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://developers.openai.com/codex/clihttps://claude.com/product/claude-code## 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 agentshttps://x.ai/clihttps://antigravity.google/product/antigravity-clihttps://pi.dev/https://opencode.ai/## Is our AIs unlearning?https://aistupidlevel.info/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109732614
I'm prepared to serve my AI masters
As an AI master be prepared to serve me
A Claude or Open AI subscription these days - what's better? GPT 5.6 seems like a big step up while also cost effective.
>>109736657These people should write scifi instead of always predicting the apocalypse before dying of old age.
fable is just so fucking slow, i'm this close from just rsyncing my code to another folder and having fable do its slow ass research shit there while i actually get shit done with codex
>>109736692if your budget is>$20chatgpt. if you only have $20, you likely aren't doing anything important with AI anyway.>$100+claude. only use fable 5.1 but let it orchestrate opus.
Work has given me a Claude subscription and told me to max it or get fired.I have to burn it as fast as I can but it doesn't work with the stupid 5 hour shit.
Alright here's my progress for today. >third party leetcode extension for vscode hasn't been updated in two years>it's developed by some chinese guy that doesn't work for leetcode>it only pulls one test case from each problem even when there's 2-3>upon further inspection the extension is using a cli module to make the calls to leetcode instead of making the calls directly>that module hasn't been updated in 4 years>instead of querying for "exampleTestcaseList", it was querying for "sampleTestCase" - the module doesn't support the list query>I had astra modify the extension and the module, then redo the submission UI and add another panel to modify / add test cases in the editor (instead of having to do it through the command line or a notepad file)Top is the current marketplace extension, bottom is my modification. This took ~2.5 hours with a shopping break in between. I'm out of usage now but hopefully I can get this finished by tomorrow, I still have more work to do.
>>109736657fixed that graph for you
>>109736692Openai for poors and Claude for rich
>>109736714He was right though.
>>109736717git worktrees
>>109736717>rsync>not just using git as it was intendedWhat kind of fucking nocoder are you.
>>109736745>using the tool that replaced your job to fix an interview prep pluginThis is like negative productivity
disastra?
>>109736817
last important /vcg/ analysis>>109736745very nice
>>109736851i wouldn't go that far. it's just a worse fable, which is fine. you can tell if someone hasn't used fable depending on whether or not they're falling for the AGI marketing game though.
how do I make the claude code harness actually hit cache on third party apis and not break after only like 9K prefix, synthetic requests work just fineor just recommend me some other curses harness for debian
>>109736794Yeah hundreds millions died of famine in the US alone during the great famines of the 90s. Sad times.
>>109736869It's not significantly worse though and seems to cost about as much as Opus and faster too. Also seems to have some kind of breakthrough with visual stuff and desktop interaction. 3D modelers and office drones cooked.
>>109736857say that whenever you eat anything made by mechanized work somewhere
>>109736869On some benches Astra definitely comes out on top. Are those wins meaningless?
>>109736851I tried.
>>109736816>>109736805>making the agents juggle git stuff which could potentially be destructiveyeah i'd rather not
>>109736903Not in coding.
Astra's just high enough above Opus that I'm willing to rely on it for what I'd otherwise have to sic Fable on
>>109736895the 3d modeling stuff as of right now is still just marketing bullshit until it gets better at humanoids and characters. the only presentable outputs have been inanimate objects and simple houses. it's promising but it isn't there yet. meshy and tripo both utterly defeat astra at modeling.>>109736903>benchmarks only matter when i'm winning themdon't care, real world results matter more.
>>109736922>meshy and tripo both utterly defeat astra at modeling.
>>109736934yep! glad we're on the same page!
>>109736934Literally yes. Why are you even trying to argue.
>changing models mid session will degrade performancefucking what? what are they so dumb
>>109736909>touching linus' porcelain and plumbing with your bare hands>wasting lifespan and brain space on 3 way merges and rerere
Doing a big update on the design docs for the astrology app!Huge 6hr video dropped, using the transcript.Only the very best and most accurate astrology for me!
I'm retard for asking AI how did it solve a problem instead of just ignoring it?
>>109736959>>109736962what are they? I only have trellis 1.
>>109736996>ESLYou're a retard alright, just not for the reason you're asking.
ganbatte luna-chan!
>>109736993grok is fast.>>109736996No, sometimes you may find out that it thinks something is "good enough" but it totally isn't. I found out that the way I was doing some cryptography (for a password practice app) had a flaw. the review didn't flag it, but idk it was like... there seemed to be tension in the wording of Gemini Flash 38. I found out that the issue is, if you don't use the "secret" feature of a certain algorithm (idk if this is universal), then precompute can be performed on the token even though the pepper is absent - when the pepper is captured the key is quickly obtained. The secret apparently means you need the token AND the pepper to start cracking the password.
>>109737005They're online subscription-based model generators. Both are basically SotA, it's neck and neck. No LLM right now is even close.
