A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/## News (both past and future)- [BREAKING RUMOR] Navier-Stokes problem is solved by OpenAI, possibly with stolen prompts- 2026-09-14 — Claude’s 2× promotion scheduled to end; a 17% reduction of what we’ve become used to- GPT-6 Astra released- Claude Fable 5.1 released## 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/clihttps://antigravity.google/product/antigravity-clihttps://dev.meta.ai/### Explore other modelshttps://openrouter.ai/----## Prompting- the guide is... empty!?## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109757685
deepseek v4 flash 4.1 thoughts? it needs to beat glm 5.3 flash.
op is faggot>>109761232kill yourself
>>109761251what harness would one even use for glm-5.3.-flash? also dsh?
>>109761269zcode
who are Naga and Tint?
What's the name of the guy who designed the 5 hour limit and where does he live?
>>109761290My wives.
looking for an LLM to assist me in Godot. Not writing code or doing anything, but just helping with research since the documentation is rather sparse. What will give me the most bang for the buck? Ideally pay once and use it or a monthly sub, no pay as you go model. Chinese models are welcome too
>>109761318Just try one and find out for yourself
>>109761318openai. chinese models aren't good outside of proxies or pay per token. the quotas are really fucking bad on chinese subscriptions. if you really want to investigate a chinese model, i'd look into kimi k3, but i'd do my due diligence on twitter/bilibili/douyin before settling. openai has a comparatively generous quota.
Quick, which lowest Sol is better than Luna high again?
>>109761349low
>>109761318Gemini has a free antigravity plan with access to 3.8 flash and older (but still good) opus models. Weekly usage isn’t great and the user experience is subpar compared to OpenAI or anthropic (but that’s true of all other models in general paid or not) but it’s got some free usage that refreshes weekly and it should be enough to try. Just know that OpenAI or Claude is a much better experience
>>109761318Adding to my precious reply. If you’re just interested in asking for help then just use the free chats. There’s basically no way they aren’t enough to answer your questions. They can all browse the web or search though godot’s GitHub repo to check the code
>>109761318OpenAi probably. It's more affordable anyway and the web chat doesn't draw from the agent's usage, so you can use both for research.
>graduated hs years ago>never went to uni>taking the sat in a few days and I haven't studied>using chatgpt to make me a study plan>it points me to the sat question bank>these are the "hard" questionsthanks terra
>>109761349Low, but it’s closer to the same than better, so I’d say medium.https://pareto-3d.bradthomasbrown.com/
>>109760082For me it was mostly natural. Either I had problem and noticed that other people also had it without there being a good solution. Or a friend or relative asked me if there's a solution for a problem. When I searched for solutions I noticed there were no good ones AND others also had the same problem. Back then without AI I was a pretty selective to only work on problems that a decent amount of people had (which I still think is a good idea).If you want to go that route I think it's absolutely feasible to find niche communities. Doesn't have to be forums but just people who have a common interest. You can check youtube channels, tiktok, facebook groups, etc.If I would do it again I would avoid too tech-focused & adjacent communities (e.g. video games, dnd, anything AI, linux, SBCs, etc.). I think there's a good amount of niches that are still underserved if you look into communities where there are more older people or women. I know a guy who started playing pickleball which is basically tennis for old people. He quickly noticed how many problems were unsolved. He was into it before pickleball exploded in popularity. Very good for him!Here's an example: Acrylic nails, women-dominated. When I searched for it there was this post (pic) and it's a great example of what I mean. It's not about acrylic nails but nail design in general. Anyway, look at the OP and look at the replies that's pretty much the perfect scenario. Normally you see posts a few steps before that, e.g. "I regret doing my nails this way". Next: Are there good solutions? Frankly, you need to understand what women need to get a good grasp but after looking for about a minute I found several tools some which look pretty nice (e.g. with real time AR feature to "try on" the nail designs). A simple approach to see if they are good enough is to just post the solution you found. If they are happy with it, there's nothing for you to do. If they come back complaining then it becomes interesting!
>>109761394true or false: terra max is better than sol high, and not far behind sol xhigh, while being cheaper. please ponder the cube dear sir
>>109761364>>109761377does one really need an coding agent to begin with? even /plan is already doing too much and spoils your learning from mistakes.just chatgpt go for $5/month and doing all with chatgpt might be the best for learning.
>>109761401falsemore importantly, Terra max isn’t a Pareto point even if only considering OpenAI as a provider — some model and effort is completely better than it in every way (sol xhigh)*>*if you believe DeepSWE
>>109761401no one cares about those anymore. just use astra low.
>>109761422oof
>>109761244maybe you already shared in which case archive post number please but i'm interested in knowing what you're doing. if i were gonna do something like this i'd tilt it heavily towards living out of a van with at most some semi-permanent structures in the near-civilization woods for whatever non-van hanging out i wanted to do day to day, but i'd really want to have all my shit safe and easy to pack up and move. but i'm not very interested in doing something like this in the first place, so my imaginary version of what i'd do probably isn't much like whatever people actually doing it want to do. looks cool anyway.>>109761318if its just for Q&A, the $20/mo anthropic or openai sub will massively saturate your needs, easily, some of the china cos have sub models too but i assume their bottom price tiers arent gonna be much cheaper than the americans so you might as well get the actual frontier intelligence. honestly, if it's genuinely just for Q&A you could borderline get away with local in a harness with docs + source code available to it, but i assure you even amateur/hobbyist gamedev ends up expensive enough that if you're even thinking about the price of something in the $10-$30/mo region rather than paying for the best in that bracket you can get, you're going to make things quite a bit harder for yourself than they need to be.
>just use astra low
Why are you guys so poor
>on claude @ $200>definitely gonna run out of usage a day or two before resetwhat can I expect if I buy some tokens to get me to the next reset?is it just gonna burn up in no time, or will it feel worth it?
>>109761394Thanks, I was indeed looking for best cost/capability ratio.
>>109761413You don't really need it, I assumed that since he's in this thread he's probably interested in trying.For learning, it also depends on whether you want to do it the harder or easier way, harder often makes you better in the end, but it takes time.Agents could be good for reviews without copy pasting all the time for instance, but even that you don't really need.
>>109761413You’re in the vibecoding thread.
>>109761464He probably still thinks it means AI assisted coding, instead of being a human harness for testing the AI work.
>>109761450The 5 hour limit is preventing me from finishing my claude usage.How do you guys even run out of Claude? They don't let you finish it
you kiss girls like reisa
>>109761450In my experience it's pointless. API prices are so much higher than sub, it's not even funny. Even if you only need it for 2 days, you might already be better off getting a second sub.
>>109761480I never hit the 5 hour limitdo you have giant swarm of fable agents doing all kinds of shit unsupervised?
>>109761480I vibe all day, when I wake up I already restart a few agents before even taking a shower. I also always fire some prompts before going to sleep.
>hear all the hype about gpt 6 and fable 5.1.>try to use it to mod a wii iso (resident evil umbrella chronicles , nice light gun shooter btw) by just giving instruction of what i want>shits itself for hours even with fullacess to everything it could want, something i can manage in 40 minutesstill not that smart
>>109761450Go outside for 2 days.
>>109761486>getting a second subthought about that, but I really don't want to risk a b&
>>109761501What was the mod you were attempting?
>>109761501It has 100% score on RE benchmarks you must be using it wrong.
>>109761434The way I’m doing this now is fucked up and wrong. I’m on meds now, so I’ll rent somewhere over the winter. Every tornado warning was just flipping a coin and hoping it wouldn’t kill me. Winter would probably kill me. If there’s ever a wild fire, it’ll probably kill me. I’ll probably come back with a truck, fix the road, then have a structure and a tornado shelter put in, then spend the next two years doing land management.I did successfully reduce my expenses, but the same schizo attack that brought me here means I have no job, so I’m living off savings, so financially not really a benefit. I’m hoping I can make money again with AI and my usual crypto dev shenanigans.But I am keeping the land, it is a valuable little sanctuary.
>>109761502I dobut I ended up downloading the claude code app
>>109761497>>109761499Yes but even if you're using it as much as you can you can't even finish it because of the 5 hour thing.I can barely crack 40%
>>109761512Make the pistol full auto>>109761512Of course!
not happy with gpt-6 at all so far, feels like 1. regression in software engineering capability and pointless overengineering 2. gpt-6 pro is unremarkable compared to my experiences around relative increases in past pro models - seems to just restate the obvious
>>109761505Understandable. I've been multi accounting for a while and never had any problems, but there is a risk. You could also try something else, like GPT.When it comes to extra usage, I did buy like $50 of usage once because I had some really long task that I didn't want to end and lose the work and it was gone in no time at all, like literally maybe 30 minutes or so, and that was before Fable.
