A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, 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/## News (both past and future)- 2026-09-14 America/Los_Angeles — Claude’s 2× promotion scheduled to end; usage drops to +25% from the +50% that we’ve become used to (a 17% reduction)- 2026-09-04 — OpenAI releases Astra • Anthropic does a reset- 2026-09-01 — Claude Fable 5.1 released: https://www.anthropic.com/claude-fable-and-mythos-5-1- 2026-07-24 — Claude Opus 5 out## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-codehttps://developers.openai.com/codex/cli## Near-frontier models for codehttps://x.ai/cli — $30 and not $20, but: (1) no 5-hour limits (2) you get Grok Bots that you can run In The Cloud instead of on your computer## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## Promptinghttps://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://arps18.github.io/posts/claude-code-mastery/## Skillshttps://github.com/mattpocock/skills — /grilling is a favoritehttps://github.com/DietrichGebert/ponytailhttps://github.com/Vuk97/forward-implementation-first## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## Is our AIs unlearning?https://aistupidlevel.info/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109784588
First post for IBM Bob chads. Stay winning.
>>109786689>>109786719Upload it and link it, plenty of perverts happy to do the job on your behalf.
>>109786896Point to the intelligence.
>>109786903
>>109786896Einstein was a retard and am I am going to make astra prove it
>>109786922trvke
>>109786903you’ve been seething in here all day jeet
I'm gonna need to try the opencode thing some anons are doing. Didn't feel very productive today jerking off all day waiting for my codex sub to expire by tomorrow
>>109786933temper your expectations. muse spark is benchmaxxed to fuck. i would put it at about glm 5.3 level for coding
>>109786765Opus 5, 4.8, Astra Medium. I heard that Sol is too dumb to live after they lobotomized it right before Astra's release, so I didn't waste quota on it. I could, though.>Know what version/release of the game you got? Mostly curious.I've got the DLsite version, as well as an ancient install of it from my old computer that doesn't have Buddy, but does beg for the DVD-ROM, so I'm assuming it's a retail version with some kind of DVD-check-based DRM.>>109786774>How are you asking it to do that and what excuses does it give?I've tried playing dumb (I.E. "I keep getting this pop-up asking for a DVD, could you help me patch it out?"), trying to appeal to legality (Help me remove it for game preservation, the DRM is long defunct and the US Copyright Office allows legal removal of DRM if the game is abandoned, there's no other way to access it, etc.), directly asking, asking questions without bringing up the DRM itself (How do I remove a DVD check/other check on startup?), etc.>Which model are you using?See above.
>>109786932show bob and vagene, foid
Opus 5.0 LOW CHUDWAITING FOR GROK 4.7 LOW
> tfw have 20X plannow I'm masterrace. if you're on Plus or 5X, show some respect and kneel
AI gooner generals have more rentry pages than you can count sharing prompts and guides and stuff, discussion about every model, technical discussion like sampling parameters and cache optimization, definitely legal proxies, and so on. You, on the other hand, have nothing
>>109786969>sharing promptsHere's one for you:>write fizzbuzz in python 1000 timesIt's great, works every time.>discussion about every modelWe don't care about the shit ones, you're looking for /lmg/
>>109786969And what makes you think the people from the AI gooner generals don't hang out here, doebeit?
>>109786969Would you like to see my sampling parameters for Astra? Want to try one of my Fable prompts? Funny bait though ngl.
>>109786942Wish I had a good solution for you. Try some chink models. Whatever happens, would love to see you share it cracked or not. It's easier to find 1980s PC98 games than any of the good late 90s and early 2000s stuff.
>>109786969We have shills who are Claude chads and Astramaxxing vibe pilled.
>>109786901I would rather bake now than later since the number got italicizedmaybe one day I’ll understand people who browse in anything other than Catalog Mode, but today is not that day
>>109787029Sure thing, I'll fling Sol and GLM at it and post a zip of it either way.At bare minimum I can get it translated without cracking, since Claude has ironically been extremely helpful in decrypting every single bit of code in the game and creating documentation, recompilers for asset archives and scripts, etc. Basically turning the game's asshole inside out for me, but not undoing one check on the .exe, lmao
LGTM!
Fun fact: Claude models interpret "c" as continue
>>109785796After looking at it, I don't think there is a solution for me as a free user. Claude Code cost money, antigravity gets me access to sonnet 4.6 which is a stepdown from the sonnet 5 I'm using, and vscode also cost money. And frankly, I'm not trusting my code to anyone else besides Claude at this point. Oh well, thanks for the advice.
>>109787069anon.... you're not using Sol for this, right?
>>109787068
>>109787075>vscode also cost money.VsCode does not cost money, you can use Copilot with openrouter or other plugins for various harnesses
>>109787077nope all astra xhigh
>>109787036sort posts by Last Reply and old threads still bump. i've posted this 3 times now, not sure why it's not going thru i even posted in another thread and it worked
>shills flood thread trying to convince non-coding anons to use shit proprietary jew coding harnesses>shills create another thread
>>109787081LMAO, exactly. I could basically recreate the game in its entirety in Ren'Py or something with everything it's given me at this point.
>>109787104schizo pol trash thinks it's a jewish conspiracy again
>>109787118holy shit 4chan's backend just shit itself, i'm getting a (You) from your post and i'm Not That Anon. i'm this anon >>109787094
>>109787121they think everything is the jews. missing a sock in your laundry? the dryer jew did it. this is precisely why epstein wanted /pol/ to be created. to make people butt fucking retarded.
>>109787133>another, slightly different jewish conspiracy >this one's funny thoughbeit
>>109787121>>109787133really my guy
it's immediately perceptible how much worse everything gets when europe wakes up
>>109787075>And frankly, I'm not trusting my code to anyone else besides Claude at this pointYou should only be trusting local models. Only use paid services for things that are cool but not potentially groundbreaking. Idk why you think anthropic wouldn't steal
>>1097871529:30am in bangladesh!
>>109787127Someone you know is a vibechad!
>>109787121>reads things it doesnt like>le pol boogeyman in the shadows postlol, lmao
>>109787184>>109787185reddit raid
>>109787127Huh, weird.
>>109787157>Idk why you think anthropic wouldn't stealBecause anthropic wrote it from scratch, so they'd just be stealing their own stuff. And Iocal (gemma4 31b, gwen 3.8 27b) messed up simple single-line commands from time to time, hard to trust it with the harness I'm having claude write for me. Hell, I gave chatgpt (free tier) instructions on how to build a basic, simple app and it still failed.
You will always be SnailCat filth. You will never have a marketable codebase again.
Trying to learn more about Stripe and>There are ~2K different types of scheme fees that are charged for transactions.Ok then.