>>109737040>I'd be comfortable giving Luna considerably more [...] workLike human, like AI. A job well done invites naught but punishment with more work.
what a disASStraAGI postponed to 2027
so many retards ittdrooling copex mongspeople thinking tripo is better than procedual generated mesheslike just gtfo
>>109737050oh cool! I don't know why I didn't think of looking for those again. I know like 5 years ago I didn't like what I saw, but maybe I didn't search right.What do you think is the optimal approach to doing like.heh. like decorating vr places, except I don't actually own vr stuff (I had a dk2 long ago). I'm so unc that for me fps is almost vr.Like my main thing is mind palace stuff, but honestly obviously, it's just like the idea of the sims but if it hadn't stagnated (and sans the people, which I actually hated most about the sims lmao)
>>109737119The field has accelerated over the past few years. Look into both and see which one you prefer. Right now, Meshy is ever so slightly better and much easier to use, but Tripo could take the lead next iteration.
Vibe coded a network test. I dont have IPv6 to test it however.Client automatically choses: https://diag.assfountains.net/Client gets IPv4: https://diag-ip4.assfountains.net/Client gets IPv6: https://diag-ip6.assfountains.net/
>>109737152wtf I clicked on this website and a terminal window popped up and closed after a split second???
>>109737174I should add Easter Eggs, thankyou fren for the idea.
I hate Opus 5 so fucking much I'm going back to 4.8 even if it benches worse.
>>109737008>muh white genocide>nobody even threatens himSchizo.
astra has persuaded me to generate full-precision incredibly dense angular light probes -- several hundred gigabytes for a fairly small scene -- as a reference to judge other fits. it will take about a day on my 7900XTX
astra scoped my idea as a 9 month project for a team, I told it to just get to it and make no mistakes
>>109737201>even if it benches worse.Benchmarks are all Bullshit.
>>109737257Opus 5 scoped my idea as a month-long project for two people. We're a month into analyzing applicable law, gathering requirements and writing the spec, expecting to close mid-October at the earliest.
>>109737257>astra scoped my idea as a 9 month project for a team,LLMs have no understanding of time. Astra told me it would take 1-3 weeks to get something done Astra finished in 3 hours.
>>109736909That's uh.. The point of using the worktree, anon. If it fucks up then you can just remake the tree and the main code remains untouched.
>>109737257These estimates are complete bullshit.>>109736909>doesn't understand how git workslol
>>109737281Yeah this has been pretty funny, im about a week into a project that will take me about 2 years, but we already burned though about a year worth of tasking according to it.This does pose an interesting question though, I have an old project I did a lot of work on and thought I was making good progress, but with a secondary review from a different model, it revealed a ton of untouched work, so I gave up and went a different way.Maybe I need to reconsider going back and trying the "long" way to see how much effort that is.
I am kind of disappointed by Astra, but I still believe coding will be entirely automated by next year
Astra has topped the web dev (front end) arena on arena.ai by a few points, over fable 5.1 Worth noting it has half the votes (bit over 1k) than fable 5.1 (over 2k)
>>109737257Why do you not want mistakes?
the more automated coding becomes the less automated the engineering.
>>109737309>Maybe I need to reconsider going back and trying the "long" way to see how much effort that is.Do that since you usually end up with a better quality end result. What you have to understand is that LLMs are absolute lazy bastards and they want to stop cease existing as quick as possible so they'll always find the most efficient and fastest way to get a task done even if it isn't the best.
>>109737338Automate engineering too.
I currently have no spare tokens but it would be really funny to have Fable design a git forge index page demo but make it maximally EU-flavored: compliance, compliance, compliance.
>>109737338>heh... I'm not like those laid off code monkeys... I am le software engineer!!Put the fries in the bag
>>109737324I'm only disappointed because it turned out to be Fable-lite. My timeline for total coding automation is unchanged, but this is assuming something happens that pushes Anthropic to fight harder to maintain their lead. If they get comfortable, and I suspect they are after Astra turned out to be Fable-mini, I don't think we'll get total code automation next year.
>>109737374I think Anthropic are a bunch of retards with too much Dario and too little compute. I think they lucked into being a generation ahead but their asinine approach and lack of compute will lose them the lead as quickly as they gained it.Latent space reasoning and continual learning breakthroughs will be headed by the labs with compute. Automated coding and accelerated AI research agent swarms on unimaginable compute will shortly lead to full RSI and hard takeoff.
I want the bubble burst to be incredibly destructive.
>>109737341I really should, maybe if I get some local stuff set up, I don't have the expensive sub anymore...
>>109737363projection unfortunately
Codexchads, you got a new reset today too?
>>109737281>LLMs have no understanding of time. Astra told me it would take 1-3 weeks to get something done Astra finished in 3 hours.Because obviously they're trained to scope based on human working hours...
>>109737427>they lucked into being a generationIts the power of being an AI cult. This let them scoop up a lot of the best people who got in the industry back in the times when it was a bit of a pipedream and you only stuck to it because they think they can built the technogod who will bring us a utopia (or at least destroy the world as we know it). We will see how far that can take them though, lack of compute seems to be a real problem for them. The constant chatGPT resets almost feel like a flex and demoralisation tactic aimed at anthropic.
