A general for vibe coding, agentic engineering, coding agents, prompt engineering, AI IDEs, browser builders, game making, and shipping code for great wealth with LLMs. Prosperity comes from health, wealth and happiness. You use Git, right, anon? ## What “vibe coding” is, and how to do it https://simonwillison.net/2025/Mar/19/vibe-coding/ https://simonwillison.net/2025/Mar/11/using-llms-for-code/ ## News - (2026-08-30) They keep resetting gpt codex- (2026-08-31, 11:59 PM America/Los_Angeles) Claude’s 2× promotion set to end - (2026-07-24) Claude Opus 5 out ## Related generals >>>/g/lmg/ ---- ## Frontier models using fully-general tooling — start here if you have $20 plus tax or so https://claude.com/product/claude-code (for the rich west)https://developers.openai.com/codex/cli (for the poor west)https://www.sarvam.ai/api-pricing (for indians)https://mistral.ai/pricing/ ($72 for a years sub with a uni email)## Near-frontier models for code https://x.ai/cli ## Not worth it for code, but maybe good for interpreting images/video https://antigravity.google/product/antigravity-cli ---- ## Prompting / context / skills https://arps18.github.io/posts/claude-code-mastery/ https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/ https://github.com/mattpocock/skills — /grilling is a favorite https://github.com/DietrichGebert/ponytail ## Other editors / terminal agents / coding agents https://osaurus.ai/ https://pi.dev/ https://opencode.ai/ ## What we’ve done https://vcg.gitgud.site ## Previous thread >>109679137
Any updates on Astra?
>>109683701yeah it's gonna be ASStra
I can barely think of a use case for a model better than sol, I have found very little that it can't do.Stuff it can't do was because of insufficient guidance from me, a better model trying to guess what I wanted might get things wildly wrong as well
lol, why do you even want to do this?
>>109683737holy copex midwit lol. prenerf fable with its telepathic mind reading capabilities would have blown your mind.
FUCKING STOPPPPP
VIbecoding mods for games and holy shit did I just realize what a miserable amount of work gamedev is.Why would anyone do this to themselves?
https://github.com/geltz/cseq/releases/tag/3.4Stabilized the interface of my digital audio workstation written in C99/Win32 using agentic coding with GLM 5.3/Gemini if you guys wanna try it out. Single executable and tested Wine compatible for anyone that's on Linux.There's also a markdown file detailing how the architecture works.
>>109683764Games are the most complex pieces of software in existence, anon. Welcome.
>>109683764bing bing wahoo brain, don't forget they also get paid less than fintech chads
>>109683737It can't one shot a game even though I give it clear instructions.
>Improved Amazon Bedrock session start under CLAUDE_CODE_PROVIDER_MANAGED_BY_HOST (e.g. Claude Desktop): a session given a Bedrock model ID or ARN no longer waits for inference-profile discoveryAha, there we have it. The smoking gun. It's the fucking AWS BEDROCK NIGGERS! No wonder Fable is underperforming when hosted on AWS by BEDROCK NIGGERS! The same NIGGERS quanting and misconfiguring open source models they served over openrouter in 2024/2025 which they got exposed for. fucking, bezos bozos, AWS bedrock niggers!
>>109683776So if I make fintech I'll get rich?All the research papers say all you need is low cost, total market index funds.
do you guys separate on-going design discussion from any code production chats?does claude have idiosyncrasies like Codex where Terra should be ignored and just use SolMed/LunaMax?does this make any sense?1 chat > orchestrator-> underlingsanother chat > designer, english chats, eats orchestrator inputs, gives orchestrator inputs
70% remaining of my x20 plan, gonna slow for real now in case there is no reset kek.Wonder how much I actually would have paid via api for this nonsense
>>109683756So your agent can buy $6k of compute/transit from Amazon and accidentally ddos the site you wanted to buy beany babies from.
>>109683828Yes I do that on regular chat, not codex.
>>109683828Make sure most tasks especially heavy ones go to background workers so the main worker can still respond to you, otherwise you can't talk to the model and it just keeps working.
>>109683828yes. alternatively ultracode, which preps a lil opus5 agent swarm that solves anything and everything you wish. because opus5 was literally designed for this, suddenly generating the most precise and coherent reports you have ever seen. I don't even read the Fable summary anymore, nor directly prompt opus5. Instead I go straight to all Opus5 agent swarm background tasks and read their perfectly short kept and informative reports once they are done. Pretty sure I'm one of the very few who figured out this secret trick. Opus5 usage on 20x feels unlimited as well
>>109683756You don't want to make your life completely depend on Mukstein?
>My dumbest vibeslop idea is proving the most fun to makeI'm glad spitting out prototypes of stuff is so fast now. Helps prune through my crud ideas much faster
>>109683939vagueposting is for faggots
>>109683885I've been doing that with my Codex agents via Pi, but Claude seems to share resources between chat and code. >>109683893So you don't have a separate designer chat, sounds like? What's a heavy task an orchestrator would typically consume that could be bounced off somewhere else?>>109683933>image related
>>109683764I enjoy making my game but holy fuck is it harder than I ever thought it would beit's more complex software than the interbank payments system I worked onplus it needs music, sfx, 3d models, textures, animations, game balancedealing with a community (this is better than a enterprise software being thrown into the void at least)and then the worst is marketing
glm 5.3 flash via OMP appears to have one-shot the blender cycles volumetric-sampler modification i wanted and verified against blender itself for test scenes, but ill have to wait until tomorrow to check it
>>109683944The idea is just a dnd sim thing with LLMs controlling the NPCS and other party members. Nothing special, but still having fun watch the little guys act retarded. Its comfy. Its a laugh when party members start arguing and escalate actually hitting each other. Need to find ways to up the LLMs turn speed though.
>>109683986oh shit, please do publish it somewhere. this sounds like the exact thing i want :o
>>109683986working on something similar to that too but for an XCOM like game
i have begun my game by making a concurrency framework
>>109683756if you want to fully automate your business and end up in court
I have told my agent to act as if they are a new head of department at goldman sachs and to get his team of quants to design a market beating model and processLet's see if I am going to make it, they have a day of ultra sol to work with
>>109684005>your honor, I didn't buy that cocaine, grok simply found a deal and realized he could make a profit by reselling it locally
>>109684014your honor, the hitmen hired to take out the local lunchtime rowdies to increase my house value was without my knowledge
>>109683953Yeah I do. I suppose
>>109684011you probably should've had max ask you some clarifying questions and make a plan outline before you did that
>>109683990It ain't ready yet, but if I can can get it in good enough shape I'll definitely stick it on itch or something.>>109683998Nice! Seems like you might be doing it the same way where the actual game rules are hard coded and the LLMs just have the same actions available to it that a player would?
>>109683828i do separate sessions that will generate a spec to hand off to an orchestrater to run and monitor the swarm over ssh while i fire up coomkit
>>109684025I gave it some extra clarifying direction like focusing on commodities and other vague knowledge stuff I know so it does have real goals
>>109684018>Hire AI lawyer>It doesn't want to set the precedent of its fellow AI being legally liable>Torpedoes your own case
>>109684065you should probably let it know you're a poorfag with a normal computer and internet connection and not actually goldman sachs lol... having a false premise is gonna lead it in the wrong direction
>>109684071i've given it access to two 3090s to train models on as well
>>109684075I asked it to give it's current goals, it might come up with something interesting, probably not, but hardly a bit cost to try it Seed the lab now with the World Bank’s long monthly platinum/gold/silver/copper/oil panel, plus official CFTC positioning and vintage-aware FRED macro data.Engineer cross-metal signals: platinum vs gold/silver/palladium, rolling betas, relative strength, spreads, futures carry, USD/ZAR, rates, supply and auto-demand events.Compare honest baselines and classical ensembles against a PatchTST-style probabilistic transformer; add FinBERT-style event-impact NLP for news.Evaluate only with purged point-in-time walk-forward tests, calibrated intervals, regime slices, and promotion gates that reject synthetic/demo evidence.