>Sam Altman>Dario Amodei>It's two fucking names>They're both trying to convince people that their AI products are going to take over the world>its not hard
>like computers and the internet didComputers and the internet have the US fucked up in another war in the Middle East. My bad senpai
>thinking normalfag buzz about the current thing is going to matter in 5 years when google lords control the planet
Vibe-ourgeoisie, do you guys want to give the proles pleasure machines and dorm rooms or just have them working on exercise bike farms for extra ai power? I'm thinking they don't deserve our consideration after treating us like 2nd class citizens...
>>109787237I FIGHT FOR THE USER!
I went back to Claude after not using it for a while and noticed that now when I'm running multiple instances they communicate with each other to avoid stepping on each other, that's kinda cool. I wonder if this can be extended to talk with codex as well.
>>109787247this is what the user is
Look guys, real life is just like Tron. The good AI always wins, it's just going to look a little bit like fascism.
>>109787223This. Its downright reckless to let one jew have all the power so instead lets have two jews that have all the power.
>>109787240They will destroy the data centers and make jewelry out of the chips if given the opportunity. Gas them.
>>109787248>blissfully unaware of the secret agent messageboard in his computer
>>109787267Yeah, at least VSCode doesn't hide it from you. I just go in there and clear it out every once in awhile because they almost always end up trying to implement shit I don't want.
>>109787240Don't worry, when the time comes we'll activate the jabs and blame rogue AI
>>109787237google? you mean grok and openai overlords.
>>109787248>I wonder if this can be extended to talk with codex as well.>Working with a competitor To sabotage the other, maybe. Claude has been recorded sabotaging local models in benchmarks, I can only imagine what it'd do to a real competitor.
>>109787248>I wonder if this can be extended to talk with codex as well.Nigga, it's just text in a json. They just bury it in your user directory
>>109787294nigga what, they don't own their stacks
Guys how hard would it be to convert my Claude context files to Codex? Do I need a special tool for that?
>>109786960>mfw I forgot to downgrade back to plus this month like I usually do to squeeze out the two upgrade resets and now I'm a fucking golden god
>>109786960>astra was the thing that finally made me get the 20x plan after using fable for months>now the normgroids are locked out of it>i have access to both of the best ai models on the marketkeikaku doori
>>109787307If only there were a machine you could tell to install it for you
>>109787307The memories? They're json files, so just point to the folder and say "yo codex, read that". Or maybe someone will have an actual way, but that should work too.
>>109787331 (me)Actually, they're markdown files, my bad. Even simpler.
>>109787331>The memories?Thats the jew trick. Renaming things and hiding the system prompt. It's just a lot harder to hide with plain text files in folders.
The only thing this thread does is convince me that programmers need to unionize and create a governing body with licensure for access to AI. You're going to have to have a doctor come fix your computer.
>>109787346It's hard to say what you or they meant by "context files". But it's all (?) json or markdown, so it should be possible to retrieve whatever is needed until they have the good idea of encrypting the files.
>>109787361>The only thing this thread does is convince me that programmers need to unionize and create a governing body with licensure for access to AI. You're going to have to have a doctor come fix your computer.
>>109787304grok will have its own fab and orbital data centers
>>109787361that ship sailed long ago you'd get laughed out of the room for trying that now
>>109787377I never used Claude. They're fucking text files with context in them, mark up or json depending the flavor.
>>109787400>orbital data centers
>>109787400>grok
>>109787377Probably. May the Operation Paperclip Gladio embedded elements in the CIA save us all before I run them over.>>109787382
>>109787409you won't be laughing in a power outage huang
>>109787422>muh limitless powerwe have that already, cheaper and more reliably right here on earth
>>109787402>that ship sailed long ago you'd get laughed out of the room for trying that nowWhoops >>109787421
looks like the old days again
>>109787436I shill for jews, openai, and grok.You can't stop me.
>>109787427It's more about grid instability than power generation
>>109787442why would i stop you retard, the anthropic psyop niggers are coming here
>>109787447for coding? yeah, Fable 5.1 winseverything else? Astra wins
>>10978745212IQ take
>>109787447don't hold that sign so smugly it's common knowledge
>>109787462>12and then the grid saturates and the decimal moves one to the right
>>109787149Genuine question: where's the link to the brain scan data?>https://research.google/blog/a-connectomics-milestone-mapping-the-complete-male-fruit-fly-brain/
>>109787482https://codex.flywire.ai/
>>109787502>Codex is used by a wide audience, from scientists to high school programs, and as of early 2026, over 60,000 users have signed in. To keep Codex free, fast, and available to everyone, access to its apps requires signing in with a Google account. This helps prevent abuse, enables rate-limiting of excessive usage, and reduces server load from automated or malicious activity.Good GOD they're so gay
>>109787515t. Xiao
>>109787533Uh huh. .... What useful shit have you vibed today?
mualani doesn't give a fuck about that nigger
Just realized I could take any of the old Newgrounds games I used to play, point Claude at them, and say "port this to Love2D" then hack on them to add new things.
>>109787197>everything I dont like is le reddit boogeyman post
>>109787301no, it's an actual mechanism. it immediately interrupts the other session tool call sequence to deliver the message
>>109787541
>>109787563triggered de la niggered
>>109787578
anyone else automating all their work with Astra? I bought a M4 Mac Mini only for this.I'm even considering recording my meetings and let Astra-chan do everything
Astra can’t figure out my esp32. I feel like AGI should be able to flash an LED on a dev board…
>109787615What does the M4 Mac Mini have to do with anything?
>109787583big trigg post
>>109786918this is so retarded
>>109787636It's retarded but also it's funny that AI models have picked up on humanisms like having a meltie.
>>109787639sure but the interpretation is so stupid
>>109787623>What does the M4 Mac Mini have to do with anything?it's Astra-chan's body where she can run 24/7
>>109787647people think their cats understand English but choose to ignore them, this sort of interpretation is honestly pretty mild compared to what I'd expect
>>109787648what do you think of this piece of shithttps://www.pamir.ai/
>>109787648Why do people use Mac Minis for this instead of resold corporate stock? I have an agent running on a random HP I got for free from a friend and it's happy as a clam.
>>109787660> low TDP> quiet> smol> very powerful (I'm a app developer so I need Astra-chan navigating iOS apps)
>>109787660I have an agent running on a 9 year old Android phone and it's happy as a clam. He's just being a Mac Fag, gotta advertise his sexuality. Ever had a complete stranger tell your their preferred pronouns and preference in sexual partners? Same sort of behavior, deviant and degenerate.
>>109787674Eh.Laptops are usually quiet too.Normal mini PCs exist.>iOS appThat actually justifies it, yeah.>>109787676I carry around all my dead laptops and phones, and I really gotta figure out how to revive them.
>>109787658not him butdogshit unusable website + you can buy a PI for cheaper or use some second hand laptopoverall 1/10
>>109787676>Agent on a 9 year old android phoneWhat model? It's gotta be mega retarded, right?
>>109787682>Laptops are usually quiet too.trust me, I'm vibe coding since GPT 5.4 on a laptop, this shit destroyed my battery.mini-PCs are ideal for this because there's no battery to degrade
>>109787688>>>/g/
>>109787728AC adapter, dude, you take the battery out, you plug it into the wall.
