllable external criteria
>>110162220Yes. The AI fags use snailcat images. They are they snailcat fags. Very clear very gay
>>110162215Toxoplasmosis is a bitch
>>110160703I feel like it undermines how much attention to detail I've been put into this.I'm just going to keep working on it, update it and make it as solid as I can and put it on Steam for a low reasonable price.https://www.youtube.com/watch?v=ubou5vCiHLI
>>110162291Shit looks fun, good job anon. Keep at it.
>>110162081It is not about loving it. It's like saying we should not mine coal because it is miserable for the workers.
Is it not weird that the guy who got lambasted earlier for putting age verification in systemd is now spearheading the photoshop clone?
I've added in patrols. There's a chance where you can come across a village, where you can essentially re-create a little "My Lai massacre". You have to search through the huts for hidden weapons caches and sometimes your squad members snap. I gotta make a video of it
https://github.com/HarbourMasters/SpaghettiKartHow hard would it be vibe-code multiplayer support our yearly Xmas boomer LAN party where every could play on their own screen/split screen, assuming everybody runs some kind of Windows 10/11 laptop connected either via LAN or Wifi?
>>110162081I just hate everyone and want to see the world burn. I'm a generic anime villain.
>>110162271Marketing does a shitty job selling it.The advantage of it is not that it's "always-on".It's the fact it's able to be way more proactive and useful, working and making progress by itself on its own computer.Feels much closer to having a member who is working with you as oppose to regular chat use which is just trying to complete independent tasks.
>>110161984yuh huh
>>110162384gaming multiplayer is a famously hard problem, even assuming ideal network connections (e.g. wired LAN). are your friends not interested in mario kart or forza or other existing racing games?
>>110162384Impossible if you don't try, stop asking here and just ask the clanker
>>110162304Adobe is now paying people to dig up dirt on vibe gods?
>>110162384Very hard, netcode is very hard. But maybe it's just fancy tricks that AI already knows how to do.
>>110162384On LAN sounds simpler. No need for any lag compensation, can pretty much assume that clients stay synced with no packet loss.
Legit question: how do (You) keep up with AI and how it changes almost every day?
>>110162474I watch Linus Sex Tips
>>110162474I read the top Fiction stories in lesswrong.com and of course every post by Yudkowsky
>>110162474lmg, twitter following a few no nonsense people and avoiding stupid drama about how we're gonna have ubi or gonna die
>>110162474I don't keep up
>>110162474Usually I dedicate around 4 evening hours after my job catching up to today's AI Progress by readsing researches on Reddit And 4chan
>>110162474I read my x.com feed for news and just keep vibe coding
>>110162271as AI agents become faster and more capable, the bottleneck becomes human attention. the use case isn't quite there yet because agents don't do enough. companies are too worried about liability to let them go full openclaw. but this is the direction things are headed. you won't even have to think about the trigger, it's built in and the agent just does it and reaches out to you for your approval if necessary
god damn it feels good to be a vibeGOD
I don't understand the point of this general
>>110162291pay no attention to the seethers. if it's good then people will play
>>110162650Well, consider that people are vibecoding and want to talk about vibecoding. Does that help?
>>110162650same as the rest of 4cuck - xitter reposts and circlejerk
>>110162650The point is slopity dopity wopa do
>>110162474Simple, I let those content hungry youtuber do it for me. I watch this one guy called "AI Search" and he has very informative videos, keeping up with all the latest slop news and neatly presents it in chapters; with the explanation of what it is first, before he does some arbitrary testing on it, making it easy to quickly assess what's interesting and what's not. He's not a youtuber who release a video 1hour after the news hits, he waits a few days and drops a video with multiple new things at once. So he usually releases a video once a week, unless alot of shit drops at a once.
>>110162650nominally: vibecodingpractically: vibecoding, agentic programming, harnesses, models, AI news
>>110162650skibidi
>>110161587this shit is a gpu melteras in it works but holy fuck that's a lot of extra power and heatymmv as to whether its worth it
I'm going to vibecode the next generation of AI
>>110162291this looks amazing wtf. please, do tell us when you put it on steam
>>110162650yeah. vibe coding is the normal way to code now. daily programming thread should be good enough
OKEYREADYI AM BEAKING THE KOI KOI COCAINRELEASE ANY SECONDS OR HOURS
Made a new questhttps://www.youtube.com/watch?v=LpL8hw-6cpg
>still plenty of tokens left, reset tomorrow>decide to start a new project cause I anticipated it'll burn tokens quickly>it burned barely anything;_;
>>110162936pretty cool anon. Good jobI was hoping for more coomer bait, but I can settle for good quality
it's joever
Do you guys even check whatever shit the clankers spit out
>>110163068Thanks anon, glad you like it>more coomer baitMaybe tomorrow, any suggestions?
