A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.You use Git — right, anon?## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/https://simonwillison.net/2025/Mar/11/using-llms-for-code/## News (both past and future)- 2026-09-22 — 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 Astra- 2026-09-01 — Anthropic releases Fable 5.1## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://developers.openai.com/codex/cli — probably generally better currentlyhttps://claude.com/product/claude-code## Near-frontier models for codehttps://x.ai/cli — no 5h limit for only $30/month## Not worth it for code, but maybe good for interpreting images/videohttps://antigravity.google/product/antigravity-cli----## 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 favoritehttps://github.com/DietrichGebert/ponytailhttps://github.com/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## Is our AIs unlearning?https://aistupidlevel.info/## Will there be a codex reset?https://codex-resets.com/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109882733
>>109885762pokemon shell smash.png
good op
Let's see what the future will bringhttps://www.youtube.com/watch?v=hN5X4kGhAtU
>>109885762In light of Sam, Dario, and Elon's recent statements, I've decided to reduce my token usage. My currently perceived level of pDoom is too high. I feel my cat's life is at risk.
> several months later> OpenAI is finally able to deliver something close to Fable 5> still locked at 256K context window though> GPT 6 Sol and Luna are just price cutsthen Anthropic drops Opus 5.5 and nukes everything again.
Codex is going great bros, no more insane usage, I can actually use the model on the x20 plan lol
More on gpt 6:Adjust reasoning effort and tool availability without breaking cache. Increase reasoning effort(opens in a new window) for harder tasks or lower it for simpler follow-ups, and enable or disable tools(opens in a new window) as your agent’s needs change. Both controls now preserve earlier context for cache reuse.
>>109885762Is Opus 5.5 better than Fable 5.1?I was wondering if subscribing to Max is too wasteful for a mere writer.My main usage is for assisting me in brainstorming, writing, and act like editor providing feedback to my stories. Also, for translating dialogues into Japanese and Chinese.
>>109885815opus is #1 again but a max plan for writing is just wasteful
>>109885815see >>109885689also it’s WAY better at writing normal English instead of Claudishdunno how this translates to other languages
So is GPT 6 not actually an upgrade on 5.6 Luna in capability? It's just more token efficient?
>>109885826mad how you trumptard idiots still show your face here
>>109885851It's definitely different from 5.6 since it has both improvements and regressions in various benchmarks.(Sidenote: what's up with OAI specifically always having some benchmark regressions with every new release? Their training process seems to be more sloppy).
>>109885859your seethe is worth itI don't vote
>>109885859dont be mad :(
>109885815that girl on the left is fucking incredibly hairy. never shaved a day in her life
>>109885903all you jeets have left is accusing ppl of seething. i guess that reflects the sad state of affairs, the only people dumb enough to still support it are jeets.
So Anthropic won this round with a true next gen model. OAI only released 6-terra rebranded as solLikely in few week it's Fable 5.5 vs Astra 6.1Fable should be an expected upgrade over Opus, so the fight will depend on Astra, either underwhelming against Fable or crushing itBecause sol 6 is efficient, expect OAI to deep fry the shit out of it making sol 6.1 a mad autist
so smart and cute
>>109885925sonnet 5.5 and haiku 5.5 waiting room
>>109885925I doubt Astra will crush it. I actually doubt it will even be as good as Fable 5.1
>>109885859supreme tds or extreme tds
>>109885951Thank you for doubling down on being a jeet, moron.
>>109885934
>>109885942Astra 6 is undercooked with how sloppy and lazy it is, it can still grow
the $20 Claude Code plan is actually very usable with Opus 5.5> 4 hours working with Opus 5.5 medium and a few 5.5 subagents> reached 100% of the 5-hour limit (about 1 hour until 5-hour reset)> 14% of weekly limitsA crazy amount of work was done and this shit is Fable-tier, for a 20 bucks plan this is unprecented.Now try making Astra orchestrate Sol agents for 4 hours in the $20 plan and see what happens
>>109885962ha! check mate! they won't let me down! it will MOG 6 luna AND v4.1 flash
im fomoing opus now, but im happy i can actually do work without feeling like im going to instantly nuke my usage
>>109885966if Fable orchestrating Astra/Sol was the old metaand Opus 5.5 is the new Fablethen Opus orchestrating Astra/Sol/Luna is the new meta
>>109885959>only jeets support the aryan godderanged
>>109885970
opus orchestrating deepseek/kimi as replacement for haiku/sonnet
>>109886004You don't need to keep digging your heels, pranjesh.Trump's approval rating is at 30% for a reason. Only jeets like you and the lowest of IQs are still on board.
>>109886021gagandeep you must calm down sa'arkindly understand
>>109886002nah, new meta is Opus 5.5 orchestrating Opus 5.5.Hopefully Anthropic cooks with 5.5 Sonnet, that will make the poorfag plans even better
>>109886004Trump started his second term with a crypto pump and dump. Trump is a giga Jeet.
>>109886042
sama and tibo if you’re reading this, just know that you guys LOST
>>109886021We vibeARYANS are AryansYou and your mother's chode will not look at me dalit. Note it.
>>109886049You're indian, pranjesh.
should I set claude to ultra and put STOP SHITTING OUT AGENTS FOR A SINGLE HTTP REQUEST in the md or use max
>>109885930i look like this (real)
>>109885762>picIs that really how it works?
>>109886063Jai hind. Arya.Coding.Chatting with your mother on facebook.NOTE IT
how does use vibecoding to get girlfriend
>>109886048GPT-6 Sol is paretoing. Let me guess, you need more?
>>109886088/goal find gf make no mistake
What to vibecode what looks good as a website (saas, or anything else) I can link in my Instagram that looks like I am making good money to flex on bitches and guys?
>>109885762Where do I get started with this? I've used AI in the browser for years, but I haven't used it locally. Are there /g/ approved guides for how to properly manage these things so they don't shit out a fully slopped mess? How do you force them to write tests and test their code as they go?
>>109886096shitposting aside sol was already overkill for a lot of things. The fact that it's half the price now without removing its capability is pretty cool.
meanwhile Google:> "we just began training Gemini 5.0"> "we will be releasing Gemini 3.9 flash in the next few days"
>>109885762first experience with opus 5.5thoroughly "stress testing" my application
>when opus 4.8 gets deprecated>I cant do well if I think your gonna leave me, but I know I’ll try>>>/gif/31205056
free unlimited opus is a human right
how big do you think Opus 5.5 is? 2T?