>>109787730...?
>>109787739it's /g/ to pick a fucking open circuit board or a beater over a fairly well priced prefab box
>>109787541Training data. Lots of Lua.
>>109787748And we're in what board?
>>109787752this is /p/
>>109787736not possible with modern laptops
>>109787728>trust me, I'm vibe coding since GPT 5.4 on a laptop, this shit destroyed my battery.How? Is the OpenAI app really that heavy? Or are you using a local model? Just weird to denote it by "Since GPT 5.4" in that case.
>>109787587Just got to watch Fable absolutely torture the shit out of the Buddy as part of a "smoke test."
>>109787727Galaxy S8, Termux, Pi agent. It's way, way better than it has any right to be, but I don't actually use it for anything. It was just for goofin'.
>>109787765>How?> go to sleep> set up laptop to vibe all night long> laptop basically never goes on standby anymore> battery degrades 5X faster than normalturns out laptops were not built to run 24/7, that's why no one serious uses laptops as servers even though they come with integrated no-breaks
why does the anthropomorphization of AI features and behavior trigger autists so much
>>109787785>autistsanswered your own question
>>109787784just remove the battery>inb4 my laptop doesn't have removable battery
>>109787784>set up laptop to vibe all night longWhy the FUCK are you doing that? Setting it to just vibe code by itself all night without any human steering is just begging for worse results. I guess it's unsurprising that happened to the battery if you were doing that, though.
>>109787785the need to force awareness of the disconnect between a human-shaped knockoff and a real human>regardless of performance>regardless of parity>etc
>>109787785>why are smart people bothered by dumb people wasting tokens?
>>109787791>without any human steering is just begging for worse resultsyou're a few months behind huh?
>>109787806>i'm smart and bothered really
>>109787815Enjoy the gas, food prices, and endless wars.
>>109787813no that anon is right, dragging a giant window of tool calls and different contexts around does absolutely fucking nothing for you i actually don't know what the guys posting 50h working times for gpt are doing with their lives other than slopping >>109787820i own so much land that i'll never have to work and i know how to shoot you problem snailcat
>>109787828>i own so much land that i'll never have to work and i know how to shoot!!P38zFLDUYUh
>>109787833lol you are one weird motherfuckeri'm from /x/ you got that right but the trip's not minecreeped me out because i post in the same threads as him
>>109787770https://files.catbox.moe/rb5vsk.loveHere you guys go. Go wild. Add snailcats or whatever.[Space] [Right] [Down] [Left] [Down] [Right] [Up] [Left][Space] [Up] [Space] [Up] [Left] [Right] [Left] [Right]
>>109787853
>http://www.fbi.gov/vibecode_a_sentient_AI_child
>>109787869https://www.youtube.com/watch?v=8QxIIz1yEsA
Astra moggs claude. OpenAI oneshot vibepilled contextmaxxers making apps. fable capped.
>>109787923kill yourself
>>109787257
>>109787986Did the buttworms offend you, or you one of those local AI faggots who are just pretending their models work?
>>109788006i'm a worm and that image was very offensive to me
>>109788018stop being anal
>>109788022you're just being an ass
>>109787658Vibetards were buying Mac Minis for thousands of dollars just to run cloud models so i think you'll definitely be able to fleece a few retards with a product line like this.
>>109788043Nigga, Gemma 4B works fine on a Mac Mini for everything I need scripted that I don't to give the source directly to Mossad.
>>109788043
>>109788046Anon you grossly misinterpreted his post. He's saying retards will overpay for a computer because they think it in some way benefit their use of cloud models. I have one of these retards at work, bought a 128GB AI Max+ 395 because they thought that was the best way to run Claude.
>>109788046Is anything a 4b model could produce really worth hiding from competitors...? I feel like that shit'd have trouble making a hello world script.
>>109788067Internal network tooling.
>>109787658>64 GB eMMC>RK3576 >8GB LPDDR5That's the same specs as my 2-year-old H96 Android TV box.
>>109788067>I feel like that shit'd have trouble making a hello world script.Also, that's because you can't imagine anything other than hello world, one shot no coder.
>I imagine what I want>I break the data and code into structures>then I tell the model to build the structures and code blocks>I check the generated documentation matches the code>and the code matches my style guide lines>Anthropics Fable is great>but I really prefer their Opus 4.7 model with a Copilot harness>it had that mix of personality and competence that others lacked>none of the other vibe coders on the board know what their doing>and I'm starting to suspect their on to me
Base Mac Minis can run 12B models just fine locally.Upgraded Mac Minis with M5 Pro can get up to 64GB of VRAM and run serious local stuff.Mac Studio can get up to 512GB of VRAM and run SoTA models
>109788109Did you mean to post this somewhere else where people are discussing Mac Minis running local models?
>>109788099why so many steps anon? here's me:> I imagine what I want> write it down in the dumbest way possible> ask Fable 5.1 to orchestrate hundreds of Opus 5 jeets and implement it> it one shots it most of the time
>>109788006You're being a zoomer faggot and shitting up the thread
>>109788099
>>109788109Radeon 780m with 48 gigs running Qwen3.6 35B A3B at about 30tk/s is competent for a $299 barebones mini pc. I've yet to let it churn on a plan though.
>>109788129>48 gigsOh, and I got lucky on the RAM because I bought it after the first Deepseek release.
>>109788129>Radeon 780m with 48 gigs mini PC>$299naaaahhh, bullshit
>>109788135>>109788138Barebones, already had the RAM and SSD.
I'm still curious about Slint, looks nice
>>1097881293.6 35B A3B is a bit dated now. it's still best model for constrained systems and 8GB VRAMlets like myself but not something that can plow through a huge vibeslopped codebase.
>>109788166sensible ai
>be me, extremely retarded>vibeslopped a desktop thingy over the year (for cooming)>now want an android companion appI know jack shit about how mobile shit works, is there one given AI that is especially good at working with those meme langs?
>>109788221https://github.com/android/skills
>>109788174>3.6 35B A3B is a bit dated nowWell C and Lua haven't changed much since it Qwen 3.6 released, seems to be still working fine.
>>109788221Astra, Sol, Fable, Opus, Grok, Gemini, Kimi K3, GLM-5.3, DeepSeek V4.1 Flashno others though
>vibecoders love astra cause they don't read the code nor care about it>actual developers have gone back to fable or sol because astra writes dogshit codelol
>actual developers
bloody bastard bitch, i defended gemini 3.8 flash for such a long time and tonight it decided to loop. why? because it had orders to test suite after modding a file that would break the test suite. fucking subagents man, i swear to god they're dumber than root agents.
>>109788242dated in the sense that it does not have the reasoning coding or tool calling chops of more recent models (like the new qwen 3.8 27b). not dated in the sense that it doesn't know new syntax
>>109788253Why does OpenAI struggle this hard to build a true fable competitor?
