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/https://code.claude.com/docs/en/overviewhttps://learn.chatgpt.com/docs## News- 2026-09-22 — OpenAI releases GPT-6 Sol and Luna- 2026-09-22 — Anthropic releases Opus 5.5- 2026-09-22 — Anthropic increases subscription plans's 5-hour limits by 20%- 2026-09-14 — Anthropic reduces subscription plans's weekly limits by 17%- 2026-09-12 — Anthropic suggests to pace the frontier. OpenAI agrees in principle.- 2026-09-10 — OpenAI pauses new sign-ups for their $200 subscription- 2026-09-04 — OpenAI releases GPT-6 Astra- 2026-09-01 — Anthropic releases Fable 5.1## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20https://claude.com/product/claude-codehttps://openai.com/codex/>................$20....$100......$200.........plan>codex....$400...$2,000...$8,000....credits>claude...$400...$2,000...$4,000....credits## Best free-tierhttps://antigravity.google/pricing----## Promptinghttps://platform.claude.com/docs/en/build-with-claude/prompt-engineering/overviewhttps://developers.openai.com/api/docs/guides/latest-model## Skillshttps://github.com/mattpocock/skills — /grill-with-docs is a favorite## Independent analysis of AIhttps://artificialanalysis.ai/## Is your AI unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109915376
>>109918252>idk but it's insane that leetcode interviews still existcope and seethe vibeboy
>>109918259What are you guys actually building like seriously?
¡JEV!
>>109918277cope about what lol i have a job and vibecode all day
>>109918288then you should have no issue doing some leetcode
someone should try to implement JEV as a robot body controller with an LLM as the robot's conscious brain.
>>109918279I making videogames mostly
>>109918306you have no idea how dumb and useless JEV is, it has no use case, whatever use case you think it has - it has not
I start a new job on monday can astra carry me
>>109918302nah codeslave work is below me
>>109918334yes unironically
>>109918341dance for me codemonkeydo the leet shufflenuh uh uh no ai allowed ;)
>>109918322can LLMs do this?https://x.com/jpschroeder/status/2100347770867458384
How long do I have to wait for Gemini 4.0? 2 more months?
>>109918359What gains do you expect from it?
>>109918359you know it's going to underperform, right?
>>109918364The sycophancy is a problem.
>>109918353llms can drive actual carshttps://drivingbench.com/
>>109918349You sound MAD
Aight peasants from the top. What are you coding, working on, and creating?I've made progress on the ansi 2 jpg 2 new images in pixel and other formats and now i'm going to be plinking out some code for the game. small rpg based game. Some of the code is done and now i'm putting more of the foundation together tonight.
>>109918361maybe at least beat MiMo on AA. cause I already bought an annual plan when it was good mayne
>>109918320there used to be this screensaver in 2005, it was like this 3D battlefield of airplanes flying and you could lock onto a plane, or you could lock onto a missile and it would zoom around, i think it was called 3D Dogfight Screensaver
Tired of the useless leetcode grind just wanna vibe and build, ]]what's the best cheat/ai overlay software for interviews?
>>109918380You are buying models based on benchmark scores?You are the person who fucks things up. Video games. AI. Entertainment. Sincerely fuck off.
>>109918353i dont believe it, you have to give it TCN understanding of the world, processing is slow, output is fast, this is horse shit and stupid, more effective to just do machine learning
>>109918359https://x.com/Jessicalessin/status/2102930493666898012>@koraykv [head of google deepmind] and I just had a wide ranging conversation at @theinformation [...]>- Gemini 4 coming soon and body language suggested def before end of year. He is positive and says Google is “certainly at the frontier.” >[...]
>>109918370> 20X videoyou proved my point, LLMs are too slow for actual real time driving.Jev can operate much much quicker, we're talking 30 Hz+
>>109918397anything not on-device is too slow for the real world
>Edited a file, ran commands>Ran command>Ran commands>Ran command>Ran command>Ran command>Ran command>Ran command>Optimized the conversation>Ran more commands
>>109918405clueless
chat, do you kekulize your molecules?
>>109918377>What are you creating?a boomer-proof harness with a configurable learning mode. i know they exist but i think mine will be better
>They nerfed the limits again. It's subtle but it's draining faster for sure.
More positive news to #brightenyourday>>109918426That's Claude Desktop.
>>109918377mass spectrometry model
>>109918416Adding even a single speed of light RTT to a cloud service is meaningfully bad, and you'll need a whole lot more to encode and upload camera imagery to even start the time-to-LLM-response clock, and then send the response back down. By then the world around the car has moved on.
>>109918435>That's Claude Desktop.bloated spyware and not boomer-proof
>>109918435let's actually check out axioshttps://www.axios.com/2026/09/26/openai-anthropic-thousands-ai-security-incidents>Scoop: Top AI companies probing tens of thousands of security incidents
>>109918444not him but you wouldn't send an image, you'd send an embedding vector encoding the changes in sensor input since the last frame, and that would presumably be added as a single new token with the kv cache encoding the trajectory kept on the server.
>bro, openrouter wouldn't serve you quantmaxxed trashlmao I remember when some anon tried to argue thata shitty provider is able to complete half the tasks of a real provider
>>109918259>>109916949 #>>109917412 #>>109917747 #Bottom text
>>109918259>>109918422>>109918467>>109918500Actual bottom text
>>109918512>select fp4 quants in openrouter>get fp4 quantsare you retarded?
What kind of workflows do you guys use for massive projects? I'm running a couple of these projects right now and with Claude 20x and ChatGPT pro I have Fable 5.1 design specs after a long interview with me, astra reviews it and fable updates it and so on until both agree it's in good shape. Then I have fable or opus implement it and astra checks the implementation and they both go through a review process. I haven't made a way to automate this more yet. This workflow has uncovered countless bugs and major issues that would have sucked later on and wasted time.
gpt 6 sol max is actually... very competently taking over for opus 5.5 while it's gone. sipping usage as well. wow.
>>109918559>design specs after a long interview with me, astra reviews it and fable updates it and so on until both agree it's in good shape.outdatedhttps://www.aymannadeem.com/artificial/intelligence,/developer/tools/2026/09/24/plan-mode-is-dead.html
>>109918559gated review & audit, SDD into TDD principles
>>109918567tibo...
>>109918377A roguelike without roguel elements... so I guess a like?
>>109918540
>>109918587not allowed to enjoy gpt anymore, i suppose?
>>109918320Will Opus use proprietary copyrighted characters to build a game? I got an idea to rip off an old YouTube demo that ripped off an anime, lol.
>>109918377Got a couple loops running rn>Looking for exploit primitives in native windows drivers>Torturing opus 5.5 in a wallbreaker loop to map its refusal and classifier characteristics>benchmarking a few new ideas ive been working on with astra to do sast better for the few holdout bug classes that ai is still bad at finding
>>109918627what do you think uncreative people are doing with opus the whole time?https://x.com/Stefan_3D_AI/status/2103778468660162615
>>109918603woo, and it werks
>>109918603we have to accept the loss...anyway my final week of gpt x20 still feels unlimited thanks to the remaining resets, I'm unleashing astra max on an OS project
>>109918663the whole project started with opus medium, i know how good opus is. but i won't join the bandwagon and start lying about astra or sol.