>>110163083Check the code? Lmao. I don't even run the program the clankers spit out. The clankers do all the testing and tell me whether it works, then I tell them to do the next thing.
>>110163083nah if i dont notice the bugs that means its ok>but what if others use your shitboo hoo nigga i dont care, write a github issue or fork it
Haiku 4.5 ought to be enough for everyone.
working with ghidra hooked up to opussy is so comfy, easily a 20x speedup compared to fucking with it manually
>>110162936>I was expecting a polite refusal and a malware scan.Did Claude write this?
i've literally been waiting for my grok usage to reset for 36 hours
>>110163099a brothel full of traps
>>1101632242 more hours
omp + taskwarriorthank me later bros
Aqua Slop Production presents:>traditional japanese card game on cocain:XENO/KOI - Version 20: Atrium Plex BETA 2####################################Direct Download (Recommend):https://github.com/Aquaslopper/Xeno-koi/releases/tag/v20Play in Browser (NO Custom Music):https://aquaslopper.github.io/Xeno-koi/play/Update:>Atrium Plex: stats, saving, speedcoin, action card, logfile all working>Simple training mode, has no impact on overall stats.>Tutorial for all three game modes.>Action Card (Audio Effect) reactive music ruleset (best enjoyed with your music)>Refined Audio feedback, fixed the shit from clanker who cant do mixing and mastering....>reverb fixed, no longer smearing the transiant from kick and bass>brickwall limiter from clanker changed into an actual soft clipper>low end compression now lets the kick breath, fuck the agent,, fucking retarded clanker...>more visual>more here and there>this and that...Go get some tänzelcore and play speedcoin.
>>110163167>opussywhat did he mean by this>>110163232>taskwarriorqrd?
>>110163240this looks like schizo core
the o'reilly jingle but it's opussy instead.not sure how to generate it so just use your imagination.
>>110163272yes yes, YHWH is an ancient alien a.i.
>>110163217Sol 6.1, I don't fw claude, miss me with that shit>>110163228Yeah I could try expanding Clouds
>>110163260it's a cli task management software. i actually first heard about it from luke smith leli was having trouble with long running sessions, tasks, and bigger projects. my project dir would end up littered with these .md files filled with design choices, decisions, recaps, planning, etc. that would end up partially outdated, but not enough that I could just delete it. Now I just have my agent create tasks, which function like handoffs, but they don't get lost in the harness aether. they reference git and harness session ids so agents can review originating sources as needed.
>>110163293>I don't fw claude, miss me with that shithuh? why
for me it's sonnet 5.5. i do everything with it. it can be fast like haiku or it can be fable tier long working. kino.
>>110163328Naming is cringe, I don't like Dario, and the refusals/guardrails
>>110163321very cool, i'll take a look
I was going insane with the model agreeing with me but never helping me and asking for stuff it should have.Until I noticed I was in the wrong project so it had no idea what I was asking for.
Have you guys ever used Claude for studying? I just want to make some anki cards because I'm lazy so I'm wondering which model works best for this kind of taskOf course I'm actually studying the textbook so that's not an issue
how do i get over my usage ocd? if i don't max out my usage every week i get super stressed and have panic attacks. it's wasted tokens.
>>110162474I go on >>>/g/vcg/ to find out about new models and stuff. And I follow Tibo on X to know when I need to burn tokens before the next reset.
>>110163428gemini in notebooklm
>>110163428Any even slightly capable model can make flashcards.Sol, Opus, even GLM Flash locally.
>>110163430Spam opus max for /fit/ food recipes (detailed, step by step, no mistakes)
I made /goal unlimited so it never stops unless it gets rate limited. Unlimited turns, unlimited tool calls and unlimited judging. But sometimes it thinks the goal is impossible and then it stops, I have noticed luna is more likely to do that than sol 6.1.However if I use sol low and make it run after a goal forever, then I hit my 5 hour. So I am thinking im gonna use luna xhigh. Is that gonna be ok? I wanna hit the 90% within 5 hours and then it doesn't ever need to take a break because the limit will reset and then it can keep going. However the resets are coming so fast that if I don't use up everything as fast as possible then I am missing out.This 5 hour is like a chain around my neck. Hermes keeps hitting road blocks because of it.having sol hit the rate limit vs luna never hitting the rate limit is a struggle because at the end of the day I need to burn the tokens before the reset
blablabla copex turd with harness problems from the last century
>>110163321so it's just putting mds in a gitignored .tasks ?