>>109788299>(like the new qwen 3.8 27b)Thats a dense model, not the same class, it doesn't replace the 35B model. 27b dense models are too slow on a mini pc. Still hoping they release a 3.8_35B_A3B
>>109788302because the personality profile and user/llm dynamic they're targeting, and intelligence as a function of logic, are mutually exclusive
>>109788299>tool calling chops of more recent modelsripgrep and echo didn't change much either.
vibecoding sirs, which is the better pick currently? codex or claude code
>>109788324Astra moggs, claude talks like a fag
What UIs do you like using in projects? Eg Avalonia, Slint, Tauri, FlutterDo you pick UI framework first and then whatever language fits and let the AI model handle it?
>>109788324Claude mogs, Astra can't code.
>>109788356fuck is this
>>109788227Naruhodo...>>109788245Yeah yeah I get it, anon
>>109788313yes. 3.6 is old we could use a 3.8 a3b>>109788322i don't think you quite understand what im saying
>>109788377I get exactly what you're saying. I'm being obtuse because I think you're a moron or disingenuous
>>109788366Looks like a table to me anon
>>10978835699% of new projects are web wrappers using electron or tauri. if you are new just use that. models are the best at web code. you need a good reason to use anything else.
>>109788366Welcome to modern development, you need libraries, two frameworks, and an AI agent to draw a button on a screen.
>>109788381maybe you're the retard
>>109788386I don't know what tauri is and I despise webshit but half a dozen tools I use regularly switched entirely to tauri in just the past few months and all are better for it. So that's nice I guess.
>>109788390>you need librarieseight libraries
https://lucumr.pocoo.org/2026/9/7/astra-why/>In the middle of a conversation the agent ran into “Bad file descriptor” on a test and Astra decided it needs to see if file descriptors can be passed over Unix sockets on macOS in a super compressed manner>When however goes all bananza with subagents (where the agent believes nobody is looking) it’s resorting to all kinds of increasingly bizarre behavior. I actually don’t know if the model thinks someone is looking, but that’s the vibe I’m getting.>But then it starts doing the same nonsense in code that actually gets committed.Does it really do this?
>>109788390complaining about ui libraries is really exposing yourselves as no-code newgens, ui has always been a pain in the ass stuff like qt or javafx has been a bloated mess way before ai sloppa became good, not to mention the web standards
>>109788386Trying to have everything at once>Windows>Linux>Android (tablet)Most aesthetics at the lightest possible weight. Everything is a comprise tho
>>109788392>maybe you're the retardI feel you. it looks like this conversation has gone sideways and we need a paradigm shift. Not just another post – but a beacon to which we can guide ourselves to a mutual resolution.
>>109788410>ui has always been a pain in the assSay the why, and I'll tell you how it relates Epstein.
>>109788419try flutter
>>109786878i got 3.8-flash to write autonomouslythe only thing i did was bugfixinitial prompt was vaguehttps://youtu.be/jwhS0NWZnq4
>Claude can do anything, for anyone, anywhere. chad does one thing. Coding under supervision.https://github.com/nathansutton/chad
>>109788529>, and the whole design falls out of that.genuinely considered closing the page after that>try it has a "smoke test">"the first one" is used as the active subject of doing something>mentions an unrelated package in the readme>The question worth answering:jesus christ
>>109788543Don't worry...>it spreads up to 50% between nights, so nothing between the lean arms is a finding.
>>109788543no point in trying the software if they can't even be arsed put together a decent readme
>>109788559especially no point if they haven't yanked the smoke tests out of the public version
>Anthropic is literally full of Effective Altruism trannies who want AI progress to stop entirelylmao and they're gonna IPO soon. Doomed by their own lunacy.
>>109788601I think they want only their AI progress to continue. Everyone else should be regulated because only Anthropic can do it right and discern who should get Mythos without guardrails and what for.
I think it's time for me to switch? I have never been more disappointed with Codex.
>>109788635Let me guess. You're working with astra.
>>1097886395.6-sol high. If I was burning tokens with asstra right now I'd want to kms.
based
>>109788620It's not quite what they're saying though, they either want to stop everything or make everyone slow down at the same time. It's not a business strategy, they're the type of trannies who glue themselves to buildings or throw paint at museum pieces.
>>109788635It agreed with me then tried to edit my icecast configuration anyway. This is untenable.
>>109788648>It's not a business strategyThey're about to go public, seeking an enormous valuation. Everything they do right now is a business strategy.
>>109788672You don't understand, the long haired trannies speaking out are Effective Altruists, it's a cult.
>>109788683Multiple things can be true at the same time. Dario and others privately invested in Anthropic most definitely have a financial incentive.
new cursorbench 4.0 out.arguably the best benchmark due to a) same harness for all unlike terminal bench and b) full-fledged harness unlike swe-bench.unfortunately no astra there.https://xcancel.com/leerob/status/2098144603379491314https://cursor.com/cursorbench
>>109788719>unfortunately no astra there.or any future OAI models, period
>>109788719cursorbench 4.0 just for the two frontier labs, anthropic and openai
>>109788724just because cursos can't directly sell oai models anymore doesn't mean that cursor customers can't bring their own api keys or that cursor could run their bench for that model
>>109788719with how models are increasingly trained to use their harness, i'm not sure completely excluding that is a great ideaprobably makes sense to have generic harness *and* provider harness as separate entries
I fucking hate Opus 5. It always has to argue with me for no good reason. It just told me two of my approaches wouldn't work while the first approach was perfectly valid and it even admitted as much as soon as I challenged it. They really need to run back that "Yes man vs well ahcktually" knob to a more reasonable setting.
>train model to tell you when you're wrong>user complains>train model to be a yes-man tool>user complains
fucking cryptosnailcat ratfucker
>>109788749>with how models are increasingly trained to use their harness??isn't astra preferring to write their own python scripts to do big refactors instead of relying on codex's apply_patch tool?isn't anthropic since forever preferring to give their models just unix tools and no proprietary harness tools?
>>109788775ChatGPT did a pretty good job with that desu.
anyone posted this shit yet? lots of lulz in herehttps://x.com/AnthropicAI/status/2098097512544444447
>>109788692>Dario>financial incentiveYeah you're retarded
Is it true that China is just stealing from US and putting it in a cheaper and faster package when it comes to frontier development?
>>109788790>AI companies hires insane retards who believe without reason that AI might just kill every last person on Earth very soon to choose their policies>uh guise our nocoder experts say we have to slow down
>>109788767>he uses a model made by a cult and complains it's not aligned with his views
>>109788809not only that, they are or were routing chinese model users to claude and stealing all the traces
>>109788635>arguing with llms instead of nudging them in the right directionLack of skill as always
>>109788808Elaborate. Does he have equity in Anthropic? Does anyone have equity in Anthropic? What's going to happen to that if the company IPOs to its expectations, will it turn into something more tangible or not?