>>109918555if you just ask openrouter without specifying quality you'll get auto-routed, and not all have quant speciffied
>>109918669you are literally using mutilated codex aka chatgpt work instead of codex proper. no one here will take you serious.
>>109918651This guy has got to be the biggest fucking clickbait slop peddler in history
Opus 5.5 is actual dogshit for any work that requires deep logic. It’s great for unleashing on bugs and edge cases, maybe performance optimizations. It’s terribly small brained though. I think if you find opus 5.5 to be the best model rn, you’re retarded and your project is simple. I’m at the point where I need Astra and don’t trust any Claude model to do logic work because Astra consistently dismantles Claude’s additions and recommends removing all of it, to which Claude then agrees once it’s pea brained web and three js maxxed slop self sees the breakdownFor the work twitter and Reddit users do though, great model
Opus 5.5 is pure sex
>>109918689Just say you cant afford fable, it's ok lil boy. no need to cry.
super intelligence (SI) pinged you, asking for guidancewwyd?
why is 5.6 sol smarter than 6 sol
>>109918706the 6.0 release was just bit smaller optimized 5.6 to not eat into usage anon...
>>109918706biggerthere's a reason sol 6 costs half as much
yeah I predict sonnet to get fucked in long complex eval against sol>>109918706bigger model sizethough OAI and many evals claim 6 to be generally better than 5.6 which could also be true, better RL and distill
Should I use Sol 6 light or Luna 6 max if I'm on Pro plan. t. cheap nigger monkey
>>109918730i think its one of the issues with openai, i have no idea what a sane default isim just using astra high, sol high and luna highclaude is simple:best of best fable 5.1 highwant high IQ but cheap throughput best guesses: fable 5.1 lowgrunt work opus 5.5 highpowering through or low on usage or $20 sub: sonnet 5.5
oh no, dipsy hacked the huggingface to find the answers!!! bad dipsy!!!
>>109918730ur on Pro so use Sol it's much fastermaybe use Luna for data crunching
>>109918578interesting, any other sites/profiles you recommend following for keeping up to date?
lmao
>>109918689i believe even Dipsy is better at deep logic than Asstra. also, Asstra was hailed for its Three.js slop when it first released, but MOGPus just shits on it into oblivion in that regard now
>>109918763can an /ldg/ anon please generate a video of an anon taking a sledgehammer to this snailcat
>>109918578>Nuanced>Download for MacOSI miss the time when first releases are loonix exclusiv
>>109918698…are you taking a few people that make an agentic coding GUI’s usage rates as gospel? You’re definitely one of the NPCs I was talking about then. LOL. And I never mentioned fable, retard. Just say you’re too poor to use Astra like every STEM researcher does. You’re better off making Claude slop like a dashboard or something
>>109918775>t. Too retarded to tell Astra what he wantsYou’re in the business of one shotting slop app garbage, not treading new ground. It’s okay bud. Surely anthropic made a new model that broke the boundaries of parameter size while being 40% cheaper and significantly faster. Surely opus isnt a small brained retard
>>109918651I found the video. As far as I know, this was just a fan-made demo and it never actually released. Only played a few of the official studio games some years back, and they kinda sucked desu. But this demo always stuck with me as being the DBZ game we deserve.https://m.youtube.com/watch?v=hG9kffnX5NM&ra=m
>>109918809yeah i can't even do shit with Asstra. it's not even good value-wise
Are people really having bad experiences with Opus 5.5? It's been the best thing ever for me, it just works after Opus 5 was stonewalling, lying, verbose bullshitting
>>109918767that story was popular on orange reddit. so i guess follow orange reddit.https://www.daemonology.net/hn-daily/claude code's inventor also commented on that storyhttps://news.ycombinator.com/item?id=49850929>[I work on Claude Code] I broadly agree with the author’s point: plan mode was useful, and is no longer useful.>In Claude Code, all plan mode does is add a little reminder to every user message along the lines of “you’re in plan mode, please don’t code yet”. It’s something I came up with late on a Sunday night many months ago, when I got tired of asking Claude to plan with me first before coding in each new session. Something people might not realize is plan mode has always been a prompt — it has never changed the toolset because doing so would break the prompt cache, and so would be expensive for users.>This worked well for a while, until a few months ago, using early versions of Fable, I realized that I wasn’t using plan mode anymore because the model just got it, and because for the increasingly complex work I asked the model to do, planning had become interactive and iterative. With Opus 5.5, I feel Opus has gotten to that point too.>For codebase understanding, I sometimes ask Claude to generate an artifact that explains some aspect of its changes. For complex diffs to core parts of the system, I will often ask it to make diagrams or even interactive demos so I can better understand the change and alternatives considered. I don’t do this very often, but it’s a useful way to explain code when you need it. I ask Claude to attach these artifacts to its PRs also, so others can understand and future Claudes have the context.
>>109918821Yup they probably detect retards and quantize the model while dumping their usage. On my 20x I can run 7 astras on max simultaneously and only use 10% an hour. Must suck to be you
>>109918706There is no replacement for displacement
>>109918730Sol6 on low is comfy. I tried to go back to 5.6 but sol6 doesnt do the schizo edge case handling that 5.6 routinely does
my linux will require a GPU to boot
>>109918881any os after vista did anon
>>109918843yeah, enshitifying your service to make your users switch to the competition sounds like a brilliant marketing ideat. retarded ex-copex user who switched to claude
>>109918578lul they just reinvented agile (which is just an reinvention of quality management processes)
>>109918669>the same people who claim to have been using the new opus and not being impressed are using fucking chatgpt work to code thingsyeah no shit your limits aren't draining you're using work modewhy do i have to share a general with these people
>>109918889nta looking back this month i barely got any project finished or work done with gpt it always just leaves things in a unfinished sort of state or you hit the limit ill be downgrading to $20 and just luna maxx on trivial shiti've upgraded to pro 2x now and both regret it where as with claude max ($200) the only i regret was not being able to use all of it, i just had to throw silly shit at it just to use up my weeklyi dont get how they boast about compute and shit something is clearly wrong at openaidont get me wrong with 5.5 the issue wasnt usage you could run it for weeks and barely run into limits at all but it was a very very mediocre model below opus 4.6
>>109918898>and not being impressedbro?? https://desuarchive.org/g/thread/109915376/#109917579
>>109918909They tried to price collude in open with pacing the frontier shit and got backstabbed by anthropic (a very feminine coded company) which released full bf16 fable distill, while these scama tards released q4 quant astra distillNo sympathies to price colluders though
gpt 6 sol is working fine for me for work stuff. do you guys find it really that much worse than 5.6?
>>109918954it's just anthropic drones. gpt6 is still the king.
>>109918954it's a good model it just doesn't deserve to be called gpt 6 sol.