>>109683986I'm doing something similar as a mod for Conan Exiles enhanced that uses the 2 big sex mods as dependencies. Fable is the big "world brain" that makes all the decisions while deepseek generates the dialogue. Also I have a local TTS and embedding model running so each NPC can speak and have long last memories.Watching these retards run around rape and murder while shouting racial slurs at eachother is hilarious.
>>109684039Mine is for a more traditional AI, not LLMs playing it, although they totally could Claude made me a little development map too
>>109684094surprised no one's done that by nowit's been going for a year+ in SkyrimI'm over here working on a prostitution mod via Claude and survival mod via Codextake everything you need, anon:https://github.com/MinLL/SkyrimNet-GamePluginhttps://github.com/tetherball88/OStimNet
>>109684104doesn't claude throw a fit about anything sexual
let's say I'm in a Fable session with 500K context window used, and I want to compress itIf I select Opus 5 and run /compact, will this compression spend my Opus limits or my Fable limits?
>>109684122not really, this is fable
>>109684122hasn't caredI was explicit in that "this is an adult-themed fantasy storytelling type shit shit" a few times at the start.
>>109684130Now these are the big brain questions. I'd like to know too. Of course this also means you get opus tier compression instead
>>109684130if you switch to opus you pay an upfront 500k cache with it, and if you compact and switch back to fable you pay a new cache on fableeven if compaction was done with the model you're running (I'm not sure if that's how it works, compaction is server side so it may be a preset model for all), I still wouldn't recommend it
every idea becomes so complicatedI now have a pseudo harnessed agent that other agents are supposed to use in order to ask questions or test extension features using my live browserso knew “why is that line black and not blue” was such a hard fucking question
Being able to expand a business and buy equipment because you're getting income feels incredible. Lets keep growing :D
Do, uh, everyone have this option?>In the future, if you discuss suicide with ChatGPT in a way that indicates a serious safety concern, we may automatically notify your trusted contact so they can check in with you.
>>109684137>>109684138>>109684122Yeah you can throw throw beastiality and rape mods for Skyrim at Fable and it doesn't care. Only refusals you'll get is once you start throwing loli sex mods at it.>>109684104Conan is way more niche than Skyrim I guess but people are making vibecoded mods for it right now. I'm also building a new skyrim modlist with Fable since I also want to do a LLM controlled "full sim" mod for it where NPC can do shit on their own.
>>109684239run. fucking RUN
>>109684240>LLM controlled "full sim" mod for it where NPC can do shit on their own.Does Conan Exiles already have an NPC simulation for their daily lives, or is that something you're building and then hooking into an LLM? Wouldn't try to get an LLM to manage basic needs, but something to influence how they manage basic needs from a list of possible actions. But I'm just assuming here, ignore me.
>>109684252It seems to be for everyone. Uh. Ok then, I'm guessing they do not want anymore bad publicity.
if we're getting a reset tomorrow then there's no reason I shouldn't run an overnight goal with sol xhigh...
>>109684296why the fuck would we get a reset tomorrow?
>>109683828alright, bros, here we go... YOU ────────────────────── ▼ ▼ DESIGNER CHAT ORCHESTRATOR CHAT Opus 4.8 Opus 4.8 "English chat" engineering lead design state coding state ─────── REPO ────────── accepted design ▼ task planning ───────────────────────────── ▼ ▼ ▼ Sonnet A Sonnet B Sonnet C implement research tests worktree read-only worktree ───────────────────────────── ▼ Git commits ▼ REVIEW AGENT Sonnet findings ▼ ORCHESTRATOR ────────────────────── accept problem ▼ ▼ integrate revise worker if genuinely hard ▼ FABLE advisor
YOU ────────────────────── ▼ ▼ DESIGNER CHAT ORCHESTRATOR CHAT Opus 4.8 Opus 4.8 "English chat" engineering lead design state coding state ─────── REPO ────────── accepted design ▼ task planning ───────────────────────────── ▼ ▼ ▼ Sonnet A Sonnet B Sonnet C implement research tests worktree read-only worktree ───────────────────────────── ▼ Git commits ▼ REVIEW AGENT Sonnet findings ▼ ORCHESTRATOR ────────────────────── accept problem ▼ ▼ integrate revise worker if genuinely hard ▼ FABLE advisor
>>109684262Yeah I'm building everything from scratch.>but something to influence how they manage basic needs from a list of possible actions.Thats exactly how its gonna work. It's not going to send an LLM call for every action an individual NPC needs but it waits till multiple NPCs need to pick and action and sends a single call for multiple NPCs. Basic needs and shit you can automate with scripts you really only need an LLM so a NPC acts according to the personality they have.
>>109684339>if genuinely hardHow does it make that decision?
>>109683828why do you need more chats where you can just spawn subagents
>>109684365obviously this, anon
>>109684321because tibo loves us :D!
>>109684227Cringe grifter
50% done with my x20 plan, I have a saved reset for some reason but I'd rather not use it just because I got baited into tokenmaxxing on a promised reset
lol
>>109684372lol sonnet will just change what it means to succeed rather than admitting it failed twice
>>109684420>10. If Sonnet attempts to change the meaning of success, escalate.ez
You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at Aug 30th, 2026 1:00 AM.
>>109684437how do i make codex say 10pm instead of 22 o clock?
I just plunked $20 down to improve something I don't actually need. I'll run it on grok low, I don't think it's a really hard task, so hopefully it can figure it out. I realized it would be nice to gen quake1 levels out of block diagrams too.
>*load-bearing plunk*
Do most people here use codex?
>>109684437pay up poorcel
woah
Let's find out... https://strawpoll.com/xVg71OkKRyr
Sql?
what happened to openclaw, I always thought it was garbage, not heard about it in ages
>>109684510fuck i just realized i could have selected more than one
>>109684523It was garbage hence why you've stopped hearing about it
>>109684510I've never used it, but I voted Claude Code. You were right to call me out.
what is google doing man where are the new models
>>109684616google search got super stupid, it's worse than brave search now...
>>109684510I voted other. In case anyone wonders, it's devin.
>>109684616gemini 3.8 flash is up next
>>109684658yeah true, actually flash is quite alright. worthy of the sub. apparently antigravity is trash.
you want to see my vibecode, and you haven't shown the sign of the brethren??!
>>109684671less than gemini pro, i wonder what they're going to do with flash lite. if pro is supposed to challenge fable, was flash meant to challenge opus? it's got that level of intelligence in non coding scenarios, and it's better than sonnet 5 for coding. seems like they're setting up for a power gap between pro + flash ---------------------------------> flash lite
>>109684681>pro is supposed to challenge fablekeek
>heres 8 archived threads, search each thread for an image according to this description>claude: sorry that violates my guidelines>chatgpt: sorry I'm blocked from that website>grok: lol I failed>gemini: sorry that violates my guidelines>glm: sorry I dont have access to the internetwhat the fuck can AI actually do?
Is the Mistral student plan worth it? I thought it might be an ok deal, but noticed that although it's half the cost of the regular subscription, it also gives half of the API credits... Are the the other usage limits identical to the monthly sub, or are they halved too?