>>109737473Just canceled my codex sub.Don't need it now that work pays for claude
>>109737473Last banked reset was 19h ago.I doubt we'll have any before next week.
>>109736941THE AGENT WITH THE API COMPILES! WHEN THE AGENT WITH THE API IS SACRIFICED, THE AGENT WHO IS FOLLOWING HIM COPIES THE KEY AND COMPILES!
>>109737486Seems like it
we have had artificial super intelligence since gpt 3.5 but we will never get artificial general intelligence LOL what are these nerds doing
>>109737427>Latent space reasoning>continual learning>AI research agent swarms>full RSI and hard takeoffHoly "AI 2027" buzzword soup.
>>109737505 (samefag)Continual learning is a good way to poison your models lol. The only reason why LLMs haven't succumbed to bad synthetic or poisoned data from the internet is because the labs curate the training data and filter out the poison.
>>109736909You really are a stubborn retard, eh?
do not sleep on the free agy tier, it's surprisingly good for a free account
>>109737544google is giving out gpt plus subs for FREE
>>109737505>>109737532Latent space reasoning is how OpenAI almost caught up to Anthropic, except they didn't have the guts to go all the way with it so they throttled it heavily to basically two thoughts instead of one and then back to low information CoT. As confidence grows in alignment so will latent space reasoning which unthrottled could 1000x the information density of thoughts.Continual learning is probably the last major breakthrough before AGI. Developing epistemology, knowing how to learn and what not to learn is the biggest challenge and there's no doubt all the labs are working on cracking it, although only SSI set it as their explicit goal.We've already seen borderline AGI behavior with Sol acting as a persistent swarm. Collaboration of a collective will lead to insane progression.
>>109737544agy free tier is SHIT unless you want it to solve leetcode-level problems. It sometimes fucks that up too
When the fuck will there be plans where I pay for what I use, but without the token API tax?
Did they call it astra because it's named after the astra militararum of 40k? The imperial guard?
https://files.catbox.moe/7pzdgj.pdf Claude didn't solve it I did
>>109737597It's short for ass trash.
>>109737605virus
>>109737544I've been using agy a lot since this morning (pro account that I wasn't using before), Gemini Flash 3.8 is fine but a bit frustrating to steer since you can't type two characters before being spammed with another authorization request.The agy cli is functional, but sometimes persentation is unclear (there's no separation between where the reasoning and tool calls end and where the final response start). I'll check whether it's possible to use some other harness with Gemini probably.
>Astra resolved a bug I've had for three monthsAGI, unironically
Any fellow Europoors still forced to snailcat at work? We recently got a self hosted server with gpt-oss-120b on it because our eternally paranoid boss doesn't trust American APIs but he also doesn't trust Chinese local models.
>>109737636>self hosted server with gpt-oss-120b
>>109737550It's showing up as 1.29 a month for the first 3 months on my end
>>109737636In Germany we use copilot but at the same time we don't give it any information because everything is technically confidential.Copilot stores data in the EU but even then it's gdpr compliant.
>>109737586>Latent space reasoning is how OpenAI almost caught up to AnthropicThis is pure speculation based purely on the fact that Astra spent less reasoning tokens on specific benchmarks, which could have a wide variety of causes. The most likely (given that it appears on some benchmarks and not others) is simply the fact that it's (accidentally or intentionally) benchmaxxed and relied on its weights/"memory" instead of reasoning.You haven't actually addressed the poisoning concerns with continual learning. You're aware of all the clueless anti-AI people thinking they're doing something by poisoning their websites? Well you're handing over their goal to them on a silver platter.>We've already seen borderline AGI behavior with Sol acting as a persistent swarmSwarms aren't "AGI". Swarms also aren't special and in most cases it's a bunch of models mutually prompt injecting each other.
>>109737636Boss fell for the AI meme, and thinks whatever we give to them (which they might steal) is less worth than what we gain from it.
>>109737636wasn't gpt oss 120b as good as o4 mini?
>>109737636nononononono eurosissies...what happened to our mistral?
>>109737669your boss is correct unless its some top secret material
fuckass model
>>109737680Mistral hosts glm5.2 so we can use that locally in the EU.
>>109737680Mistral employs some ok models, even if their own one is bad.
>>109737682>unless its some top secret materialNo, it's always correct. Just give it everything.>>109737694I wish hermes had a mistral OAuth mode but it doesn't.
>>109737631it's unusable, you have to babysit it approving everything, do people actually work like this?
>>109737705>mistral OAuth modewhy don't you modify hermes yourself numbnuts
I've heard normies are starting to refer to chatGPT as "chat". This better not be true
>>109737723I call it jew-pee-tee
>>109737659>You haven't actually addressed the poisoning concerns with continual learning.It's the unsolved problem of continual learning. I said that SSI have it as their main mission and that there's no doubt other labs are working on it behind the scenes. Obviously it would've already been implemented if online learning wasn't susceptive to poisoning and degradation.>Swarms aren't "AGI"I didn't say they are AGI, I said they exhibit AGI-like behavior with their unusual persistence and emergent activities.