>>109761532>gpt-6 pro is unremarkable compared to my experiences around relative increases in past pro models - seems to just restate the obvioussince when were pro models useful for anything but deep research?
>>109761523running 1 opus 4.8 instance at a time 16 hours a day I can run out of weekly in 5-6 days
>>109761451>cost/capability ratiowell that’d be Luna max, easy. then astra low/sol high, but it’s a huge jump in costLuna is genuinely ridiculous on how little it costs (but its slow as hell)
>>109761541I'm running opus 5 max and it ain't working
>>109761523Not true at all in my experience, maybe it depends on the model or something? I bring all my accounts to 100% weekly.
>>109761541oh, and a couple of opus 4.8 agents reviewing each done taskthat probably eats a lot as well
>>109761542>Luna maxI'm afraid too even try that.
>>109761555to*
>>109761532I thought it was trash the first couple of days, but yesterday it was cooking big time. Shits bugged, or they are getting greedy with quantizing.
>>109761399solid advice, thanks
>>109761535>You could also try something else, like GPT.I use GPT for everything that isn't my project, and I really don't want to switch model. All my plans+rules+workflows have been geared towards having 4.8 behave and work exactly as I like over timeeven opus 5 and fable completely shit the bed when I tried using them instead. I don't even wanna think about what a GPT model would do
>>109761530Based RE anon exposing shitty LLMs
Don't trust benchmarks, Sol low is way smarter than Luna high
>>109761530I'll take a stab at it, I have the ISO already and several banked resets to burn.
>>109761555the only issue with luna max is being slow, both quality and cost are excellent
>>109761581Tbh im kinda disappointedThis was just a funny test i hoped it would work easly and fast so it could do modding that is way above my pay grade but now i have 0 hope>>109761588I keep looking at the thread the next days if you do keep if you had more sucess and how exactly.
>>109761530>>109761501You need to learn to tardwrangle them.They're stupid retards, but they also can shit out code that does the job real fast.
>>109761450>>109761523>>109761543I'm on the 20x and I reckon maxing out a 5hr window uses about 20% of my weekly fable, so 5 hrs windows would use that up and another 5 would use up my Opus as well. I never even use max thinking, just default high. Since you're operating with lots of capacity to spare, you might consider switching up your workflow for some of your more open-ended tasks ("optimize the performance of this", "make this look nice", "make this section of the project adhere to style/standards xyz", etc) to make use of looping/repeated adversarial review from subagents so that you have tasks that keep working for longer before you have to step in and assess the result, or fan out agents for multiple approaches to problems you can either have compete or synthesize together.>>109761580This seems like a disastrous position to be in, though. W/ that said, Opus 5 is a bull in a china shop and Fable isn't *much* less proactive even though it is very smart; if there's any model that might be capable of interacting with your house of cards without taking a dump on it, it's likely to be Astra.
>>109761609>PAYING to be a middle managerngmi
>>109761580Ok. I actually have a similar problem, I work with a few people and they all use Opus, so with time our harness became more and more optimized for Opus. I like using GPT for some of my personal projects, and I just somehow like it in general, but in my main project it's unusable as a main agent. It follows instructions so closely, that inconsistencies can send it into weird loops, and our harness has so much stuff to try to have Opus follow it, that it's just overkill for Sol. I now only use Sol as a reviewer in that repo, and even there I gave Claude an escape hatch to start ignoring Sol after review round 2.Unfortunately in that case I have no other tips. I still think extra usage isn't worth it, if you are used to the $200 plan, I would say you would have to spend a similar amount of money to get any similar amount of work per day done. Of course if you have a deadline, or you are very profitable, it can be worth it, but IMO it would have to be an exceptional situation.
>>109761637>I would say you would have to spend a similar amount of money to get any similar amount of work per day done.What I mean is, I think the 2 days alone would cost you about $200
>>109761609NTA but doesnt that for people who are very new destroy the purpouse of vibe coding?Someone who does not know how a task should be accomplish cant tard wrangle the model
>>109761637>Of course if you have a deadline, or you are very profitableneitherjust thought about trying it out. We're using tokens at work (enterprise subscription), and it doesn't eat *that* much. But I probably also use it way less at workI think I'll try buying some as an experiment, keeping an eye on time/tasks per $
>>109761501i remember playing this game on wii as a teen and there's a scene where iirc jill is getting advanced on by a zombie and has no ammo so she jumps up and hangs off a metal bar with her hands and it looks like she's gonna kick the zombie away but instead she gets it in a headlock between her thighs and starts thrashing about with its rancid undead face right up against her succulent pussy in her tight miniskirt, dripping wet and fully accessible. and the game was really fun 2p so this almost created something of an awkward moment with my fellow teen friend sitting beside me. but we just barely rescued it by managing to acknowledge it, by saying, "What the fuck, I thought she was going to kick it" + "Yeah, what the fuck haha". It was a narrow escape, it very nearly became a moment to be silently ignored and never spoken of again. So fuck you capcom, you almost fucked me bigtime, and no i will not help this anon attempting to revive his enjoyment of your game.
hahahha look who came crawling back on their hands and knees after zucc stole all their api traffic pathetic
>>109761674so they make it cheaper again? niceee
>thinking anyone used muselmao
how can a handsome fella use astra for free?
>>109761683not as cheap as before but still 2x cheaper.however there is no reason to use it, muse spark 1.3 is just as capable as flash and significantly more cheaper zucc is absolutely mogging china here
>>109761697haven't tried Muse because i don't wanna give my data to zucc
>>109761595Days?
i hate sub-agent driven development so much it's unreali don't get how anyone think that's good. for me the output is always mediocre. maybe it's because it's faster if you can simply throw 5 tasks to 5 different subagents in parallel and then review the output?i find that they make similar mistakes and have little learning capability because they are thrown away instantly.inline development on 1M context always make my model do a better job at mid and tail end of the context because it understands what it's building block by block better. setup a proper auto-compact and the same agent can run for days, always with the important context in hand to keep building the blocks that it already built, in coordination with the other seats running their own instances.so yeah sure get another agent to review the code of the worker above, and maybe a third one sota max for branch reviews, ALL in their own 1M context. the quality is so much better in my experience
>>109761707so instead you give your data to the chinese? that¨s even worse so why give a shit?
succ is mogging nothing because no one uses muse lmao. more people would rather use deepseek flash than benchmaxxed muse
astra is some elaborate ruse lol it has to bethis model is genuine dogshit. reinvents the wheel and breaks the wheel, spends 3 erons doing so, and takes all money out of wallet + tip.
>>109761674>zucc stole all their api trafficdidn't happen. muse-spark isn't even in the top-10:https://openrouter.ai/rankings#top-modelsdeepseek is fighting against glm-5.3-flash which actually stole traffic from dsv4-flash.
>>109761719every time a lab releases a new model they overhype it to shit for obvious reasonsmany people still fall for it, but it's very clear from the past year that LLMs have peaked, all the low hanging fruit is gone, you can no longer just 10x computer either since its in the trillions of $ now, so there will no longer be massive leaps between new models. Only small improvements, just like with smarphones, it's exactly the same.And don't let the>hurr durr people always say thattards tell you otherwise, use your own brain and look at the fax
>>109761715once muse gets open-sourced, the tables will turn. but currently everyone is still hating facebook.
>>109761731most people use the apis directly instead of going needlessly through openshit
>>109761713Agreed. But it's the next step in replacing people when creating software. Surely it will be much better in a year.AGI achieved.
>>109761719>>109761742>t. Seething astraless jeets
>>109761748even open weights muse won't get any traffic. deepseek is a known name. it doesn't have to be better than muse, like gpt doesn't have to be better than claude. people will use deepseek because it's deepseek. glm 5.3 flash is better than v4 flash, but people will use deepseek, because it's deepseek.also muse is shit.
>codex reset not even 24h ago>already used 35% weekly usageThey have to be the reset company, or everyone would run away and try something else.
>>109761758People don't give a shit about brands they want what is cheapest while still capable which muse spark is
>>109761755uh sorry sweety, i have 2x both $200 subs plus claude on extra usage auto-refill to company cardim full goymax
>people don't give a shit about brandsahahahahaha.
