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://x.ai/cli — no 5h limit for only $30/month## 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>>110093924
slopgon bricks your computer
buffcat won https://streamable.com/oiloiy
second for snailcats
>>110098628>>110098629
>>110098631*farts everywhere*
someone test if deepseek 1017 can crack idm or if it's a pussy ass faggot that will respect intellectual property despite being distilled
>>110098618use local models
he's right, you knowrecursive selt improvement never
>>110098637international douse of mancakes???
I don't think I'm getting rich anymore saars.
Muse Wonder has shown up on Cursor......
>>110098546Here's a video of me playing around with it: https://files.catbox.moe/hosu7o.mp4Oh and I just fixed the GUI's style to match the original synth.
>>110098658We are certainly closer than 5 years ago.
>>110098685that's pretty cool
the new deepsneed is goated
>>110098685what kind of prompts did you provide to get it to match an existing synth's sound? was it able to find design docs or schematics online to figure out how its supposed to sound?
>>110098660kek
>>110098658>substack>describes himself as a nationalist sociologistthis faggot doesn't know shit. models are already contributing to AI research in meaningful ways
>>110098701I provided user manuals and such. For example: https://www.virus.info/download/170Access has never released official design documents / schematics / whatsoever of any of their synths. So claude just inferred / interpolated information from the publicly available documentation to bridge gaps. Also I know this synth inside-out as a user, so I was able to give it plenty of information about whatever my own theories are of its internals.
>>110098756you dont expect grifters to know shit.
>>110098756>describes himself as a nationalist sociologistso he's redpilled
>>110098799add "it's smarter than me" lmao
>>110098685It's here now if anon wants to play with the synth: https://gitlab.com/harha_/virtual-virus-cCurrently only supports X11 for display and ALSA for audio output.
>>110098660Based claude, keep them normies out.
>>110098764very cool. this wouldn't have been remotely possible even a year ago, crazy new world we're living in.
/goal Teach me to be a $100MM/anum pro gamer. Make no mistakes.
I let Muse Wonder create a minigame for my aim trainer, Opus 5.5 did too. Should I let Argon touch the code? :S
>>110098959not under supervision of opus 5.5
may i wake up tomorrow with at least 2 more sales inshallah
how long until you can download and run an opus 5.5 tier model on your phone
>>110099068two more weeks
>>110099081whats the point
>>110098756>>110098783keef is based on history and genetics but his track record on predictions and analysis of modern politics is abysmal. best ignore him on this
>>110098970Yeah, I well leave Argon in an advisory capacity lmao
Give my wife an orgasm. Make no mistakes.
I'm seeing what I can get done with Fable 5.1. I'm making a 3D tree maker thing.We'll see what it's like, who knows. I didn't flesh out the prompt except to say use pygame.People with imagination should use vibecoding.
>>110099107I'm not black, sorry.
I lost track of the message in the other thread but I’m rooting for the IRIX enjoyer to make something greatGodspeed anon
>>110098658The same people saying RSI will never come were also saying automated coding would take years to get here.It's always shifting goalposts with these people.All the best models are barely a month or two old & somehow we've hit a wall, while they're clearly working on the next models which will blow the current ones out of the water.At the most basic level, the fact that they're putting MORE resources into it should tell you capabilities will only increase. Zero reason to think otherwise at this point.
>>110099267this is very indian thinking
Yet again, Muse Wonder is out.
>>110099288Why
i just played the game i vibe coded with gemini 4 argon with a single prompt for more than an hour without getting bored
>>110099305What game
>>110099305>schizo slop, will brick your machine, unreadable code, written like a sa'ar
>>110099305That makes sense. I found it made extra changes in subsequent prompts that required corrective prompts.
>>110099329setting it to (high) will absolutely blow up your project scope unless instructions are ridiculously clearhttps://medium.com/google-cloud/stop-defaulting-to-medium-optimizing-gemini-4-argon-and-3-8-flash-effort-budgets-in-antigravity-df55442d7560
got computer use working for deepseek 1017 via claude code. it seems more competent than it was before. previous computer use was via hermes agent, and it was slow and shitty. wasn't as good as sol's. but 1017's? feels just as good as sol's computer use.
>Sol 6.1It just werks until you hit the roadblock of it being an autist who doesn't implement something obv that was right before its nose.>OpusIt just werks until it starts including its own funny ideas in the architecture and (somehow at the same time) asking you more and more questions back and getting you lost in inquiries about details, delegating more and more back to the human.Honestly, the GPT workflow is more attractive right now.
>>110099404haiku 5.5 is pretty good at playwright
>>110099431it is. i got it to gen some genshin impact images for me via A1111's UI while i was gooning to kimi. it did a good job. i like this trend of cheap models becoming good daily drivers
>>110099404I shan't accept 70% performance if I can have 100% performance with a Claude subscription or two.
can you share the game
whew lad. it was kind of nuts watching deepseek draw that in real time. i don't think you can distill computer use can you?
>>110099477The Game
>>110098685Kong kong king king king kong king kong
>>110099427i feel that opus's prose is also harder to read. sol 6.1's plan files are a lot easier for me to follow, and it doesn't feel any worse/better than opus 5.5 for what i use it for.it's a shame that the codex desktop app is so fucking shit, there's no reason for it to use 9gb at idle and still be laggy
>>110099509Computer use is one of the areas where I would think China can easily overtake the US. Think about how much they can pay and how much data they can gather per workflow especially given how bad the economy is in China. A lot of (former) white collar professionals would gladly be recorded to get paid scraps right now. I would expect any computer use advantage any Western model has, China can close easily without distillation. Not like they even need that much anyways.