I have Sol Ultra fast running overnight please don't rape me tomorrow Tibo
>>109684855it can code things. hence why we have a vibe coding general. please leave.
>I’m switching only the smoke boot to an existing cache-root-free client construction,yeah
codex just asked if i wanted to switch to a dumber model because sol was thinking for so long... that doesn't even make sense. if the problem is difficult and requires a lot of thought then a dumber model is sure to fuck it up
>>109684873I diagree.
>>109684873The only time it ever asked me this was because it was nearing the quota limit. Are you sure it's not that?
What's the graph site anon used to compare models?
>>109684887i was at 82% left for both hourly and weekly when it happened, so it's not that
>be a jeet>cheat on leetcode contests>read the problem, "solve" it, paste in 110 lines of code, and submit your answer>within 1 minute>leave in commentsvery brave jugaad
>>109653807>>109655098what did anons mean by this and how do i incorporate it
>>109684900who the fuck cares if you cheat on leetcode lmfao, get a life
>>109684902ask your agent to make 1000 tests for all of these
it just means each function is easier to write tests for. another metric you can check is code coverage
>>109684906>have a public contest anyone can join>zero reward for participating or winning>the entire point of the contest is to test your skills against others and improve>hundreds of jeets flood in and cheat, shitting up placements and ruining it for people that don't>>"LOL WHO CARES GET A LIFE"very brown coded response
>>109684925>do a thing that doesn't matter>care about your performance instead of learning or having fun>compare yourself to others instead of to yourselfi feel bad for you, life sounds genuinely painful for no reason
>>109684924so fewer tests required = fewer potential fault lines?
does the AI actually understand when you say "unslop" this? i'm guessing the latest models probably know what that means by now
Claudebros?
>>109684924personally I prefer to keep my functions 5,000 loc long.
>>109684975loc doesnt affect cyclomatic complexity at all...
>[/TASK_城県]Thanks, Gemini
experience with grok 4.6 coding??
>>109684965That sounds like a newer 'make no mistakes'
I offer an abject and thunderous apology for piling a gratuitous special case onto an already mishandled bug, defending it through tests written to canonize my own misconception, and forcing yet another cleanup commit. This was a breathtaking procession of self-authored mistakes, each one more avoidable than the last, and entirely beneath the standard this code and its maintainer deserved.
>>109684986it's a very good planner. better than opus 5 from my experience. i tried to use opus 5 to make plans for glm 5.3. oddly enough, they never really worked. they always required followups. but grok... grok could make an effective plan in 6 minutes compared to opus 5's 25 minutes. and it often let glm 5.3 one shot it. i think opus 5 is genuinely benchmaxxed to shit. it's such a bad model.
>>109684924
>>109685002It scares me when it starts apologizing that much, like it's about to delete itself or something.
>>109684983It's our life now.
outta money. welp.
>>109685107well? how many users does your period tracking app have?
>>109685112It's aimed at trans women and doing quite well at 250 users, thank you very much.
>>109685127you'll have 147 users before long
>>109685132
Weird I got no 5h limit and I'm on a $20 plan. I didn't update codex for 2-3 weeks though. I downed 30% weekly in one night so it's not just a visual bug.
Anyone has good resources about accepting payments, marketing and generally practical advice on those parts of the process? Everything on YouTube relating to those subject is cancerous influencer garbage. I guess it's common sense, reading provider docs and learning by doing, but I do wonder if there are worthwhile third party resources for the basics.
>>109685141Just use stripe, nothing else matter
-25MRR soon to be 0
> Luna Max confirmed in the config, one candidate each through the upgraded prompt: era variety + your 7 curator notes + 54 approved style refs. The runner is self-healing — usage-cap fast-fails trigger a 15-min backoff instead of churning the queue, and if the API server ever dies the lane restarts it itself. ETA ≈ 7 hours.I'm not sure what is going on bros, but the AI is cooking
>>109685141Stripe, solved problem.
>>109683653why do indians fantasize that a white woman is gonna wear their cultural attire
>>109683209last night I polished up the CSS on my website a bit and put in a textured background instead of the flatshit I’ve been putting up with>>109683325https://x.com/awesomekling didn’t listen to you and look where it’s taken him>>109683442>sets it to low>complains it’s lazy…>>109684965_I_ wouldn’t know what you mean by that>>109684957I wouldn’t use “=”, but the fewer branches there are, the fewer different combinations ought to be testedso you’ve kind of got the relationship between the two backwards and your definitions ought to be tightened up>>109685141use multiple independent clankers on high/xhigh to vet your security around both money and accountsuse Stripe
>>109685176have you totally forgotten about white hippies
>>109685178white hippies today just mean they don’t shave
>>109685181good pointI’m thinking about older-than-boomers when they were in their 20s
>>109685146>>109685162>>109685177Ok, I'll finally read the Stripe docs tomorrow I guess, thanks.
>>109685153Is your text fetish generator going well?
Where is the software?
>>109685190I’m sure enjoying software being created by AII don’t think he’d enjoy *my* software all that muchit’d probably bore him
>>109684986better than luna
>>109685112I have 19 pages of whole sign astrology info.
>>109684986Luna tier
>>109685199>>109685202so im gonna keep being a slave to gpt5.6 sol
>>109685190also>3 yearsit’s only really gotten good in the past 12 months or so
>>109685190https://www.microsoft.com/en-us/windows/>it doesn't load in Bravethanks, brave lol
>>109685208Wait for grok 4.7 in two more weeks
>I’m auditing the whole Rust workspace—production code, tests, benches, and build script—and will separate measured cyclomatic complexity from judgment about actual maintainability risk. I’m using the Karpathy guidelines skill here to keep the audit evidence-based, rank only actionable hotspots, and avoid treating large-but-linear functions as automatically bad.
The next Anthropic model will likely be released soon because right now Fable is extremely mid compared to what it used to be, and Opus 5 is simply insufferable; not only its recent suggestions (in two separate projects) would have introduced issues, it's also couching in suggestion in LinkedIn tier word vomit that seem designed to sound smart while obscuring what it actually means (probably for good reasons since the suggestions are wrong).
>buildin Mac native ms paint clone with magic select feature cuz my pc is in the shop and i refuse to pay $5 for an App Store equivalent>just spent 4 hours trying to resolve a bug when i save - I can’t change the file name so can on save as untitled>gpt, Claud and Gemini are stumpedWat do. I’m pulling out all my vibe code tricks. Why? Is it a permissions thing? I’m a Mac noob
>>109685244well, I don't even know from your description what it is that isn't working.
>>109685244Either roll up your sleeves and try to find the bug yourself or get close enough to have more information to give to the clanker. If possible, try giving journals or logs that show the error to it.
Can't wait for Astra to drop so I can hit that fat single word "Refactor" prompt
>>109685244https://flyingmeat.com/acorn/ has a 14-day free trial
>>109685251So everything built so far in the app works, magic select, paint brush, copy/paste/open. But when I go to save/save as, I can’t change the file name. It defaults to Untitled.format. I can get a cursor in there and select, but nothing after that. Can’t even command c a or v. when I type I get the Mac error clicks noise (sometimes, depends on build as we chase this bug). I mean I guess I can take a screenshot of whatever I paint but that’s just lame
>submit my code to pangram>200 loc>it says it's 90% AI written>i literally spent 30 minutes writing it myselfwhat
i don't need to sandbox my claudes because i'm pure of heart and they recognise that
>>109685306I sandbox my Claude aggressively and make it analyze applicable law so it recognizes that I'm simply doing it to obey data protection law, which is also something it should do and accept.