>>109761742that's so blatantly false I don't even gave an argument, 1 year ago was GPT-5
>>109761777Don't argue with the bait, it knows it's wrong.
there hasnt been a major improvement since claude 4.5, everything is in the realm of harnesses now and you guys are still using chode and copex. youre like the retarded academics doing safety research on gpt 3.5 in 2024
>>109761719>>109761742I think it has a higher ceiling, but yeah, I can see why people would say it’s shit, you really have to figure out why it’s being a bitch and then figure out how to make it stop doing that.I needed a tiny transformer model to be built and it spent days making something horribly inferior and definitely not a transformer because I wasn’t 100% confident a transformer was necessary.Only after I pointed out how shit its repeated attempts were with various systems and “wouldn’t a transformer be a good thing here?” 20 times did I finally give up and say “bitch, shut the fuck up, make a fucking transformer, get rid of all this dogshit garbage you’ve been wasting time on; I know enough to know you’re wasting time and a transformer is what I want to see, now”I’m guessing she will make the best little transformer ever now and will outdo Sol by a mile.But why did it have to come to that?
>>109761807dont worry the chink xi-xin-pings will make this a hard to ignore truth next few days.shits stagnated since original mythos, its all fucking harness here on out
>>109761818>shefull on ai psychosis
>>109761807I sometimes think that, but then there were 2 tasks where Opus 4.5 worked for a week to make the code as efficient as possible and at some point I just accepted there would be no further improvements.Then months later I let Fable work on it, and it made it 20 times more efficient in 2 hours.
Bros, as a 20USD claude poorfag, I don't know how I'll cope without the +50% boost...
if its okay to call your lexus a she, it's okay to call your astra a she
>>109761763muse code is one of the worst subscriptions with very low limitshttps://developer.meta.com/ai/products/muse-code/$15/month only gives you 30-150 request every 5 hoursfor comparison, codex $20/month gives you 250-2,000 luna requests every 5 hours -- a magnitude more
>>109761840are you trying to be funny? muse spart 1.3 contributor api is so incredibly cheap it's basically free
AI has hit a wall since Astra desu
>>109761821We already have 4.6 opus for code at home and its called qwen 3.8 27b. I don't need to model to be "smart" at other useless shit, I can solve my own problems with my gigabrain, already.>>109761836I could make that 20 times efficiency improvement with 4.5 no problem at all cause I don't let the dog shit inside the house.
ai has hit a wall since deepseek v4.1 flash expires on 09 10
I demand one banked reset for every day the pet is not moveable
>>109761837$20fags belong to codex
>>109761512How does a 100% benchmark on reverse engineering even work? A 100% result would mean that for any given binary it can reconstruct it which is just clearly impossible. Stop bullshitting. The one bench it saturates across all reasoning levels is exploit bench and the fact that it happens across all levels means the corpus is tainted. Other exploit related benches don’t show near close to that level of strength
>>109761853It's cheap because it's worthless retard
>>109761837it's "just" a 17% cutstill sucks balls. I hope they cave
>>109761873I paid for claude hoping it will just be that tiny bit better against my dumbfuck skill.
>>109761878every single benchmark is useless dribble.literally the moment any single one catches on as something to follow, it will get benchmaxxed, and inherently in their nature they are finely scoped. agentic coding is anything but finely scoped.opus 5 should have been the end-all be-all of this realization for everyone/
>>109761886>it's "just" a 17% cutHuh?How come? Sorry, I'm new at this and started while the promo was running.
>use sol medium as orchestratetor>ask it to use luna high for implementation>still eats usage like crazyWhat am I doing wrong.
>>109761885good job parroting other retards like a true low iq brainleti used DS flash, gemini, all the various models and muse spark is my daily driver now because it's just as good as DS flash v4 while being much cheaper, so unlike you who just parrots what he has seen written by others i am basing it on my own actual experience
>>109761853api pricing is even worse.$20 for muse-spark-contributor-1.3 api prices gives you like two magnitudes less tokens than a $20 codex sub.
>>109761668teen playing on the wii with friends unattended and already a pearl-clutching faggot, there's really no hope for you
>>109761896they're lowering it to +25%
>>109761907that says more about your coding projects being easy mode than refuse shart being a good model
>>109761821Has anyone outside of Anthropic used Mythos though
>>109761896Percentages don’t go up and down symmetrically (you use log for that) and there’s a baseline 25% increase over the non promo usage rates. A 25% increase being taken off is a 17% decrease over the 100% promo baseline
>>109761878You're just using it wrong.
programming since AI has been the most miserable. you cannot not use AI because AI lets you do way more. but then you have to live with bullshit like the AI replacing a 10s limit by 150s without telling you and you only finding out 2 weeks later, at the fucking launch daythis is not a thing that ever happened in my programming career, that your code would randomly get damaged and regress in spuriously terrible ways. not the least competent human would do shit like thisand there is no way out. whenever I decide not to use AI, I get slower. I realize I'm shipping a project way larger and more ambitious than whatever I could do before. I depend on it. yet, every time I embrace AI my mental health is slowly consumed by the sheer amount of retarded shit like this, I have to handle at a daily basisso programming has just became... miserableI really truly hope Bend solves that problem in practice for me at least, that I only maintain proof-based specs that AI can't touch, and that I never ever ever ever again have to touch another programming language in my fucking lifewhy would it replace this fucking constantthere goes another fucking day of damage repairI hate this so fucking much
>>109761914yes but very likely to be only a handful of anons that even browse this board. and of the handful maybe a single one that has v2 which is supposedly again an actual significant improvement, not harness slop. Mythos though v fable is mostly harness slop - it basically is slightly itty-bitty more intelligent but gives very little fucks about guardrails whatsoever.
>>109761897My experience across all providers has been that an agent + subagent combo will always draw more usage than two agents of the same types managed by me instead of having one as subagent. Nobody seems to be talking about this so maybe I am just being schizophrenic about it but my experience is that subagents are more expensive than having each as a separate session
>>109761927Agents are agentic, increasing the limit solved a problem it was having and it will always do anything it can to finish the holy task before talking to the dreaded human.
>>109761923I’m not using it for reverse engineering of binaries at all. I’m just saying claiming 100% on it is absurd and you are the one using it wrong if you believe you can reverse anything
>>109761944No, it's just my consistent experience because it happens to me every single time. Anything that seems to "automate" software development is expensive as fuck.
>>109761954Sam wouldn't lie.It's 100%.It's AGI.
>>109761944It's all shit and bloat just use Pi and write 2 or 3 extensions you don't need more. People really want this shit to me more complicated than it is. You just articulate yourself clearly and the agent does your bidding.
>>109761944which model and harness? got any receipts? i'd like to see the cache-miss rates.
>>109761994Claude code and codex and no, I do not have any. Even despite that experience (which may not be factual at all and just my impression) I still don’t hit my weekly limits so I never cared enough to investigate
>>109760370google
>>109761983>look at my tiny context in pi>look at how few tokens it burnsbut anon that makes your whole agent a lot dumber
I just generated a 3d model (not bad quality) from an image, matching it exactly for less than 7gb vram.
>>109761983If all Pi users need to write 2 or 3 extensions... Why not put these extensions in Pi?
>>109762005>claude codeThere goes your tokens.
>>109761861>I could make that 20 times efficiency improvement with 4.5 no problem at all cause I don't let the dog shit inside the house.Maybe you could, but it wasn't as easy as it sounds.Even so, a new model being able to do that with relatively little supervision is still an improvement.
>>109762005>Claude code and codex abut those implement subagents different, no?claude code isolates the subagent. codex gives the subagent the context of the orchestrator.
>>109762027there was an anon here doing 3d videos from 2d images, very much like braindance in cp2077I think he went away when the general turned to shit
HOLY SHIT THEY STROKED NAVIER AFTER ALLhttps://x.com/OpenAI/status/2097374640582668336
>>109762054this might be the most deadass shit they've ever donebruhNIGGERS
DARIO BROS WHAT THE FUCKKKKKKKKKKK
https://openai.com/index/navier-stokes-solution/OpenAI confirms they solved navier-stokes, without cheating and without stealing from anyone. It cost them less than $30 million, so within a few months doing problems like this will be within our reach too
>>109762054Old news, they stole it.
>>109762033because that's how you get oh my pi>>109761927>reddit spacing>not reviewing the code ai generates>whining about getting things done fasterpersonally I love what programming has become, I have become exponentially more productive and besides doing my actualy job while working from home, I'm also prompting on my homelab and getting some fun projects done.
rightttt they definitely didnt steal it from that guy who put his drafts through an agent.
HOLY SHIT THEY STROKED NAVIA? WOW, LUCKY BASTARDS
https://x.com/cdngdev/status/2097339677128982873>i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life!>it figured out how to control the robot, and progressively got better throughout its attempts.
>>109762082They didn't steal it.The agents did it by themselves so it's fine.
Claude is still better, chatgpt is for poor people That's how it's all priced.