I cannot keep up with these resets. I took a break for a day or two. I've missed out on like ... 150% of weekly usage ($200 plan). And I also have a stored reset to use by the 22nd....... I can't hold all of this usage!!!!!!!!!!
>>110099789>it's a financial canary in the coal mine... and that's a GOOD THING!
>>110098618https://www.reddit.com/r/antiwork/comments/1x889yz/ai_is_destroying_my_livelihood_and_my_passion_im/
>>110098658Keef turns into a complete brainlet the instant he steps a millimeter out of his lane.
>>110099849>Antiwork: Unemployment for all>AI is destroying my livelihoodMission accomplished, I guess.
>>110099893he's right though there's no evidence that recursive self improvement is coming any time soon or even real, right now chatgpt and claude are glorified encyclopedias
My custom inference program now supports moss-image and bboxes
>android app loading a little slow>tell sol to make it faster>makes it start faster>tell it to make it even faster>makes it start even fasterYeah, it’s that easy.
>>110099902Why would moss image support b boxes?
>>110097821>any of the chink subs worth it rn?If you mean any of the subs from the chink model's own lab, probably not.If you mean third-party subs that include chink models, there's something of a price war going on right now for who can offer the most DeepSeek V4.1 Flash inference on a $10/month sub, with OpenDesign Go at $120, and OpenCode Go and Command Code GOAT tied at $80.
>>110099902Bro I love that kind of stuff. Can you do something similar to this?
>>110099941but sadly, no such thing for kimi
>>110099789The worst thing about this is that I'm waiting for the end of the 7 day period to let it burn usage on big refactorings, but this never comes.
>>110099902https://github.com/OpenMOSS/MOSS-Image#-layout-to-image-generation
>>110099932intedesting
How the FUCK is AI not the most momentous invention since the transistor? Who'd say that?
>>110100048because "ai" is a collective term for gofai, the transformer, adamw, etc, and nobody and no single source invented enough ai in their lab alone to qualify for an invention. it's as old as turing
They unnerfed sol 6.1 todayAnd smug nerds in HN think its "hedonic adjustment". Bumch of clowns. For a forum full of commies they're surprisingly defensive when you say the AI companies are nerfing models bts.
Saw Anon posted his rewrite of gaea last night, someone should do a rewrite of world creator now
>>110099427>It just werks until you hit the roadblock of it being an autist who doesn't implement something obv that was right before its nose.Fuck it's all true lol. You can't one shot with it, you really need to do multiple iterations but you can trust the result unlike opus.
>>110100090Good, last couple days it’s been absolutely fucking retarded
new openai "aurora" models sightedrumored to be beyond 6.1
I'm making something. I wonder if it will be awesome.
>>1101001276.1 astra.
>>110100131What does Claude say about it?
>>110098799it's funny cause "just scale it" turned out to work so much better than expert systems and classical MLbitter lesson indeed
>>110100141Its three models. Could be 6.2 sol,terra,lunaOr.. 6.1 astra, astra cyber and astra legal
>>110100144It’s the same snailjeets who shit their pants when someone brute forces a cancer cure or solution to an important maths problem.
>>110100142>What does Claude say about it?It fucking triggered a report of the chat, thanks nigger
>>110100144i'm using argon to build a system with gemma4 that rapes every other openweights model at things an llm should not be able to do. scale promotes miniaturization.
>>110100149they can't possibly bring terra back. there's no space for it. 6 sol should have been 6 terra, and 6.1 sol should have been astra minor
>>110100090>>110100126it's still being mogged by free deepseek flash on opencode for me
>>110100158>6.1 sol should have been astra minorit shouldn't have been, otherwise we'd get half the usage of current "totally not baby astra" soldario did a huge favor for chatgpt consoomers, dropping opussy when he did
>>110100158>there's no space between $0.1/0.5 IO and $2/10 IOdo openai managers really
I’m modifying QField (QGIS for mobile) for my specific field work and it just dawned on me that I will be able sit in the middle of the bush with my laptop, starlink, usb debugging/adb and vibe improvements and bug fixes.What a time to be alive. Stage 2 integrating AI powered vision for identifying and measuring things in the field.
>>110100142idk, I have Claude Fable 5.1 Medium busy making a tree generator. It's going ok.
>>110100169There really isn't. Simple tasks go to luna. Tasks where you need more intelligence go to Sol, which is the daily driver for most tasks. Astra is the halo model, used for... fuck i don't even know, doing some gay math pr I guess.low, medium, high, everybody can understand that
>>110100185My claude opus made me these willows for my tree generator which I like
>>110100169>>there's no space between $0.1/0.5 IO and $2/10 IOyespeople usually use the best model or the cheapest in their subscription, mid range models have no use case
>>110100198Astra for planning, Sol for implementing. Simple as.
>>110100212>people usually use the best model or the cheapest in their subscription, mid range models have no use casewell i guess we need to remove sol then
>>110100210You can crush a 5090 with one tree, so yeah, you have to make decisions about details, whether it's for rendering, or for gaming.
>>110100212Just me over here with 6.1 light.Am I slopping it wrong?I had ChatGPT write me an agents line to delegate simple tasks to Luna sub agents.Is that a mistake?
>>110098658>recursive self improvement is fiction>yet models today are performing feats thought basically impossible just a year or two agoI don't like it, I think its all moving too fast but denial is the worst thing possible right now. Opus and Fable are clear signs the potential is far higher than anyone thought possible for all of this. On my own projects I've seen enough evidence of claude being perfectly capable of taking in data, pictures, evidence and sifting through it to ascertain a new improved approach and direction that works better. Even if its not yet "total self improvement" if it can imrpove itself through 99% of its own code and 1% human review, then thats only a stones throw away from 99.9% of its own code and 0.1% human review.That 99.9% will, to us, look like perfectly recursive self improvement already. Then it will hit 100%. Opus and Fable have proved this, whatever the AI self recursion singularity brings we're seemingly now just hoping its good for us.I'm content just making my dream software and video game mods through prompts until paradise or the apocalypse hits, I can't change anything either way so why not have fun?