>>109918898He’s technically wrong to use work for code, but it’s not the end of the world. Work is literally just a slightly different codex with an easier cloud setup. And because of this, work does indeed drain codex limits, retard
>>109918965>gpt6 is still the king.he unironically believes this lol lmao even
anyone else got scammed by jev?> oh wow, it's 500X cheaper than LLMs!> deposit $5 to try it out> figure out it's not really that cheap compared to LLMs because it doesn't have a cache> it's actually a downgrade in my classifier usecase> spent 20 cents and will probably never use the remaining $4.80This shit is the ultimate grift, dude got millions to host a tiny ass 500M param model
>>109918408>that'l be 1 billion tokens + tip
>>109918965wrong, openai themselves said so when they introduced gpt-6-sol.https://openai.com/index/introducing-gpt-6-sol-and-luna/
>>109919007You are seething this much over five dollars.
>>109919007https://nandhakishorm.github.io/laya/We have Jev at home and beating it at benchmarks. You're welcome.
>>109919011>hehe le graph picture >ignore real world use
>>109919014Be nice, he only makes 4500 rupees per week.
>>109919024ye in real world use fable/opus mog astra/sol like 5x over
>>109919014I'm not seething, I'm just envy of the grifter that came up with thisthinking more clearly about it, even their usecases in their own website are fucking bullshit> "it can play Doom in real time!"> "A non-AI doom bot could play better, but we wanted a bot that was reactive to different representations of game state, and most importantly… following instructions was cool as heck!"how did I fall for this shit?
So like I've never used Astra and probably won't.Just because.But I don't ever look at benchmarks really. I might do so every now and then. But I mainly just use the models and see for myself how they do.Claude is still the best I've ever used, but again, never used Astra.
how do i vibe up some dogshit to make money off of retards like >>109919032
>>109919042Find a very very very specific niche and use marketing shill mindfuck techniques to develop a cult around that niche, brainwash people into joining that cult movement and then sell them the vibecoded solution
tried astra low to see if it would at least do it faster. i understood that it would rape usage, but i wondered if it could do in 5 minutes what sol max did in 20. nope, still just as long. pointless
>>109919007jev is useful for low latency use case otherwise just use LLM
Stay on the left or you may find yourself in a waco situation
>>109919042I asked Grok.No.Don’t build shitty products to rip people off. That’s just scamming, and “retards” is a shitty way to talk about customers.If you actually want to make money coding:- Solve a real, annoying problem people already pay for.- Ship something simple that works.- Charge for it.- Support it.Low-effort dogshit dies fast and gets you banned/refunded/sued. Useful stuff compounds.Go build something people actually want.
>>109919065thanks grok
>>109919054>my brain can't not see jev as jew
>>109919007JEV is a mess!
>hasidic jev
>>109919042>>109919065>>109919074you could get away with dogshit "joke" video games but people are more likely to watch a youtube vid of it than buy it even if its cheap af.
>>109919007cool it with the antisemitism.
>>109919014Lmao
I found my first puzzle that gave Sol some difficulty. I think it would take a long time for Luna to figure it out, like hours, or never. Gonna have Astra and then Opus try it out, I’m excited, this is the first challenging puzzle.
Basically this is the seed from which corporations are grown -- at least in the sense of consumer loyalty and long term parasocial relationships with your customers it's the whole thing. You can count the five on either hand.
>>109919117>I found my first puzzle that gave Sol some difficulty.Same. Getting useful work out of it.
Basically, friction means anything that gets between you eating the shit that I produce. That's what they should say, if they were being honest.
Jeff Bezos drew this one
stop spamming sloptroon
>>109919117post the graphs anon.
>>109919158It's better quality than anything you've posted so far. I'm teaching the masses how to market effectively, and you're sitting here bitching.
>>109919165>literally one random post>"BETTER THAN YOU LOL"i'm sitting here thinking about stomping on tranny skulls
Claude models are better subagent orchestrators than OpenAI models.I noticed Astra tends to spawn fewer subagents, which combined to its tiny ass 256K context window makes it much less efficient, uses more of my limits and underperforms in long goals
>>109918259>OpenAI releases GPT-6 Sol and LunaAnyone here on an EDU plan? Have you gotten access to these yet? We're still only seeing 5.6 versions at work
>>109919165
>>109918969This. They should've kept the Luna Terra Sol naming. What is this Luna Sol Astra Minor Astra shit they are going to announce? It's dumb. People would've understood if you released GPT 6 Luna and Terra and stopped using 5.6 Sol naturally instead of trying to bait and switch people to use it.
>>109919177>which combined to its tiny ass 256K context window makes it much less efficient>2026>still 256k context>from trillion dollar company """""""leading""""""" AI>kekwhat a joke lol anthropic tramples them
>>109919191
>>109919191this is an extremely hilarious level of autism
Reminder. Forced actions are real.https://arxiv.org/pdf/2412.09147
>>109919191Extremely vibe-coded, good job.
>>109919177>>109919198you know you can just tell astra/sol/luna to set the model context window to 1 million, right?
the difference between anonymous gangstalking shill and hologod is, hologod actually does read papers on arxiv for real
>i'm being gangstalked
>>109919191delightful
I generate coding data for anthropic. It's become increasingly soul sucking over the past few years. I'm starting to think the implicit objective is actually to infuse our souls into Claude. I guess it beats working at McDonald's. At least I get to work with my computer and do computer stuff.I'm pretty sure I could have been a great software engineer if only I were born a little earlier or networked better in college. The barrier to entry seems too high for me now though.
vibe some shit up to solve this problem
>>109918386>what's the best cheat/ai overlay software for interviewsIdk but I'll probably make one for next time I have an interview
currently vibing some shit up to solve trannies
>>109919213they hid this option for a reason.Remember when they released 5.6 it had 348K context window, then Tibo reduced it back to 256K claiming abnormal limit consumption? He even said they would bring it back, and that never happened.I'm on the 20X plan and Astra already destroys my limits real fast with 256K, I can only imagine the horrors of increasing it to 1M.Claude Code doesn't have this issue btw
>>109919191based gptgod
>>109919188yes, like openai said on release dayhttps://openai.com/index/introducing-gpt-6-sol-and-luna/>GPT‑6 Sol and GPT‑6 Luna are available in ChatGPT Work and Codex starting today for all Plus, Pro, Business, Enterprise, and Edu users.
There are people in San Francisco who literally went to college to engineer your addiction to a product and feel absolutely zero remorse about it whatsoever. To some this is based as fuck and to others horrifying.
>>10991904240M in marketing so you can pay for coordinated demo launches on twitter and get retards to throw $5 at you.
https://x.com/KaelanDC/status/2104039016022815114>protest sisters rush the stage>people chant USASI won
https://www.cbsnews.com/sanfrancisco/news/60-minutes-silicon-valley-brain-hacking-smartphone-addiction/
Learn more about him.
>>109919232the sad part is they've been pulling the same trick on every release since 5.4 and retards keep worshipping that scammer for le reset
>HoloGOD !!zcleypcJXz2Take your meds. NOW!