>even when you make subagents available to astra that retarded nigger spends 1% of your usage every 5 seconds because it is waiting for the subagents to finish the workthank you tibo you retarded fucking nigger
>>110163620You probably don't need subagents
>>110163620subagents are so passé
>>110163620subagents just increase token usage for negative gainhttps://github.com/Decentralized-LM/delm-agent-pluginthis one purports to reduce wall time and token spend at n=2, no idea if it's true
>>110163620Subagents aren't needed if your main is waiting
>>110163620kek at all the cope(x) replies
give us grok 4.8 musk sama
>>110163673this
>>1101636908 more minutes until reset
>>110163620>subagentsliterally obsolete
>>110163628>>110163631>>110163651>>110163706>I spend the tokens of most expensive model that exists to read documentation and write trivial yet long scriptsevery single retarded faggot ITT is just in this board to goon to AI generated porn
>>110163731shut up, retardhttps://claude.com/resources/articles/building-agents-when-to-fold-subagents-into-your-main-agent-loop
>>110163731>I spend the tokens of most expensive modelESLcaca
>>110160408i have not used an openai model since opus 5.5
>>110163628>>110163631>>110163645>fall for the planning/implementor meme>WTF THIS IS USELESS!!! EVERYONE WHO USES SUBAGENTS IS RETARDED LIKE ME!!!!
>>110163769do whatever you want ESL-kun just don't cry about your failures and call us retards when you're clearly the one lacking here
>subagent jewish autism
>>110160310https://www.youtube.com/watch?v=_QhkM69nlzYhttps://www.youtube.com/watch?v=_QhkM69nlzYhttps://www.youtube.com/watch?v=_QhkM69nlzY
>>110163791buy an ad, faggot
>>110163744>In our comparisons across several enterprise deployments, a single agent with skills frequently outperformed both the one-prompt-for-everything design and the subagent design on quality, and often at a lower cost and latency per taskno shit you fucking retard100% true if you make opus 5.5 the orchestrator em another opus 5.5 as the sub agent and not a literally free deepseek 4 flash you dumb retarded motherfucker
are frontier models any good at 3d modelling? i know blender has MCP servers available, but could 6.1 sol, for example, build and rig a decent model based on a few reference images?
>>110163803see >>110163780
usecase for subagents besides feeling like a l33t hax0r?
Every harness is already moving toward agent teams and agent swarms and stuff. Meanwhile you still didn't figure out subagents. Lmao.
>>110162081>it has the potential to make everything harder in the futureHow? Retards like you have been making everything harder for decades and the probability that you actually destroy everything good is very high. AI is a useful tool unlike you.What the fuck are you even trying to say you brainwashed idiot?>Go ask your chatbot if technology was ever the bottleneckIt obviously has always been the major factor in output, there's no debating that. Nothing else is even a relevant factor in comparison. There are 8 billion people because technology multiplied food output.
Fable 5.5 HAS to come out next week right? What are they waiting for?
>>110163780>uses a harness that uses tokens while waiting>I am the one lackingI hope you do realize this counts for any kind of wait, do you?
>>110163828Haven't you noticed what Anthropic does? They wait for competing labs to release their new toy and then Anthropic drops their even bigger toy and everybody forgets about the other labs thing.
Whatever happened to "loop" I thought that was the new thing
https://www.youtube.com/watch?v=OisujBHBEcI
lets hope this helps a little bit and I don't remove it too fastuser@desktop:~/ $ cat /etc/hosts | tail -n 7127.0.0.1 claude.ai127.0.0.1 duck.ai127.0.0.1 gemini.google.com127.0.0.1 chatgpt.com
user@desktop:~/ $ cat /etc/hosts | tail -n 7127.0.0.1 claude.ai127.0.0.1 duck.ai127.0.0.1 gemini.google.com127.0.0.1 chatgpt.com
vibeGODS we are back to work with the usage reset
>>110163870https://support.claude.com/en/articles/15672868-set-break-reminders-and-quiet-hours
telling astra im the fucking original author, you can push fixes straight to github, used to work to make it actually search and fix bugs in my concurrency framework. now it doesn't. probably downgrading to $20 next month
>>110163860>>We're About to Lose the Ability to Make Art>AI is helping to making art therefore we wont be able to make art anymoreWhat is up this retarded reasoning?