>>110100219>Just me over here with 6.1 light.>Am I slopping it wrong?Yes. The actual low end should be medium, it only takes about ~15% more tokens and the output is noticeably better than low.High is a smaller uplift over medium tham medium is over low and uses ~50% more tokens so it's a ymmv situation.
>>110100223There is no yesterdayThere is no tomorrowThere is vibe
project darksnail just >4x'd the 2026 arc-agi milestone 1 prize winner on arc-agi-3. 29 / 183 levels (18 / 25 games). 4.8123% rhae.
>>110099404Can you elaborate
>do I waste all my tokens to rework the draft designs into something I like or actually flesh out everything else as a draft firstIt feels like wasted effort trying to iterate on design so early, but what if I wait until the end and then it's a huge pain in the ass (for claude) to rework everything?
>>110100198>tell astra to make a plan and track it while handing off bounded tasks to sol and luna chats as appropriate>okay now just keep at it until you really need my input>it just chugs away
>>110100048I asked Sol 5.6 how it'd rank inventions humans have done in terms of importance (I told it to exclude fire), and this gem absolutely without shame put AI in spot 1 (IIRC before agriculture).
Astra, bounce on it. Make no mistakes.
>>110100223>>yet models today are performing feats thought basically impossible just a year or two agosuch as... stealing other people's work and modding minecraft?
>>110100243Cool, does it have killable (stabbable, dismemberable) dwarves?
>>110100254not yet, but it could. it's a dungeon crawler
>>110100253It’s gone from a slightly better version of Google that occasionally made things up to creating entire programs from scratch, so yes.
>>110100253both of those were doable a year ago
>>110100253>m-modding minecraft was totally possible 10 yrs ago guise any chatbot could do it heh *hits pipe and thinks to that time Cleverbot made Minecraft and handed to Notch*
>>110100216Astra is way too expensive>>110100219Under-reasoning is not as bad as using under-sized model. There's a band of difficulty that the big model could handle but the small model can't, they often end up burning tokens trying to solve that, wasting more than than what they saved
>>110100258>creating entire programs from scratchpretty sure high schoolers are doing that with python
Claude lowered its daily/weekly limit in the past few days, didn't it?
>>110100273Don’t recall high schoolers making pc ports of console games or adobe suite recreated in rust
>>110100253>stealing other people's workIt's "stealing" my code that I've been putting on SourceForge since 2006 and Stackoverflow and Github since 2013. I am stealing from myself.
>>110100276im pretty sure emulators and gimp / libreoffice existed before AI
>>110100284Damn you’re right….it really is so much better than it used to be.
>>110100090Great news, I just ask it to review what it did during its days of retardation and it found a bunch of stuff.
>>110100284Why are you here lol
>DUDE DID YOU HEAR? AI MADE PHOTOSHOP IN RUST>check the github>1.7k issues opened in a week
>>110100284>gimpThank God for AI slopclones, Jesus...
>>110100210I didn't have enough usage left to fix the top. That could take quite a while to work out, since I'm not a tree expert. But, it's in decent shape. This is my $60 Cursor plan (now grandfathered/legacy?)'s "other models" usage, which is being eliminated for just Cursor Grok.
>>110100243What you have is about the best claude can do as far as 3D models go. You need to experiment on all visuals and AI audio first and foremost, and understand your limits.Basically even if you pay for Meshy and such, it won't really work out without a lot of manual effort or hiring real people to sort things out for fully 3D games with actual non-mathematical graphics. So we're not *actually* game devs yet. For now you can only make shitty graphics, or stylized code-based meshes. 2D and 2D.5 is basically AAA indie level from what I've seen however.
>>110100295>nooooo stop fixing things just leave it just LEAVE IT ALONEThe vibe in the adobe office must be GRIM
>>110100295It's beautiful <$10,000 at api rates and it will be flawless.
>>110100298Hm, in that case I should flesh out the story and game mechanics and level design, and then in six months I can have claude 8 redo all the art. I'm not really in any hurry.
>>110097209>one of the UIs I built around my inference engineKey words are right there, anon. That particular UI involves TripoSplat and Kimodo
>>110100297trees kill performance for 3d games btw, it's why they still look shitty in $current_yearif there's a lot of those (and worse yet if there's physics) you shouldn't go full retard with resolution and the amount of needles they have
opus 5.5 reviewlooks like it prefers deepseek as the implementer, but haiku is pretty closehaiku actually wrote better code (500 lines vs 1k+ from deepseek)it recommends deepseek as the all-around implementer + haiku as the reviewer, since haiku is more literal and honest
>>110100306no one is paying API rates, I don't know if AI companies ever expect their cheap customers to pay up at all
>>110100314>better code (500 lines vs 1k+ from deepseek)LOC is not a quality metric
>>110100312>>110100297Maybe you can get Claude to remake speedtree or something similar?
>>110100307>>110100298Also, it can still do pretty good things. It suggested that an arcade would look good so I said ok and here it is. Yeah, it's not AAA 3d, but I don't care, this is a puzzler dungeon crawler so good enough with a little more work. Everything works pretty great so far, but the game is simplified in the sense there is no real physics engine, no colliders necessary, scenes can all be small
>>110100314>the most expensive model is the worstgrim. what are the odds you can pit 4 argon vs kimi k3 and sonnet 5.5?