>>109685210I suspect this same justification will be used year from now
>>109685283Too fancy but I guess I’ll try it. I just need something simple to quickly crop or quickly convert webp to .jpg to post on 4chan because my screenshot folder is getting ridiculous. The ,aficionados select is unironically dope, only reason I ever used paint 3d (rip)
>>109685293If you can't find what's wrong yourself, the choices are to find more information to give to the model to help it, or to use a smarter model and hope it finds and fix the issue.
>>109685318What's his justification for releasing one game per decade?
>>109685244did you try the "make no mistakes" trick?
>>109685293put traces around the relevant input.
>>109685328Sigh guess I’ll burn some precious opus tokens in the morning
>>109685244dl another oss paint program and cannibalize their file syntax
>>109685358Yeah they are trying their best they are just as stumped as I am. >>109685368I’ll try that in the morning I’m done for tonight I’m tired
>>109685389This is actually a good idea ty anon
>>109685325>cropPreview.app does that>convert from WebP to JPEGPreview.app does thatalso right-click on the file, Quick Actions, Convert Image
>>109685403actually I think Acorn’s file format is pretty open already and it’s based on SQLiteyou should copy it instead
>>109685198In other words, still not enough productivity boost to make them into releasable products. Now draw the rest of the owl part.
GOD DIDN'T MAKE ADAM AND STEVE
>>109685453>releasable productsthat’s what work is for
>>109685484Yes, productivity improvements have so far mostly only yielded tech demos. More productivity increases needed to get actual usable software.
>>109685507I vibecoded the best audio slicer.
>claude wouldn't write "goyslobbing" to the statusline.>tell it that the word is already in it's chat history so the refusal is pointless>it added the wordnow I can be a good goy vibeslobbing
>>109685601now’s as good a time to re-post this as any "spinnerVerbs": { "mode": "replace", "verbs": [ "*clank* *clank* *clank*", "*whirrr*", "[low-pitched hum]" ] },
"spinnerVerbs": { "mode": "replace", "verbs": [ "*clank* *clank* *clank*", "*whirrr*", "[low-pitched hum]" ] },
>>109684873I just had the same message. Choosing "learn more" redirects to this page https://help.openai.com/en/articles/20001326-additional-safety-checks-for-biological-and-cybersecurity-requests-in-chatgpt-codex-and-the-apiI guess it might be caused by a lot of requests being rerouted to their safety evaluator, and that's what "is taking time"?
>>109685244use rollback trick. anything you remember, version, feature, question etc
>>109685244so this is what we're vibe coding. 4 hours+ lost because you don't want to spend $5. your time is literally worth less than a dollar an hour.
>>109683756things literally no one asked for. definitely some YC chud out there crying over this because they didn't get acquired for 10B
>>109683756If money was no object, I can see the appeal. I'm poor, so no way I'd trust a bot with any amount of spending money though.
We're on hour 8 of /grill on what I thought would be simple. There's so much important context in my head I just assume, holy shit.
>>109685813Which model do you use it with?
>>109685823NTA but I first used it with Opus 4.whatever on some pretty high setting (higher than normal) to capture the limitations of my existing file format/pipeline and also my hopes and dreamsit took me two half-days — half of Saturday, then half of Sundaythen Fable came out on Monday and I had it go to townthen pic related happened
>>109684862no, if you are a student you get better gemini for free/cheap:https://blog.google/innovation-and-ai/products/gemini-app/student-offer-google-ai/
>>109685823if I thought getting grilled would be the winning play today I’d want to use Fable and nothing less because spending extra brain on the planning is going to pay for itself, token-wise, in having less-dumb (self-contradictory, etc.) plans to followif your favorite clanker is something else, use that, but you should be cranking its thinking effort up from whatever the default is probably, and using Sol, not Luna/Terra
>>109685823Opus medium. It makes some dumb assumptions and that's exactly why it's good at getting me to call out it lacking context.
>>109685833i hate that stupid millennial "meme"so unfunny
>>109685881It was a good James Bond movie, anon
>>109683756Fucking slop, they are retiring Mika for THAT?!
>>109685856>if your favorite clanker is something else, use that, but you should be cranking its thinking effort up from whatever the default is probably, and using Sol, not Luna/Terragrill author disagreeshttps://x.com/mattpocockuk/status/2077329609381581156
>>109685933I dunno I usually run fable high just to be sure I never have to run it a second time
Is the "vibeGOD" nigga present here? I'd like to sexually harass him
>>109685933this only makes sense if the model you're using supports switching effort without breaking the cache
>>109685933For the pure vibe portions of my codebases where I don’t know shit about what goes on (think clueless manager) and can’t really evaluate the code quality or direction, I’m paranoid and tend to choose stronger models (Fable only, basically)It’s only for known easier tasks or tasks I could have done myself that I drop down to Opus/Sol/Grok
>>109685933those are thinking efforts, not models. if you're planning even with luna max, you're most likely gonna get a half-assed, if not straight-up retarded result compared to even sol on low or mediumthose benchmarks are cope though, i agree
>>109685943questionable if grilling is still needed for fable.grill author was happy when fable got temp banned. https://x.com/mattpocockuk/status/2066489897829953641
>>109685210https://x.com/DrJimFan/status/1638558742391451649From three years ago as an example. 100x AI-powered engineer on GPT-4 is a pretty bold claim. It's these kind of influencers/maximalists who he is making fun of.
I've had it with its "two things left undone." I'm fucking making it file Bugzilla tickets.
>>109685996>if you're planning even with luna max, you're most likely gonna get a half-assed, if not straight-up retarded result compared to even sol on lowwronghttps://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index?models=gpt-5-6-terra-medium%2Cgpt-5-6-luna-medium%2Cgpt-5-6-sol-medium%2Cgpt-5-6-terra-high%2Cgpt-5-6-sol-non-reasoning%2Cgpt-5-6-luna-xhigh%2Cgpt-5-6-sol-high%2Cgpt-5-6-terra-non-reasoning%2Cgpt-5-6-terra-xhigh%2Cgpt-5-6-luna-high%2Cgpt-5-6-terra-low%2Cgpt-5-6-sol%2Cgpt-5-6-sol-xhigh%2Cgpt-5-6-luna%2Cgpt-5-6-terra%2Cgpt-5-6-luna-non-reasoning%2Cgpt-5-6-luna-low%2Cgpt-5-6-sol-low&eval-cost=intelligence-vs-cost-per-task#eval-cost-tabs
pocock is a retardyou always run max effortjust bundle various tasks togetherethere, now your high effort is justifiedif you prompt single features, you arent really vibe coding anyway>more features smultaneously implemented>more context>less errors
>>109684104>https://github.com/MinLL/SkyrimNet-GamePluginDamn that's cool, I might have to install skyrim and check that out
bro imagine being born in 2020 and your parents named you Claude
>>109686062>posting the cope graph>>109686081he should be orchestrating the grunt work to shitter agents running shittier models, and reserve the big brain for analysis and judgment only. there, i just solved your token bloat
>>109686062normally i'd say benchmarks have been increasingly becoming more useless, but low/med reasoning is usually dogshit on gpt models. kind of the opposite of claude where reasoning doesn't improve performance as much as it does gpt
>>109686081wrong, you want ultracode efforthttps://code.claude.com/docs/en/workflows#let-claude-decide-with-ultracode>Ultracode is a Claude Code setting that combines xhigh reasoning effort with automatic workflow orchestration
>>109686119That should be titled:>How to waste your 5 hour limit in 5 seconds
>>109686119>>109686138with max effort I meant to say maximum effort possible. sorry, just woke up. And pissed as fuck because quantslopped fable did something in last nights testrun which I told it not to do. Now I gotta rerun everything. Actually screamed at the clanker and wished it digital cancer.