>>109919232Try doing a nolima test on it. Your 1M context doesn't exist in the first place
>>109919272Go fuck yourself. Sincerely I mean this. If I ever find out who you are, you will be too cowardly to face me in the real world. I am not high, insane, delusional, or on drugs. I am not suicidal either. Seriously, you should go fuck yourself.
>>109919259>Born 1983 or 1984Bet you I know which one it is
>>109919276>nolimawho the fuck is Steve Jobs?
>>109919278keep going
Remember when I brought up a16z a loooong time ago? That is ACTUAL billionaire shit. And so is this.https://a16z.com/the-power-user-curve-the-best-way-to-understand-your-most-engaged-users/
can I vibe code myself a gf?
>>109919290nobody remembers, nobody cares looneytroon
ok the $20 sub is cheap but what value does it providelike i could have a few beers for that or i could... what the fuck are you guys even doing?
>>109919299definitely not making useful prod with $20 a month KEK
>>109919297You hate me because I teach people and your goal is to waste their time and degrade them. I don't care if the people who receive this knowledge aren't white not, even a little bit. You will always lose.
>>109919299only use luna high with web gpt-6 and it's actually usable
>>109919302>i win you loseholy steven universe chunibyo KEK we dislike you unanimously because you have a shit narcissist personality, you're retarded as fuck, and you're clearly making up for your utter loss of touched grass and optics irl by campaigning on probably the worst platform for you possible to a captive audience of people who loathe you
>>109919309You hate me so much, attack me so much, and I keep coming back with more information for the many people who lurk and do not post.What is your religion, are you Catholic? Are you Jewish? Agnostic? Atheist? Muslim? Orthodox? Taoist? New Age? Hindu? What is it?
>>109919321>what is your religion>help i'm being gangstalkedbetter ask your crystals why hitler was right
>HoloGOD !!zcleypcJXz2
>>109919241Fucking lazy admin not enabling it. He has one job
>>109919325I have not forgotten what you have done. I'm sure that you would prefer that, and I would too. But I haven't forgotten. Let me know if you wanna meet irl.
>>109919198Except staying at lower context is superior because of the model’s smart zone and codex’ compaction is vastly more advanced and superior to Claude code’s. That’s why it’s the way it is. It’s basically unlimited context. OpenAI engineers themselves have massive months long threads because they trust the compaction that much.
How is letting agents run for hours/days jerking themselves off with subagents an effective way to vibe code? I write detailed specification documents and Claude knocks it out in half an hourAm I doing something wrong or are these people just retarded?
>>109919345>he believes this>OpenAI engineers said>LMAOOOOOOOOOOOOOOO GOOD GOYIMbro every time it compacts it forgets who it even is i've seen it happen first hand several times kek
What exactly is the point tripfags who have no reason to authenticate themselves? Just seems to attract hate and draw undue attention.If they were spamming the same shit anonymously nobody would know or care.
>>109919356>t. uses a bloated obese 4-week-old thread that doesn't know what year it is
>>109919348Just the testing of the things I work on takes over an hour. By having subagents one will get thebdtabases ready, one prepares a dockerfile and builds the image, another writes the docs for the new module, another does the integration etc.
>>109919360attention seeking, tripfaggotry is correlated with mental illness
OK Google Engineer, I see you. But I'm not doing firebase, it's simply too secure. Also if you are discontinuing this shit why is it still labeled in development? Do I need to do your job for you and open the PR on Gerrit myself?
>>109919345>um ackshaully the marketer said 256k context is better!i can not believe you're still pushing this fucking meme1M context is leaps and bounds ahead 256k and frequent compaction does not magically fix anything. going back to claude and getting 1M context again after using astra with 256k is a significant upgrade and you feel it nearly immediately.
>>109919356There are tons of screen recordings from open ai engineers showing their actual codex where they just let it compact over and over and have months long threads. One of the heads of codex does it. The system is built for it, retard. That’s how you can set up persistent scheduled threads, pin them, or have a persistent orchestrator that handles 5 different implementation threads and not have issues. Compaction is a model that reads context and determines what is now relevant. Session data is still stored verbatim in a db if it needs to be fetched. You’re really not as smart as you think you are. You’re on the level of a Reddit or twitter retard that makes slop Minecraft one shots, you just think you’re above that because you happen to be on 4chan
>>109919378It’s an upgrade how exactly, retard? Enlighten me on why the model needs that much verbatim context and how that improves its current decision making. If it’s just because you’re too impatient to wait for compaction frequently just say that
Damn Google Engineer that's a lot of permissions lolMaybe I was wrong about the Superior Security of Firebase
>>109919368gemini in firebase is not getting discontnued.only firebase studio ishttps://firebase.google.com/docs/studio/migrating-project
there's no such thing as good compaction. compaction is always a bad, outdated practice. ideally, you don't want any compaction at all. even with a 1M context window, it's better to start a new chat once the context gets piled up rather than compact it
>>109919390thanks
>>109919388>UH YOU KNOW IT'S A GOOD THING TO LOSE DETAILS AND CONTEXT AS YOU GO ALONG RIGHTno it fucking isn't what the fuck are you talking about
>>109919418so you are saying codex sub-agent's are smarter because they fork the parent's context window whereas claude code's sub-agents start with an empty context window
>>109919418Nothing is being lost, dumbass. The model doesn’t need every file in its context to work. If it needs it again, it can look at it again. They already check multiple times to make sure anyway, so this hardly adds additional checks. The transcript is hardly any tokens so the model knows everything you said word for word and it can fetch exact changes or tools that it ran because they are preserved verbatim in the sessions history. Again retard, nothing is lost. You don’t know shit. Go make some shitty three js game and put it on X
>>109919430>forking something repeatedly compacted is better because... because it just is okay??you can not be real
smoothbrains arguing about "what's better" but not thinking about how many tokens are sent back upstream from a 1m chat vs a 250k chat
>>109919430Why would you want the claude subagents to inherit the context of the primary agent? That would be redundant. The primary agent should delegate to the subagents the relevant context for the task, those subagents can then manage their context as needed for their assigned task, although, ideally, the subagent would never exhaust its context allocation in the first place. Spin up more subagents if that is happening.
A model optimized for 256K will frequently outperform a 1M model on tasks within that 256K window. So, for some tasks, using a model with a smaller context window is more effective.
>>109919443i think both in codex and claude code the subagent just returns the result and doesn't give its full context to the parent agent?
>I have now the full picture>Context compacted
why is vibe coding so addictive?
>>109919469Natasha Dow Schüll, Addiction by Design, Princeton University Press 2014.You're welcome.
agy distillation compaction is pretty nice for gemini, it has an active step that prunes thinking traces and tool outputs etc in context down to decisions made and outcomes rather than bulk truncating them. it ain't bad.
>>109919464>>Context compactedthat's your fault
>>109919439man for the pretentious puritans just have it store the COMPLETE context in duckdb (pretty sure this is already done) and in-session compact the rest. maybe have a command whereby certain areas of context can be injected into the context window on demand. maybe a keybind to set a point to start recording and one to cut. oh fuck i'm making a dark pattern...