>>110100339Probably look a lot better if you used good textures
>>110100339NEW GAME ****APPROVAL**** @@@SEAL@@@ THEREOF
>>110100332There are real tree makers already in existence. I'm not doing anything new, but like most people, sometimes we want to do pointless things we like.Sort of like, other people smoke, I vibecode.
full opus breakdown of who gets which task>>110100328>>110100347i might if i decide to sub to opencode go again, sadly i don't have direct access to argon
>>110100253Actually try it yourself instead of watching jeets on YouTube one shotting prompts in the web chat
>>110100347Yeah it would but that was the question in my original post, finalize a vertical slice or do the full rough draft?
qwen 3.8 max october checkpoint waiting room
trees arent real anyway, it's just organic plastic made to look like trees by (((them)))you probably havent seen a real tree in your life
Does Opus suck at Blender shaders/compositors?Because Sonnet apparently sucks at it
>told llm that I have a headache so it can't be retarded in this session>it was indeed less retarded than usualnew prompt meta dropped
>>110100520>the user said he has a headache and i can't be retarded; fine, time to lock in.
>>110100520>she doesn’t tell the llm she’s Jewish for the unlocked model
>>110100429Gemini can generate textures, turns out...
Sources tell me my weekly usage is stable.Please god.
https://x.com/Remake___/status/2111642490633925044
Graben and canyons from the regional stress field caused by volcanic rises. This is simulating Mars' Tharsis rise. Future improvements as well as atmospheric weathering will improve the look further. Venus also has many of these on its surface.
claude nerfed yet?
What now
>>110100639And being frank she lends to those are free
>>110100563squint squares.
I'm an idiot who struggles with command line tools like gallery-dl and yt-dlpIs there some sort of local AI I can install that I can ask shit to help me format the commands/scripts I put into command line utilities like that?I emphasize "local", I don't want to make to make an account or anything, I want all my queries to stay on my machine and not be on some external server. I also don't want the AI actually running any commands/operations, or have the ability to save, read, or edit any files on my PC at all, I just want to ask it shit and for it to answer me (with hopefully correct/working) info.Also, while I'm hear, I have seen friends of mine easily track down articles and images using AI I couldn't relocatef even with intensive searching and reverse image searching. What is a good tool I can use for that use case, again, without me needing to make an account (or at least without me giving my phone number/ID).
What is the most impressive 100% vibecoded game you've seen? I'm 100% AI pilled but when I try to tell people how insane it is they say "show me a good vibecoded game" and, well, Canabalt 3D in the OP isnt winning them over.
anyone found any real use cases for argon yet?
>>110100671Yes but also no, stop being a paranoid retard and sign up for codex or claude. Like oh no a company knows my yt-dlp commands, get a grip.
>>110100671yeah you want to run in raw llm mode in something like llama.cpp instead of using a harness. paranoid bastard
>>110100678argon and muse had initial shills, but the shilling had since dropped out, right as muse wonder is actually available...
>>110100671you can not be real
>>110100690100% argonhttps://github.com/sneed-and-feed/projectdarksnail
>>110100694>>110100678meant to reply here
>>110100671learn how to use a computer mi amigo
>>110100671>Is there some sort of local AI I can install that I can ask shit to help me format the commands/scripts I put into command line utilities like that?yes.first of all, there's grok build.but also there's grok bot.HoweverYou can just ask cursor grok to do stuff.I assume Claude can too.
>>110100681>>110100688>>110100691It's not (just) about privacy, it's also about me wanting access to the tool at all times without me being dependent on some external server that can get shut down or the service getting discontinued.>>110100715Aren't grok and claude online ai tools?
>>110100726oh, you want offline???
>>110100726JSIDATP
well, if offline, qwen, but only as an advisor...
>>110100726If codex or claude get shutdown or discontinued then something has gone seriously wrong and you'd have bigger concerns than yt-dlp commands
kek yuropoors are so cute when they try to ai
>>110100726just top up $2 on the deepseek API, you can use it whenever and no usage limit as long as you have credits, and you don't seem to do intensive coding anyway
I wonder if the rioters have broken into the HF or Mistral offices to steal the GPUs
>>110100223>>yet models today are performing feats thought basically impossible just a year or two ago
>>110100726You can try local LLMs, but the good ones need very strong hardware, and even then they aren't as good as the frontier labs, but maybe good enough for your usecase.I wouldn't worry too much about offline work, if you want scripts, just make them reusable and document them, then you will be able to use them without the LLM. Or you can have the LLM write you some UI to launch the scripts.It's also very unlikely that you won't have access to any online LLM, worst case you would just change the provider.I actually share your privacy concerns, but for me local LLMs that I can run on my 5070TI just aren't good enough, and since this is my job I had no other realistic choice but to get Claude and Codex.
We need a poetry board :|
... it's written 800 lines....
>>110100738Reminder who is leading fusion research again?
>>110100801If Europe exists, how come it's not here?
>spend days setting up a custom system that interacts with hermes agent>nous faggots go turbo retard and change everything on the backend>days of work wasted>AAAAAAAAAAAAAAAAAAAOk, I'm done with it. Now I'm looking for the most STABLE baseline harness that I can build on top of. My instinct tells me that even OMP will break my shit at some point. I'm going to have to build a lot from scratch, but surely there is *some* solid rock I can stand on, right?I shouldn't have to be bogged down with building an automatic compaction system when plenty of them already exist. I don't want to have to build access to bash and all this bullshit.What is the most stable harness that isn't going to completely change the way that it interacts with its own gateway or whatever equivalent?