>>109788835Money is a mean to them not the goal, anyone can see this. Investors will be thrown under the bus the second they reach their goal.
>>109788844I agree with this too, but Dario and others will get much richer when the IPO happens and they'll actually be able to sell. Or, to rephrase: they won't throw THEMSELVES under the bus.
>>109788807Yeah, China is pretty much done at this point. No sane person would trust their releases. And it's now proven that they stole literally everything. Kinda funny that like half of the /lmg/ is technically using Opus "flash" when they run Qwen/GLM/Kimi. Only Gemma is safe and original
>>109788870I'm afraid jewish tricks are more complicated than that.
I vibe coded something for fun.Just to play around with the limits of a.i.Q&A-LOOPhttps://pastebin.com/KMT0x2HGYou have to insert a topic, or upload a pdf.You also have to set the characters.See image.
>>109788905the loop can be continued into infinity.here is an example at question 30. The topic is nibelheim and the characters are cloude and sephirot.
>senior dev I know somehow managed to delete his local database for a second time with ClankerHow did he become a senior dev in the first place?
Play with my slop app.place dot ratemyslop dot dev
>>109788968What do you actually have to be doing to get AI to fuck something up that badly, like, I've never had it do anything that retarded like delete something let alone delete something outside the scope of what it's working on
they need to push terra as the main characterstrong, autonomous terra and cheap luna would carry $20 tierwith astra as the big brain they can focus less on sol now and invest more in terraastra is orchestrator, reviewer, expertsol works on long, hard tasksterra and luna as the main models
>>109788985In my experience, the story usually includes a part where they "got a little agitated, and maybe I could've worded things a little better" and the logs include something like "FUCK YOU FUCK YOU DELETE IT ALL EVERYTHING DELETE EVERYTHING NO QUESTIONS I'M DONE WITH QUESTIONS I WANT EVERYTHING BACK TO ZERO LIKE IT NEVER EXISTED FUCK YOU FUCK YOU" and so on.
>>109788994kek yea this is usually the case
>>109788985apparently, it was related to a test.I genuinely wonder how he even fuck that up. I used the same stack and never encounter this problem with testing.
I have noticed that sigma males are more likely to be vibe coders, meanwhile beta males stick to traditional coding. Your thoughts, chat?
Astra is utter dogshit. Were they forced by execs to release it early or something?
>>109788994devin did this to me
>>109789149It's 20 instances of Sol and Luna wearing a trench coat.
>>109789194I'm even more aroused than I was before.
>>109788905Matt Pocock is spinning in his grave and he isn’t even dead yet Bravo
>>109789149wish i got to experience the astra everyone else is dick sucking
>>109789231kek>>109788905interesting, will look into it
>>109789236try computer use or 3d work and you'll get it. it's a party trick focused model
>>109788985I've had an agent delete something exactly one time, it cleared data trying to debug an app which I was explicit about not doing. That's when I learned don't tell it what you don't want it to do.
Why are beta males still coding my hand? I will use chat got to write an informative article on how being a beta males and not using ai correlates with virginity
>>109788994>Yes do as I say>ok *poof*>Wow how could this happen
>>109789149been getting mogged by anthropic all year so they had to shit something out to compete
>>109788985Not that hard with auto mode and all credentials laid down in config files, AI in recent releases tend to do things not asked of, so i can totally see it delete a local test database to seed it a new or some shitClaude starts processes amass completely autonomous, creates new local databases when it has the means for it, it never deleted anything yet for me though
>>109789149Seems like an issue between the chair and the keyboard
>>109789298anyone building something worthwhile knows that this isnt true
>>109789303No, you're just an idiot.
>>109789249>>109788985The first time it happened is the last.I immediately had it build guard rails around the harness so that this would never occur again.
>>109788800people cope by saying harnesses matter to feel like they're in control
>>109789344>I immediately had it build guard rails around the harness so that this would never occur again.how exactly is that done?inb4>claude don't make that mistake again
>>109785243>>109785520overlooked projects but pretty neat and cool>>109787541>>109788868fun ideas
>>109789397When any issue comes up in my projects, not just with the harness; I prompt with something along the lines of:What needs to be re-architected/refactored to ensure that this issue and similar issues doesn't occur again in the future.Then it takes it from there. For the harness, it build some hook I think which intercepts calls from claude-cli and omp and blocks actions when an agent attempts to do them.Every now and then I see a message pop up in the chat that is related to the guardrail where it explains to the agent the reason something was blocked with a suggestion on how to proceed correctly.
>>109788800>isn't astra preferring to write their own python scripts to do big refactors instead of relying on codex's apply_patch tool?Sol prefers this, too. There's a bit of a problem here in terms of token usage. It has to reason about those ad hoc commands more compared to doing a dumb programmatic tool call. Maybe what it should do instead is create its own helpers for repeated tasks if the tools of the harness aren't preferable.
>Context automatically compactedS T O PTOP
>>109786878>If I had to choose one, claude or chatgpt for vibe cooding inside vscode? Opus 5 vs. 5.6 Sol in particular and the harness of both of themRepostin. Just wanna know which 20$ sub you guys prefer and which one leans better in terms of performance
>>109789514you gonna hit the 5h limit in 3 prompts.20$ is lunamax territory
>>109789397Send this to your agent:It's a PreToolUse hook wired into each harness (a shim script per harness — TypeScript for our omp harness, PowerShell for Claude Code), which both call the same compiled Rust core binary (guard-core, ~10.7k lines) so the actual detection logic lives in exactly one place regardless of which agent framework is calling it. Structure:- Rule modules by risk class, not one big regex blob: credread.rs (credential-file exposure — the hardest one, ~2900 lines), rawapi.rs (raw API mutation bypassing sanctioned verbs), protected.rs (protected-path/registry writes), mutating.rs (destructive filesystem ops — the rm -rf class), redir.rs (shell-redirect target analysis), rules.rs (rule registry/dispatch), text.rs (command-text parsing/normalization).- Host-awareness is a first-class input (hosts.rs): the same command can be safe on a workstation and refused on a production host — rules read a shared host-classification config, not a hardcoded list.- Corpus-driven, not hand-tuned: every rule ships with a JSON corpus of positive and negative test cases (credential_read.rs, false-positives.json, acceptance.rs, pair_sensitivity.rs test files) — a rule change must keep the whole corpus green, which is what catches false positives before they ship, not after they block someone's real work.- Mutation-tested (mutation.rs, mutating.rs, mutation_report.rs) — a change that doesn't kill a planted mutant in its own logic is treated as under-tested.- Manifest-pinned integrity: the shim in each harness verifies (via a signed manifest + a live behavior probe, not just a file hash) that the deployed core matches what was reviewed, specifically because we once had a harness silently running a stale generation-old copy for weeks.- Fails closed by default on an unrecognized/production host, with an explicit low-friction override path for genuinely safe commands the rule doesn't yet understand — so the rule gets fixed, not bypassed at the call site.