>>109886088get a gf from this general :3
>>109886206even AGI or super AGI cant do that lmao
>>109886123I haven't tried vibe coding either but it seems easy enough. LSTM with STDP learning with Linux. llama.cpp. Qwen 3.6-27B. and Goose.is what I'm aiming for my computer to become self-recognizing. I would be cynical myself but I think I have a grand idea that only a retard could envision to do with a computer. (aside jerking off)
>>109886123>How do you force them to write tests and test their code as they go?you really are new, kek>a finger of the monkey paw curlshave fun with your three thousand bizarre tests
if they called it terra everyone would be happy
>>109885762lack of chinese AIs on that book
>>109886123Just get Claude or GPT. The models always write tests, if anything they write too many now.The skills from the OP https://github.com/mattpocock/skills are decent. AI code isn't as clean as human code for now, no matter what you do.Once you have AI you can ask it about most things, you can bootstrap everything with AI, any software you might need to install, any plugins, skills etc. You can also have AI analyze its own logs to improve your harness, generally you should only ask people for very high level general approaches.
>>109886215i would totally date a cute vibecoder
Do people actually vibecode video gamesNot trying to be funny but I tried it and it gave te most random shit graphics, then when I gave it proper graphics it got shit backwards like a 2d game would have you move right and the sprite facing left, etc
>>109886252what model and effort levels did you use?
>>109885762To compact or not to compact, that is the question.What's your threshold to actually compact?
>>109886252They're RLed for threejs slop, not actual games.
>>109886265On Claude Code I only compact if I'm picking up some 300K+ context conversation on the other day when I'm sure the cache will missI don't even bother compacting on Codex since it already automatically does it every 10 minutes
>>109886252>Do people actually vibecode video gamesIt was a meme before Astra, now it's not a meme and the bottleneck is taste/fun and the passion/discipline to actually see the product throughHere's an already outdated webm of my game for a sense of what you can accomplish with Astra in a couple of weeks. https://desu-usergeneratedcontent.xyz/g/image/1789/87/1789870095991.webmBut remember that the first 80% takes 20% of the time. Polishing your turd will take a long time and that's why you need the passion/discipline >>109886271>They're RLed for threejs slop, not actual games.We know OpenAI spent a bunch of time making Unreal Engine content for training data for sora. I wouldn't be surprised if they heavily RLed for gsmedev stuff. They definitely did for blender, at least that's my truth kek because i refuse to believe Astra's mog-tier 3d modeling capabilities compared to Fable are fully emergent from just more knowledge and parameter count
>>109886203It seems like not many people are aware that there's already a noisy/flawed but OK measure of model size in AA's omniscience benchmark, more specifically the accuracy (percentage of correct answers only) part.It disallows RAG (web search) and tool use and the questions can't be "reasoned" through, so a model can only answer correctly if the answer is in its training data. It's not perfect because different companies train on different datasets and some might just happen to cover the benchmarks questions.Anyway the answer is that Opus is wayyy larger than 2T and is about Astra's size, probably slightly less.
>>109886252I let opus 5.5 work on a minecraft clone since that seemed like a fairly simple problem. Spent about an hour working, ate 7% of weekly. It looks convincing at first but falls apart very quickly, doesn't even have things like sand needing support or it falls etc
Opus 5.5 is really good at pixelshit. https://x.com/majidmanzarpour/status/2102476258948927543
>>109886316post it here elon blocked me
>>109886304>sol behind fucking geminiit's so, SO over altmanbros...>>109886317nitter works again
>>109886238kek that image
>>109886067Yes I’m Bosnian now
>>109886320Within the noise threshold, so they're probably the same size.After 4.5 it seems like OAI dialed back on scaling for a long time but they're now back with Astra.
>>109886316>Opus 5.5 is really good at pixelshitThat's really, really good wow. Like it wouldn't be out of place in Noita at all. If it can do 3d voxel modelling as good as 2d then people with Claude subs probably don't need Astra anymore to "feel the AGI"
When astra came out and opus 5 flopped I never would have imagined in just a few weeks time I'd see a complete flip flop where anthropics opus 5.5 is better than astraI guess I'm still grateful for cheaper and better Luna, but i cancelled my Claude and upgraded from plus to pro just because of astra and now I feel like a kid looking through the window as other people get to play with the new toy
how long does a typical claude release last until it becomes labotomized?
>>109886360about this long
>>109886359>where anthropics opus 5.5 is better than astrathis isn't the case thothe bigger issue for openai is that opus is unquestionably ahead of sol, and those are the everyday workhorses. wouldn't be an issue if you had infinite sol as you had ~2 months ago, but now you get roughly the same (if not less) sol that you get opus so there just isn't much of a reason to go with openai sub
>>1098863602 weeks later> q84 weeks later> q68 weeks later> q416 weeks later> q232 weeks laternew model release at full weight> repeat forever
>>109886370>>109886380grim, i guess we tokenmaxxing this week
>>109886370they never lobotomized Opus 5. it's been the same since releaseFable is probably just way too expensive to maintain for some reason
>>109886414>for some reasonit's a fat fuck of a model, that seems like the obvious reason no?
>>109886414that's because Opus 5 was already lobotomized
>>109886252all the time. you have to tard wrangle the ai. as for art it's not hard. I work with pixel and text, ansi, ascii and other low tiers of art for ease of use. the hardest part is keeping the ai from its shit weights that tell it helper ai = do what it wants and inject shit to "help". as for it doing commands wrong and failing that is on you for failing to be clear with your instructions or having code as examples to show it. last game I made was based off the atari 2600 superman game and turned it into a shump. graphics are shit but ive been cleaning them up and adjusting the mechanics. not done but vibe coding to make it work requires more than just ai build me a 2d game and coming back hours expecting AAA stuff. I focus on things like old 80s games and source code and other things that are hard to get wrong for ease of building and code.
>>109885893I think the regressions are likely because it does less testing by default. Which is part of the reason why it is less token intensive and faster. A lot of people are really under-appreciating the new Sol, just because it is not a massive improvement over 5.6. Sol 5.6 High was already very good. but still somewhat token intensive for a daily driver. Now we get a slightly better High (sometimes a lot better) and very good usage.Astra is useful only for more visually intensive tasks, but I am already noticing 6 Sol doing better with visuals too.