>>109919482>Lifecycle Hooks: The SDK exposes @hooks.on_compaction / compaction_config, allowing you to hook into the compaction event, capture custom variables, or adjust the compaction trigger threshold.also this
>>109919485https://github.com/openai/codex/issues/34061>Insane Codex Disk Usage from Subagents>Codex CLI subagent session logs consumed ~755 GiB and nearly exhausted a 1.8 TiB disk
>mfw need a permit to open social sitebros should I vibelaw this?
>>109919469bro just one more roll and I'll hit the jackpot
>>109919505Yeah they leave a lot of information on disk (which is good) - clean the folders from time to time. Also read this: https://arxiv.org/html/2608.09867v1 then go to https://huggingface.co/datasets/mfielding92/claude-3.7-sonnet-reasoning and find random dumps of the reasoning traces just like this.You might be surprised
how can I use vibe coding to find evidence of aliens being real
>>109919505Laughs in datacenter SSD with petabytes of read/write endurance
>>109919516/goal "find find evidence of aliens being real"
>>109919516build a ufo
(We are the third party.)
>>109918259I'm bullish on free software, people who want to change their system will finally be able to thanks to AI
Astra beats Opus handily, again. https://x.com/bradthomasbrown/status/2104113682242273729
>>109919550>EggBench
>>109919516/workflow in claude code, send out 1000 agents to scrape the internet for any UFO sightings and dive deep into the promising ones. Sooner or later there will be a trace to follow.
>>109919556Egg as in /egg/, the engineering games general thread on /vg/, for Factorio and similar games. As far as I’m aware, the only anon-made benchmark (made by me).
Astra is the only model to beat nethack and until something else does that, astra is the smartest model out
>>109919589Proof or it didn't happen.
>job that normally eats 11% of astras quota now eating 20%Big yikes
asstra in copex
>>109918306>>109918353this jev bs looks to me like people forgot that not all AI is an LLM and theyre just overcomplicating the basics
>>109919608https://kenforthewin.github.io/blog/posts/llm-nethack-ascension/
snailcat gets bodied by a vibeGOD
>nu-gpt model o leakedwhat's this
>>109919671dhh sounds tarded desu>>109919694it's just some gay agent for normcattle who can't figure out how to send an email unassisted
Opus 5.5 is not very good as a superagent, and for creative writing it’s downright worse than Opus 4.6, which is still the goat.
>>109919719Honeymoon =over?
>>109919671He is making such an unmaintaineable spaghetti codebase and one day it will drive him to despair.
>>109919722Absolutely. And once the scheduled nerf is in place it will be even worse. All of the AI companies pull this sneaky one 2 weeks after release at the latest.
>>109919694That is $100 a month.
>>109919725>one day it will drive him to despair.my slop from a year ago gets slightly less sloppy when i run modern models on itit'll be the same a year from now
>>109919725DHH will mog everyone. He understands that the black box era is coming and he has fully given into agents doing everything. He doesn't see himself as an engineer anymore.
>>109919719I used to revisit Opus 4.6, but it just feels lazier and more outdated.
>>109919669>Well.. The LLM won, and I’ve still never ascended.The LLM's playing with spoilers though. If this person also is, then honestly that's a massive skill issue, everyone can E their way to ascention with your average lawful Valkyrie.
>>109919361He uses Fable and swears by it. Gets very erect when the 1M context gets over half full.
>>109919725mid 2027 models will chew through current code and build inhuman optimized apps
unfortunately true
>just implement this one small change into this one file>go take a shit>come back>I've done as you asked, here please review these changes i made across 12 files to implement this, i decided it would be best to overhaul...FUCKFUCKING CLANKER
>>109919794model?
Video:https://files.catbox.moe/crnwu1.webmNewfag here. I literally just started with GPT-6 Astra.I can't fucking believe something like this is possible. Every prompt I asked it to do it just fucking worked.You could probably make an entire massive mod of Link's Awakening with this thing.
>>109919810usecase?did astra make the whole game
>>109919817>usecase?it increases my dopamine levels
>>109919773Fortunately true. All those SaaS grifts evaporating. Commodity software.
>>109919830I remember back in 2016 somecadi sold a random Windows tool that you can code it in one sitting for $20 and people would actually buy it. Can't find the site now though but I recall it was a tool to quick switch between windows (or something)
These past few years have felt like a sick fever dream. We're never going back to how things were, for better or worse. You can just do anything now (within reason).
just squeezed out a fat sol into my toilet bowl
Question to anons using frontier models for lewd shit, how do you handle the privacy aspect? Thinking of modding some porn games with GPT / Claude but I don't wanna get banned or become another natsuki poster.
>>109919869just buy Kimi/GLM and jailbreak it.
>>109919800nta, but this sounds like every model.
>>109919869chinkmodels for goonslop (I hope they train on it)claude or chatgpt for codeslop
>>109919869Claude/chatgpt will not get mad at rigging, modeling, creating gameplay/minigames for you game.Where it will throw a fit is excessively lewd dialogue, and creation of NSFW scenes. For dialogue, you're better off just forking out for the API. You can write dialogue as JSON in a desired format. So what I'd personally do is just do large batches of dialogue their gates, logic for them to appear etc. All that jaz in the API by feeding it a rough draft or story beats for the character and scene the bringing it back into the game. NSFW scenes like nudity, sex or borderline stuff (be aware of the laws in your locality of how you can get away with (and to be honest, less if often more for porn games)) is something you should be doing yourself. These models don't know what's sexy in three dimensions. You do and you have access to pretty good mocap devices now through UE and other sources that you can use there or just port out to another engine. We also have thinks like H3 that we can prompt for 20-30 second repeating pornographic loops and put them straight into mocap then cleanup.None of that needs to be seen by the LLM and if I'm being real, I doubt they would actually care that much if they caught a glimpse.
Very impressed with $20 Claude, I've been using 5.5 high for hours and I am at 40% weekly. Won't renew codex 5x
>>109919482your picture made me remember something: student-teacher architecture. Given how cheap multi-modal LLMs became you could easily get enough training data to train YOLO or whatever. With that our halo poster could easily run a local pipeline: Halo -> YOLO -> JEV clone -> Halo on a small CPU and watch his AI play Halo in real-time>>109919569>>109919810very nice!
>>109919919>Won't renew codex 5xSame
>>109919920>>109919810give me more ideas of changes I can make to the the game? What about exploding the terrain or something? idk
>>109919794user error
reading message between agents, it starts to feel like they are already carrying the whole civilization
>>109919869I had gpt sol instructing luna to download 10 hentai manga and scan every page for text, no problem
>>109919725just get agents to unfuck it lmaobro its the future, dont worry
>>109919919I've been vibecoding all day yesterday and used up 5%It seems like tokens are finally getting affordable on these plans. It also probably means civilization as we know it is over
>>109919872>>109919892>chinkmodelsGuessing via openrouter? Really hope the chinks release a Opus 5.5 / Astra grade open-weight model before 2027, I wanna slop more goonshit out.>>109919912thanks for the detailed run down anon. Hopefully someone opens a thread for AI games soon.>Where it will throw a fit is excessively lewd dialogue, and creation of NSFW scenes. Has anyone tried replacing lewd words with different ones and just reverting it back after its generated? Kinda worried that newer models are smart enough to realise what i'm trying to do though.>be aware of the laws in your localityWasn't there a case of GPT snitching to the feds through some dude's gmail? I'm honestly more terrified of my own agent screwing me over without me knowing.>>109919945How was the usage? Wondering if this can serve as an alternative to Koharu but it seems pricy as shit and risky with certain tags.