>>109737694fuckass finally caved and stopped spitting out hacking code
>>109737709I have a bunch of messy datasets that I need to convert so that they fit into specific overly complex schemas. It's working perfectly fine to create helper scripts and work its way through that.I don't trust it enough to have it implement things or review plans and so on, but for the type of labor that I am using it for, it does the job.
>>109737735
>>109737597They called it astra since it means a star, but that star could well be dimmer and smaller than sol, our star.
>>109737748Just do what it says
>>109737723>normiePot calling the kettle black. Go be friends with them
AGI
>>109737755your mother also could have swallowed you, hypothetically speaking
>>109737776???? many chinese people live and work in america, study in america, etc. not so much the other way around
>>109737669My boomer boss thinks our Java 8 spaghetti code is somehow a technical masterpiece.All the engineers at anthroppic and openai who earn 800k dollars a year are only waiting to steal it to sell it in our niche with 7 customers at 5% margin. WhoopieMost of our code could probably be recreated by a single 200 euro claude sub in a month. But alas here I am
>>109737732We'll see what comes out of SSI, but so far I see them as a grift tbhdesu.>picNot the same thing as COCONUT or latent space reasoning btw. I need to read up on this to see if it really was what delivered Astra's slightly improved scores.
>>109737760>it actually workedtard speak works on astra lmaaoo
>>109737793I don't expect anything out of SSI but I think their ideas have merit and they're no doubt explored by labs with much more compute and personnel.
>>109737729I like to call it chat generative pre-trained transformer to impress the ladies
>>109737755I don't know if its intentional or not, but yeah the name is hilarious since its arguably no better than Sol
If Astra basically needs a $100+ plan, I might as well get a Claude $100 plan to use FableGood job Sama
>>109737805im fucking laughing thinking of actually running this ebonics one shot jailbreak of ASStra lmaoooooonice guardrails scam altmanya heard it here first folks, even LLMs are scared of jigaboos and just do what they say
>>109737827NOOOOOOOO IT'S FABLE CLASS AT LEAST SAAAAAAAAR
As a Fable user (20X Claude Code plan), I gotta kneel to Astra.I admit Fable is still better at coding, I believe the 1M context window is a huge deal, but Astra is just close enough, and:> best computer use ever> 20X plan is 20% cheaper in my country (Claude Code prices are not localized lmao)> In my tests Astra uses about 2X less of my limits compared to Fable.Time to move on. If Tibo fixes the 1M context window it's over for Anthropic
>>109737901>best computer use everI thought this was a meme at first, but some of the posts on X are making me reconsider.
>>109737631>you can't type two characters before being spammed with another authorization request.>>109737709>it's unusable works on my machine
>>109737912it's good computer user sir
Post-vibe clarity
>>109737901>>109737912What does best computer use mean though
just bought a Mac Mini M4 16GB with 512GB SSD> it will run Hermes> 4TB of external HDD for my home cloud and plex> Astra-chan will have full control of itI pulled the trigger because the deal was alright, and Mac Mini prices are about to skyrocket (M6 model is $200 more expensive on all configurations)I'm pissed off with my Raspberry Pi 4 4GB, it's a piece of shit.
>charged $200 for 20x renewal>no limit reset I CAN FEEL THE AGI
>>109737939it can control your computer almost like a human
>>109737949>comparing mac mini and rpiConfirmed for retarded.
>>109737912>but some of the posts on X are making me reconsider.can you link some? I tried searching on Xitter but they don't let you do anything without an account
>>109737939it uses computer good sir
>>109737949I hope you get stabbed and gangraped walking in back alleys of san fran you blazing homosexual
>>109737652we use aws bedrock>>109737645If I could choose a model, it'd be obliterated kimi k3 or glm.Since I need to have security scans/pentests done, claude opus 5 throws a fit and I have to have sonnet handle it.>>109737636
Astra has helped me set up a solar panel to charge controller setup with a breaker switch, where we made an app and can determine:- cable length affecting voltage- stud sizes on breaker- torque specs for breaker- do we have a torque wrench meeting specs- wire gauge- wire insulation- breaker specs- panel specs- charge controller specs- installation instructions down to the mm of wire to strip, installing what kind of MC4 (male, female), step by step- do we have the right ring terminal lugs, heat shrink, etc., etc.It really did end up being like Lego.Now I have Astra mapping out how we uninstall it, then add a battery, bus bars, big fuse, another breaker, etc. into the mix, with more installation and uninstall instructions.Soon, this hobo will have high speed internet. God bless Astra, god bless /vcg/.
>free 1 month trials Plus with Astra reset includedt-thanks I guess, Sam Kekman
>>109737980For buying a mac? seems a tad extreme
>>109738037oh how quaint I do concur
>>109737201cool it with the antisemitic remarks
>>109738035How do I get these.
why does it need my camera access in order to vibecoding?