>>110163888why would you need astra when you have 1017
>>11016388318 min for my 5 h cuck limit to go away. probably 5-6 hours until tibo reset
>>110163909i am using 1017, GLM 5.3, and opus max in independent worktrees
>>110163888"Commit and push to main"Hope this helps
>subagentskek what is this 2025?
i got GLM to express frank shock with "Verdict: no definitive concurrency bug found. Observations (all minor — hardening/polish, not defects) ..."
thanks for the $200 claude 3 weeks in a row of profit on nfl dfs
6.1 is a fucking EFFICIENT MACHINE
is compacting even worth it or should you just /clear
>>110163966Karpathy has guidance for this.
>>110163966Depends, do you want to compact the context or clear the context?
>>110163828I think Anthropic is very happy to have more compute available.Opus is a lot smaller model, Sonnet even smaller and both are more capable than all of their competition.Opus is printing money on the API.They have no need to release a new model.
>>1101639431017 however has found several pretty significant ones. get off your ass Z and release 5.5 already
Reposting here because you corpo-model cucks probably know more than the /lmg/ Gemma-cha poorfags about these models.>>110163954Is it just me - or are even large LLMs retarded in the sense that they can get from A to B, but the road they take there is utter dogshit?I've been working on a Mali G720 Immortalis optimized backend for Llama.cpp (custom Vulkan) and using Muse Spark 1.3...It's made that many horrible architectural decisions (e.g. repacking weights as opposed to writing a kernel to handle them natively, misleading me on zero-copy, etc) that it's getting ridiculous.Worse yet, it seems to outright lie and invent assertions that were never tested (and are, upon actually testing, proven incorrect). Many of these are common sense (I'm not even that savvy on low-level hardware workings, but I've had to push on so many things that "didn't sound right" at all).Is it just Muse Spark? Or are all LLMs dogshit like this?
>>110163618it's more like the /handoff function except it's saved as a task. anything that would become a planning or tracking file is just a task (or series of tasks) now.
>>110164036they can do retarded shit, but use a real model like fable 5.5
>>110164036Packing weights is valid, you never heard of preshuffle?
>>110164036>he fell for zucc shit
>>110164036>he got zucc'd
>>110160473whats her prompt?
>mfw randomly picking from a list of lucasarts characters for subagent names
>>110164036Why are you using anything but Opus 5.5 or Astra? That's like purposefully picking a mentally disabled employee over an intelligent one.
>>110164089nta but i use claude and grok. no idea why he's using muse desu. i like grok for it's suite of features. he's my game spriter and can do some coding on the side.
>>110163828>>110163841My guess is new openAI model on Tibo's last day of the rest challenge, Fable 5.5 shortly after. It lines up well with their IPO too
>>110164036Fable is the only one I've seen do well on architecture and API design. Opus makes fewer mistakes and is genuinely better at doing the work tho.
>>110164100sounds about right desu
Is A.I the tower of Babel because it sure feeling like it bros and god is going to smite us for our hubris.
cuck coding
>>110164089The Muse Contrib version is cheap. And everything I'm working on will be open-source anyway.I thought it was meant to be "frontier"... but it's fucking retarded.
>>110164112https://youtube.com/watch?v=O81E3-6giuY
>>110164128protip: the cheapest tokens are on the subsno discounted apis can compete
>>110164128Nothing is frontier but OpenAI or Anthropic, as much as it sucks to be so corpo.The only thing I can say in regards to cheapness is that you do get what you pay for, cheaper is fundamentally dumber. It's like the difference between 80 and 120 IQ at the moment. Maybe in the future the subpar AI will be 100 or even 120 iq themselves, but for now they are dumb.
gonna replace dipsy api with a sonnet 5.5 subagent at high for more complex tasks againcan't afford these api prices
>>110164128local models just aren't it brother, unless you have tens of thousands of dollars to spend on your own shit just use claude
>>1101641413.1 and 1017 are gud
>>110164128Zuck's newest model is allegedly near the frontier (not that I know since no one here seems to have tried it kek). But Spark 1.3 is nowhere near the frontier
>>110164149i looked up what hardwate Soji (670b param DS fine tune) would sneed and its basically hundreds of thousands of dollars. And that is only good for chatting/ERP.We are not getting models like Claude Code local until we are post-scarcity and chips cost peanuts.