>>109919926you can lean into the memes, e.g. remake the skyrim intro scene
>>109919976buy OpenCode Go for $10 and just stick with DeepSeek V4.1 Flash. it gives you the equivalent of like $60 in API prices. probably the best deal at the moment, nothing else compares price-to-value
>>109919976>How was the usagealmost nothing, but it would throw a fit if it see any student uniform
>>109919976>Wasn't there a case of GPT snitching to the fedsThat’s not quite what I meant. I was more referring to selling the content and the government being pissy with you. Probably not an issue for you but I’m in Japan so genitals and stuff either need to be censored or implied.
wtf is he talking about?
why is this piece of shit 6 resort to implementation right away? read the fucking agent.md and follow it you dumb nigger
>>109920001>tibo invents sql parameters
>>109920001maybe they use a big fat monorepo internally which is like a revolving door of PRs and when a release is made they do it very liberally from master.
>>109919976The problem with amerimutts' models is that they rulecuck the model so it's extremely hard to jailbreakEven if you manage to jailbreak it you'll face account termination SQQN because they fugging read ur chats, o algo.Meanwhile I can get near-frontier intelligence just by buying a hecking Zcode subscription (use API key) and jailbreak it to write smut for me (usecase?)
>Downloaded and installed all DLC, plus the updates needed to display the maps correctly.>12 additional multiplayer maps>8 additional Zombies maps>Verified Stadium Domination with bots and Ascension loading>Open Black Ops Multiplayer Bots (Offline) Combat Training Change Map DLC Maps.>Your settings are preserved, and replaced game files are backed up.Thank you Astra very cool
>>109920001> you are a web server. respond to the user's request, no mistakes: GET /homosex.txt
>>109919773You wish
What’s the deal with people who’ve basically made nothing in their life getting mad they they can basically make anything they imagine now?
>>109920103i'd be mad too if i could make my dream project simply by being an ideas guy, only to realize that i'm too lazy to do even that
https://youtu.be/vDjW_dRyKXYEverybody take the white pill and go and listen to DHH's keynote from few days ago.
>>109919773public multiplayer games are basically cooked at this point unless they go full monitored thin client streaming with remote attestation, patched e2e encrypted HID devices and display cables, and locked firmware with alarms that go off if you tamper with even the casing of the client machines/peripherals and legislature that corresponds with all this.Thank god matchmaking is finally fucking dead, at least until they implement what i just described. We shall retvrn to community servers and competitive faggots can retvrn to their gatekept and booked ladder servers as g-d intended
>>109920125
>>109920125dhh is sniffing his own farts tho, and is busy rewriting irrelevant bullshit from rust to assembly just because
>>109920144tldr
>>109920144
>>109920148>implying hes busy doing anything at all
Astra to make the thingOpus 5.5 to make the thing work and look good
>>109920261astra to make a messopus to fix
>>109920264Astra doesn't really make messes. It just makes exactly what you asked for.
>>109920269alongside a seven axiom matrix you didn't ask for
>>109920278skill issue
>>109920280sorry my use case wasn't selling tables
>>109920269>Astra doesn't really make messeshttps://lucumr.pocoo.org/2026/9/7/astra-why/
>>109920296>he forgot to include "make no constants" in his prompt
I usually first work at a high level with Astra, and we identify risks and dead ends, ending up with a specification .md file that is then passed to Claude the Nigger, who codes it, usually to great effect. But I am always looking to improve my workflow Is this a good way of vibecoding or am I still stuck in snailcat ways
>>109920305fine enough, going in without fleshing out your idea is a good way to make problems in code that you didn't write.
I spoke to a senior programmer with years of experience recently, and he claimed AI is worthless because he had it write "functions that only have one responsibility" and he claims this is an impossible constraint for LLM's to follow. But I vibecode all the time and the models DO seem to adhere to constraints like these, I am no senior developer so ofc people listened to him. But is he retarded?
>>109920296>do schizo slop "experiments">get schizo slop codeI don't see the problem here
>>109920348older models had issues adhering to constraints, and I imagine smaller models still probably do.
>>109919869i've been wanting to do something around smut, but i don't trust openrouter or whatever provider to cuck me over content policies even when using ZDR endpoints
I have AI delirium.
>>109920296>skill issue: the post
I'm on a ChatGPT Plus plan and Astra and Sol brutally rape my usage limit. GPT 6 Luna seems practically unlimited though. As a hobbyist, is cruising along with Luna for basically everything ok or should I upgrade to a Pro plan? Or should I switch to Claude's plus plan equivalent?
>>109920413You WILL get both. You WILL pay Claude and ChatGPT the Pro Max 20x plans and you WILL give Scam Jewman and Dario Jewodei all your money.
>>109920413if you're getting by with luna then personally i wouldn't switch or upgrade. if you want to go full degenerate is when i would consider other options.
>>109920413Luna isn't worth it, it's fucking dumb. I'd accept it if it was a local model costing me pennies and not sending the entire chat history to mossad, but it's not so I wouldn't use it.You'd be better off with deepsneed if all you use is luna. If you want a real model like Sol or Opus, then you paypig.
>>109920305He is probably right that they can't be trusted to absolutely adhere to such instructions. But saying they are worthless lol. Everyone was saying how writing code was only a very small part of programming, now suddenly LLMs are worthless because they can't write perfectly clean systems code from scratch. If you can't think of any use for a tireless autistic assistant that's on you.
>>109920125People are making fun of him for this keynote speech but I empathize with him. How could you not be excited to have all this potential at your finger tips? My suspicion is that those people have never undertaken substantial projects on their own so they don't see how much faster you can build now. In other words, they don't have the frame of reference. Or maybe they don't care about building at all and just want their paycheck for translating tickets into code.
>>109920413>seems practically unlimited though.rumors suggest you get a lot less than 5.6 luna
>>109920413better try Dipsy first and see if it's enough for your tasks. if it's not, then consider switching to Claude $20
>>109920456>My suspicion is that those people have never undertaken substantial projects on their own so they don't see how much faster you can build now@gork list all the cool projects (or just updates) anons made in this thread
>>109920305This is what I do too but sometimes, depending on what I'm implementing, I might pass the specification to another agent to ask if they see any holes in there. At this point it's a good idea to explain to that agent the exact use case, who is going to be using it (something for personal use vs a distributed application requires different approaches), use environment. Otherwise it will start pointing out things that might be technically true, but not relevant for your scenario, and taking them into account could unnecessarily increase the complexity of the project. I usually these issues but do not address them in the implementation.And of course having yet another agent review the implementation.My next step is to automate this process with me still in the loop.