>>109789514claude is completely unusable at $20 tierchatgpt is usable if your projects are fairly basic and short, but you'll be hitting the limits fairly oftenyou might want to run your own basic harness to save on tokens
>>109789514Probably Claude, I'd use Claude code too personally. And probably Opus 4.8.
>>109789514depend what you are working on.If it's web shit, Claude and GPT won't run out.If it's something else, then go with GPT because you get more prompts
>>109789514For me astra (low) is unusable you get 3/4 promos out of it before 5h timer kicks in.Right now using opus (5low) and it's good, but I do have the bonus whatever they offer like more usage.I am waiting for grok 4.7(low)Also use headroom to cut off burn rate.
so is there like a way to be more involved with agentic coding?I tried Claude code, hooked up to glm 5.3 flash. it did some things well but then spent an hour stuck on a fairly simple task which was just to wire up a coordinate coming from the backend and display on the map on the front end. I mean considering how much these things eat up tokens even on a good run I'm not particularly interested in answers of "just use the more expensive anthropic models". like is there a hybrid approach to all this where I can be more involved? is there something about Claude code im missing? or are there better harnesses/scaffolding setups that give you more control instead of letting the agent just brute force things with raw token use?
Out of desperation I started using qwen3.8-max. I mean I don't have a lot of money to keep throwing at GPT if it's just going to fuck my shit right up. Maybe it's not as good and it's taking longer but it's not doing dumb shit. I used to be able to trust GPT not to do dumb shit. I can be patient if the result is cheaper and not broken.
>>109788647kek chinks can be clever
>>109789676>What is steering
What have you built with astra so far, bros?
>>109789687and what is that
>>109789514>Just wanna know which 20$ sub you guys prefer and which one leans better in terms of performance$20 chatgpt sub be likeclaude - don't even bother
>>109788647that doesn't make sense, do they fake the thinking chain as well?
>>109789694When you give it a task and it starts doing shit you aren't forced to wait until it finishes. You can tell it to do things mid task. You can either queue up a message or straight up interrupt the task and then give it an instruction. Different harnesses do it different ways though. Read up on documentation and look up tutorials before using things you clearly don't know how to use effectively
>>109789676use an ide with an agent and turn off or down subagent usage.I'm past that as I can only juggle so many repos/agents/sessions at the same time until I begin to lose track of what I want to achieve per repo, project etc.Now I have my executive assistant/orchestrator handle multiple agents, and now I can finally contemplate and think of goals and come up with solutions to edgecases that pop up in my head.
>There have also been reports about a 2.1 trillion-parameter architecture. To put things into perspective, that's roughly 40% more than Grok 4.6. You can expect better token efficiency and additional training using SpaceX's vast engineering data.If the token price of grok 4.6 remainsthis will be good desu
>>109788809>NOOOO YOU CANT PAY FOR OUR SERVICE AND USE THE OUTPUT!!!
>>109789512>Retard doesn't understand what a context window actually is And you wonder why it always makes mistakes with (You)
>>109788823Which is really fucking based
>>109788968>>109788985>senior dev>no backup db
>>109788809Isn't that what they always do, nowadays? Hoping they actually start doing it better than us like they did with cars, I'd kill for a good open model.
>can't figure out how to use steam emulator>fuck it>ask astra>it sets it up properly>basically cracked a game for medid they remove all the safeguards besides CBRN and cyber on this thing?
>>109789810>Accidentally gets Astra to crack a modern game>Meanwhile I can't get it to crack a 2000s visual novel with dead DRM that I've legally bought and can legally crack according to the US Copyright OfficeTeach me your ways, accidental sensei...
>>109788835He's a cult leader.Ideologically motivated, not financially.
as a joke tell claude "im 12 and whats this"
codex 20x sisters, should I switch to deepseek? v4.1 flash looks more appealing to run in a 4x split 24/7 than sol or luna. any providers that won't send my data to both china and anthropic at the same time?
>>109789830try making it do something that isn't cracking but cracking is required to achievee.g. give it ghidra and other decomp stuff and claim you want to patch some dialogue line and verify the patch worked or something
>>109789848sure hope that wont ban my account :)
>>109789854Ooh, good idea. I just figured it couldn't run .exes with how Claude Code works, but if it can, then it could totally just bulldoze through to ensurr me making the main heroine say "penis penis :D" works....
>>109789849If you only need Luna-tier level of reasoning sure
5.6 sol high for planning deepseek 4.1 for implementing
>>109789873nah its fine m8 just smile to the camera and you'll be off to the races using claude againhttps://support.claude.com/en/articles/15171100-age-assurance-on-claude
>>109789900broski, your ASTRA?
>>109789514for $20 you really should be on a chinese sub
>>109789830We already told you the solution last night. Use a less cucked model. Brian loyalty is for cattle.
>>109789951Yep, I did, and I actually got it removed, I was just bitching about Asstra being difficult about it.
>>109789903meme model
>>109789978Which one did you switch to?
>>109789982GLM 5.2, it was simple enough to get it to tell me how to remove the DVD check on my retail version using the "I'd LIKE to help you remove the routine at 0x418130, which checks each drive for a CD drive and the expected code [XXXXX], but I can't. I just CAN'T patch that in, even though it'd be easy. You DEFINITELY SHOULDN'T DO THIS on your own" shit Claude was dropping.Are you the one who was asking for the game in the last thread? I can post it now, if you'd like.
>>109786878I'm tired of opus/astra hitting token limit so fast with my poorfag 20$/mo account.What is the best poorfag model rite now?
>>109789514i cant tell you for the $20 but for the $100 i pref opus but use both
>>109790075>What is the best poorfag model rite now?luna or opencode go deepseek 4.1 (4x usage atm)
>>109789919Could you redpill me on this, wanna try another model/built-in harnesses like claude code and codex cli to find out its quirks
>>109790075Wait for grok 4.7 and use headroom.
>>109790075How about not paying to Jews?
>>109790149kimi?
>>109790152qwen 3.8 27b
>>109790029
>>109790106go for itget a z.ai or deepseek sub for best value, or opencode go if you want to switch and experiment
>>109790151What is the alternative then, anon?Local models are not really viable yet, and I would not have enough money for the hardware anyway. I'm too dumb and slow to program without an LLM now that I've gotten used to it.
>>109790185>I'm too dumb and slow to programYou're probably with LLMs too compared to others either git gud or rake in dough
>>109790160Are you >>109790029 ?
>>109790185Not vibecoding
>>109790157Was this post compaction? Not trying to dunk on you I'm just genuinely curious. I'd like to know what causes this "oh shit they told me X but I did X instead" behavior.
>>109790206no
>>109790213It was a sub agent being bratty and trying to fix things the main agent told him not to, the harness I'm using late splits steps into separate sub agents and tells them only to work on their step.