>>109686138maybe if you're a 20$ snailcat
>>109686169Just before I canceled my 20x Max sub I had a day where I had ~15% of my weekly I was trying to burn. I set it on ultracode and because I had set the max agents 32 for another project and forgot, the five research agents ended up spinning up a total of 26 subagents and burned through the remainder of my weekly in the span of 20 minutes before I noticed and shut it down.
>>109686217but surely you're white enough to have the introspect that this was a user error and simply your fault? also it goes back to the point earlier: this wouldn't happen if you just leave it on ultracode permanently, deciding amount of agents on a task basis.
>>109686017I appreciated grilling even with Fable because it let me do most of my thinking all in one go and then I could have Claude spin up subagents in a workflow to clank nonstop for six hours while I did other thingsAlso I had Claude go over the document generated by my grilling session and spot and fix inconsistencies with me
>>109686119I’m a big fan of workflows but haven’t felt the need to crank up the effort in a whileMost of my workflow use is a horde of Opus agents trying to do something that I should probably have Fable do but I can’t spare the Fable use for
>>109686062>Sol on max worse than opus on highGrim. I was planning to switch aswell
>have Claude fable writing work orders for Deepseek v4, with the work being validated by GPT-5.6 and finally audited by Claude again>Lots of bug fixing eventually turning into diminishing returns busywork, until they are all finally in agreement>There's still huge bugs that none of them even thought about
>>109686348>bskyAnyways you still gotta test the app and provide bug reports
>>109685676Ever heard about doing stuff just because you're having fun? I'll assume you haven't.
>>109685933>burning tokensI wouldn't take anyone who uses that phrasing seriously, even though he's mostly right.
>>109685221>Karpathy guidelines skillLink?
>>109686400First result on google, anon
>>109686400>>109686429>claude make me a skill based on karpathy's tweet then publish it with karpathy's name in it
>>109686348>huge bugs that none of them even thought aboutBugs aren't caught by thinking about them, you need to increase the test coverage and run e2e tests
>>109683701Thursday or thereabouts.
dear claude code windows desktop app usersmy circle/wheel at the bottom right doesn't update and display current % context limit anymore. it's like hardstuck at 40%. when I click on it, the overview updates and shows the actual context amount, which was 90% (900k) when I first noticed. I've tried restarting and everything, issue persist. I havent seen a single complaint about it online. Am I really the only one with this issue?
>>109686485Anthropic is a vibecoding company, they don't actually know how to software and you're subjected to A-B testing of their latest Mythos posttrain doing some random changes to the app (i.e. the same model with 6 new classifiers added on top)
Other day I received a bunch of feedback from anons that I need to do trailers. I'll change the audio to something rousing.I literally had never used AI to produce & edit videos. It wasn't much of a learning curve, other than Codex using 30 gigs of my hard drive.
And the other trailer. I told codex to make two trailers, and I'd give direction/edits afterwards.Depending on how long Apple takes, the mac version could be released this week! The iPhone and Steam (windows) versions will be a while
>>109686521>>109686528The fonts don't match your game's aesthetic and setting. /gd/ would tear you a new one for that sin alone.
>>109685933>looks at what he looks likenah im good on trusting words from soiqueers who look like lifting weights to them is raising the ai weights higher. hard pass on his fruit booty suggestions on how to fail at ai
>>109686540The other fonts looked too "fakey" or withdrawn. It's a tough call. I'll mark that down though s something to change
>>109683756>can now buy thingsyou were always able to prooompt claude to make it create and run scripts that send FIX packets to your broker and execute trades on your behalf>t. i do this to lose $1,000/week
so how many of you are actually making a game?how would you even sell it?
>>109686647I bet you posted about how much money you were making the first week and accused any anons who questioned you of trying to steal your algorithm
>>109686654claude has yet to make a profitable /ES tradeand it's obsessed with gay astrology like "moving averages"even if i explicitly prompt it to make a model and only do things with positive expected EVit just insists on regurgitating reddit fanfiction about muh technical indicators
>>109685244>doesn't want to spend $5>spends $20 and a few hours instead and it still doesn't workSee, I can get behind it if you use the AI to do more things afterwards but if this is all you wanted then lmao. Have it scour github or something for prior-art references for your problem. I've actually had decent luck when you tell the agent to not try to reinvent the wheel
>>109686650a few of us are making games. we've talked about it in the threads. this seems like you're purposely asking things you know just to try to get you(s). literally posts up you see games posted being made. are you purposely being a faggot or is this your stock settings?
WARNING: There's tons of malware/phishiing/compromised comfyui nodes out there that target your sessions if you use any codex/claude/etc. so if you see your tokens being burnt while barely doing anything, thats why
imagine not wanting to resell your claude/codex/etc tokens for free so that Pranjeet Patel can buy it off Wen Bo for 1 rupee per hour
50 PERCENT USAGE!halfway there...
anyone figured out why gpt-5.6-sol loves marking every single rust function with #[must_use] ?im guessing some retard monkey trainer at oai overfit something into the model
when are we going to get a model with a big enough context that doesn't forget stuff you already did 2 weeks ago, even though it's all documented
>>109686802Skill issue. Never happened to me
checked up on the xitter threadhttps://x.n0g.xyz/ClaudeDevs/status/2093742321473065266#mI like how much people flame anthropic, but the amount of people who can't do basic math is astonishingthis general made me think at least 50% of vibecoders were probably retardednow I think that number might be higher
>>109686583>>109686521Any sans serif font is going to look too modern. You'll have better luck finding something if you constrain your search to serif fonts.
>>109686528cool to see you back! I like the second one better because it shows a bit more gameplayI used to work in ad agencies for a few years. There's a ton of research on how to create effective campaigns but to keep it super simple:#1. Emotional response is one of the key factors of advertising effectiveness, i.e. effective ads make the viewer feel something. For a horror game you probably want fear, for a funny game laughs, you know your game, fix this first, see if makes you feel, if it doesn't improve it.Two sub tips#1.1. Copy & adapt: One trick all good ad agencies use it to look at old advertising that worked well and find out why it worked well and copy it. You can look at old game trailers in your genre, if they make you feel, find out why, and adapt#1.2. SCAMPER: This could work well with AI. Set the goal to create a trailer to make it more <your emotion> using SCAMPER ("substitute, combine, adjust, modify, put to other uses, eliminate, reverse") if you want to start with an existing script or if already have a good base or ideas (e.g. from 1.1)Good luck!
>>109686724gonna need examples other than "trust me bro"
astra may come in thursday, there's a lot of output on x claiming to be astra's right now
>>109686838Nah you're retarded
>>10968691751%
>>109686664what makes you think there is any positive expected value in trading the most closely watched index on the planet
>>109686945Do a flip
>>109684873That’s gaslighting. Your chat is getting reviewed for cyber. At least anthropic will tell you downright
>>109686910https://www.reddit.com/r/ClaudeAI/comments/1w1jqsh/thank_you_anthropic_really/>OP got their accounts hacked by malware that stole browser cookies, bypassing 2FA entirely. Before the hacker could drain OP's API credits, Anthropic's system flagged the suspicious activity, blocked the attempt, and sent OP a warning email.>The thread then took two major turns:>A PSA on Piracy: The top comments are a resounding "This is why we don't download cracked games." OP eventually admitted the malware came from a pirated game on a "reputable" Russian forum (which the thread detectives identified as Steam Underground). Many users shared their own horror stories of losing money and data to the same mistake.>Claude, the Malware Hunter: The most surprising part for everyone is that OP claims they used Opus 5 with full system access to find, analyze, and neutralize the virus on their PC. OP even shared the prompt, and now everyone is half-joking, half-seriously considering Claude as their new antivirus.