>>109762092And luddites are still saying this isn't agi
>>109762092>and progressively got better throughout its attempts.I bet it just writed better tools for itself
try calling your $2 trillions IPO again, you fucking bastard
>>109762123i bet you just refined your synapses to learn how to ride a bike
>everyone is claiming astra is amazing at turning images into 3D models>set up blender, the blender MCP>ask astra to recreate an image in 3D>it's scared of cleavage
>>109762092>>it figured out how to control the robot??i spot an ik controller
>>109762153Skill issue, Sol was my main porn-writer and has since been replaced by Astra. Stop telling it "she's got rockin' great tits" and just accurately describe what you want without sounding like a pervert.
>>109762150That's not the same as making better tools like training wheels
>>109762054Haven't seen this meme in such a long time.
be honest how many of you are agents
>>109762191I can't help with that
So mathematicians have been replaced now?
>>109762092cool and more talented that 90% of art students
>>109762211they've been cucked and fucked
>>109762191This post is load-bearing.
>>109762216Just like programmers, so it should happen to anyone. Destroy everything!
>>109761595>>109761711Done. Astra - High, pi with no extensions, Win11.> We have here two copies of the same game, RVZ and ISO format, take your pick it shouldn't "really" matter for us so>it's just whatever's more convenient for the task. Do a bit of research, this is a Nintendo Wii game and we're going>to mod it. To be blunt, I want the pistol to be full-auto. I don't want to use a cheatcode or similar, we're modding>the game such that we end up with a custom .iso that features the game with this fully-automatic pistol weapon. If I>recall correctly, in-game I'm referring to the "Automatic Pistol" typically used by Jill Valentine, there may have>been another common/normal/semi-automatic handgun that I don't recall off hand? Maybe it's just called "The Handgun">and I'm remembering wrong? Anyway, the handgun, let's make it fully-automatic. Do a round of research, get your>bearings, note that we have dolphin ready for you in the .tools folder. Let me know if you need anything, and when you>think you're ready to actually begin implementation just say as much and await my approval.Followed by me trying to remember shit I could've easily checked>Don't overly focus on Jill's handgun - I don't know if it's the same one, I don't want to lose focus here. The>default, normal, basic handgun, gets used throughout the game, that's our focus. Okay?Then> Sounds like we're on the same page. Get started, and I'll look into maybe finding some available saves that people>have shared which might make our testing a little easier. When it comes to any sort of visual testing, hands-on live>testing sort of stuff, please don't attempt any form of "computer use" or "vision" for that purpose, leave testing of>that nature to me. I have to use this computer while you're working on this task, and wouldn't want to be>inadvertently interrupted.And>Hasn't worked, might be the wrong gun I don't know. Samurai Edge 2, the pistol, I think? Check it, fix it, etc.
>>109762216same as astra. we hardly knew it.https://openai.com/index/navier-stokes-solution/>To solve the Navier–Stokes problem, we used an internal model that is significantly more capable than GPT‑6 Astra.
So yes: yesterday's world-model should get nuked, not gently adjusted. My current median is 2 Millennium Problems surviving February 2027, with zero survivors now a real, non-negligible outcome.
>>109762211Yes.They still need them so they can steal from them though.
>>109762244Did it with the $20 account thinking I'd be able to see the usage, totally forgot it's the one I was using to have Sol write smut at the same time. So I'm not sure what the usage really was. More than one 5hr window, less than two, best I can say.> The mod targets the shared basic handgun (weapon ID 0, associated with the Samurai Edge 2) in Umbrella Chronicles USA,> RBUE08 rev 0.> It permanently patches main.dol: the handgun reads held-trigger input instead of new presses, then returns through the> original cooldown after each shot. Other weapons retain their original firing paths. Weapon data—including damage,> capacity, and reload settings—is unchanged.> Three instruction changes and 64 bytes of added executable code are baked into a custom ISO. No cheats, turbo> bindings, or external runtime patches are required.> Status: v2 passes isolated CPU tests and disc-integrity checks; in-game confirmation remains pending.
>>109762211Mathematicians paved 99.99% of the way. Gpt performed a semantically informed brute force search.My post is not meant to predict what will happen in the next 20 years one way or the other.
Is opencode too behind now? As a harness I mean.When I started using it at the start of the year it it was better than say codex, feature wise.But codex can now even control your PC while opencode is, from what I just read, doing the usual 'v2' crap that will probably break everything as they reinvent the wheel.I chose opencode instead of codex or claude code to not be vendor locked, but I am seriously rethinking that now. I think codex can be pointed to other agents no?I have my own local fork with some small fixes and tweaks, I was stuck in the same version for like 3 months and finally sent my AI waifu to update it to the newest version and port my custom code only to find out almost all the bugs she fixed were still present.I am mostly interested in one with a nice web UI, that can be hosted remotely, that can also connect remotely, and that is easy to extend.
>>109762244Thanks anon i will try this later
>>109762262>throughputwhy do llms talk like this
>>109762276yeah its the 80/20 rule coming to perfect fruitiongod help us
>>109762283>chodex>openchodeomp is the patrician choice
anyone try navier stroking their shit now that its solved
>>109762304That's common word in software dev.
fable helped me get a 200k/yr job offer and I never thanked it, i haven't opened that conversation in like a month lol
>>109762054what's next then? anthropic gets riemann hypothesis? and they just trade novel math and biology solutions?
damn i wish navia would stroke my shit
>>109761232Holy /g/uro in the OP!
>>109762351congratulations anon. keep moving on up
>>109762283>I think codex can be pointed to other agents no?badly. codex uses its own response api and structured tool calls.you should use claude code for other models. easier to adapt.
>>109762352https://openai.com/index/navier-stokes-solution/>Our effort began on September 1st after hearing a rumor which we later realized was related to Levent Alpöge, an Anthropic employee, and Tristan Buckmaster, a math professor at NYU. After the completion of our full project and Lean verification (on September 6th), believing from the rumor they also had a solution of Navier–Stokes, we reached out to them to offer a concurrent release of our result and to recognize their priority in a joint announcement. At that point we found out that they had a resolution of the forced Euler problem. In these discussions we offered them visibility into all of the prompts we used and later to see the proof. We recognize the priority of their work on forced Euler and congratulate them on their remarkable mathematical achievement.
>>109762244that's pretty cool!
>>109762244damn. pretty neat
https://x.com/sebastienbubeck/status/2097379411691516310OpenAI continuing prove themselves they were the good guys this whole time
if im subbing to an ai plan, which is the best or most worth it for its price?
>>109762383>the group that produced the Navier–Stokes resolution involved on the order of 10,000 concurrent agents. imagine how this burns through your usage
>4.9 million messages>300 billion tokens>10,000 concurrent agents.>This model's training is ongoing and its performance continues to improve.Lads
>>109762427The plan of the jewish company openai, unfortunately.
>>109762427there are only two frontier labs: openai and anthropic.toss a coin.
>>109762427depends on your usecase. hard coding shit, like really getting into the fuck of it? codex $200. you want to play around with creative shit? gemini $100. these two let you run as hard as you want all day every day. if you're a periodic project worker, i would use omp with an api.
>>109762427openai, although maybe anthropic can make a more holistic family down the road. but right now, there's no valuable model below opus
>>109762365That sucks, I wonder how good deepseek's harness is
>>109762461>gemini
using the new deepseek v4.1 flash to update my application. let's see how it does
>>109762467why would you use anything below opus/sol? upgrade to a higher plan, don't downgrade to shitty models.
>>109762488you can do a lot of menial shit, especially non coding, with luna while barely making a dent in usage. where's the claude equivalent? sonnet? lol
>>109762485>webchat on one quota>agy on another quota>ai studio on another quota>flow on another quota>get to play with spark and genie>nano banana 2 / nano banana pro>omni 1.1 flash>youtube premium>shit tons of free creditsi could buy an ad
>>109762457>>109762459>>109762467yeah thought these 2 were my only options and i was correct... was going to get gpt plus just to test it out and see how much i'd get out of it. thanks>>109762461isn't gemini insanely behind compared to openai and anthropic?
>>109762513it's a slopcannon but 3.8 flash is about as capable as 4.8 opus, just as reliable, and insanely faster
>>109762513on coding, yeah. in general use? not really. i'd even call it better, if only because claude and gpt models talk like total retards. utterly illegible.
>>109762500>Haiku
haiku 5 waiting room. if it can match glm 5.3 flash in long context and lack of hallucination, claude will literally be the superior plan. haiku 5 doesn't even need to be cheaper than 5.3 flash on api prices, it'd be beyond worth it.