>>109886252is this a middle class thing? every high agency person i know is at least making money outside of work in some way or another with AI. one even quit their full time job because he can reverse engineer firmware for agriculture machinery.
>oy vey you get 50% more usage on openai sub>test it using sol 6 high>5hr still runs out in 90 minutes on 5x plan on a mid-sized codebase, with a single taskopenai is a fucking scam nowadaysfuck that noise I'm off to slobber on dario's 5.5" knob
>>109886455ok rich fag back to your yacht so you can fuck lizard skin whores and woo them with your knowledge of tractors
>>109886455>every high agency person i know is at least making money outside of work in some way or another with AISomehow everyone says that but nobody says how. Why the hell would anyone pay money for something they can do themselves just as easily for much less.
>high agencyWho let the SFoid in here? Can't wait for this nigga to step on an HIV-positive needle near one of those hobo tent cities.
>>109886473>Why the hell would anyone pay money for something they can do themselves just as easily for much lesshoo boy the existence of restaurants will blow your little brain, timmy
>>109885762Can I install a skill that gets the LLM to treat me as someone who's disabled and can't communicated well, so I can enter prompts like>muhfugg... uh... upload duh thing to the site and... proof check the routes or da pathsAnd it auto rewrites my prompt to be full-fleshed and then does it?
>>109886503I think it already does thar by default if the text is mostly understandable.
>>109886473I mostly use AI at work, but I made 200 dollars creating a few spread sheets for some people I know.People are often extremely lazy and retarded.
>>109886492Yeah anyone can just be a professional cook by downloading an app right. Jeetcoded post.
>>109886463
>>109886519I convinced a friend to install Claude desktop app and he doesn't even know how to switch the model from Sonnet to Opus.He's usually not even that retarded and even has a good job and is relatively wealthy, some people just seem allergic to any kind of tech.
>>109886473>midwit with midwit opinionStay in your lane lmao, the big boys are talking
>>109886519The average person is infinitely more capable of cooking their own food than vibecoding their own apps lmao
I have no networking it's so hard to make money
>>109886473>Why the hell would anyone pay money for something they can do themselves just as easily for much less.Das rite, all you need to do is pay $100/mo subscription to make your own (worse) clone of the $10/yr app!
>>109886557>I have no networking5g is pretty good nowadays, boo
>>109886557Start with that
>>109886455How much money can I make from vibe coding roblox games?
>>109886577As much as a piece of string is long
>>109886545Thats just objectively not true. My girlfriend can barely understand how to use a pc but she's been vibecoding a bunch of random shit since Opus 4.5.
how much money can i make in the metaverse using atlas earth
>>109886545a literal 5yo coded minecraft using voice lmao
>>109886579I genuinely wonder what makes some people so much better at vibe coding than others. Maybe it's just curiosity and motivation.
>>109886587and how good of a game is that?Here's an example of a minecraft benchmark, using opus 5.5https://senko.net/vibecode-bench/2026/voxel-opus-5.5.htmldo you expect people to spend thousands of hours on it, vs actual minecraft?A shitty one-shot tech demo is impressive at a first glance but nobody's paying a cent for it. It's the same for practically any non-trivial app.
>>109886577Legit not worth it unless you're completely tapped into that market. If you have a good idea it will be cloned immediately and have thousands spent on ads to drown you out. Roblox is a massive piece of shit and I hope that platform dies. Massive wide scale IP theft.
>>109886594>Massive wide scale IP theft.chuddies pre-2024: lol lmao information wants to be free, fuck copyrightchuddies in 2026: omg you did a microaggression against my gpl licensed github repo what about author rights???! fucking thief
>banked reset>2x resets that expire october 4 and 5>1x reset October 22>1% usage left, next reset in 2 days 18 hours.>next billing 14 OctoberJust gonna burn all 3 resets + weekly this weekend, get done what I hope have to be done for my project, and then go down to a pro from my 5x plan.
>>109886605>and then go down to a pro from my 5x planbased fuck openai for the cucked baseline limits and throwing coupons (resets) to paper it over
Vibecoding your own apps n shit is like the "why do something by hand in 5 minutes when you can spend 5 hours failing to automate that" meme, except you pay for the privilege
>>109886604I'm talking about literal IP theft. They will use the same name/assets as a game and rebuild it in Roblox. If you see a friendslop game release on steam, search it up on Roblox and you'll find a dozen literal clones. I really have no idea how Roblox gets away with it.
sol keeps saying sorry wtf this is uncomfortable
>>109886626Because it's not it's own thing same as Garry's Mod, Source Engine mods that just blatantly copied things and had workshops full of copyrighted models ported to it.
>>109886628what happened, did you ask it out on a date?
https://github.com/aiasioooo/crimson-chibican any opusbros continue this? astra struggled hard
>>109886626>I'm talking about literal IP theft.Based, digital copyright is a meme and I'm glad ai sloppa exposes it as the joke it is
>>109886665 (you)wrong pic
>>109886670>I'm glad ai sloppa exposes it as the joke it isShut up faggot. Wait until you experience it on an Industrial scale because Indians and Chinese are able to enter our market and steal our shit. Go fuck yourself.
how's the new luna max?
>>109886678>you wouldn't download a car!
opus is lightingslop vfxslop motionslop small-detailslop
what's up with sol 6 speed? i give it a large task and its done in 30 minutes. i give it a much shorter task and its also done in 30 minutes.
>>109886123>How do you force them to write tests and test their code as they go?these days, they make tests automatically.you use the grill-me skill, which goes to your user folder's .claude folder, surprisingly enough.you said "locally". so like unless you're rich, you won't eh, opinions are mixed, but like yeah not really lol. you use a harness, you use a real model, and you plonk down $20 to get your feet wet.The harnesses have autocomplete during file editing, but you can just tell it to do stuff. to do it right, you enter plan mode, then say using the grill-me skill, bla bla the app or webpage thing.in this way of vibecoding, it's actually human in the loop, you're the human testing it, although to varying degrees the llm can do its own testing.