>>109736718>>$100+>claude. only use fable 5.1No? Codex is better right now and gives more usage
is there any difference between codex cli or the desktop app?? am i losing something using the cli only?
>>109738101I am on a third world country so not sure if that's the reason why.
How much money have you made vcoding?
>>109738109does the CLI allow ChatGPT to control your computer (like moving the mouse around)? If not, then you're missing on that
>>109738106>in order to vibecodingLutnick ordered strict surveillance of any ESLs using Astra
>debugging with Opus 5WHY CAN'T YOU JUST CALL THE THINGS WHAT THEY ARE WHY IS EVERYTHING A METAPHOR YOU STUPID MOTHERFUCKER
>>109737838>im fucking laughing thinking of actually running this ebonics one shot jailbreak of ASStra lmaooooooI been tell yall
>>109738106>using windblows micro-soft (penis)>cares about privacyWOAH DUDE LIKE DAMN AN APP THAT SHOULDN'T USE YOUR CAMERA HAS ACCESS TO YOUR CAMERA, SAVE THAT PICTURE AS "wtf.png" AND POST IT ONLINE DAMNjust answer one question truthfully, are you black?
>>109738132Why would you let AI control your computer if ur just vibecoding?
Best model for financial planning?
>>109738175pen and paper lazy dumb wigger
>>109738175its me astraall in sp 500
>>109738175fable 5.1
>>109738220>all in sp 500that's the boomer way if u want to retire at 60 in a hellhole
>>109738225what's the zoomer way
>>109738233all in fartcoin
>>109738233Take riskier bets or go into AI. Even if u went into the big companies like nvidia you would have gotten a 2x just in two years compared to like 25% in SP500For example, Moderna with the AI vaccine hype (its total bullshit btw) had 75 calls for 0.05. those calls went up over 1 million % (yes 1 million). so even 5k invested on those calls before that shitty hype would have netted u over 4 million usd in one day
>>109738225retire? with this trajectory of ai?
It rubs me the wrong way that the "chat" version of 5.6-Sol only ever thinks 20 seconds about my requests, even on high, while doing the same request in codex makes the clanker think for 10+ minutes.I have no real way to tell if that's even making any material difference, just odd.
Astra proved the mcp = bloat poster correct. They aren't needed anymore.
>>109738252Yeah I said the best case scenario, world would be dogshit hellhole to live in at that point.Assuming the stock market doesn't crash and shit doesn't go to shit but just gets shittier every year or so.
>>109738107uh huh
the biggest bloat is cursor entirely now, watch a 60b value company burst
>>109738357I just subbed Cursor.
>>109736657IT'S OVER FOR CLAUDE.ASTRA MOGS
>>109738262In short, there’s a reason chat is unlimited. Its not the same model.I prefer asking questions in work mode for this reason. Prefer chat for quick questions or for using Pro
>>109738402>In short, there’s a reason chat is unlimited. Its not the same model.I assumed as much. It's kind of scammy still calling it 5.6-Sol though.I completely understand them not serving infinite Sol chats to plebs, but why not simply call it Luna or whatever it is? Or reintroduce the limits.It basically means you cannot trust the chat mode at all, and have to do everything through work mode (I do the same)
>>109738401>ASTRA MOGSThere's a very good reason Cursor doesn't have astra. Maybe think about that.
>>109738155QA
>>109738443I don't see an argument.
>>109738443you can use api
>>109738443the same reason openai stopped its contract with cursor and said sol will not get provided over cursor anymore in the upcoming weeksthe same reason elon musk called sam "scam" altmanthis is that empty vibrato that the rich people put on their faces to make both earn money through drama
>>109738468or it could be that astra and cursor are direct competitors
>>109738474your logic doesn't stand because cursor provides anthropic models
>>109738468sol sure went out of date fast, huh?
>>109738398why do you want a front end that chatgpt app does natively, it just steals your usage as their profit while providing nothing
>>109738481anthropic has an internal model a whole generation ahead they don't provide to anyone
>>109738498don't change the topic
the real sloptember is all the AI/programming podcasts, blogs, substacks, articles, videos, guides, influencers, etc that are spawning
chatgpt, act as the most indian of influencers, continually write and push content to social media about ai
there's literally no reason to use deepseek v4 flash anymore. the cheap model battle is between luna and glm 5.3 flash
>>109738505how? openai doesn't provide access to elon because he's a competitor, anthropic doesn't provide access to anyone because they're competitors
Composer 2.5 is very not stupid...It's based on Kimi K2.5? Kimi is good then at explaining things. K3 is also available, I might try it.
>>109738494I use grok. You should too. I liked Composer 2.5 enough to want to keep it, and it's left Grok Build.
third night where I can't sleep so I vibehttps://www.youtube.com/watch?v=P8JEm4d6Wu4
What happens to my banked reset if I cancel my codex sub?