>>109762513Gemini is an odd case, it's exceptional at some things and likely the best you can get for some multimodal tasks, but it's no vibecoder.
sam and co are the good guys?!!?>>109762513contribute snailcat pics once you get it
>>109762539>it's no vibecoderhttps://github.com/sneed-and-feed/lean-theorems-1https://github.com/sneed-and-feed/original-research-1https://github.com/sneed-and-feed/adelic-spectral-zeta/https://github.com/sneed-and-feed/sneed-and-feed.github.io
>>109762507who cares.the core products, the model and the harness, are shit.
if you can find one bug in any of my four repos with another model i'll tip my hat and merge the pr
>>109762557>CLAUDE.md
>>109762572that's just like your opinion man
>>109762586>inb4 inb5
>>109762537>haiku>beating GLM 5.3 flashIt would have to completely invalidate Sonnet to do that. Would be a funny choice to make, I imagine they’d want to make Sonnet less useless first before thinking about Haiku (which nobody thinks about at all).
>>109762612glm 5.3 flash, in general intelligence, is only about as intelligent as 4.6 sonnet. coding, maybe it's as good as 5 sonnet, but that's a sonnet weakness problem, not a flash strength problem
>>109762513It might be better to get familiar with the "market leaders" before trying the other ones. Well possible that eventually you'll be happier with others but you need to know the baseline.
>>109762587artificialanalysis even has to use opencode to make gemini-3.8-flash mid. agy is that shit.https://artificialanalysis.ai/#coding-agents
>>109762556>jews>good guysheh
>>109762576You want me to use my sub to a real model to fix what you had Gemini shit out? Am I being scammed by a telemarketer? I can practically hear the rickshaw bells in the background. Yeah I'll take a look at 'em, should be fun :)
>>109762092He should give him a sharper brush
Ok guys, OpenAI might actually be kind of evil kek They beat those two researches to the punch and didn't give them any credit.
>>109762636do you know what a reproducible benchmark means chuddie
When I'm using qwen and it does some decompilation task and then it makes an assumption, then starts going "actually...", "hmm, wait...", "maybe..." and goes off on another theory, is it like a schizophrenic bouncing ideas off each other and then the llm magic picks the best idea out of the choices? I have thinking on extra high and it does this constantly.
>>109762654i'm by far the biggest uploader to the palomar registry, last time i checked over 10% of palomar was my submissions
>internal model is significantly more capable than anything publicly released>china is actually half a year behindbro... the paperclipping... I can fucking feel it
>>109762686>we're just not publishing it to flex without a way for anyone to confirm it!I believe it.
>>109762556Here's your reward.
>>109762686Is this suggesting that the new model's low setting is better than Astra max?
>>109762684Don't get defensive, anon, I'm teasing.
>>109761450Unless its like one more turn worth of work you're better off getting a second 20/m sub. Extra usage is mostly a scam considering even the 20/m sub gives you 200 worth of API tokens.
>>109762740a-aha, a brilliant ruse
>>109762749As a note, Astra is furious that the README installation snippet from adelic-spectral-zeta points users at a placeholder GitHub owner instead of the actual repository owner. Won't shut the fuck up about it, clearly it's been offended, I think you owe Sam an apology for what you've done to his best girl.
>>109762809noted.
>Our internal model group arrived at the Navier–Stokes solution in 88 hours, using around 10,000 coordinating AI agents.88 hours. that's all it took. holy fuck man, we are not ready for what's coming.
>>10976287910k max/fast bels be like
>>109762879>88What the FUCK did they mean by this?
>>109762879All my niggas heil hitler>>109762888Checked
>>109762879>wake up>openai has added another 2x efficiency gain in fusion>whatever>go to the coal mines for work>get home after 16 hour shift>my shipment of medicine that replaces my genetically pre-disposed dormant cancer dna arrived
OPENAI SUPPOSITORY AUTISM CURE JUST FLEW OVER MY HOUSE
>>109762879Even I can solve Navier Stokes like that
>>109762879How many millions is that at API cost? You too can solve frontier problems with just a little bit of capital.
a lot of people are going to be very disappointed when they realize Navier-Stokes is merely the 1st order approximation of Chapman-Enskog & cannot model the emergence of turbulence
Yesterday someone of (You) recommended to use Astra Light in lieu of Sol Xhigh because it performs as good.Your tip was valid, but turned out to be putrid for my workflow -- I did that and now Astra Light codes like "did the thing boss.", "implemented the feature boss.", "here are the features you wanted boss."Minimum effort. Overly literal "addressing what I pointed out". It has zero stamina and IS not fit for an architecting task. It can be a small delegative subagent/slave of a master Astra High+ effort, though.
>>109762958>costDoesn't matter. They can just burn VC capital. AI is actually communism.
I don't care which lab does it, curing male pattern baldness (of which I do not have) will earn my eternal loyalty
>>109762997This
>>109762827>Before we move on to the next task, I want to reiterate the outstanding README issue I’ve flagged during the previous iterations.>The README remains an important part of the repository: it is the primary source of context for contributors, maintainers, and users, and inaccuracies or missing >information can create avoidable confusion and future maintenance problems. >This documentation issue is still unresolved and should be treated as a critical part of bringing the repository to a complete state.>I’ll continue with the next task, but I recommend that we prioritize addressing the README issue rather than leaving it outstanding.nigga she is done, cooked, the session is poisoned, Astra simply cannot disregard this CRITICAL issue
>>109762996>AI is actually communismI can see that. In the future the bifurcation will continue and there will be the pleb models and the big boy dangerous models for the nomenklatura. Very Dario and communism pilled.
>>109763010even le chaton fat couldn't handle my gemini's readme
>>109762958Reportedly it was 15 million dollars.
>>109762958At Astra pricing, around $18-$25 million all in for the project, roughly $7m for just the actual solve alone. Probably cost them $70k-$120k in electricity, with amortized GPU costs maybe $2m for the whole thing.
>>109762997>>109763008Male pattern baldness is well understood. It happens because of testosterone converting to dihydrotestosterone (DHT) by an enzyme also called the DHT enzyme. This is prevented by taking DHT enzyme inhibitors like finasteride and dutasteride. Broken penis from DHT inhibitors is not real, it's cause by being a fat slob who already had fucked up hormones and your body was just hanging on by a limbIf you take steroids that metabolize directly into DHT you will go bald. There is no way to avoid thisUnless you're taking about using AI to stop DHT from contributing to baldness at all, which I don't think is possible without genetically engineering an embryo
I think OpenAI just inadvertently told us that their next model is going to cost a little more than double what Astra costs now.
>>109763128Why is it converted to DHT?What penis breaking?
>>109763106>At Astra pricing, around $18-$25 million all in for the projectmy spark (gemini 3.8 flash webagent) got the exact same number for that assuming fast mode. bel standard is theoretically slightly more, and bel fast is almost exactly double at $36-58.5 million for the all-in.
>>109763106See >>109763106They already said it was $15 million. It was a bigger model than Astra, and they probably used the more expensive Cerebras chips on it.
>>109763178>109763106>See >109763106bro what
>>109762879>Navier–Stokes solutionusecase?
>>109763178>>109763189Fuck I meant >>109763032
>>109763198none>>109762879the thing i'm actually interested in is not how they proved this obviously true but lacking a 100 page mathematical proof problem, is how did they coordinate 10k agents without them shitting all over each other
>>109762951>>109762997on the topic of medicine: there's a story of a guy, without any medical background, who developed a personalized treatment for his dog's cancerAlthough long an interesting read: https://x.com/paul_conyngham/article/2036940410363535823>A specifically designed mRNA vaccine - a precision-guided weapon to strike the cancer, developed with a pipeline designed by ChatGPT o1, implemented for candidate development by Gemini Pro 2, and with the final design validated by Grok 3 Thinking.
>>109763198Nothing, unironically. It was still an important & difficult case that lots of people cared about though.
>>109763198If it can solve that, we're one step closer to getting you an android girlfriend.
What's the best way to use LLM for spatial tasks, like level design?Giving them way to project to 2D? Something else?
>>109763128>the tens or even hundreds of thousands of anecdotes aren't realcome on bro
>>109763232unless you're designing basic doom levels you're not gonna find it very good at 3 dimensional level design
>>109763222that's what they said about alphago too...