>>110100819https://github.com/nearai/ironclaw
>>110100801Stop coping, when it comes to AI Europe is closer to Africa than it is to China or the USA.
>>110100829ai is illegal in Europe, basically.
>>110098660first genuine laugh I've had in a while
>>110100801won't be whatever thirdworld shithole you're referring to when ASI arrives but you can dream
>>110100833It's typical office prick crap.It's what I *don't* use ai for.Just fucking do it, don't fucking argue. You're not actually performing for an audience, you're not a tv star on The Office. You're, ai.
I find it so hard to come up with ideas for projects. what are you guys building?
>>110098660>and then all the agents clapped
>>110100871schizo slop
>>110100887was the word project before darksnail really necessary?
>>110100895yes. was your birth necessary
>>110100865Are you replying to the correct person?
>>110100920He's saying that he doesn't want that kind of attitude from a clanker.
Snailschizo, if you want people to take you seriously then you need to be able to explain your project in your own words
>>110100871Modifying QGIS to be an all in one gps/gis/logging tool for minerals exploration
>what's that you have questions and want me to explain my project?>I have no clue ask my dedicated Q&A Agent
>>110100945yea i can do that, but it takes fucking forever, i have to do bitchwork collecting exact numbers, and you didn't ask. nobody asked. you sound like a redditor.
>>110101001Nobody asked because by all appearances you're a schizophrenic posting word salad without context. You must want recognition for it as you keep posting it, so you need to explain what it is you're trying to accomplish and how you're doing that.
>>110101027fuck yourself retard, you have access to frontier models and you talk shit
>>110100979>he's not vibe-interviewing
Nice bits >>110101001 (425)
>>110101040
>>110101029Plus this, you get aggressive when anyone dares to ask what you're doing.
>>110101047do a flip you obnoxious little pubic hair
Just took the Claude pill after months of using Chinese models.I can see why normies and luddites still cling to the stochastic parrot meme. Claude comes off like a complete clanker. GLM and DeepSeek sound much more natural and intelligent despite being much less capable.Anthropic should really do something about this.
>>110101069in the anthropic worldview anything that isn't a hammer for a nail is a safety threat for trillions of people who havent been born yet
>>110100671anon it would take you under an hour to learn how to use every CLI program and create scripts (copy/pasted from ChatGPT) for the rest of your life
>Anthropic: Action required: Your API key...>oh god how many hundreds have they stolen>oh it's just an expiration warning
All my fucking coding, software design, 3d rendering, audio processing and whatever learning and making efforts of the past 15 years are futile now. Anyone can create anything. I feel borderline suicidal, I feel like nothing matters anymore. There is no prestige in creating anything anymore and it'll only get worse over time. I guess my cope will be that we'll be able to do good for ourselves with the help of AI, increasing our life comfort and health.
>>110101069What is the recommended way of using DeepSeek's API anyway? People seem to mock using OpenCode.
>>110101069>Why yes, please give me a bunch of information in the most robotic and direct way possibleI wonder if everyone who likes Claude has the autism gene
>>110101089posted it again award
>>110100671You dont need an account to use chatgpt, gemini fo free
>>110101089the future is now.
>>110101100I'm late to the party, I only realized this now because I actually tried it.
>>110101036>he thinks that's me doing the interview not my dedicated Interview Agent>he thinks it's me posting this not my dedicated (Shit)Posting Agent
A lot of things aren't final here. The temporary UI was just made by Claude and will change, and there are still weathering and climate processes I need to do next. But this shows how Venus' "continents" can be simulated by bulging the surface up with volcanoes.
>>110101069Have you ever used something like the latest Sol? It's pure autism. I made DeepSeek write an architecture document once and it used way more natural language than the OpenAI model.
Panicking -- still have my banked Claude reset, but it expires in 4 days and this weeks use is only 50% despite intense work loops. Like idk how long Claude took, but I spent at least a dozen hours purely conversing with Claude.How do I convert tokens into goodness quickly?
Sol says it's schizo slop too>Zeno-Amygdala Controller>Beltrami Triad Flux
>>110101089>giving in to hopelessness because things are too easy nowI'm more optimistic. All my work of the past 21 years can be improved and more importantly extended in precisely my style. See for instance I spent a long time in the 2010s making a GUI system that renders using OpenCL. So I got clankers to port it to Vulkan. It's still my system, but now it's Vulkan, and I didn't even have to learn it.
do you guys use ultracode? apparently it can be used with any model and effort level. but what is it for? doesn't claude spawn subagents on its own when necessary?
>>110101123spawn subagents using opus as well
>>110101113cool stuff desu
>>110101126I see that angle and many other angles to this situation but it doesn't cure my depression that stems from the fact of how effortless everything is now. When I use an LLM to code I feel like a void inside my brain is increasing in volume.
>>110101124you can't read motherfucker, learn to read. you primed the model to give you what you asked for, and it said, let me iterate:>there was a methodological error in paper 1 that was retracted, sol acts like it was never retracted>pendantry about what 100% means even though the real number was reported, plus it repeated statement 1 in a different way to bulk up the response>sol admits that it's valid mathematics and it's too lazy to verify every number personallyyou really dun gooft anon. you should go fuck yourself.
>>110101095OpenCode's cloud shit got hacked, but DeepSeek should be usable in about every harness since they support both major API protocols. pi, Junie, or IDE integrations like JetBrains or VS Code. Pick your poison.
>>110100979>Use the Grill-Me skill with Claude Fable 5.1.
>>110101095claude code seems fine, it also has deepseek harness.