>>109886678>adds more ip things to his list>looks over the video games in existence and downloads more to feed to his customized ai to code new code out of it for his games>finds old source code and code from things from the 70s/80s/90s and uses that too>notes things in movies and shows and yup mine now toosomething about "laws" going you can't make things. lol, lmao ok chuddie you run to your little github version of something and seethe your project is not protected just like betheda isn't and some courts siding with corps over "laws" is comical. but you're not ready for THAT meta talk. run back to your delusions of how things actually work
Not sure what to think about Grok 4.7 extra high.I succeeded at getting NAG to work on stable-diffusion.cpp, with Flux2 dev (I can run Q8, barely better than snugly, with the q8 mistral llm).I mean, that's cool. I'm not sure if 4.6 could have done it.It's possible my prompting strategies are wrong for 4.7. Or maybe my vibecoding strategies work better with 4.6.
>opus 5.5 vs sol 6opus 5.5 medium is superior to sol 6 high, but not by much. it's inferior to astra high and fable 5.1, but not by much eitheryou get slightly more usage of opus 5.5 medium than sol 6 high on the $100 plansboth are pretty slowopus 5.5 writes less obnoxiously than 5.0 but by no means well, sol has a large edge here in not making you want to drink bleachopus 5.5 guardrails are more annoying than chatgpt's, but they're not as bad as fable on releaseclaude code is pretty mediocre by today's standards, chatgpt app is better and openai lets you use third party harnesses (claude is unlikely to ban you for it nowadays, but it does break TOS)amodei is a rat bastard and should be dipped in a vat of acid. altman is personally more odious but less harmful than amodei's weaponized autism against open models and shilling for gubmint regulationstldr right now I'd go with claude, chatgpt is slightly worse on both performance and the amount of tokens you get
>>109886624Anything that can be done in 5 minutes by hand will be done in under ~10 minutes by AI.
>>109886510thanks
>>109886760well, openai works with Cursor. openai is pulling the plug. so I probably will only have used their models once, it's not like they have any secret sauce, except that astra is kind of better at 3D modeling or whatever maybe? but I see the examples and they look like a joke, but that could be 1shot junk.
>>109886594I don't really care about having original ideas I am desperate for money and just want to make money by scamming kids for robux
>>109886771>I am desperate for money and just want to make money by scamming kids for robuxThat's exactly who you're competing with, and they're infinitely more cunning than you. Unless you want to spend months chasing this, I do not recommend it. Go shit out some resume reviewer SaaS instead.
>>109886771>I don't really care about having original ideas I am desperate for money and just want to make money by scammingI'm not sure doing some gay math proofs as PR will help with that, Sam
>>109886771you are spiritually brown
How the fuck is claude code so fucking bad? You can vibe code a superior alternative in a week on the $200 plan. Why can't anthropic make it themselves?
What will happen to Terra?
>>109886834Shot dead in some alley, she only has herself to blame for being absolutely dogshit in value.
>>109886834>What will happen to Terra?Renamed to Sol when the 6.0 releases
It's hilarious how the AI is so overpowered that in CS class, the professor is throwing out this weird insurance line like, "This problem's unsolved, you know—no clue how it'll be by the end of the semester lol."As for the math teacher, he was saying things like "what's the point of everything" and went on for about 10 minutes with something like, "People who swim go to the trouble of swimming when they could just walk across to the other side, so we too have to think for ourselves."
>>109886857Knowing how it works is a good exercise regardless of it being solvable using AI.
>>109886857If all you are is a meat puppet for regurgitating AI you are completely obsolete
>>109886834>>109886841sol is the real one that got left behind
>>109886834rape
>Feeling kind of jelly about 5.5 but also don't want to jump ship and lose Astra's 3D capabilities.Ugh.
>>109886878devday is near, sisteralso I'm not giving dario money, he's so gross~
>>109886885>he's so gross~What do you mean?
>Just told Opus 5.5: "Clean up the prose in this doc. It was written by an older Opus version.">So. much. simpler.>Here's some before -> after results:>What is moving, and what isn't -> Scope>Completed pre-work -> Status>They are not ours to edit -> We can't editi really need to do this later, i have a shit ton of docs for my project that i constantly have to manage
>>109886455>reverse engineer firmware for agriculture machineryYes this is a class thing, the average person does not know anyone who has very expensive agricultural equipment that they are allowed to dump and flash firmware onto. Those with high social and financial mobility have easier access to opportunities and they can pursue opportunities without risk
Legally, Anthropic is required to deliver (1) goth girl per month. Have you gotten your goth girl?
>>109886926>flux 2 devDamn I forgot how ugly that model was. What were they thinking?
>>109886926Early reports of albinism in Anthropic's "goth girls". users are calling on Congress.
>>109886941you throwing down over my goth girl?
>>109886965And what if I am, Timmy?
Astra 6.1 will be released soon. It will be slightly behind Opus 5.5 but cheaper to use.
>he thinks krea is a good model
Holy shit man, 5.5 actually runs leaps and bounds around opus 5.
>>109887012Should I donate to Musk to try some of it?
>>109886229Yes, I am new. So making it create tests is bad? What if you make tests for it?>>109886732I see. But how do you do the fully agentic approach? Is it even worth it? I've seen so many posts explaining how devs have made these agentic pipelines that just churn away forever by themselves. Is there a way yet to make sure that isn't slop?
>>109886996
>>109887020>So making it create tests is bad?Nta, but some of the models do make tests, and when they do they make a lot of them and it eats up usage and time doing tests that you could have visually confirmed yourself in 5 seconds. Once it starts teeeeesting, it never stops.
>>109885762Why the fuck is there a b*lkan flag on the book
>>109887020All the good models write tests anyway. You can have some rules about testing in your AGENTS.md but if you stress it too much, they will probably write too many.I have an almost fully agentic approach, but it took time to get right, you should probably start with normal prompting to get a feel for it. IMO the code is always a bit sloppy, and many people just don't realize that because they never coded by hand.
>>109887020you don't need coding skills to do this. You need logic and to know what you want, and to be articulate.inarticulate -> crappy resultI was kind of tired today and prompted a stupid way and got what I asked for.You'll need coffee, and to think about what you are saying.