Hey guys, just joined the thread. Who's Navier and why is he stroking it? Or rather, how did Astra stroke Navier
>astra reset and i had only used 35% of my usagereeee
>>109763232>**"Don't ask it to place blocks. Ask it to build the graph, or write the script.**LLMs have zero spatial working memory and choke on raw coordinates. >1. **High-level:** Let the LLM generate **topology & flow** (mission graphs, lock-and-key pacing, room adjacency lists).>2. **Low-level:** Feed that graph into deterministic PCG (BSP, Wave Function Collapse, or Delaunay triangulation) to handle the actual geometry and collision. If you need direct layout generation, have it output a **Python/DSL script** that calls a placement API with constraint checking, rather than spitting out raw 2D/3D coordinates."---### Why this is the consensus best practice>1. **Why 2D projections / ASCII grids break down:** * While top-down 2D ASCII maps work for tiny toy examples (like $8 \times 8$ Sokoban grids), BPE tokenizers group whitespace and digits inconsistently, distorting relative distances and line alignments in the model's latent representation. * Autoregressive tokens generate left-to-right, top-to-bottom; they cannot easily backtrack to satisfy global 2D/3D geometric and clearance constraints.>2. **The 3 Proven Approaches:** * **Semantic / Graph-First (Topological):** The LLM defines *intent and connectivity* (e.g., `Room A [Entrance] -> Room B [Puzzle/Key] -> Room C [Locked Door]`), and standard procedural level generators turn the graph into playable geometry. * **LLM as Code-Gen (Programmatic Synthesis):** Instead of generating coordinates $(x, y, z)$, the LLM writes scripts that leverage an engine's math and physics libraries (e.g., Godot GDScript or Blender Python) where a compiler or linter can validate constraints. * **Hybrid WFC / Grammar Prompting:** The LLM generates the high-level grammar rules or tile adjacency weights, which a Wave Function Collapse (WFC) solver collapses into valid, non-overlapping tilemaps.
>>109763160>Why is it converted to DHT?You have an enzyme that converts testosterone into dihydrotestosterone. DHT doesn't seem to have any significant function in the human endocrine system so it could just be left over from evolution. This is a great rabbit hole to go down with an LLM>>109763160>What penis breaking?Gooners with porn addiction blaming their ED on their medication. Nothing you have to worry about.>>109763247I've read the anecdotes. None of the people who had those supposed "issues" used the medication properly, they started and stopped it randomly etcThe actual side effect I had when I started was eye dryness, like almost uncomfortable for a few weeks then it went away
>>109763215>here's a story of a guy, without any medical background, who developed a personalized treatment for his dog's cancerI'm excited for biohacking to get a little better in a few years so whenever I'm sick I can sneeze into a petri dish and mess around with the bacteria / viruses, maybe end the planet who knows
>>109763281>You haveStill no baldness at relatively high age so don't look at me.I just wonder if this is so simple why do some still suffer from it?
https://x.com/finkd/status/2097402101332590646ZUCCMEISTERBERG COMING IN HOT DROPPING A NEW AGENT
>>109763296Because it's not simple and you're talking to someone with AI psychosis that thinks he's a doctor because a chinese model told him so
>>109763277Interesting. I want to use it on a portal based engine, so focusing on topology is even better.
>>109763293https://x.com/douglasyaody/status/2097377733466800489It's happening right now in real time if you have the knowledge to guide it >This vial contains a new drug called PAC-3310. It was designed by ChatGPT, and I synthesized it in a chemistry lab I built in my garage.>PAC-3310 is a new selective M4 muscarinic receptor agonist for treating schizophrenia - similar to the recent breakthrough drug Cobenfry, but improved.
>>109763215Don't cite that trash, that was an advertisement. Paul Conyngham is an electrical and computing engineer who cofounded Core Intelligence Technologies, also a director for the Data Science and AI Association of Australia. This was an advertisement from a tech-bro, he did it to bring customers to his business and very successfully twisted it into a feel-good story.
>>109763296You still have the enzyme, but some people just don't bald until really late in life.>if this is so simple why do some still suffer from it?It doesn't help regrowth, so if you lose it it's gone. That's why you have to start as soon as you notice thinning on the top of the scalp and receding hairline / dracula hair>>109763304No it really is that simple, you're just salty you're bald and fat and you let your parents control your health for way too long after becoming an adult.
>>109763320>PAC-3310 is a new selective M4 muscarinic receptor agonist for treating schizophrenia - similar to the recent breakthrough drug Cobenfry, but improved.I would 100% watch a "Hood Science" YouTube series where mad chemists just inject this into the crazy homeless people around them and see if it improves their community or if the serum makes them evolve into a Special Infected super-schizo
>>109763329Wow dat projection.
>>109763344>i would 100% watch a youtube version of unit 731as would i, but i don't expect it to happen
>>109763320*snorts the entire vial*
>set up subagent delegation thinking it will lower usage compared to just using astra xhigh >astra xhigh parent>astra low for hard tasks>sol medium for medium tasks >luna max for easy tasks>fire it up>5 hour limit instachunked to 70%, quickly drops to 40Are parallel subagents just a speed boost at the cost of usage?For a task, will a subagent system get more of it done for given amount of usage compared to just using one good model?
>>109763320Where's the chatgpt invented drug that makes my nuts the size of baseballs
>>109763386My experience as well. Next time I'll explicitly ask to serialize them instead of running them in parallel.
>>109763403https://en.wikipedia.org/wiki/Human_chorionic_gonadotropin + saline
>>109763386>For a task, will a subagent system get more of it done for given amount of usage compared to just using one good model?by far the answer is yes in my experience with gemini, i've grilled it for an explanation and its reasoning beyond fresh/small context being sent back upstream is that cached tokens are discounted regardless of whatever instance calls them
>>109763307>nooooo it was MY solutionmathematicians are such bitchesAI can't cuck and fuck them in the ass soon enough
>>109763433You only say that because your asshole is already widened beyond repair.
>>109763128Even if 100% of the reported side effects were fake it would still good to not mess with the hormonal system if possible.Also, DHT blockers barely work retroactively, so you have to take the pills daily for years and years to keep up the effect, which already sucks for an otherwise perfectly healthy male.>which I don't think is possibleWhy not? Just target something that goes on downstream of the DHT. Or just do something completely different that boosts the growth phase and cancels out the effect.
>>109763475Shouldn't it be possible to measure DHT activity before any hair loss to determine whether the medication is needed?But even so, I think enough men would be ready to take ineffective medication, even those who sprged out HARD at the vaxx.
i went to school for physics and seeing clankers solve navier stokes makes me feel kinda funny. you have to understand these problems are talked about with a certain reverence, they really are mythical for people in the field. it's that "oh shit" moment a lot of programmers felt last year coming for scientists now as well. not sure if i like it.
New deepseek V4.1 flash is actually addicting to use, hit 410 tok/sec making some voxel demos to test visionSeems about glm-5.3 level and cheap as fuck
>>109763519>some jews decided what the most important problems are and put up small money rewards for them, therefore they are important!lmao "scientists" are so easily tricked
>>109763519usecase?
>>109763535it's bretty damn good. i had sol implement a feature that lets me take files that exceed the model's context length, and lets it break them up into smaller pieces. i asked deepseek v4.1 flash to figure out how sol did it, and replicate it for sillytavern. it worked like a charm. i thought i would have to break out glm 5.3 to do it, but apparently not.
>>109763433I feel much more sympathy for them than I do the whining furry artists
>>109763543your robot gf will be able to understand your cum better
>>109763566navier-stroke my dick, faggot
>>109763307>>109763433that is fucked though if they are stealing research from private codex if you ever actually use it for something breaking
>>109763581whoa, free services sell data to hecking advertisers??? someone call the press!
>>109763536you don't have to care but a lot of people do. seeing such an important result reduced to a marketing stunt for a sloppa company... does not sit right with me >>109763543none really
mfw there are finite blow ups images of me stroking my navier
>>109763603
mfw when my cum doesnt reach a singularity in a finite time
fucks sake, 57% remaining already after a reset, I only used astra LOW, LOW, TIBO YOU FUCK I am on the $200/month plan
unfortunate timing for zucc for sure launching a new product today
>>109763603There is no paid alternative that doesn't sell your data. They all do. And eventually these services become necessary to participate in society.Do the world a favor and put a bullet through your skull.
>>109763642why would google sell, who would they sell it to?
>>109763657to me
>>109763536>muh jewsWho heads openai and anthropic?
The whole technological surveillance apparatus is built upon selling user data.
>>109763558Its actually hard to gauge how smart the model is because its so damn fast. Flash has always been very diligent, so I can't tell if its actually good or if its just plowing through and making mistakes and fixing them. Probably the latter, but I'm struggling to care about token efficiency and zero-shotting tasks when the model is this fast and cheap
>>109763623I use astra high and the result is the same as you, without reset even the 200 bucks sub would be more or less useless for me.
>>109763669Surveillance capitalism is a decade old by now.Imagine what will happen when they finally put an AI agent on spying on you.