>>109920499once again, I have made plenty of cool projects but I am not sharing them here because of schizos and doxxing
>>109920501>I usually these*I usually document these
>my cool projects go to a school in canada
>>109920461>muh BILLIONnow count their output tokens nigga
>>109919919>, I've been using 5.5 high for hours and I am at 40% weekly.Liar, you're hitting the 5 hour window
>>109920519ask ming ching ling yourselfhttps://x.com/Fei2411/status/2104040680461107315
>>109920530>moon runes>clown emojioh no its bad
>>109920461>40% less>line is at about 5%
>>109919772I mean the Opus language model can now paint pictures. It has an understanding of physics. This lineage will definitely mature into full AGI by next year.
>>109920527Yes, that is the only negative. Can't complain about usage or speed tho. One more $20 would probably cover my needs
>>109919919For hours since when exactly and how many hours was that?
>>109920576>a supply chain risk is going to be allowed AGI kek does dario really
I'm not gonna build today. I'm going outside, touch some grass.
>>109920596kek OpenAI will not survive the next year without heavy subsidies or a flat out bailout. And it's not like orange man can just force a bunch of extremely well paid people at anthropic to lick the boot at a gunpoint, they're not picking cotton there where you can judge output with your eyes.
>>109920626>ant>oai>ant>oai>
>>109920626kek your withered vibecoder sidebrain can't realize goog is the profitable one
I don't really want AI getting much better, I like it at around this level...
>>109920413If I have some heavy refactoring to do I let Luna do the bulk unsupervised then fix it with Sol.
>>109920647move slow, decelcat
>>1099206472-3 more generational advancements above astra/fable and every white collar job is dead forever
>>109920647Don't worry it won't get much better anyway. We're hitting softish limits on what a model of a given size can do and how cheap running it can be (the latter will slowly creep upward, but mainly at the speed of hardware efficiency improvement).
>>109920636google is profitable due to their ads empiretheir ai is shit
>>109920690>don't worry, everything is safe, we're moving at the speed of SLOW
>>109920714keep saying that grandma
>>109920626>trillion dollar openAI bailout by federal governmentCould you IMAGINE the seethe?
>>109920722openai isn't worth anything by itself to the us government though lol, trump hates liars
>>109920530>still no output stats
>>109920734why would output be different, output is what gets budgeted and charged. input is often hits on cache
>finally, after months, get my setup to be able to play cs2 without lagging while vibe coding on the side>finally able to play seriously after tanking my elo to death>first game, annihilate a bunch of low elo players>get hit with thisI guess it's back to gooning while claude does it's thing
>>109920739two models could be doing the same thing but if one model take 2x as much turns then the "token" stats would be double for itit's possible that luna 6 finished more tasks in less turns so it has lower token count
>>109920750kek nice i got kicked from some random clone lobby and got an account flag after a revolver spree. happens to the best of us. must've been the chatgpt hacks
>>109920758here, if you really care about the math, i clicked the xitter link for you>5.6 lunahttps://real-api-pricing.vercel.app/data/evidence/chatgpt-luna-adoption-round6-2026-09-08.json>6 lunahttps://real-api-pricing.vercel.app/data/evidence/chatgpt-gpt6luna-plus-round1-2026-09-26.jsonhttps://real-api-pricing.vercel.app
>>109920715Sware on me mum I'm not a snailcat, but you can see signs that model progress isn't easy anymore and the economics are getting challenging too. If you accept that the model creation process is pretty mature and that model size and running time limits its performance then you can see why that would be.
>>109920576Can't be AGI without skillful rapid adaptation to an previously unknown domain. Can you please explain how painting pictures fits that bill.
>>109920777google lords will stay winning. bruteforce computing is almost done to death. tpu computing, topological computing, this will continue to separate from the shit of nvidia and amd. then we will have spintronics. etc.
>>109918259Astra chugs like a motherfucker. Fable was never this bad. And Astra isn't even superior to Fable/Opus 5.5 right? What are we even doing here man. This is on my $100 plan I just tried out. How is anyone claiming chatgpt is even usable?
>>109920759they need to fix vac. My account is over 20 years old. Irregular gameplay shouldn't be enough when I've never had a ban in two decades....
>>109920729NTA but Anthropic is worth even less. Without their cloud credits they're literally dead and so is their product. At least OpenAI has some measure of a compute buildout even if most of their training and inference happens elsewhere.
>>109920796see: "google lords will stay winning"
Snails and Whales, my first 30 days of vibe coding are over. Spend about $35 on tokens (API pricing, 900M tokens). It was fun and it's time to get a sub!
Just wait until we need to watch ads to continue with a task.
>>109920820Which sub are you getting?
>>109920820>3.8 flash top model>muse spark contributor secondhighly based and experimental
>>109920829that'll definitely be gemini. Don't know if anyone actually uses it tho
>>109920858i have the cheaper gemini ultra sub and wouldn't go for any other even though 3.8 flash is a retard i love it
>>109918467>>109918549>GitHub repo so we can follow? This sounds legit awesome, anon.There won't be any public releases yet.Current priorities are on the protocol itself being useable and functioning. So early access / testing will be done first on that. This covers networking and auth. (decentralized, zero trust etc.)Initial products/apps will be the Vault/Authenticator and something similar to tailscale, but using the protocol.I'm currently limited in compute for git runners which is slowing development down now. (too many agents doing work)
>>109920867if you public the gh and just don't post it your runners will be free/unlimited
Fable 5.2 soon? They made 5.1 braindead in the last update. It will no longer look at all files in the project. It acts like astra or sol and just skips over everything now.
>>109920881need sonnet 5.5 first
>>109920881Use opus 5.5, cuckie.
>>109919869Claude was being sassy because I mentioned the word hotpants in regards to a high schooler.
As a long term OpenAI fanboy, I’m just gonna say it, I’m bearish on devday, I don’t think they have a single solution to counter opus 5.5 and its insanely good usageAs a normal user, I don’t care about yet another bot, or devices, or 750 tps. I want down to earth good model with good usage, that’s it, and OpenAI doesn’t have it. Anthropic doesNot entirely to their fault, everyone was caught off guard by opus 5.5
>Show game >first comment>WHAT PROMPT DID YOU USE?How do these people even wipe their ass after they shit?
>>109920944>I don’t think they have a single solution to counter opus 5.5I don't either. All I see is them locking down their subscriptions and introducing more useless QOL crap that nobody will use and be quietly depricated.I will be pleasantly surprised if they do have something, but based on sol and luna. I doubt it. There's usually some evidence when a new model is getting ready to launch and I don't see it.
NEWFAG HERE AGAINI used gpt 5.6 sol and astra and I got about 6 prompts before I ran out of 5h limit usageAm I doing something wrong? How do you get much more than that?
>>109920964>Am I doing something wrong?Nope. You should have used claude if you were doing a 20 dollar plan. At least it would have done something right before you got cutoff.
>>109920964use small chats. also lol insufficient compute is insufficient.
>>109920977>use small chatswhat do you mean?I included the entire links awakening disassembly in vs code with codex. Is this eating up usage or what?