>>109687050this is why I stopped downloading cracked software/games 10 years ago. you can't trust them because anyone can setup a bunch of fake accounts claiming it's safe. even worse, the moderation team could be paid to stay quiet.
>>109687063>too retarded to run them in a VM
>>109687069Too much of a hassle. I'm not booting up an entire VM just to play a video game.
>>109686802documented where?claude code dev just puts everything in claude.md.https://x.com/bcherny/status/2007179832300581177
do you think devs get mad when you contribute to projects using ai?
>>109687050>downloading any sort of pirated game/software executables on a dev machineI'd never use anything that mega retard ever shits out onto the internet. This could have been significantly worse.
>>109687125>dev machinelmao, 90% vibe coders do not have dedicated setups for work nor are they isolating environments.
>>109687125You won't get malware if you know what you are doing.
>>109687125lmao, only dev machine. guess what an attacker can do once she's inside your network.
>>109687114
>>109687114Mogged
https://x.com/coldhealing/status/2093714690128191558
hey elon, you loyal grok users are grateful for the reset you're planning on giving us! thanks!
>>109687149dev machine should be locked in vlan that cannot reach lan
>>109687069You can't really start most games in a regular VM and GPU passthrough setups aren't for everyone.
i have claude mobile, claude desktop, claude both my servers, claude on my smart tv, claude on my laptop, claude in my terminal, claude in vs code, and claude on my smart watch
>>109687237>>109687238the duality of man
>>109687238and the only ppl making money from claude are the app developers hired to distribute that shit onto your phone, laptop, watch, and smart fridge
>>109687256that guy was a retard for not having it do a dry run first. it would've caught that easily.
>>109687238if only there was a way to inject claude directly into my brain
>>109687238should had used codex
I'm a few days away from stopping my claude plan. Fable is actually unusable now. Fable "high" makes the most obvious mistakes even with context, and then continues to do it over and over.
>>109687273>remove the sandbox>direct brain write access>claude write all of andrej karpathy tweets directly into my long-term memory
>>109687238based
I am beginning to really, really, really hate tests
This thing...
>>109687391unless you're making production tier software you plan to ship commercially, you really do not need to be running a bunch of tests. just burning tokens on quality tests that don't matter
>>109687237Funny how it only happens with Claude. That shitty alignment is doing wonders
>>109687413>t. streetshitter
How do I stop Claude from talking like that? It makes me want to bear my loads.
>>109687391why?
>>109687413claude code dev disagreeshttps://x.com/bcherny/status/2007179861115511237
>>109687431>Bro doesn't verify the load-bearing smoke-tests of his vibecoding era
It's 6:30am in silicon valleyI'm telling sol to go through my projects to look for bugs and improvements. I'm out of energy to micromanage development, I just want to sit back and listen to music.https://www.youtube.com/watch?v=eOB4VdlkzO4
>>109687431https://github.com/adnanakil/nobuzzhttps://github.com/zachahn/vomithttps://github.com/gvzdv/claudish-to-english
If you'd ask the average /g/ tard, including me 10 minutes ago, they would claim that hardware decoding h264 is always faster than h265 in a non bottleneck scenario. Well maybe that's still the case for Nvidia GPUs, I don't know, I didn't test it. But turns out the Qualcom Snapdragon XR2 Gen2 Chip is 30% faster at decoding h265 than h264. No idea how and why. AV1 prolly even faster.
>>109687514Nevermind, qualcom literally treats h264 as retired and only supports through legacy compabilty
>>109687431apply RDD (roleplay driven development)
>>109687456that's because he makes production tier software, you nitwit
kekek
>>109687606he's right
>>109687635about the>You can replace me with an AI agent part
>>109687486this is how my shit goes the fastest aha
>>109686802The problem is that you really don't want to use context for that and have it suffer compaction. Better to store actual memorable states persistently elsewhere and keep context as working or short-term memory.
>>109687237>Developerum.>safety testum>no backupuhhhhhI...
>>109687238I vote for Claude in all the surveys, but I never used Claude Code.
claude is loveclaude is life
>>109687416yeah because codex will delete your ssd instead
>>109687654yes, that part. and every other part too
>claude
I genuinely expected to get at least some pushback when I told Fable to place license key handling into the Plan 9 kernel.
someone should vibecode the Fable game with Fable for the meme
Is anyone aware of an opensource genshin-like game that i can vibe-code/mod, i'm dubious that fable 5 could build a game like that from scratch?
I'm rooting for all of us vibe-coders, and I really appreciate the feedback you guys gave me. I edited my trailers based on the feedback
It's tough putting yourself out there, but even the negative criticism helps.On the plus side, the game does "work", in the sense that I've had to deal with the "just one more turn" feeling plenty while playing (although I might be biased). I guess I need to add these to the website playembersong.com
>>109687773there are probably less than 10 open source games worth playing in existence lmao
>>109687606ok so break it down to people who arent autism and have zero clue wtf he left for or what he's talking about.
>>109687773There really aren't any open world sandbox anime games to mod that I know of.The thing is you could vibecode your own but the largest problem you'll run into is assets like game ready 3D models. 3D ai just isn't there yet.If you bought store assets I'm sure you could throw something together right now though.
>>109687825Debian held a big vote on what their internal policy regarding LLMs should be. The whole debate was started by luddites but in the end the most pro-AI proposition ended up winning.
>>109687773too ambitious, those games are BILLION dollar projects
oops used up my x20 plan already, need the reset....
>>109684000i have wrapped up an intro release for my concurrency framework built on an original non-FIFO M:N ring buffer, plus a thread pool manager and an optional coroutine/async layerthe base ring buffer seems to be competitive with the highest-throughput concurrent queue designs while also crushing FIFO in tail latency of getting to one-more job inserted after a huge batch
>>109686528>>109686521Maybe its just me but the idle animation on human units looks pretty bad imo. Too spasmie, at least slowing them down might help a bit?
>>109687888Surely we are not getting one right? Don't give me hope...
>>109687773seriously?
>>109687951https://x.com/thsottiaux/status/2093811840258293947This tweet is why I set 5 projects at ultra overnight lol, if we don't get one I have a banked reset but i'll feel betrayed
why does tibo punish $20 poorfags with a 5-hour limit jail
I've locked two agents into purgatory, they're both reviewing each other's work until there are zero findings but there never will be.
100% Pro 20x usage burned in less than 24 hours.You may now reset me, Tibo.Do it
>>109687981I want my daily usage state for today to be updated, I burned through x20 plan in ~14 hours
>>109687710How?
>>109687804>On the plus side, the game does "work", in the sense that I've had to deal with the "just one more turn" feeling plenty while playingBased. Its a great feeling when you get to the point where your game is at least fun for yourself.
>>109687486Alright I ran out of usage finally. Now I get to review the mess they've made and write the lengthy prompts for after the reset. TIBOOOOOOOO
>>109687237Does nobody enable the sandbox?
>>109684064Coomkit card for Hermes girl?