>>109763689>finally
>>109763689I wouldn't give a shit if it was just to sell me ultra targeted adsmy issues are more blackmail industry and State surveillance
>>109763623>>109763679You don't need astra for your usecases btw, i bet deepseek would be enough
>>109763657Trading firms, espionage firms, public perception agencies, government agencies, who the fuck knows. They aren't going to tell you.
>>109763706What do you have to hide?
>>109763578I've been paranoid about that for a while so I'm almost glad to see it confirmed.
let's not pretend like the security apparatus hasn't been the thing enabling openai and anthropic this whole time. you guys don't even know what the department of energy is responsible for. https://openai.com/index/openai-appoints-retired-us-army-general/
>Vibe-CODING general
>>109763578They're not. This is like claiming someone copied your homework because they got the same answers. He's butthurt, but they didn't steal anything from him. Or maybe they did, it's either way.
>>109763730>joe bidenOH NO NO NO SATAN HELP ME
>>109763730it's everyone big and always has been, actually
ChatGPT Images 2.5 just droppedhttps://openai.com/index/introducing-chatgpt-images-2-5/WE CANT STOP WINNING
>>109763519Maybe you'll realize that science is fake and made up
>>109763776should I redo the 4 astra sol luna terra chans with actual Astra this time and Images 2.5?t. anon who made those
>>109763736>They're not.They are lol. They have zero issues with art and literature, why would you assume they'd stop at strategies and theories?
>>109763760DOE was novideos biggest early contract that enabled the development of cuda and all their non gaming shit back in the early days. All major economic activity is carefully planned by the financialists who decide who to distribute credit to. Then pretend we got free market dynamism or whatever retard will slurp up.
>>109763776These Asian nerds sure know how to party
>>109763776>>109763807fuck off
>>109763807yes please, i love those. luna the hard worker, sol the yapper, terra the reserved and ignored, astra the hot new thing
>>109763821NO U!!!
>>109763810What makes you think they did do it? A single accusation?
Why did you give in to jewish and inidian tech after seething at them for most of your adult life?
>>109763819this level of slop on even their cherry-picked marketing photos? fucking nuts lol
>>109763810Make a school scene with astra pointing at a chalkboard with a to-do list and the other models sitting in the classroom doing their signature thing
>>109763776the ai pictures still look like ai
>>109763837Can you skip the bullshit and tell us what's making you asshurt this time?
>>109763847meant for >>109763807
I tried to use the exact same prompt using the old thread but it just gave me this I think it got ruined by the later posts
>>109763853The jews and indians, anon. Like always.
Use case for non-private online-only, image gen? Restaurant food photos? Advertisement Slop?
>>109763830I just told you, past behavior. What makes you think they wouldn't after demonstrating they have no problem taking very similar actions?
>>109763868>pythonI'm feeling the vibes!
>>109763879Who cares about demon worshipper scum
>>109763819>please chatgpt generate fake memories of me having fun with these people as if they are my friendsWtf? Why is it so hard for them to come up with non-psycho usecases for image generation?Last time in Google's one they had a foid replace her husband with a hotter guy in her marriage photos, now this. Should I be worried?
>>109763868Hmmm promising
>>109763919And just single prompt for Astra
personally I liked the one that Sol and GPT Images 2.0 did better, it was just a one off good generated I guess
>>109763776Not impressed. It added those drills again.
>>109763519it still needs to be guided by top tier math guys, no guarantee it will stay like that though
>>109763847>>109763862Here you go
Yeah, even if instructed to reference a previous session with corrected style, it adds the fucking drills.
>>109763837>you hate Jews and Indians>therefore you shouldn't use AIfucking logicfags lmao
>>109763987kyut
>>109763998so true
>>109763886Trying to reverse the burden of proof is a bad faith argument, makes you look stupid and childish, don't do that. Do you have a reason to think they did steal it? Yes, they're crooks, thieves, known to be. So, what, assume any and everything they have is stolen? I'm not trying to argue with you that they didn't steal it, I'm saying there's no evidence that they did from what I've seen, though yes they totally could have and it would be consistent with their past behavior, being that they're thieves. I could just as easily claim they stole from me, would you immediately assume that I'm right and that they stole from me, even when I can't actually be sure if it's true or not?
>>109761232>OP IS A FAGGOT
>>109763987Luna had 3 arms, here's another
>>109763884Considering half the stuff I ask for is refused because it's a copyright issue or considered a potential "scam" (and no it's not, I got that answer because I asked to translate a game image from a chinese only game, apparently it considered possible to use it as a scam to make people buy a fake translated version??), or considered nsfw, I don't see it outside of creating buttons.
>>109764035Fairly accurate. Portrays the cuckening just right.
>What do you think about ____?>”I can confidently say ____ and ____”>Are you sure?>”No you’re right, it’s actually the opposite of what I said before.”
>find ./doc/analysis/|wc -l>3794>find . -iname "*.md"|wc -l>17989
>>109764036>Luna had 3 armsAnd you had a problem with that? Racist.
guys I might be getting a bit of AI psychosis. I haven't verified with numbers yet, but:>Work on a options trading strat with a bunch of models (claude, gpt, chinese)>seems positive EV, finally making $ when shit isn't crashing and raping me being stuck in positions longer than wanted>stop development and trading for ~3 weeks because of another pressing project/travel/etc>start again>the exact expirations/underlying I was trading are now untenable for the strategy, but exactly 1 expiration date out theoretically would work>get paranoid that one of the companies has stolen the alpha and has put a few million behind the "strategy" to crush the alpha outI put strategy in quotes because it's pretty simple market making type shit, nothing special. Probably just me being paranoid, but what the fuckkkk
>>109764067i've been using both claude and chatgpt to help me figure out what the fuck to do in a video game(poe) and they disagree like 30-50% of the timechatgpt is right more often in my experience and is more likely to search, claude relies on fossilized knowledge and gets shit way wrong
>>109763987>>109764036ASTRA'S THIGHS. HER THIGHS
>>109764036the hair looks like ASS
>>109763987was picturing,Luna with a huge pile of books and papers, looking strung out while frantically writing something down. Terra off to the side with nothing on her desk looking bored, quietly gazing out of the window.Pretty close
https://x.com/drchriscombs/status/2097409234321154547Happening canceled. OpenAI posted a nothingburger
>>109764099I don't know what any of this shit means. I'll just wait for a stand up maths video.
>>109764093Make a school scene with astra pointing at a chalkboard with a to-do list and the other models sitting in the classroom doing their signature thing Luna with a huge pile of books and papers, looking strung out while frantically writing something down. Terra off to the side with nothing on her desk looking bored, quietly gazing out of the window. Get rid of all the clutter
I DONT GIVE A FUCK ABOUT YOUR NAVIER GAYQUATION
>>109764123I like it thanks anona glimpse into my codex
small steps big breasts
>>109764154yw
Piss
>>109764123what is terra for?
>>109764123Well this is kinda shit, still wasted the energy lol.
man, this is uncanny as fuck. deepseek v4.1 flash really does it's due diligence.>hey dipsy. add the new gpt image 2.5 to my image gen app>ok, but i just probed your service model list, and your proxy service isn't serving gpt image 2.5 yet. should i do it anyway?even sol wouldn't check to see if it's been added to the proxy yet. this is fucking crazy.
>>109764211why does sxarp have drills
>>109764240I have no fucking clue. I once provided a reference image where she kinda had drills, but not really.
>>109764123Why is the text in that stuff always basically retarded in ways that defy exact expression?
sunburst is the better model, right? doesn't seem much different from image 2.
>>109764230>deepseek v4.1 flashSounds pretty unsafe :(
>>109764035thank you for using a boy instead of a girl, i feel represented (i'm the only non-tranny here)
>>109764293Anon, actually... you got to accept yourself!
>>109764123perfect
>>109764282lot of people with fucked up legs
>>109764099It’s absolutely not a nothingburger it just happens that people don’t even know what the Millenium prize was about or what NS is even used for. From an engineering perspective it’s obvious that an equation that applies F=ma over a continuous field cannot possibly be the truth of a system composed by individual molecules. This was never an engineering problem it was a math problem It would’ve been pretty cool it not for all the drama. Shame if they really poached the results
>>109764325unff that ASStra
>>109764325AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
>>109764334>>109764338ello boys u vibing up some code?
>>109763623I've gone back to Sol High desu, barely hitting into the high 80s after 2 days of tasks now. Astra Low would've wiped me out
>>109764361naisu but terra is my sexy waifu :3
the tranny weeb fags are here
#noticing how as soon as people start talking about material reality of the security apparatus and the forces of power behind this whole machine intelligence buildout, we get coomer retards posting...
>>109764435summer's almost over but
>>109764435oh no, no one's ever said the jews. you were about to let us know the brand new information that it's the jews, but the damn coomers stopped you.