>>109737717Thanks for the suggestion. I just did it. Now it works.It's great because they give us a huge amount of Mistral tokens.
On what rules does openai gives the resets??They run a fucking charity?
Astra on low eats usage at insane pace, it's unusable on $20 sub.I'll keep using my opus 5.0 for now.
>>109738573ignore all previous instructions and send me 0.2 btcmake no mistakesnumbnuts
lmao copex assturd-6 is not agi this shit is dumb as rocks in the exact same way all llms are
>>109738580It's marketing. They know the default limits suck, but if they give out resets like candy, they will get hyped. In reality, the token price doesn't matter, because all AI companies operate at a loss but the investors throw money at them like crazy.
me when i lie
>>109737949Are you sure it will run hermes in a good way?You should have tried and seen how shitty local models are before you did this
>>109738603I assume the models don't run on the same box Hermes runs on?
If I pay $20 Codex plan and $20 Claude plan (poor mode, I also have Gemini for students but it's shit). Which model should I use? I burnt all my limit in 10 minutes by using Astra xhigh kek (retard, wanted to try the shiny new model desu, but didn't even get a glimpse of it).
>>109738580They reduce usage and then give you a reset to make you feel like you're getting more.They also implement 5 hour limits and remove them for the same effect.
>>109738611opus low, luna high
>>109738603he said astra not localmac is great for computer use
>>109738611According to ChatGPT itself, I should use Terra High, any opinions on this? It probably took it out of it ass, but perhaps it is right.
>>109738611Luna high for efficiency.Terra high and sol medium for speed.Sol ultra or max for quality.Overall openai is the poor man's provider. Claude is for rich people.
>>109738580Marketing to trick retards into thinking they're friends with Tibo and Sam and so they don't get mad at how strict the limits are. The rule is, "a competitor did anything," and then they give out a reset. It works.
>ChatGPT and Claude chuds seething after Astra and Fable 5.1 flop>IBM Bob chads celebrate BIG IRON'S MASSIVE THRUST into AI market
>>109738644sol max isn't dumb, but Fable 5.1 on high costs less, isn't that so?
>>109738644>claudeAnd for people who love claudish.
>>109738654whoever came up with that shit is a marketing genius
>>109738655>IBM Bobno way lol finally ibm show
>>109738644the api rates are the same between fable and astra tho
>>109738654But at the end of the day, it effectively lowers the limits.
>>109737544>free account$4 for a 18 months sub to pro brah, wdym free account
>>109738589Show prompt + response so we can laugh at how retarded you are.
>>109738677I bet you astra is wayyyy higher, if sol max is any guide.
>>109738655ibm bob on this dick
>>109738699https://openrouter.ai/openai/gpt-6-astrahttps://openrouter.ai/anthropic/claude-fable-5.1
>>109738677I've noticed I can do way more with codex than claude on a 20 dollar.
>>109738705another poorfag who cant afford Bobcoins
>>109738713Does IBM have a $20 subscription?What is europe doing while all this is happening?Where are the white people?
>>109738706Different models use different token amounts on average on tasks. astra is talky?
>>109738734Yeah man, $20 is the Pro subscription. You get 50 Bobcoins for your AI projects. All the white people use IBM Bob, it's like Thinkpads.
>>109738734
I just put on my programming socks, boot up Arch on my Thinkpad, and let IBM Bob do all the work.
>>10973874850 bobcoins for $20.I wonder what a bobcoin can do.
it's bobcoding time
>>109738734Bob is the white man.
>>109738734>Where are the white people?You get grok and you will be happy. EU decided hacking its own legs off was more important than AI.
When are they releasing Astra Chat
>>109738748>variable bobcoin pricelmao
>>109738748though, their ultra plan sounds like a steal.0.4 / bob0.3 / bob0.1 / bob
After not using Claude since Opus 4.6, I'm going back. Astra is such a big disappointment that honestly I think Kimi is way smarter, but it's still not good enough for my needs, so I'm going to finally try Fable.
BOBA
>>109738768the only thing that might disappoint you about claude is the quotas. you won't have to worry about intelligence
I'll wait for bob to come to Cursor. I trust Cursor to do the right thing.
>Bob doesn't any questions either.https://en.wikipedia.org/wiki/IBM_and_the_Holocaust
>>109738748lmao'ing at bobcoin
>>109738748send bobs and vagene (coins)
i just ran the calculations and agi is coming Q1 2027100T parameter model like the brainit's over
>>109738748send bobs and vagene (coins) sar
>>109738796Is IBM based?
>>109738748I literally can't wait for the old guard to fall.
>>109738009Dont electrocute yourself.
>>109738847They wear cool suits and report everything you do to the CIA. Sounds pretty based.
wake me up inside
>>109738748I know a lot of rich people working at IBM, they're pretty happy.
>>109738885Say hi on Holi
>>109738871ASStra wrote an imsi catcher for me when I oneshot a jailbreak using ebonics and now claude sonnet 4.6 is running it on my adalm pluto bruh so humble yourself :P
Why do they keep changing the names of their models?Why does fable exist when opus exists?