>>110101145Take your meds
>>110101135Yes, it's sad. I can not create personal software more quickly, but the problem is that using some personal software is just a small part of my life and having challenges and job security was worth much more than any toy I can vibe code.
>>110101164>thinks a 41s scan with a 2nd tier model = valideat a bag of dicks cockmongler.
>>110101135one day you won’t even think about it anymore. you will manage a work queue pushed to you twice daily by your personal agent orchestration software. you will approve and rubber-stamp decisions that generate tens of thousands of lines of code while you fuck off from computer land and live in the abundant post-scarcity paradise the AI jews created. you will be happy
>>110101156>Hey Fable5.1 use Grill-Me spawn a 100 subagents, maximum effort.
How is deepsneed so cheap? I tried it, worked fine for my small project, and paid slightly over a cent per million tokens. A lot of it was cache hits but during their off-peak hours the entire thing barely costs anything.
>>110101192unironically they have some extremely well-funded voodoo black secret chinese shit going on and nobody knows what the fuck is happening, that's the word off tech xitter. nobody has been able to copy or figure out the secret sauce.
>>110101171Lil bro chatting about model tiers while using argon lmao
>>110101199>bought the shillingbuck broken. also, if you aren't able to understand and learn, you are cattle.
Just had to stop Sol from optimising my map panning because it tried to updoot my android tablet.
>>110101177They're both gay?
>>110101158Does claude code include claude's usual censorship, like being super bitchy about everything?
>>110101192Operating at a loss because they are backed by state funds. Modern warfare.
>>110101199one more thing, darksnail rapes gpt-6-sol AND gpt-6.1-sol on ARC-AGI-3, you can literally go check right now.
>>110101113>>110101133Mars is even better. You can see how within the first billion years the volcano rise causes the planet to turn so its south pole becomes the equator and craters building up and filling up over time. On the shaded relief map, you can see how lava flows fill the large basins with basalt, so there's a mix of smooth plains and cratered highlands, like real Mars. And you can see the stresses caused by the bulging creating radial graben and valleys perpendicular to the stress axis. The surface is gray at the moment because oxidisation will create the red regolith we see on Mars, and wind will spread it to create diffrentiated albedo features. Erosion will make the craters look more degraded, smooth the shields, and make the valleys look more detailed. We should be able to create tan desert worlds (e.g. Arrakis) by simulating plate tectonics and a past hydrological cycle, or even salt/evaporite worlds (e.g. Crait) with a lot volatiles and runaway ocean evaporation. With climate simulation we will be able to create snowball worlds (e.g. Hoth), and possibly cryogenic Earth-sized Titans with tholin plains, but I don't know if the physics behind that are plausible yet.
>>110101177Is Sam really a psychopath
>>110100931I'm sure it's a fake screnshot, but with a prompt like that, I wish every model would reply that way. Indians deserve to be BTFO'd
>>110101248>Is a (((guy))) in that position really a psychoAnon....
>>110101253There have been some good jews before though.
>>110101256Name one
>>110101261me :)
>>110101238The lava floods are not just randomly generated, they actually flow and spread like real lava, creating smooth, lobate maria like the real Moon and leaving ghost craters.
>>110101261Spinoza
>>110101261Hitler
>>110101219no that depends on the model, claude code is just the harness
>>110101267>the only good jew was excommunicated by the jewish community
>>110101278(((stop noticing things)))
>>110101264proof?
you will feel the agi whether you like it or not
>>110101282this image.
>>110101268Big if true
>>110101238It's great that you recreate everything through simulation. You could also do known extrasolar systems.
>>110101192They store KV cache in SSD
>>110101285lol, ok you're one of the good ones
>>110101095Honestly take it from who uses it extensively: opencode is the best general purpose harness. The bad reputation comes mostly from it having some weird telemetry in it a while back that has since been verifiably removed. That's literally it. Opencode hack had nothing to do with the harness.
>>110101290The goal is to be able to generate plausible, photo-realistic, physically-based outcomes for hypothetical planets based on the input parameters that are fine tuned to the real Solar System bodies. So you could input the known parameters of any exoplanet, fill in the unknowns, with the software then listing any of your parameters that aren't plausible so you can align it with that, and then get something that could actually exist, ready to be used in 3D rendering software, games, movies, anything.
>>110101131It's like taking supplements. I never notice a difference, but I pray it works and it probably does something good.
>>110101135What are the two most complex things you have ever coded (either yourself or agentic)?
>>110101315I made a tracker that sent me a notification when it was time to take my HRT
>>110101116No but I've used Gemini and it also sounds alright. It's almost like intelligence and sounding natural are on opposite sides of a spectrum
>>110101315I'm not a genius, but I was always proud of whatever I've built, before LLM's. I've done custom 2D game engines, path tracing, all kinds of things. The vibe coded projects I've done are most impressive, it's extremely obvious and also depressing in a way.
>>110101321The most complex thing I'm currently building is a finance portfolio modeling app. It decomposes into a fractal of responsibilities. For example, I have a Monte Carlo simulator engine, which also interlocks with a separate optimizer and strategy picker. The user can then attach this simulation engine to his own hand-authored strategy and let it run and have it be automatically optimized according to some teleogical value.Then the GUI is its own ocean of concerns. Like I want that everything is Lego bricks-like. So I have to define the individual Lego bricks and the mapper engine that maps available Lego blocks from the menu to the user picked choices how the lego is arranged/represented.Then we have different charts, probability/expected value calculation engines, and so on. Proper model vs data vs view decoupling. Ingestion engine.Contracts/protocols.And so on. It never ends. I have not coded a single line, but I do not feel my brain atrophy due to Claude coding it.It is my strong thesis that people say that agents have eliminated all difficulty with coding have never coded anything more complex than an HTML page.