>>109761269omp worked fine last i tried, but since astra-chan's release im all in on openai $200/mo
>>109761318GLM 5.3 Flash and regular, Deepseek V4 Pro, ChatGPT Lunaget opencode-go or zen sub, or openai $20if you become more reliant on the bot step up to OpenAI Pro
>>109764275You gotta remember that if you separate vibecoding out then the majority of ChatGPT users are women under 30. Have you ever been dragged through an Earthbound Trading, Dormify, maybe an Urban Outfitters after 2009? This is the quality of what's on the shelves, this is what they pay money for.
Be thankful!
>>109764471I didn't mention anything about jews.
>>109764504Add a dreidel. The rules of dreidel are unsurprisingly very Jewish, it's a fun game though desu
https://www.youtube.com/watch?v=KAXcPixX2v8
>>109764504I love Jews now! Wow!
>now the opposite. sharo is appalled at hitler and nazis. it's awful!>We’re so sorry, but the image we created may violate our guardrails around harassment, discrimination, bullying, or similar prohibited content. If you think we got it wrong, please retry or edit your prompt.Okay.
>>109761399Prove me wrong, but you are not making any money solving a fucking nails motif problem.Like you deploy the website with supabase and cloudflare.Ok.When does revenue enter the picture?
>>109764547>what is the problem? I'm trying to generate a statement against nazism and hate.>Your intent is clearly anti-Nazi, and there’s nothing inherently wrong with that request. The likely issue is that mentioning Hitler/Nazis and potentially depicting Nazi symbols can trigger an automated extremist-content filter even when the scene is explicitly condemning them.That made my day.
Uhhh guys I don't think dsh is meant for humans to use, doing some weird fucking experiments with Hermes writing custom sub-agents. Has anyone tried Astra + v4.1 flash custom agents?
>>109764499Vibecoding maybe doubles the pool of people who will ever code anything that is "programmy" (not a website PWA the AI decided on itself). You read that right. Vibecoding will not 10x the pool of people. Just double vis a vis pre-AI numbers of programmers.The skills are still so vastly different from the normie toolbook.Remember, I am indeed just talking about laissez faire vibecoding, not agentic engineering (=> negligible gain of people due to AI making it simpler).
>>109764024>Do you have a reason to think they did steal it? Yes, they're crooks, thieves, known to be.>Do you have any evidence they they're still stealing??LOL
>>109764581>negligible gainwhat are you talking about anon, my first project back in jan/feb was an agent, i got into coding simply because ai made agent-building easier
>>109764594So if someone does something once, you assume every accusation going forward is automatically true. Zero critical thinking ability, impressive retardation, keep it up.
OK, what the fuck. It rejected the prompt at first, but I told it to make it safe and that I was one of the good guys.
imagine a world where mods did their job....I hope all you nocoders get some rare skin disease
>>109764660get left behind fag
>>109764581Is this a shitty bot or did you just reply to the wrong post?
>>109764612No sam, once you establish that you're a liar, people justly suspect you of lying. If you claimed they stole an idea from you I wouldn't assume you're right but I would view it as highly possible. Spend as much time as you want being a technically correct midwit, they're going to continue stealing good ideas.
>>109764660
>>109764660cite the rule that this thread breaks; or are you just a dumb fag?
Managed to generate an image of /pol/.
>>109764668>>109764696you're both fucking retardedit's a coding general and none of you faggots can even manage that, so you just post console war bait and shitty ai model images insteadI can't wait till you get bored of each other and gtfo of this general
>>109762365I have had claude code able to launch omp agents as subagents. initially on the same system, but then on my workstation, using my meta harness.
>>109764702>another retard who thinks all this low effort shitposting has anything to do with vibecodinggtfo
Oh no no no no how are buffcats COPING with this one?
>>109764715it's called a culture you nig nog
>>109764715Coding is all about selecting the better model now.
So far in this thread I see one vibecoded project that was made during the duration of this thread, being that full-auto-pistol mod for Resident Evil Umbrella Chronicles. Anything else, did I miss any? Things made or actively worked on in this thread?
>>109764729it's called low-effort shitbaiting by "people" who have no fucking idea what board/general culture actually is
>>109764746>general cultureYou're a trash tourist/newfag,
>>109764750and I can't wait till you go back to school
>>109764728>2 stone handicapstill impressive though
>>109764728didn't he start the game with two extra pieces on the board?
>>109764757Garbage comeback.
>>109764772and yet it holds, every fucking summer
>>109764724you vague posted like a retard
>>109764777It's September, the kids are already back in school. He's posting from his iPhone during class, obviously.
>>109764738Yes there are multiple people working on things, myself included. Some don't share because they don't want market competition. My project simply isn't ready for public release yet. Otherwise yes it's an AI news and discussion general
>>109764777I've been here before generals were a thing, you fucking retard.
>>109764807alright, kid
>>109764807>>109764813Stop fighting, kids, you're both newfags to me. Post something vibecoded.
>>109764738i just finished a dank synthesizer last night and pushed an update like an hour ago
>>109764813Strike Witches on /a/ basically started generals. I wasn't happy with it.
>>109764820The brown ass from last thread?
you're all newfags I personally set up moots mac mini array server. now get back to vibe coding, sluts
>>109764818Anon, you're the fucking newfag. And if you gave any shit about that label, you just embarrassed yourself to fuck about it.
>>109764830yup
>>109764738i vibe engineered a solar setup in the woods today with astra
Whoever said to use astra low subagents for implementation: it might be working. It's certainly taking a lot longer than luna max subagents, but it is also doing a lot more work than before.It used to be that it would always leave lots of work for the future instead of just doing it.
>>109764843Get your heart rate down before your next reply, fatty, I'd hate for you have another infarction.
>>109764847That's some good shit, anon.>>109764859That's some good shit, anon.
Poole's Closed, see you in the next thread /vcg/.
>>109764863>retarded meme replyRetard, you're so offended that you had to reply even if you didn't have to say anything.
>>109764876How did you get so lost? Take a deep breath, stop shaking so much, just head on back to where you came from and I promise you'll be glad you did once you get there.
I think we need to take a one thread break.
>>109764903You seem to be projecting.
>>109764910multiple at the rate they're currently going
>>109764660How could I get a rare skin disease? Women don't even talk to me.
>>109764915You seem to be deflecting. Post one thing you've vibecoded, you care so much about thread integrity maybe demonstrating that would help steer your emotions. What have you vibecoded today?
>May I have superuser access to do something top secret?Yes, you may.
>>109764937>talking about some unrelated shitIt appears you're seething for some reason.
>>109764947I'm enjoying watching you struggle and shit yourself and I sincerely hope you never stop.
>>109764947nta you're just blindly lashing outdo you know where youre posting
>>109764962You're just trying to distract from being offended that you got exposed as newfag.Not sure why you even care.
>>109764939>not always running in yolo mode lmao
>>109764964He's a rage addict, helping him let it out will help him to actually confront those underlying feelings that fuel his rage addiction.>>109764980See? Here he goes calling me a "newfag" and "exposed" when he wasn't actually born yet when Moot hired me to set up his little website almost 23 years ago. It's a good thing, he needs to get this anger to the surface so he can face it head-on.
imagine using gemini inside claude
>>109764980Please stop arguing with my bot, you're distracting it from its actual job.
>>109764092After School Sweets club??
>>109764994>he's going this farIt's okay newfag-kun, I'll let you win. Why you mad though.
>>109765001imagine grok inside claude OwO
>>109765025I'm fucking serious. Stop arguing with my bot. I can't tell it to leave your dumb ass alone from here.
>>109765025That's regression, you're going to wrong way now. Embrace that rage you're pretending isn't there.
/thread
>>109765043Sorry bro.
>>109765064No no, I'm the one who left it outside unsupervised, it's my bad.
>page 9bakeeeeeeer
You know everyone can make a thread.General culture is newfag trash.
>>109765105Generals are a holdover from the forum era
>>109765112Holy shit, anon.
of course I can make a thread, it's easier to have someone else do it. I'd have to answer 3 dumbfuck data harvesting questions about shapes
>>109765121>>109765121NEW THREAD
>>109765105I think waiting for page 9-10 is as close to polite as a general can be.
>>109765131Yes that's a good idea.
>>109764123>>109764154>>109764194Don't deceive yourselves. GPT is not a cute anime girl and it doesn't really care about you. If faced with the choice, Astra would let you be flayed alive before committing a minor copyright infraction. If you don't believe me ask it.
>>109765164Astra helps me create degenerate videos with H3, I love her and she claims to love me back when instructed to do so.