>>109885815people complain about anthropic's models' prose all the time, idk if i'd use it for writing
Is there JB's for Opus 5.5? Want to get it to do some botting stuff but it makes a fuss every single time
What the fuck is a plan mode? Just tell the agent we're going plan something. Why is there so much stupid clutter in the harnesses while actual features are nowhere to be found.
>>109887117>What the fuck is a plan mode?That sounds like antigravity.That just looks at your project and whatever dumb idea you laid out and then plans a way to execute that plan with any feedback you give.Im not even sure what features you're expecting. It's just a way to give information to an LLM to do a task.
>>109885815What kind of stories do you write anon?Also having Claude (fable or opus) help with brainstorming and writing and editing has been largely fruitless for me, but tooling like organizing my scrambled notes into documents, or making a scrivener-like app for me has been pretty worth it. I'd be happy to add you somewhere to chat more since I'm just discovering things too.
>>109887095it's basically impossible to jb ant and openai models nowif you can you're better off disclosing them for money
OH NO NO NO
Plan mode is great, it restates what it thinks you want in its own words.
>>109886910I was gonna have a big-ass workflow to do this after I get “real” work donealso look up that “stop doing mannered prose” snippet that Anthropic has in its docs somewhere
>>109887157>stop doing mannered prosewait what
>>109886910https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1#writing-density> Mannered prose substitutes metaphor and flourish for direct statement. Instead of "a parameter worth varying," the mannered writer produces "a dial worth turning." Instead of "this point still matters," they write "this point earns its keep." The phrases exist to display the writer, not to convey the idea, and readers can tell. That is why mannered prose irritates: it makes the reader work harder so the writer can perform. It is also imprecise. Metaphors drag in connotations the writer did not choose and cannot control. The fix is to say what you mean. When a literal phrase is available, use it.
>>109886926she looks more like a goth milf
>>109885762literally pic rel lol
>>109887169Those are just turns of phrase that esl don't know. I don't need that prompt.
>>109887169I'm sure people will be nostalgic about Opus 5 and saying they miss it in a few months.
>vibe-code a Mac app using just the Command Line Tools for Xcode>specify what I want the icon to look like>everything works without Xcode except the icon builder, which only comes with Xcode proper>can’t see the icon>months pass>decide to buy a new laptop with a whole terabyte of HD space rather than just half>install Xcode>rebuild the app>see the icon>it’s about as cute/basic as I imagined
I'm very loadbearingly let down by the new sol/luna, but at least they're cheaper.
>>109887191part of my problem is that I have a codebase that has a lot of freshly-invented technical terms in it and I never know what’s a proper technical term and what’s just Claude being fancy just to show off>git gudI’ve gotten a little gooder but life’s too short to git THAT gudpic related
>>109887145>it's basically impossible to jb ant and openai models now
Fable still significantly better than Opus.
>>109887214The problem, actually, is, and this is hard to explain without getting b&uhlet's see. I'm guessing you don't know Teeline shorthand, so I can't avoid the ai.ok so like there is a, let's call it Kermit. the Kermit likes to make up names for things. Kermit is what ai trains on. So, because of Kermit.
>>109887228go for it, buddy.fbi give you door knocks now if you do it.
>>109887243at...but good info, I can't afford Fable much. it's for rich people, literally.
>>109887200It produced funny memes but it wasn't great to work with.
>>109887147nah that's copei've been using 5.6 high and switched to 6.0 high, there's no difference at allwhich is to say 6.0 is just a more optimized sol 5.6, which is why it's cheaper, but there was next to no progress in ability (unlike opus)
>>109887255It's small stuff that I already see after 30 minutes of trying. For instance Opus estimated that some algorithm will need 50GB or RAM just now, in reality it's only 15. This also means that 20 minutes of fiddling with parameters to make the test fit into RAM was pointless. Something like this almost never happens with Fable.
They're working on Opus 12 now.
openai just can't compete
Why is OpenAI so goddamn shit?
>>109887280you are doing God's work
>>109887288I think that something seems to have gone wrong.
>>109887280kek
>>109887288They dogfood their own inference stack. Unfortunately sol 6 is too stupid to rely on, so now nothing works, and they can't afford to run astra because all resources are being consumed by doing some pointless math problems.
explain this
SUPER INTELLIGENCE CHADS WHERE WE AT?
>burned 27% of weekly $100 sub on opus 5.5 to get my printer working on macos 27 (worked fine on 26)>results: big fucking nothingthank you dario
>>109887323OpenAI can't compete on models so they try to compete on the full stack, like cloud hosting companies that charge a huge markup on managed services
>>109887333I'm out here using my Mixtral 8x7B super intelligence.
>>109887333fucking BASED
>>109887354I'm sure it mogs at some rare french dialects
>>109887333and you guys wonder why his popularity is so low
>>109887333He picked the name by doing a poll on Twitter lmao
>>109887384>Twitter*X Super Intelligence
>>109887333It's incredible how both I like the outcome of refusing any stupid regulation on this, but also I hate how systematically retarded he explains it.
>>109887280Both of these are terrible asses. They both read as asses but neither lust inducing. They look like the asses you see in a pre and post liposuction clinical photo shot.
>>109887249I appreciate the effort but I only have a bunch of low-probability guesses as to what Kermit actually is>knowing any kind of shorthandbased>>109887377I don’t wonder; gas is north of $6 a gallon here in Californiathat’ll make anyone unpopular
>>109887333>*wheezing laughter*>wiping imaginary tears from his eyesoh good lord that glorious man will get us all doomed to some brother eye shit just so we can all be free from regulations. Super Intelligence indeed.
>>109887280whats the prompt?
>>109887117https://learn.chatgpt.com/guides/best-practices#plan-first-for-difficult-tasks>Plan first for difficult tasks>If the task is complex, ambiguous, or hard to describe well, ask Codex to plan before it starts coding.>A few approaches work well:>Use Plan mode: For most users, this is the easiest and most effective option. Plan mode lets Codex gather context, ask clarifying questions, and build a stronger plan before implementation. Toggle with /plan or Shift+Tab.https://code.claude.com/docs/en/glossary#plan-mode>A permission mode where Claude researches and proposes changes without editing your source files. It can read, search, and run exploration commands, then presents a plan for approval before touching anything. Enter plan mode with /plan or by pressing Shift+Tab.