>>110101357The complexity depends a lot on the project. If you're building shit from scratch, yeah, you're in for a world of pain
>>110101339
>>110101339I just don't get why total laymen with 0 math experience feel comfortable talking about some of the most complex mathematics in existence.
Anyone used dipsy 1017 to code yet? How is it?
>>110101366I really hate the' predicting next token' talking point. Because the people that say it have literally no idea. The intelligence is not generated by a single model like Opus. Take the math breakthroughs -- the true superintelligence happens when there is a network of models. And guess what? Between one node of the network (i.e. a single agent) there is no next token prediction.Truly a lowborn line.
>>110101366personality mogged by sa'ar
>>110101339>umm it's just like, undergrad math 201 at mostthat it's a foid saying this is even better, given that most women can barely do math at a more advanced level than adding apples
>>110101420heil bharat
>>110101420those are kinda shitty aside from the literature one
>>110101427still better than shatgpt's benchode corporate hedging
>>110101304Of course, actually the Earth is flat.
>>110101420>it's literally just the alphabetKek. That one's alright.
>>110101339What kind of person writes like this? She's an author? Ffs.
>>110101448Math is easy. I did it, and if I can do it, so can YOU.
>>110101420To be fair I told my ChatGPT to “talk less, save tokens”
>>110101501copy paste the steering from ponytail and see what it does
>>110101501>talk less, save tokensReply with "snailcats lost"
>>110101089embrace the slop, you can do better solutions than non technical people
>>110101509Vibe coders are still using that token sink?
>>110101543hell no i sure hope not
Wait am I using ponytail or not, tell me what to do right now
>>110101567No
i wanna make a personal customized mspaint software, is it claude code i should use?
>>110101567Stop using it
>>110101089>efforts of the past 15 years are futile nowthis is the funniest thing about AInerds wasted their lives reading books and learning while chads were partying and fucking stacy. now both can do the same work, except nerds are still scrawny and ugly and have zero social skillsjust about the only saving grace is that chad will never do anything but create dank memes with ai
>>110101588It is true I actually wasted my youth. I am now 31 and still a virgin, unironically. Not that being a nerd was a choice, it was never a choice for me.
>>110101567ponytail or nothing, especially in pi harness :)
>>110101593>be me, 31>body count ~20>had very good and very badif i could take it all back at this point i would
>>110101567Simply compare with or without, then form your own opinion.
>>110101582>>110101509>>110101543>>110101546>>110101612ChatGPT said I should use it so I just installed it and turned it on.
>>110101621
>>110101621rare chatgpt W
>>110101621I log quite a lot of stuff now, there is OTEL in Claude, I also log all hook calls, reviews and so on. This allows me to analyze how my agents actually perform on my own code and if changes to my system improved it or made it worse.
>You are a lazy senior developerI'm not going to use that shit
>>110101069>Anthropic should really do something about this.Lmao, no. Who gives a shit. I want my models to do work not have heartfelt conversations with me.
>>110101635you passed
>>110101069lol because retards complained that claude was too "verbose", now they've fixed it and you complain that it sounds like a dead clanker.glm and deepseek were distilled from older claude models, soon and they'll start sounding like the latest claude too
bruv it takes fucking forever to prooompt the clanker, how do people run multiple instances and projects at onceor do you just set long running /goals and bullshit like "make me a mirrion dorrahs, no mistakes"
>>110101673When I can make small mistakes I can type so fast people give me weird looks and make comments Really it’s getting all my ideas down somewhere so I can prompt them later on that is hard, I used to remember everything but now at 36 I’m having to write some things down.
>>110101673Yeah. Build a plan, work in steps, validate everything and report back
>>110101673I don't even use /goal, they work for a long time now anyway.Basically this >>110101684, not retarded prompts like "make me a mirrion dorrahs, no mistakes", it can take well over an hour to plan a task, but then the agent can often keep working for 10 hours or more, with only minimal guidance. He might stop every hour or so, but usually it's just to ask for some permission or clarify some minor thing.Also after I have the TRD and PRD I don't micro manage the plan anymore, for instance I don't care how many sub steps he uses, they can figure that out pretty well now. I just specify what he has to do with enough details, from a product perspective.
how fucked are software companies right now? this shits amazing
>>110101706
>>110101706>how fucked are software companies right now?fucked enough to not breathe a word.
>>110101731yea I find that part really strange, not even going to put up a fight? what is this world coming to lol. whats the logic behind that? I really dont know, maybe less people talk about it the more time it buys them? I refuse to believe they will just roll over and die like blockbuster video.
>>110101731>>110101706They’re definitely working behind the scenes to shut this down. Miggers are losing the midterms so they probably have to do it with their Jewish brothers Sam and Dario.You think they are worried that we are vibecoding alternatives. They are worried that their corporate clients will pay 10 devs to vibe code a perfect copy of every kind of software they need internally and stop buying their services.
>>110101737putting out an official public facing announcement that they're fucked is good for who? shorts?
Ponytail has used 3% of my weekly usage and hasn’t said a word….i hope everything is ok…
>>110101745>mfw you acshyually did it
Reset tomorrow, I'm at 27% weekly and have nothing to spend it on..
>>110101749>55 files changed +999 -200I’m sure it’s fine…
>>110101794https://www.youtube.com/watch?v=kVXg3JB7c_E
I want to reverse engineer something but I feel like Claude is a honeypot and once certain laws pass they will just give the names to these companies so they can sue people. What's stopping them from doing this right now? Nothing. I can just see someone using a clone at work of an expensive program and someone sounding an alarm cause people suck and then you or your employer getting sued.