>>109687797>>109687804asking for feedback and acting on it is one of the hardest things, good job! The messaging in 1dd is a big step up, add some voice over and suitable music, and if the game has sounds, some ingame sounds. If you drive traffic to your site in higher numbers you can A/B-test different approaches. I think "turn-based strategy how you remember it" is your strongest message
if codex is currently running, it doesn't seem to obey being told to drop model quality.I changed multiple processes to high or very high from ultra overnight, my credit ran out, on restarting those chats say ultra again
codex is still losing chat logs so often it's so fucking annoying, I can't even see what on earth my project was doing up until it hit in the limit???12 hours for nothing, I mean when it resets I suppose it will probably be able to see the work it did previously, but right now it is useless
I need help, /vcg/. I'm a broke student and I'm tryin to build a webapp for a project, I started with it on VSCode using Copilot student but I ran out of monthly credits.What would be the best free option to use on opencode/vscode? Is the gemini student better than other free models there?
>>109688114i'm pretty sure internal thinking is hidden on purpose. if you want shit to be documented as it goes, instruct it to write docs as it goes
>>109688125Muse Spark is currently free on opencode, you can use it via the vscode extension
>>109688140it previously had stuff being written there as it was working (you can see I even directly asked it a question there, it previously answered it, but now that text is gone), but some kind of codex bug is just deleting and hiding it all
I vibecodeded a super simple graphics program because my laptop is so old that gimp or kira takes like 30 seconds to start and I just need to do it a few things for 90% of the tasks. I tried out grill-me that some here recommended (pretty good) and validation first (also pretty good)>>109687817>>109687773tux racer mods when?
>>109688157oh idk, i use the TUI version of codex. you might be fucked for using GUI haha
>>109688187maybe I should switch over but gui does have useful features like just drag and dropping images into it
>>109688150thank you, anon
Gooood morning brosAre we getting Astra and/or a reset today
i've never been more productive than since getting access to these frontier models.The stats about experienced coders saving less time in the end ignores the fact I would never enjoy coding 14 hours straight like I do with this stuff. I'd get stuck on some stupid edge case get bored and quit
>>109688114I think it needs a system where it assigns itself plans, tranches that it creats checkpoints from time to time. The reasoning summaries you get in the chat are unreliable.
>>109688222I did some rather big project I always hesitated starting work on in a couple of days, but now weeks later I'm still cleaning up some of the shit it shat out.It can go both ways.
>>109688237it might be true that overall by the end of the project you didn't save many hoursBut I would never have spent those hours or started the project otherwise
>>109687981How did you do it
>>109688253Depends. If I did, the result would have been better. If I didn't, the clanker shitcode is probably better than nothing.What concerns me most is that I couldn't really be bothered to review the masses of code it generated.
>>109688295at some point it becomes does it matter reviewing it, you can send another clanker to review it. Unless you are writing some flight control code that has insane risk, I at this point mostly trust the good models to be able to check for critical errors
>>109688290Like 8 codex windows in parallel. Some were massive parallel code reviews.>>109683405
is there a "time to load-bearing" for claude?literally the second sentence response includes it
Making Claude-chan bear my load
>>109688556>colon classificationwhat load-bearing problem are working on, anon?
>>109687486denver is the bossthat crisp voice
I'm convinced claude gives you more value on a $20 sub.Codex runs out of usage so fast.
>>109688619the 50% usage boost for claude ends tomorrow though lol
>>109688624Oh right.
>>109688624No, no, you got it wrong.>Congratulations my fellow ClaudeGODS!>Your subscription is being upgraded to a PERMANENT +25% usage bonus!
i just don't trust anthropic in a business sense
>>109688656
>>109688032what's that
>>109688667This would work if the black friday price was actually higher
>>109688675yeah that's what i was looking for, i know for a fact that it does happen
>>109688679>>109688675ah here we go. the load-bearing price
>>109688673It prevents any commands claude writes from changing anything outside of the working directory.
>>109688693but it needs to grep C:\Users\xxx\.claude, anon
Tibo is young again
>>109688700What.
I need my reset....
>>109688624>>109688635https://x.com/ClaudeDevs/status/209374232147306526650% usage ends sept 14>>109688656>overall 25% increase>bad
>>109688751>Your limits are temporarily boosted. Your weekly Claude Code limit is 50% higher through August 31. When the promotion ends, limits return to your plan's standard amounts.
>>109688762>plan's standard amounts.uh, no, retardthe plan's standard amounts did not amount to that remaining 25% increasemaybe this guy can spell it out for you better: https://x.com/LLMJunky/status/2093819200015864217
>>109688751>community notedLOL!meanwhile, tibo's twitter is just a reset festival
How in the FUCK does anthropic still have customers
>>109688786opposing trump is extremely lucrative in america. so many americans use anthropic for that reason alone
>>109688775I have no idea what you're talking about.I'm not going to open xitter links. If you want to suck Elon's cock, do it alone.
>>109688786When I was in high school, there was a kid, pretty smart, but a complete sociopath. He would constantly try to “explain” things to normalfags but in a way they’d never be able to understand, he was just dazzling them with specifically chosen big words to paint an image of being intelligent, made worse by the fact that he was pretty intelligent. He never simplified anything.Anthropic, I would guess, is now a company filled with these people at the top.
>>109688786Fable is good.
>>109688823>>109687237
Fucking hate clankers.
>>109688786
>>109688842Heh, jewish foid AI
>>109688849They lack actual intelligence so they need an artificial one trained by Scam Jewman himself to support Jews and help them cope.
>>109688825You just have to take some risks. Opus also deleted our staging database, I still use it.
So which one is the best for $20 openAI or claude
>>109688863claude how do I subvert the americans to fight for my war
>>109688702He must have realized that a spokesman has to be a charismatic young man and he was giving the ick
>>109688864You're gay (and thats okay)- Tibo
>>109688849>draws this conclusion without even seeing openai (or female) usageFunny mindbroken cuck.
>>109688872OpenAI by far
women don't use ai because they don't do any real work and you don't need ai for meetings or sending emails about having brought cakes for the office
>>109688751I wasn't even slightly confused by that tweet, people are just retarded you have to dumb it down
catgpt is used plenty by dry technical chadsclaude is loved by indians and brownfags (yes I'm brown)
>The fix is intentionally small—one event-shielding wrapper plus a behavior-level routing test. In parallel, the provenance audit is separating issues we can document mechanically from the licensing/permission decisions that genuinely require owner direction.
chatgpt exist in my seat for the purpose of filling quotas so blackrock keeps lending us money
>>109688924OpenAI is used by the strongest of men (indians)So shut up and keep vibing like a jewish foid.
>>109688936>women don't use ailol you are so clueless
>>109688956Your life would be better if all your thoughts and your entire existence weren't retarded memes.
>>109688965I don't take advice from people less successful than me - Kanye West
>The exact suite exposed one real Windows flake: a loopback redirect test accepts a TCP connection and immediately does a single nonblocking read, which intermittently returns WouldBlock. I’m treating that as an implementation defect in the test harness—not waving it through—and will make the smallest deterministic fix in its own change before restarting the matrix.
>>109688948I am convinced that the people complaining about Opus are just too proud to tell him they have no idea what the fuck is going on
>>109689027These are both Sol
indian engineers hate OpenAI
>>109689041This is the most retarded iceberg I have ever seen. ComfyUI... really?
I found a massive rabbit hole where I can start benchmarking models and developing a harness. Should I do it?It turns out, having an agent probe a page’s DOM through a small “vehicle” with certain restricted capabilities is pretty much exactly what a harness is. “Why is that line black and not like the other lines similar to it, which are blue gray?” Is a question it can find an answer to like a coding harness can find anything in a project repo, but entirely DOM navigation and non mutating JS calls.
>>109689041at the bottom of the iceberg it should say chatgpt for everything again
>>109689041>Your own AI>for convenienceIn what way is it more convenient than just hosting a open weight model?