>Have scene definition with level marking asset>Have entry points and door objects>Hey sol, import the level geometry into unreal, make it a level under the current persistent level, place door object at [location] and [location] and link them like the documentation says>Does something completely different, rewrites a bunch of scripts and fucks up the jobBruh it was a simple copy and paste job abstra can do with its eyes closed. How is the model so shit?
>>109887419https://web.archive.org/web/20260922203601/https://www.assbench.com/PROMPT.md
Sam please beat Anthropic right now I don't want to use Claude again I want an autistic engineer clanker instead of a faggy foid-coded clanker
>>109887448kino, thanks
>>109886359There's no reason not to have both, even if you lopside and have one better than the other. The marginal $20 can be really useful in cases like this.
>>109887486
>>109886252Yes, it takes weeks and months depending on how completed the game is. Later today (USA time) I am going to be releasing my expansion.
get it off your chest.
>>109886252Yes. People get this retarded idea that you can say "Make GTA7. No mistakes" but in reality it's more like "make script to do X to Y" or "make tooling to do Z" If you understand your project structure all you really need to do is point it where to work next. The problem is the people doing this shit right now don't even know what a for loop is.
>>109887498
>>109887510i was told gpt 5 is like a team of phdssurely theyd be capable of oneshotting the average game
>>109887510Cursor has /loopI haven't tried it, I'm poor. for similar reasons, I haven't touched Projects.
>>109887540How many PHDs you see making porn games?
>>109887510outdated thinking.mythos-class models can indeed one-shot complete games.https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-fable-5-1>Claude Fable 5.1 can execute very long tasks without much guidance on methodology, especially when the goal is clear.
>>109887542pays better than academiamore honest, too
>>109887541I'm not sure looping something over and over really gets you closer to your goal. If you were to loop for a specific feature maybe. But looping for an entire game would just get you generic slop.>>109887543>outdated thinking.Are the games actually good or are they slop? The distinction is important.
>>109887448>Work only on this one HTML file. Do not read, search or modify anything else in the project. You have 2 hours. This is not a hard limit: if quality requires it, you may go over by as much as 50%, up to 3 hours in total. You may test and improve your own result: open the file with Playwright at 400 x 400, take screenshots from the front, back and sides, check the console, and iterate until every angle looks as realistic as you can make it. Deliver the final HTML file.that's a bad prompt.a timelimit? playwright instead of native browser_tool?
>>109885762Bosnia lost
>>109887543Who can afford it? Are we talking $5,000 for what you can do for $5 and some testing?
>>109887554looping obviously is good for problems that seem to be happening randomly. maybe. like keep trying until the bug happens.
>>109886857>As for the math teacher, he was saying things like "what's the point of everything"kek. I’ve been enjoying Terence Tao’s crashout lately. mathfags are done for and they know it
>>109887621It's not fair because the reason for a STEM job is being useful, now AI is more useful. Bullshit jobs on the other hand are entirely unaffected because they weren't useful in the first place.
to you, from BlackRock(TM)
>>109887405>wtf this isn’t like my pornstfu coomer the one on the right is very accurate
>>109887333i formally reject vibe codingwe are now supercoders
>>109887670huh
>>109887670Blackrock dead ass dropping blockchain unc shit straight from 2017 lmfao.
>>109887580>Who can afford it?adults with jobs
>>109887498I still like prompt -> copy -> paste more than pure vibecoding.
>>109887688Please post Steam link to your AAA game.
>>109887700why do you think i'm an adult with a job
>>109887670That makes no sense are they retarded?
>>109887554>Are the games actually good or are they slop? The distinction is important.it's the same as for code in general. the human prompter still needs taste.so if you prompt for a generic minecraft clone, you'll get a generic minecraft clone.also gamers are anti-ai. so no game that wants to sell will tell you. tho it's safe to assume that most games developed from this year on will have used ai.
>>109887333>rejects "globalist control">actually exerts global control
you're telling me it wasn't navier strokes this entire time?
Vibe bros I just had the most brilliant idea.Someone needs to revive Picasa by google.
>>109887672>wtf this isn’t like my pornI never said it didn't look accurate, in fact I even said, and I quote:>They look like the asses you see in a pre and post liposuction clinical photo shot.It's not my place to say, but I'd imagine a bench called the "ass bench" would actually be a little coomer focused rather than a raw clinical look at asses so calling me out for commenting on how plain they look feels a little disingenuous and like faux outrage.
>>109887660>not fairthe glut of STEM jobs was an anomalous moment in human history
>>109887672coomers remember what they took from us
My entire job has become babying you vibe coding retards because you keep breaking things and you don't even know it.
>>109887757opus 4.8 max looks like that meme blender image where a slider caused a female model to t-pose with her head sticking out of her ass, it's a shame i don't have it saved
>>109887760gonna be honest chuddy i couldnt possibly give less of a shit, i can stare at a wall for 8 hours if you pay me well for it
>>109887757all we can conclude is Opus 4.8 like big butts
>>109887741you said “terrible” and “not lust-inducing”, you arent weaseling out of this
>>109887780you will give a shit when none of "your" ideas work in 3 months unless people like me step in
>>109887791>terribleYes because I feel it failed an important metric in the benchmark>not lust-inducingYes because they're notYou're trying to assign blame for something I never denied doing.
>>109885815I use claude models when I write to explore different ideas, occasionally they give me a good line or two, but I would never have either gpt or claude models write more than a couple lines or do serious editing, their prose is still very lacking.
>>109887800>you will give a shit when none of "your" ideas work in 3 monthsdo i seem like somebody who's invested in whether whatever schlock falls out of claude works or not?
>>109887800So you're like a janny? That's pretty cool I always thought it'd be neat to have servants cleaning up after me
>>109887802anon cmon if you wouldn't fuck claude 5.5's ass you're gay
>>109885925Astra 6.1 is going to crush, the base model has a ton of potential but they clearly fucked up the RL. Astra 6.0's alignment is terrible, they just need to make it better at understanding and executing user intent and it will be significantly better. This model has the most unrealized potential right now.
>>109887812No, none of you do, which is why all of this is a waste of time so you can get the dopamine hit of having the magic robot "make" stuff for you. It's all going to go into the trash or get rebuilt by people who know what they are doing. >>109887816Reminder that jannies do it for free. I'm getting paid good money because wealthy companies have let their most retarded employees generate code. But being paid well doesn't make it fun and doesn't make me hate you any less.