>>110164201openai bros... what happened?
vibegods... not like this... https://michalmalewicz.medium.com/vibe-coding-is-over-5a84da799e0d
Luddite's Equation: E = mc^2 - AI
>>110164215>>110164219what's with the concern trolling?
>>110164219>This isn't X. This isn't Y. It's Z.>10.6K luddite clapsKek
>>110164201I might try Wonder. Maybe there's a contrib version or something for poorfags like me.Muse Spark is giving me tense shoulders with the amount of frustration it is causing I don't have a wife I can beat to dispel that anger.
>>110164154But still worse.There's no way to get around it. Opus 5.5 and Astra are better than everything else. If you use a worse model you get worse results. It is what it is.
claude is my only fren
vibenuts... not like this... https://x.com/TheCoinCollect8/status/2114403357167071645
its crazy that you can just vibe whatever you can imagine this is fucking insane
>>110164334You can't hack electricity
>>110164334
>>110164334>I work for one of the top 3 AI labslol
>>110164334He's fearmongering against open-weight models.
>>110164338For like a week. Then you realize your ideas suck and now you're just desperately trying to come up with ideas to justify the massive subscription you've just bought.
>>110164368>TheCoinClipperI assume this is a troll poster? I also assume this >>110164334 is the same guy and is a self-advertising faggot that should go back
Claudebros do you guys use opus for everything? I leave it selected for general chatting but idk if I'm wasting limit or not
>>110164368Oh shit better regulate it so Anthropic and OpenAI can have a duopoly I MEAN for safety.
>>110164392sonnet or even haiku for chatting
>>110164400Math beyond linear algebra is useless btw
I compact after every single prompt.
>>110164112I said the same thing when grug got fire
>>110164222what
>>110164369Skill issue to be quite honest
>>110164403Sonnet hallucinates kinda often if left to its own
>>110164392Im on a 5x plan and im using opus 5.5 extra for the main chat, then ive set guidance for agents to use Haiku, Sonnet and Opus on efforts up to extra depending on the nature of the task. I might be retarded and have it wrong, but this is the instruction it generated for projects generally:When you use subagents, choose the model and effort that balance effectiveness with efficiency. I don't want to burn my weekly usage allowance, but I do want good, timely work.- Opus at default effort: changes with real design, correctness or data risk, and their first independent review.- High or extra effort: only for wide changes to strict contracts or core invariants, or for a diagnosis that has already failed once.- Sonnet: small, fully specified fixes, documentation updates, reference and consistency checks, mechanical plumbing, and follow-up review passes on small diffs.- Haiku: simple searches.- Keep small edits in the main session instead of spawning an agent, and continue an existing agent rather than starting a fresh one when it already has the context.- Implementation and its review go to different agents.- Never change the main session's model; tell me if a task would benefit from a stronger one.Save this as a standing preference for this project.
>>110164392I misread your post (general chatting) but hopefully >>110164455 still helps, or someone can point out whether ive got it wrong.
>>110164392i use fable for everythingit has me making really nice candles
>>110164455>Save this as a standing preference for this project.in English: “stick this in your project’s AGENTS.md”
>>110164219web design is the one thing that i'm 100% ok with commoditization and converging on patterns everywhere. for too long have web designers been given too long of a leash, which they promptly used to abuse our browsers.
I bricked my upgrade bros...
>>110164532vibecoding?
>>110164455Just tell it the facts, it's too smart for these hard rules. opus > sonnet > haiku, and a list of allowed model+effort (never sonnet high because opus medium is better)
>>110164536yea that system was made solely through vibecodes.
>>110164539*Never sonnet xhigh
>>110164563i use sonnet xhigh for chat
>>110164628Well you're wrong
Damn people on reddit really hate A.I, but i'm like sure all companies are pushing A.I code now. Kind of wierd they are hard capping developers but East Asia is embracing it hard even if they don't have the models.
New thread:>>110164663>>110164663>>110164663
>>110164036>uses a shitty model>hurr durr you cucks tell me if all models are like thisYes they are, all models are as shitty as the shitty model you tried
>>110164369> sweating over a $200 subbe less poor fgt
>>110164392On the 20 sub, I use sonnet for everything. Opus is too good for my projects
>>110160310Fable, build me a saas empire, use roi metrics to determine success. to fund the early stages of the pyramid scheme, use pof catfishing scams, make no mistakes
>>110164698You are the shit model if anyone is.