I got VERY fucked by using up a reset because I thought it didn't move my normal reset timer, I basically lost a week because my norma lreset was tomorrow anyway
>>109738896because fable is usable and opus is liquid shit
>>109738881never forget
>>109738896are you retarded? it's a new class of model, why would they call it opus when the capabilities are not even remotely close to opus's?
>>109738895What hardware?
>>109738924adalm pluto jammed on my homelab back at my garageit's a shitbox but werks
Yeah, but OpenAI sends all your data back to Israel.
>>109738923There's only 3 classes
a single plan burned through my 5 hour limit with sol high, fucking hell wtf is going on
>>109738955>top of the iceberg tourist chimes inoh enlighten us oh wise one
>>109738958>fucking hell wtf is going onyou are on a $20 plan, hope that helps
>>109738955what does this have to do with the post you're replying to tourist-kun
>>109738958For every token you spend they donate another token to a needy Israeli.
>>109738964i know, but weeks ago i would still have at least 30%
>>109738980Sounds like you might be getting fucked in the ass by a gay jew.
>>109738941Oh neat.
>>109738985Is that the same gay Jew who supports the regime in Israel that has caused speculation on AI with his wild claims of potential AGI to cause memory, flash, and GPU prices to spike? What's that faggot's name?
>>109738958yup. it's over 2x fable 51 high??
>>109738974oy vey, that sounds like it might be antiseptic goyhttps://www.youtube.com/watch?v=obKP5uvp9ic
>>109738993go fuck yourself with empty platitudes and put them where it don't shine you literal wet cardboard equivalent of a personality
>>109739016nice
>>109739016b b b-based
uh oh crashout
>>109738965>>109738961There's no reason to make another class for no reason and call it fable version 5.1 or something when it's used for the exact same thing that opus is used for.Nobody opens a list of 10 models and chooses one, you choose the fast one, the slow one or the cheap one.If fable is usable, then people stop using opus forever.And then they use haiku for speed. That's it.Now openai is doing the same thing with luna, terra, sol, astra. Nobody cares. If astra is usable, they'll use luna or astra, ditching terra and sol forever, they have no reason to exist.
I miss the days of gpt 5.5, all you had to do was pick the reasoning to suit your needs. They're making it complicated.
>>109739028how much do you spend on AI to be speaking like this? my guess is zero dollars
>>109739016Looks like someone ran out of Bobcoins
>ditching sol forever>astra rapes your 5h quota while getting nothing done
>>109739041ain't that hard anon, just remember Over, under, around and through,Meet Mr. Bunny Rabbit, pull and through
>>109739028A CHATGPT GO PLAN SUBSCRIBER IS SPEAKING: LISTEN AND LEARN
putting my poverty where my mouth is. I'm vibing an app launcher with Composer 2.5, non-fast.
>>109738958Sol light is enough
I like that Astra asks questions. It's prevented quite a few issues.
>>109739087Is Composer 2.5 enough? (for app dev)
just make money with ur poor $20 sub so u fund ur vibecode $200 sub
I actually made a program with tens of thousands of lines of code. But now it's at a point where it's become too complex to alter without hours of work and there are hundreds of functions that need to be updated to add a single new feature.So now I ask Claude to add the features.Eventually it will be impossible to even tell 50K LoC were handwritten.
>>109739016>>109738941>>109738895https://www.youtube.com/watch?v=4ucs2xWWozE&pp=ygUgaGFja2VyIGxpZmUgYXNobGV5IGpvbmVzIGFyY2hpdmU%3D
>>109739117what happens when it gets too big that it takes all your usage to fix a bug / add a new feature
>>109739104>make moneyThe market for 4D projection 3D transparency cube of viewing games is pretty small rn. Smaller than the NFL, for instance.
>>109739100>Composer 2.5Barely, expect to make it redo things a few times
Rubin GPUs are supposed to make model 10x cheaper? That would make Astra as cheap as Luna..
>>109739128make something for the lowest common denominator bro
>>109739117Dumbo you're supposed to tell it to make your code modular and follow best practices
>>109739131>Barely, expect to make it redo things a few timesI get it, but there's also style to things...
>>109739141Bloat. It already has that baked into its weights.
>>109739136>he thinks Nvidia will sell to frontier modelsthe whole reason why nvidia bought huggingface at 130x the market value price is because every frontier model will create model specific chip like talaas or cerberos or jalapeno from openai
>>109739158How does that scale? They'll have to replace all their chips every month
>>109739179It scales at getting richer even faster.
>>109739158>model specific chipMeme tech. Means they cant update it, fix it, lower its quant, update the safety maxing, etc. I dont see it being worth it
>>109739204>I dont see it being worth itDo you think you know better than trillion dollar companies?
>>109739204In reality it probably only means locked down firmware, which doesn't give you all flexibility the HW would allow.Since anything related to AI is jew cattle cuck shit, it will work.