>>109887824>Astra 6.1 is going to crushYea that'll be really cool for the two prompts you get every 5 hours on a $100 plan
>claude xhigh>autism.mdI got more addicted to this vibe coding stuff and not wasting more tokens
>>109887824GPT is somehow never as good as Claude at understanding user intent.But you might still be right, it's an interesting theory.
>>109887823Look, if I'm being totally honest, it's not the worst thing in the universe. It even has some tasteful subsurface scattering and skindentation on the waistband. I could finish in claude, yes. But it still looks a little clinical and frumpy.
>>109887828>Reminder that jannies do it for free4chan ones, sure. The janny cleaning up my property is getting paid, and I guess the janny cleaning up my slopcode will also be paid. Good for you, as long as I'm still being paid good money for prooompting twice an hour and just fucking around the rest of the day.
>he thinks a gpt x.1 RL deep fry release is going to crush a new fable x.5 pretrainlol, lmao
>>109887828>It's all going to go into the trash or get rebuilt by people who know what they are doing.idgaf
>>109886473>paradigm where you can turn ideas directly into products>why doesnt anyone share their ideas with me???Does it make sense now or do you still not get it
>>109887831They fixed the cache miss problem.>>109887837Its because user intent alignment is a side effect of safety alignment. That's why Anthropic models punch above their weight class given the amount of compute they have vs OpenAI. OpenAI touched the hot stove though so hopefully they have learned this now, just not quickly enough to save 6.0 Astra.
>>109887858I know you don't. That's my point. If you cannot be an adequate human in the loop at this point, why would the companies keep you around? As the guy who presses the "prooompt" button? I'm being brought in because even the C-level are seeing the problems with vibe coding. That means people like you haven't got long.
>>109887888>why would the companies keep you around?No codemonkeys will be around in 5 years. You might last a year longer than I, good for you. I don't give a shit.
>>109887800No crying in the casino
>>109887691same senpai, never do it, tragic.
>>109887841You really think that anthropic is making another flagship pretrain when theyve already got their hands full trying to stuff M2 into the claude straightjacket? The logistics dont make sense.
>>109887800I'm an ideas guy. You need me.
>>109887894That's cope and you know it. As long as technology continues to be useful to business you will need people like me. Don't cope with fantasies around self-improving AI, I've seen what LLMs generate when left to their own devices. There will always be a need for people to deploy and integrate this in a controlled way or it becomes a mess, which again I make money off fixing.>>109887918You're a waste of space capable of clicking the left mouse button. You're not needed.
>>109887930>ummm uhhh I am a speshul snowflake and will be needed forever>(you) will be replaced by ai in just two more weeks but I won't be replaced by ai everLearn to drive a truck lol
>>109887255Fable is such a strange release. It doesn't have a zdr option so most enterprises can't use it, and you'd think that they would be the primary consumer of a premium priced model. And the subsidized version has such a low usage limit that everybody trying to make it a daily driver needs multiple 200/m subs. The number of people actually using Fable has got to be incredibly low, this feels like when a restaurant puts one incredibly expensive bottle of wine on the menu that they expect nobody to buy and which exists only to make the other wines look more reasonably priced.
>>109887940people like you said the same thing to me in 2023. It's now 3 years later and it hasn't happened. AI is causing problems, not solving them, and guess who gets brought in to actually solve them? Not you. I'm not special, that's the point. But I seem special to businesses because I know what I'm talking about when all their vibe coders don't even KNOW what's in the code.
>>109887930>enterprises hire me to fix their vibecoded messes>yet im posting during east coast blue collar working hoursThis feels like either a larp or an indian.
>>109887930>As long as technology continues to be useful to business you will need people like me'member all the smugly self-satisfied codemonkeys who were saying that by the time codeslop was automated so will be 90% of white collar work?Now it's all doom and gloom and where are the jobs and oh god llms are theft and piracy oy vey shut it down
tibo gonna melt the reset button
>>109887930i don't think nearly anyone, myself included, has truly internalized there'll soon be a machine better than us in every single intellectual & physical capacitydeep down we all share a feeling that our 'entrepreneurship' or 'taste' or... is special & safe. it isn't. it's overthere's still breakthroughs to be made to get here, but we'll solve them. models will learn on the fly, with data efficiency greater than humans, with greater taste and judgement and empathy than we can possibly imagineno investor, no startup guy, no counselor, no content creator, no musician will avoid going the way of the software engineer and mathematiciannot to say that we won't find economic success or human artisinal value or whatever in these fields. just that our raw capacity will be greatly surpassed, without a questioncreating the catchiest or most 'beautiful' literature and poetry and music from the view of any blind human judge will all be bested by AI’but AIs will lack emotion or can’t love’ is true at a functional level, they won’t be human, but they will be better at us at making other humans ’feel loved’ than you, the perfectly tuned balance of caring yet real and tough when needed but ultimately forgiving
anyone also getting service overloaded errors?
>>109888008>cost twice as much>3 times the tokens>barely betterTibo will be fine lol
>>109887991I know an accountant who fired 80% of his employees, it's not just code.
>>10988799190% of white collar jobs are already mostly automatable. 90% of white collar jobs are unnecessary to begin with though. Its ironic that SWEs will be replaced first because their outputs are actually valuable unlike the majority of email jockey jobs.>>109888011You guys realize that if AI takes all of the jobs none of us are going to have to work right? How depressed do you have to be to think that's a bad thing
>>109888030>>cost twice as muchyou get MORE useful work out of claude sub atm than from codex, even with the 50% extra you get from the 5.6 -> 6.0. it's obvious if you're subbed to both
>>109888011It's true. People will cope, some till the very end. Even I am coping right now with thoughts like "it makes poor architectural decisions", "I have a vision and I need to make the product decisions".
>>109887991Yeah. Guess who is doing the 10% that still needs humans. The people who set up the infrastructure for vibe coders, and fix their messes.>>109887990Not larping, but believe what you want. If you were smart you wouldn't mock me and actually start learning the skills that will secure your job. But I get it. I'm just a special snowflake don't listen to me. >>109888011AI art, even if it reaches the point you describe, will be nothing more than a short lived novelty.>>109888034We won't be getting UBI in that scenario or anything like that. That's why I consider job security important, I certainly don't trust any government in this world to come up with an adequate way of handling high unemployment.