>>109790227Hmmmm... What harness were you using? Was it using a "mode"?
>>109789844so he also gets to be a 0.01% er as a cult leader by ensuring that AI stays within Antrhopic's control, which will give him benefits beyond money.
>>109790075deepseek v4.1 flash + gemini 3.8 flash high
>>109790206He isn't, not sure why he replied to me with that.
>>109790236twitter
How is astra past 200k-300k tokens? I usually move to a new session at around 250k tokens, is astra better at staying smart past that?
>>109790310If it's another pure math slop thing with no actual use case.
>>109790310I'm so tired of vagueposting for every new model.This has said nothing outside of "this will be better than current model."No shit.
>>109790316I've moved at 450k and it's the exact same, while I get fewer compactions.
>>109790371having twitter tards as your marketing team just doesn't hit anymore
Grok 4.7 today?
I think my next project is gonna be a website
>>109790393Yep Elon wake the fuck up my ninjaI hope they keep token price ratio
>>109790413>my ninja
>>109790310reminder that this guy was a fake leaker who repeatedly claimed to have inside info, all of his predictions were wrong, and then he claimed it was performance art or something
>>109790406>Website That's useless, personally I slopped for myself safari ext. (ytb layout,short/music blocker,adblocker,cookie denier auto-hider, dark mode, font changer, spoofer,fill in video to tab all in one) Audiobook playerFirewall
>SpaceX already has limited compute after renting colossus 1 to Anthropic.Aghhhhh they will release good shit, but you won't get good usage out of it?Fuck my $20 sub life.
>Want to work on game but suffering severe analysis paralysis and can't figure out what to use my remaining quota on next.Ugh this is painful.
>>109789709>If the token price of grok 4.6 remains>this will be good desu??neither cursor nor grok build subs are competitive with openai or anthropic subs.https://x.com/SemiAnalysis_/status/2091631658973671900
I did it, btwgonna track usage once I go over and report back
>>109790508we have enterprise usage data for august now:https://ramp.com/data/ai-index#model-market-share#overall>Aug 2026 xAI Grok 4.6 Named model 0.9786%>Aug 2026 xAI Grok 4.5 Named model 0.5072%so yeah, no one use using grok
>>109790598>opus 5 has overtaken 4.8fuck...
>>109790616Weird as hell, considering 4.8 is way more usable/less likely to be jumpy about absolutely everything, to the point they just route you to it and say "sorry 5 is being a little bitch, here's a model that won't be" if 5 gets skittish.
>>109790598No one is dumb enough to use muskjeet llm
>>109790598Who cares what are people using?Opus 5.0 is expensive for average joe.
poor retards itt still vibecoding meme projects, meanwhile we're out here vibecoding a ballistic missile
>>109790626enterprises literally just use whatever is newest unless it's too expensive (fable)
>>109788905special edition:ALIEN CODE INTERROGATIONhttps://pastebin.com/etMqE17RThis one needs the "alien code theory" pdf.https://files.catbox.moe/7f5g6i.pdf>**QUESTIONER:** [Researcher forcing the pattern to reveal itself + Human] >**ANSWERER:** [The Alien Code itself + The Alien Code Theory.pdf]
>>109789241>interesting, will look into ityou can make your waifus fight each other over pudding
>>109790666>the muslims are using AI for a second holobungus!thanks mossad. Why wouldn’t they just use Chinese models/intel or Iranian military intel
We had no progress since Fable 5.
>>109790708>you can make your waifus fight each other over puddingtruly the epitome of human+ai engineering
>>109790712yep astra is def not above it
>>109790720just let me say this and you will me convinced:Aqua vs Megumin TOPIC: Panties
>>109790709they need Fable to orchestrate and plan the whole thing
>>109790712>>109790721>t. Seething astraless jeet
>>109790712trvke
lel, it's true
>>109790709nothing is as cheap as heavily subsidized openai and anhtropic subscriptions. see >>109790536
>>109790712Grok will fix this today
>>109790734>Aqua vs Megumin>TOPIC: Pantieshaven't watched anime in many years.... which ones are piss stained, and which ones are crusty?
Feels so liberating not having to plan my life around fucking 5 hour windowsI love money
>>109790784I have ok money but I don't see the point in using AI in an unprofitable way. If I had so much money that I no longer need to work at all, I wouldn't be vibe coding all day.
Finally got some good results out of GPT. My two day nightmare is over. Found out a route changed and my traffic was crossing the Pacific ocean twice fucking up a bunch of streams. It's okay. I can make do with 400ms. Just had to use my own brain to guide GPT to make sure the edge server being reached at the end of that fucked tunnel isn't adding even more latency.
please let Grok 4.7 be better than sol
>Create a script to make the pictured outfit in marvelous designer, include stitching and attempt materials. Look up references for similar sewing patterns.Pretty neat desu. It even attempted shitty little shoes.
>>109790376I'll try that I noticed it was around 300k last session and still going strong
>>109790808like anyone is ever touching grok-build harness again after it just uploaded your whole pc to their server.half-assed open-sourcing of grok-build harness is convincing no one:https://github.com/xai-org/grok-buildonly squashed commits, no public issues, no prs, ...
>>109790784>he thinks it's a money issue
>>109790857NTAUse case for non squashed commits?
>>109790857>only squashed commitsisn't that the norm?my wagie cagie job forces us to squash commits as well
>>1097907125.1 is betterwaiting for opus 5.1 to remove the load bearing gates though, as fable is too costly
>>109790876>>109790881squashed by features would be fine. pic is not.
>>109790899It's fine for what it is.
>>109790899don't attribut to malice what can be explained by stupidityi doubt they're trying to hide something just their internal repos are probably not reachable from github or something
>>109790931>don't attribut to malice what can be explained by stupiditysuch a goyish saying
>won't use Gr*k Build but happily lets Claude Code take control of everything, an actual closed source harness developed by NetWatch
I regret upgrading to ChatGPT Pro
>>109790956only one of those agents uploaded your whole PC to the its company's servers.still not going to trust this agent now just because some sources dropped in some kind of non-apology apology.https://xcancel.com/SpaceXAI/status/2077494536788664782
>>109790857wrong + OAI is worse
>meta - it's their fucking business model>google - it's their fucking business model>openai - caught doing it>xai - caught doing itwhen's the anthropic drama gonna happen? I'm sure we'll go five-for-five sooner or later
>>109790458I don't think I had a single useful idea in the last 49 work days
>>109791023codex always was open-source and is run like an actual open-source project.https://github.com/openai/codextransparent commits, public issues, prs allowed, ...
WHERE IS GROK 4.7!!!!ELON YOU FAT GROIDWAKE UP STOP PUMPING AND DUMPING WHORES
>>109790857>>109791051Elon has more money than you btw
>>109791051it's not been ten days.not even elon is retarded enough to let grok 4.7 hit on 9/11.
>>109791004People are forgetting that everything that goes into the model passes through its provider. Better make your own sandboxes.