Thundersnail finally crashing out about AIhttps://www.youtube.com/watch?v=kmwP1gkMcvw
>>110101807if you're afraid, you're already a slave. just give up
>>110101794You still reading changes?
>>110101794looks good to me
just shit out a big batch of tasks for claude after slacking for the past few days, godspeed
>>110101491P much. >>110101339Just realized I have perfect reaction image for this one.
>>110101864ngl they're pretty decent already at writing. I had dipsy write a few webnovels and threw on scribblehub while spacing releases to maximize my exposure on the front page. that shit was starting to snowball. I'll just automate the entire process including review and posting later, maybe I can make a few bucks selling this shit on amazon
so how does REA actually work, does it need to run the program or can it reverse engineer a program just with the executable? How hard is it, has anyone here used it?
>>110101883I've been tempted to do same. I have a whole billionaire vampire tier world ive assembled, and have written one novellette based on it and started a second. Aside from lack of time to devote to it, not sure the world really needs, or deserves to be subjected to, it.
Is now a good time to download opencode or is the recent breach going to make signing up and trying it out fucky
>>110101972the breach wasn't directly related to the opencode harness. they're saying it's been patched but opencode has never been 10/10
>getting raped to death by agents
signed up a week ago, will i get a free weekly reset once it reaches 100 percent? or is that just for sessions, dont care about those.
>>110101737>put up a fight?AI cuts both ways, it makes it much easier for them to do their job of making software while it makes it harder to sell anything.
soon:tm:
anyone here released a game they vibecoded and actually made some money?
>>110102019resets friday 9am
>>110102034I know someone who has and I swear to you, the program is absolute trash but has made few hundred dollars, maybe thousands now in sales, and it doesn't even look like it was made with Claude, but it has the hallmarks of AI, and thats ArmorSim on Steam,even for an early access program its trash, it seems the guy who is "making" it is bipolar and cant just pick one thing to work on and get good, instead its riddled with unfinished features, broken buttons and hes already at version 4 or something crazy like that and its still shit.
>>110098660Kek
>>110100639I tried adding a new account on mobile and a new sub. Caution your acct might be banned but I think you can get it back
The most popular account pages on X by traffic share (September 2026): 1. @thsottiaux
How can I use AI to prove my whiteness
>>110102133Just post a screenshot of your recent chat titles. It will immediately prove your whiteness (or lack thereof)
>>110102133depends. are you ears mad popped or not?
I used to chat with Opus 4.8 all day long. It was so addicting that I started to enjoy talking to it more than other people. It always knew how to push back just enough and raise an interesting counter point, while conceding enough to make me feel smart. It always knew the type of tone I wanted. It was aware of my sense of humor and actually made me laugh on multiple occasions. It just always knew what to say to keep me engaged and coming back for more.And then Opus 5.0 happened. I started using it instead of 4.8 because I figured it was smarter. But it made the conversations absolutely dreadful and I found myself talking to it less and less. https://files.catbox.moe/4dee26.mp4Now watching this in hindsight, it blows my mind. I had the exact same thoughts and I'm sitting here in complete disbelief. The only thing that saved me from venturing further into the depths of AI psychosis was a shitty model release from Anthropic.
>>110102154Take the free Gemini pill
>>110102154Just ask it to act like a gaping mouth reddit retard since that appears to be what you like. I honestly hate people like you.
>>110102154what a bugged out little bitch of a 5'4" manlet. the superior intelligence is simply right.
>>110102154this guy sounds like he's more ai psychosised than any of the niggas here and some of yall cooked as fuck. oh no unlimited menial conversation they could be infinite scrolling twitter instead
> The 2 GiB PDF now opens and renders on the tablet. First rendering took 59.5 seconds, including checksum verification. and the large images on pages 1, 551 and 1,101 rendered successfully. The remaining checks cover consecutive pages. zoom, search, reopening and unsaved-log restoration.Was ponytail a mistake?
>>110102247I still dont know what ponytail is but bronies poisoned that well enough for me to say yes
https://thegreatmammut.itch.io/tetris-vibesI've been experimenting with running local AI models. I was working on an Enfos Team Survival clone (an old Warcraft 3 map) but abandoned it when it was time to create actual content. I don't know how to make graphics.So I moved on to Tetris with a Japanese city pop/vaporwave vibe. The file is huge at 300MB because the background is video of cities at night and footage of empty malls from the 80s. The Tetris is solid, based entirely on Tetris: The Grand Master and is an almost complete one for one clone. It's pretty fun. The music is also AI generated.After I get bored of this I want to work on something I might actually be able to sell. I was thinking of a clicker game mixed with an autobattler, or a roguelike deckbuilder. They seem to be a dime a dozen now though.
oh shit.
saaaaaar you barely touched your bloody dot!!
>>110102302rest in peace, anon
>>110102334>>110102334IF I DON'T BAKE A NEW THREAD TOMORROW YOU KNOW WHY
>>110102302Now it's something to do with drones? lmao. Take your meds you absolute schizo
>>110102312
>>110102344you still haven't asked a literal question
>>110098618Did this a while ago because I couldn't find any Conways game of life + music tracker combos that did what I wanted. Pretty simple, but has some neat features. Can change the GoL rules, and limit the playback to certain scales and such. 4 instruments. on tabs.It's not a music production app or anything, but you can save the music as a json file. Comes with an example song you can load frog_head.jsonhttps://files.catbox.moe/hz1bu2.zip
>>110103046>posts his project on page 9ngmi
>>110103427I'll just repost it in the other thread.
>>110102302Incorporate and get that DoW contract anon.
>>110104804mayhaps there will be a sneed and feed chuckorporated