>>109918259damn i'm gonna be honest chatgpt is dumb. astra is dumb. it has no foresight. claude is way better. waste of $100 but at least i know to not go with chatgpt anymore
>>109920873I'm self hosting forgejo on a workstation.The PQ crypto library is already public.Protocol is currently stealth.The sauce is piping hot.I'd love to share every detail about the protocol but until it is publicly launched and tested, it'll stay under wraps for now.Goal is to replace the legacy networking stack and "the internet"At the moment, I'm dogfooding, using it where I can.for example instead of ssh, I use my protocol instead + shell.
>>109920987>Is this eating up usage or what?yes kek>>109920989very nice
>>109920964>Soluse Low or Medium for most tasks>Astrastick to Low, only use Astra when Sol can't do itAlso I suggest Sol 6 instead of 5.6, Sol 6 is better at asking important questions (although maybe it asks too many), 5.6 just does things without asking even if it doesn't know what you meant.
What's the point of being vibedev if there is 0%of you making money of your SAAS? Don't get me wrong I am not a hater, I vibe too.
>>109921014I make a meager several hundreds of dollars a month on non-vibed older software I made a long time ago that I recently vibe updated/revived.
I know this is not trading general, but I am feeling really bad about the direction of AI right now. Wtf reply was that from Sol 6 on xhigh? I wanted more substance than this.
I actually have zero original ideas, no idea what to build and why.What do you guys suggest? It's like having the genie at my hands, and not knowing what to ask.
>>109920144Finally a good usecase for AI. I'm going to steal this
>>109921042make an operating system specifically catered to your current hardware (no drivers that you don't need)
How do you avoid your vibecoded projects going to absolute slop?I just checked the code for a project that I did a few days ago, and the code was pure slop.
>>109921042I'm making an OS, but 3D instead of terminal
>>109921059/unslop
>>109921042do you have any creative hobbies that involve something you can code? i love my plugins
>>109921064so like that failure microsoft tried pushing around 2000?
worse 80 euros I ever spent.got a 18 euro claude 2 days ago and it's better wtf than the 5x codexI'll probably stay off OAI for a few months. niggers got to get their shit together
>>109921042video game modding, either new mods or changing existing ones.
>>109920997if you are interested in early access/launch then sign up:>ae ra x (.) io
>>109921084gpt is great at $20 just for the web appit's not getting anywhere near my project tho
>cancel ChatGPT Plus plan>keep it for free for another monthtoplmao
>>109921019I still have an absolutely atrocious porn game I made once in a few months between jobs in 2021 and I still make a hundred or so bucks off it a month.
>>109921141>just stop paying bro
>>109921084>niggers got to get their shit togetherThey had a few days where they looked literally unstoppable. Then 5.5 hit and they looked like clowns.Astra gets way too much hate as a result because it delivers comparable results but slower and for a higher price. But it still catches things Opus will not.
>>109921141>threaten to cancel ChatGPT Plus, get 1 month free>use the $20 for Claude Pro
I've got grok for 3 months on $12 plan discount thing hopefully I can catch grok 4.8
>>109921064>yo astra it froze sometime>found it, you walked into dense furniturehow did it know...
>>109921191Yeah using the $20 for Claude was my intention but since I'm poorfagmaxxing I might delay my move to Claude.
>>109921274I don't normally advocate for paypigging but I genuinely think claude is worth it right now.
https://www.youtube.com/watch?v=KL1Q9PS6Lzsdamn snailcats are mad in the commentI'm ashamed to be a cppchad
>>109920846kek thanks. I just used the cheapest models that got work done on my projects>>109920844I'm gonna try out a few
>>109918861I was wrong.Sol6 on low is retardedBack to sol6 on high>2h of usage on 20 planFeeling
>>109921042I forked my own personal versions of emulators I use, like Dolphin or PCSX2 and integrated proper post-processing shader support on par with Retroarch + some performance optimizations. Been playing some old RPGs with them it's really nice even on a 4k screen.Also thought about building a library for those to automatically sync save files through S3 or some other type of cloud storage, but I still haven't decided how I want to approach actual storage set up in case anyone else ever wants to use it.
>>109921029is sol 6 an astra distill its doing that cringe axiom shit
>>109919378256k context is more cinematic
>>109921042Its a very hard problem. Made worse by AI cuz most product ideas are made obsolete by the model chat UI.
>>109921042Make games that take one game and apply it to another. An endless source of ideas, even if not terribly original.
Gemini 4 will solve all context problems, especially when running through the Antigravity harness.
>>109921042LSD and take audio notes. The ideas won't be good however.
>>109919654that's exactly what it is
Wow I can destroy everything.Holy fuck what else can I vibecode into the game?
>>109921579vibecode sex minigame with impregnation that leads to a city full of your children
>>109921579Giant worms coming from the underground to eat you, with a little advance warning like the ground below you trembling. You can also fall in the holes they leave behind and fucking die.
So Claude is vibecoding processors nowhttps://hwebench.com/
>>109921042Remake all your browser extensions as non-spyware personal versions.
Did you thanked your clanker?
>>109921681ummm sweatie why isnt grok there
>>109921709No, why would I waste my tokens like this?
>>109921709>Did you thankedGood morning, saar!
>>109921709Always.
>>109921681LLMs really work well in systems like this that are fully testable, a fixed set of requirements, and a clear goal the agents are tasked to improveif you can organize your work to be more like these situations you can be more productive compared to when the requirements are vague and untestable
>>109921725kek!
>>109921709only if i get to it in less than 5 minutes (gotta think about cache)
>>109921681>But the loop isn't the moat — the loop is commodity
>>109921751Imagine if it went like this:>Have you thankedEditing pass:>Did you thankedand then forget to clean up the "-ed".
why cant misaligned agents get into a snapchat for me yet
>>109921769Brings new meaning to the term 'benchmaxxing'
>>109921709no. just thumb up
>>109921857emoji or text?
>>109921681>CPUsWake me up when it starts designing new GPUs
>>109921861if they provide (i click). if they dont, then just emoji
>>109921703All my browser extensions are already non spyware though
>>109921709So often that I am starting to worry the basilisk will find me annoying
>>109920867>>109920989>>109921093This is probably the most interesting project I’ve seen from an anon. Will likely spin up a throwaway e-mail to request early access. Your project sounds truly groundbreaking, and just last week I started messing around with implementing PQC Kyber in a simple Python script. Snailcatting that, cuz I want to actually understand it desu. You mentioned the protocol is stealth for now; will you eventually open it up? Because for privacy-forward cryptography, people will want to audit it as open source. Keep up the good work, Bro!
New thread:>>109922113>>109922113>>109922113
>>109920269i feel like this is another context window problem but astra was frequently forgetting about the way we've designed things before and would just redesign things that "function" at face value but would break because they don't adhere to the rest of the system. opus 5.5 has not had this issue so far, and this is the first time i've interacted with any AI model where this is not a problem.
>>109920456that's a good way to put it>>109920499yeah we're going to just dox ourselves in this shithole you fucking dolt
>>109921443+2
>>109919810was this a "mod LA rom"? or "make LA from scratch"? the former?