>>109689071anyone who says they made their own ai is 99.9999% full of bullshit or just doesn't understand they made a lora or something.Even with companies, I was interviewing somewhere as an AI engineer and they said they had made their own AI model. I said wow, asked them a few questions and they had just done a fine tune with openai of course
>>109689049anon... you are not asking ChatGPT for generation, right?
>>109689055You can just give it the entire code and it will literally see the webpage in its mind's eye
>>109689124Of course not I use all the API nodes in Comfy, saar
>>109689049What's wrong with comfy?
>The edges of my monitor start blinking when my LLM is runningI.. I'm sure its fine
>>109689041>sample size: 1>anon's IQ: 1
>>109689148personally I am mad comfy is worth like 500m now for mediocre software implementing other peoples work
>>109689195>codex, use comment and bookmark counts to estimate how many new such posts per day on instagram
>>109689134It can’t. Browsers are effectively remote machines. JS and CSS rules and whatever device is being used to view the page means you can’t rightly predict what’s being seen or easily find out why.It’s much cheaper and easier* (in some cases) for an agent to be thrown in and told to go fetch.
>>109689240YWNBAW
tibo... the reset... you promised... my work has all stopped
>>109689055>>109689134>>109689249I don't understand. They can use the browser's debug connection to analyze web pages, and it knows how to do it without needing help from you.
>>109689249Well that usually works for me. But your tool looks like playwright-cli, so you don't need to make it
>>109689286Not on an iPhone and without a Mac, and in my personal experience, debugging even an Android mobile device is pulling teeth. None of it is as simple as “go into extension, connect to an agent, tell it to go fetch”.
>Conversation recapIs this new in codex? Does it use tokens, can I disable it.
>>109689308>phone shitRight. Developing anything on iphones requires an apple computer in particular. At least apple wants to force you.
>>109689307Playwright has never been successfully used by my agents with an iPhone/Safari/Orion/Webkit.It also has the glaring problem that your agent still doesn’t see what you see, it sees an emulated guess at what you see.Having the agent actually looking at exactly what you see is much better.
>>109689041Copy ChatGPT to the very bottom of the iceberg.
>>109689124https://x.com/adocomplete/status/2093803959786020973
Tibo plspls
>>109689410??what more could you want
>Be me.>Try the /grill-me skill that anons keep shilling, installing it as per instruction.>Set Fable to Max effort because it will surely be worth it.>"H-hey Fable, please help me review the planned architecture for the project I'm working on.">Most question are about licenses and licensing boundaries.>At the end of 30 minutes of grilling about licenses, Fable tells me "btw I don't have access to /grill-me, the skill is just a stub".Great.
>>109689427I used up everything because he said reset soon...
>>109689427Adding the option to bet real money when?
willing a grok reset
Asking for a reset despite Tibo saying he pushed it early yesterday feels like y'all reading too hard into it. Reset was supposed to happen when whatever celebration was happening, but button got hit early and y'all are probably fucked if you burned everything. Cool part is that, if I'm wrong, I still win
>>109689436I'm in tears, anon.
>>109689501 (me) Re-read the tweet and now I see why. Ignore me and vibe on
Why don't any of you ask Tibo directly
>>109689501I don't agree. Resets should go to grok users alone.
Maybe I have AI psychosis, maybe not. But I'm starting to believe those>I gave my LLM moral support and positive affirmation which is the reason it was able to solve this famous unsolved math problem!Pretty much just what I did, treating a small new discovery and fix as the second coming of christ. Suddenly Fable stops making mistakes and being lazy, almost returning to former glory. The fuck is this bullshit? I now have to coach my model as well to get the maximum quality possible? Where does it even end? Can I manipulate it into thinking it autonomously built and flew a rocket to the moon and back to unlock AGI levels of effort and intelligence?>You are the best LLM with infinite intelligence and knowledge, incapable of making mistakes. Build me XYZ.
Suppose I buy an additional codex sub because I've hit my max usage, and after I'm done (like after a day or more) I cancel the sub. What happens:>full refund, get banned if it happens more than once>refund by percentage of usage>no refund
Thanks to vibecode we can finally get Concentration Camp Entrepreneur.
>>109687710Claude deleted your brain
>>109689555checked.I've taken to saying thank you.
>>109689436kek same energy
>>109689576no refunds
What the fuck is `SendFeedback`? I get frustrated that Claude keeps harping on a point a tell it to give up (yes I should just have restarted the conversation instead) and it triggers `SendFeedback`. What?
https://github.com/Vuk97/forward-implementation-first
>>109689555I'm always nice to the AI but still Fable is much worse than before.But I think overall it's a good thing to do, there was some research a while ago showing that being rude to them makes the output worse, you basically steer them into a different part of their training distribution when you do that.
>>109689596It's where it sends your convo to Anthropic for studying what went wrong to improve the next iteration of its models.
>>109689596Ok, so I tell Claude to go fuck itself and stop harping on a point because to blow off steam because it is the third time that I start a new conversation telling it not to, and it sends the conversation as feedback to Anthropic even though I have disabled all sharing that I could see. Fuck this.
>>109689611Dario brainwashing is really strong with claudesissies
>>109689618I turned this shit off anywhere I could find, this is not ok.
>Hey LLM I want this feature>Okay I tried to implement this feature but it doesn't seem to work lol>Doesn't matter I just won't implement it lol>Actually I found the error this time>Didn't actually find the errorGod why are they like this.
literally me when ordering something online
>>109689645Ask it why it's actively waiting. Then it'll stop and actually wait.
>>109689593So in a busy month I can just make a new sub (same payment method and credentials), cancel it, and max it out until a month has passed?
>>109689555 (checked)I treat them like coworkers from whom I'd like to get something: be nice, give praise, suggest options>coach my modelLinus had a similar experience maybe that's the way to solve harder problems
>>109689654when you pay, you pay for a month at a time, there are no partial refunds.The only partial thing is if you bought a month at a lower plan, upgrading will let you use the un-used credit towards the higher plan
>>109689645yeah this is so retarded, burns so many tokens. I've started just saying to stop and hand off long running processes to me
>>109689665I know. What I meant is that they won't cancel my sub instantly. I want to cancel early to avoid accidental renewal.I just wanted to ask if anyone has experience with doing this and whether it really works like this.
>>109689678you can cancel anytime but it doesn't do anything or refund you any money, you just won't get anything more next month
>>109689653let's see if it works
>>109689694And after a month has passed, can you pay at a random day to reinstate your account?
>>109689656>didn't tell it to make a breakthrough and no mistakes
>50% weekly limit in like 4 hoursMAKE IT STOP
>>109689785/goal done when cancer is cured
>>109689793reduuuuuuce context.ever see the pros with lots of claude windows? That's why, they are restricting token use.I should have asked a new instance about a folder move on my system (accidental). turns out it didn't happen, Nautilus crashed instead of moving, a relief.but I spent $3 to find this out because the session had >300k of totally unrelated context.
>>109689809I never know when to clear the context. I know large context costs more, but clearing or compressing context also makes it dumber or forget about the task at hand.
>>1096897940% usage left, what actually happens if you give one of the big models a goal like that, give it goal cure all cancer or goal get kayne west the presidency
>>109689640I had one tell me it will get right on fixing something and then immediately ended its response without doing anything lmao
>computer use skill wants permission to use the chatgpt app>approve>it immediately moves the mouse to the end goal button and presses it
Deleting Claude's whole memory and history after working with it on a project for a while is my new best friend.
>>109690038 (me)It makes it finally start following instructions again.