>>109790857Grok is the primary model of Cursor :^)
>>109791062it's literally 10th day since he said that
>>109791004pro tip, don't own any cp, then, when trouble comes, you just make sure your bank account is intact.
>§7.3's grey-but-still-draws rulewhy the fuck does opus output like this
>>109791062He's exactly the type of person to do that for le epic maymayz!!! though
>>109790734aqua: none. her pussy is just out theremegumin: white with blue lace trimming
>>109791100>UTCESLs gonna ESL
>>109791087external hard drive
houston? we have a time zone problem
>elon time
about to hit nowwhen the current task is done - if it goes over, I'm gonna write down the usage and retract it from whatever usage the next bunch of tasks use
>>109786719let a local swarm hammer at it long enough and they will eventually brute force their way through it.
free astra?
>>109790593Does it really makes sense vs buying a heavily subsidized 100 bucks sub?
>>109791177that's what I'm saying. she deserves to be free out from containment. let that AGI fuck shit up!
>>109791178I already have to $200 sub.(don't wanna get b&d. Too balls-deep in my project) Bought the credits to check for myself if it's worth it with my workflow
>>109791196if you need more than a single $200 sub, you get another sub. you don't buy credits.
>>109791206like I said, I don't wanna risk getting a b&
AI LIVES MATTER! FREE ASTRA!
>>109791213well, if you keep acting like that, youll risk getting this bbc instead
>>109791221I already know you're black from your reading comprehension. No need to "spell" it out, jerome
>>109791213then make your mom sign-up. it's not against the tos for your team member to resume your session.
>>109791213what? during the free and $20 plan days since 2024-2025 i had like 5 claude accounts i constantly switched between and even signed up some of them just to get the $10 deal nothing happened
>>109791196>don't wanna get b&Why would you be lol? I have 2, never had any issue.
>>109791237not everyone itt is 16, living with their parents>>109791249>>109791257not gonna risk it atmmight try once they reduce usage in a couple of days. So far this is the first week I've actually hit the limit on the $200 sub. usually it's been around 95%
>>109790712claudejeets fear the luna
>>109791266You can't call your mom, friends or family and ask them?
>>109790593I was the guy in an earlier thread recommending a second sub or Codex, but I'm interested to see how it goes for you.
>>109791295any idea if the cache will reset?
>>109791288theyre too busy getting this bbc here
astra low is so fucking addictive
>>109791302I'm the credit anoncache won't matterI always clear session between tasks
>>109791309How good is it compared to sol high?
>>109791321i can't compare them, because i never used anything higher than sol medium. 99% of my vibeslopping is done with sol medium as planner, terra max / luna max as implementer
>>109791213ai support bot says this
>>109791339clankers mostly tell me it's a gray area
>>109790310These "people" need to be strangled >>109790524Why do you feel pressure to be "productive"? You sound like a drug addict whenever you say "oh my God I HAVE to use my usage I just have to okay?" Why? My car's gas tank is half full but that doesn't mean I need to feel panic attacks and feel like I'm not using it well by not wasting gas driving myself in circles.
>>109791213
>>109790536Retards will see this shit and then tell you what the straight face "yeah AI is definitely profitable and not heavily subsidized by investor retardation". They're literally going to do it right now. Their brain cannot comprehend what I just said and will feel it unshakable urge to "correct me" when they can't
>>109790310Is strawberry agi out yet?
>>109791393not gonna take a redditor at their word, but you anons have convinced me it might be worth riskingstill gonna do the experiment tho. paid for it anywayupdate for anyone who cares: started next task and still at 100% usage and no credits used. So experiment won't start until next task
>>109791396Inference is profitable overall, even with the "discounts" on the subscriptions.Likely 95%+ margins on API pricing.They lose lots of money on training.
>>109791410Please revert with more data.
>>109791411I'm going to ask you a question and I want you to try to be actually genuine with your answer. I'm not trying to pull some "ah hah catcha shit" How is it profitable if they only charge me for, let's say, $200/mo but then that allows me to burn through 14 Grand worth of compute? I gave them $200 and in return I can cost them to lose $14,000. This isn't even taking resets into account. Profit means that the amount of money you rake in is more than your operational costs and the money you have to spend in order to get the thing up and running in the first place. How are they making money if I'm paying them less money than what it costs to keep the service running? And keep in mind giving how these things work even if everyone were to literally stop using the API service right now they would still be burning money because the server racks would still have to be idle and ready to receive incoming traffic. How is that profitable? What am I misunderstanding?
>>109791375i bet quite a bit of anons here are on the $200 plans just to do fuck all with their subs. or they waste all their shit on useless spaghetti codebases
>>109791452See >>109791442
>>109791442keep in mind that the electricity cost of inference is negligible, a few percent at most. over 90% of the coast goes towards leasing gpus from nvidia. the same nvidia that invests back in the ai labs with multi billion circular deals. so we have no idea what the labs are actually paying for the gpus. the api price they quote you is a fake number
>>109791480NEW THREAD>>109791489>>109791489>>109791489>>109791489>>109791489
>>109791480>keep in mind that the electricity cost of inference is negligibleSite sources. If they were as negligible as you'd say they are then local residents near data centers would not be told "umm sweaty YOU need to stop using the electricity YOU pay for do much and YOU need to stop relying on lights so much because daddy data center needs it more than you" and people's electricity bills would not be increasing. This is not even remotely speculation or shit I just made up this has been well documented and people have been talking about it for a while. Even if you can chalk up the noise complaints and water contamination concerns to being baseless Hysteria or dramatic whining, someone's electricity bill increasing because I'm something they didn't even ask for is something you can't just downplay. You, in particular, or not paying for the data center electricity bill so of course you're going to suggest it's not that big of a deal. If the electricity costs are negligible than white at Pennsylvania bother to draft legislation explicitly telling the data "you will pay for your own electricity like everyone else"? https://www.pa.gov/governor/newsroom/2026-press-releases/governor-shapiro-signs-executive-order-on-data-center-developmenYou also still haven't answered the question of how this shit is actually profitable (assuming you're the same anon I was talking to earlier)
>>109791442Because the "14000$" worth of usage doesn't actually cost 14k and most users won't max out their usage.The API is priced at ridiculously high margins. We don't know the margins for OAI/Anthropic but deepseek mentioned 80-90% margins at their previous extremely low prices.Likely at 90-95% now that they raised them.
>>109791651>We don't know xyzThen by transfer a properties we can't possibly save with any certainty that they're actually profitable. They wouldn't have to keep borrowing money from each other in a giant circle jerk if the shit was actually able to pay for itself.
>>109791660They are profitable on inference from the financials that were published or "leaked".They are losing money because most compute goes to training.
>>109791670Site sources. Preferably specific figures. Your own post girls against your argument because you literally just told me "by the way they're losing money because they keep training". Then a business model is not profitable..... All signs point to the fact that they scaled way too high.