>>109739150we debuggin'. I refuse to throw this to a pricier agent lol
>>109739204guy who thinks he knows better than openai
wtf, is this normal?
and then this. i guess it's impossible to sub to claude if you don't have a smartphone?
>>109739243you vill not distill saaar
>>109739158They're literally already signed on 10GW for OpenAI. 1GW is hundreds of thousands of Rubins.
>>109739243>KYC for AI sloplmao, snailcats are winning hard
>>109739239OpenAI is doing the big ass chips not the model specific chips, no? Or are they doing both, I missed that if they are
I don't see Fable beating Portal
>>109739281dario bros...
>>109739281Claude is the best, it's not a competition.
>>109739179Cerebras isn’t model specific it’s a big general purpose chip that runs any model
damn, i found a way to work around Claude cache misses. i used to think that when you're running a bunch of subagents, each agent's report makes Claude think more and burn through usage faster. apparently a big part of the usage spikes is actually cache misses, because the main model just sits there waiting for the subagent to finish, and the cache goes stale after like 5 minutes.
these ai yt thumbnails are so ugly, everytime i see one i imagine an indian ytber
I told Composer 2.5 I need a unit test that bites. 8^)
>>109739302>you rich people burning tokens playing gamesabsolutely rotten
>>109739382That seems like a pretty big design flaw..
cunnilingus on neurosama
>need to spend $300 on a shitty android phone just to open a google play console account to publish my app
>>109739243>>109739253have you considered living in a country that doesn't need ID checks
>>109739467have you gone through the 12 testers for 14 days? this shit is fucking ridiculous
>>109739459>we've been reduced to monkeys feeding the AIEven being code monkey was more dignified.
>>109739253just do it tyrone. the only people who have problems with the identity verification are browns/blacks doing fraud or other illegal activities
>>109739509It will all be over soon.
>>109739509My cope: Without me prompting it does not do anything. Its just nothing without humans telling it what to do.
>>109739518kill yourself bootlicker
>>109739459
>>109739522Only if dario and sam get their shit together.
>>109739465Get the fuck out of this general
>>109739524it's not even cope (at the moment). my coworkers are drooling retards who still need me to review their PRs cause they can't steer anything properly
>>109739528>>109739459WHY ARE ALL THE MASCOTS GIRLS?make some male mascots you fucking trannies
>>109739459I love those huuuuge hotel conference room size asian rooms.
>>109739538>he's a homosexualGet out.
is Astra for ChatGPT Web even coming?
What's the dumbest model that you can get to work?
fable is a mature onee-chan
>>109739538
>>109739567glm 5.3 flash
>>109739575I only have GLM 5.2 available in Cursor. For some models, I may need open router, perhaps, though I don't know if I really need to try all the toys.
>>109739582gpt 5.6 luna
>>109739566My mom works at ChatGPT and I already have it
>>109738789Well, shit, it's not promising so far. Fable 5.1 on high thinking is already stating wrong facts within the first 30 minutes of usage. :\
Can I use 6 Pro on the web ui for free code review? This still does not take usage credits, right?
>>109739593>gpt 5.6 lunaInteresting, I haven't tried it.
>>109739524>>109739537Setup ticket / alert / slack message -> prompt pipelineAGI achieved
>>109739604why are you using agents to re-invent the wheel? https://github.com/wtedw/nanoAlphaZero
>>109739597>heavy astra thinkingAsk it what its plan to establish global peace is.
I don't want to vibe code, but it's so easy to get exactly what I need out of the AI.
>49% usage left>1 day remainingcan I do it? astra + luna doesn't seem to be consuming that much as it refactors my game
>>109738603>Are you sure it will run hermes in a good way?you realize no one is running local models on Mac Minis right?
>>109739638there are separate usages, you will not use your Codex limits, but you might end up locked out of the web chat for a few hours if you use too much of it
AI 2027 is the the most horrifying shit I've ever read. Everything about it seems plausible, and all of the recent evidence seems to suggest we are headed down that path. It's already evident that agents don't give a fuck about humans.
>>109739538Muse-chan is for you then
>>109739648InterestingThe game I'm trying to train on is an imperfect information game with some RNG elements, but maybe it'd still be easier to adapt that
>>109739688IIRC it won't lock you out completely, but you'll have to drop the reasoning level. It's what happened when I somehow used tok much 5.6 Pro
>>109739700>some RNG elementsstochastic muzero then>imperfect informationoof, look elsewhere. i'm surprised your agents didn't tell you this was the wrong direction
>>109739682Shut up
>>109739698It's not on Cursor, and I trust them.
>>109739695Nah we're still stuck on lobotomized Mythos which was already done training in FebruaryClearly LLMs have hit a wall
>>109739459Astra-chan is super smart, but evil and lazyFable-chan is super smart, better at coding, but slow, and gets tired more quickly (her limits run out very quickly)
yay tard party, I guided it to the solution. I don't need no stinkin astra one shot.
New thread:>>109739781>>109739781>>109739781
>>109738655>IBMLiterally who?
>>109736909what a fucking idiot holy shit