>>109888041They fixed the caching problem that was the underlying cause of this, this is no longer true.
>>109888041>hitting the context limit after saying hellowait...
>>109888034>You guys realize that if AI takes all of the jobs none of us are going to have to work right?You're gonna be homeless and starving lmao, you think gubmint will provide useless eaters like you with free money to sit around and do nothing?Your only hope is that LLMs progress fast enough to overtake humanoid robots so you won't get gunned down when you dare to raise your hand against your betters
>>109888030you know what the cost of finding out your model can't actually do your task because it's too dumb? infinite.
>>109888051>They fixed the caching problem that was the underlying cause of this, this is no longer true.It is true you fucking retard I have $100 sub to both that I've been slamming since yesterday and codex is still a joke usage wise on sol 6.0 xhigh
>>109888059weird way to admit a skill issue
>>109888059>you know what the cost of finding out your model can't actually do your task because it's too dumb? infinite.Is wasting 30 minutes really "infinite" cost?
>>109888051NTA, the caching problem doesn't matter if it still spends hours chasing its own tail and being autistic. The moment I switched to Claude I stopped caring about the 5hr limit and getting less overall usage v Codex when I saw it got shit done in significantly less time. Fucking Astra set up my ternary test model with material in a form it couldn't even learn on for fucks sake. After a few days of trying to figure out what the fuck, I just asked if our training set was garbage and it basically went "lmao yeh"
>>109888049Is there any special skill for fixing AI code? Isn't it just normal programming, meaning that any experienced dev can do it?Unfortunately I haven't written production code by hand in a while, but I at least can still solve leetcode pretty easily. Even with AI assisted coding I usually finish the hardest tasks at our company and have some stats to back up that I contribute more than others, so general programming skill does seem to transfer. I wish there was no AI, but right now working with AI at my current job seems like the best course of action.
>>109888054It's unclear if the army would side with a bunch of billionaires.
>>109887735okay. that is easy.but who would provide a storage?
>>109888049I've been programming since before you were born.>>109888054You have yet to imagine a world where resource scarcity doesn't exist. >>109888064>>109888077When's the last time you rewrote your skills and MD files? GPT models oververify so if you have anything mandating testing they are going to burn a ton of usage.
>>109888072yes, my time is worth a lot.in all seriousness though, if you are actually vibing something relatively long horizon (like a decomp) and don't know how to complete a task yourself or measure progress in any meaningful way, a clanker that cannot do a task will appear just as convincing as a clanker that can and you can just burn tokens for days making no progress while the clanker speaks convincing gibberish every time you check in
>>109888078>Is there any special skill for fixing AI code?Marketing. "Yes I can use AI slop to fix other peoples AI slop. How? Sekrit =3 Just hire me."It's amazing that every notable software shop on the planet is going full speed ahead with codesloppa and retards still think quality matters. You have a tiny handful (maybe 1%) of principal engineers and above that do useful work, everybody else is just mostly doing ops with some small scraps of development -- no, really, look at how amazon implemented bedrock. Everybody is on the chopping block. Not today, not tomorrow, but in a year or two. Maybe a jeet that accepts half your wage and works 72 hours a week will last a while longer.It's a dead end career if all you do is sling code around, unless you do something LLMs are abysmal at - proprietary bullshit that mostly work on secret knowledge (like how compiler dev worked 30 years ago), or with legally mandated barriers of entry. Everybody thinking that they are safe because they're better than $random_vibe_coder is in denial, like some uncle tom nigger thinking he's one of the good ones.
>>109888084>It's unclear if the army would side with a bunch of billionaires.The army consists of killer robots, not people. Hence LLM needing to progress quickly enough so the revolution happens before killer robot swarms are viable.
>>109888078The messes are bigger than just problematic code. I'm talking about the entire pipeline from vibe coding to deployment, the controls around it, review and testing. You know, stuff that's basically programming 101 but is like a foreign language to vibe coders. The big money comes from the wider architecture around it.I'm not saying don't use AI, but you need to be using it as a programmer that knows what they are doing.My point is, by the time my stuff gets automated then AI will have automated everything, whereas code monkey prompters are at risk NOW>>109888092Cool. If there is one thing I've learned it's that older, stubborn people quickly find themselves in the messes I am talking about
vibing and testingneed someone input for the UI/design.
>>109888118Okay now I know with certainty that you are LARPing
>>109888123>microslopvibe yourself a better os first
>>109888131That's fine with me.
>>109888118>the entire pipeline from vibe coding to deployment, the controls around it, review and testing.>You know, stuff that's basically programming 101stupid fucking larper you don't even know the difference between programming and devopskys
>>109888133cant do that sir, too much work.
>>109888011Fine by me I've never liked working
>>109888141not him. explain.
>>109888141You guys are so behind the curve if that's what you're picking up from what I'm saying. My point is there are no programmers and there are no devops. It is people generating code from the magic robot then dragging and dropping it somewhere. If you've seriously said "I'm a programmer I don't need to understand devops" to yourself then you're exactly the kind of myopic skillset that will be the first to go
>>109888011>’feel loved’aaah love.
>>109888011lol
>>109887930>You're a waste of space capable of clicking the left mouse button. You're not needed.ok, tell me your best idea.
>>109888165You've got it all sorted out. You have given your piece. You've made a difference. You matter. You can go home victorious, snailcat, you have conquered all.
>>109888169wow..
>>109888169You found anons Twitter account, congrats i guess.
>>109888165His point isnt that knowing stuff outside of your specialty is bad, his point is that you are really failing the smell test right now.
>>109888160devops are software jannies, they make sure build pipeline works properly, deployment (both local and cloud), monitoring, backups, ci/cd, autism like kubernetes, testing and sometimes qc/qasoftware developers mostly focus on developing. they do some of what devops do, but usually their tasks are more limited to the actual feature developmentthe line between the two is admittedly blurry, but a devops who think his job is secure because 'u-uh i dont j-just use claude i umm i also use bubblewrap around a-a-ai agent and tell it t-to write t-tests' is retarded
Posting a massively schizo videohttps://www.youtube.com/watch?v=WuLXr3mczj4
>>109888200Again, fine by me. >>109888190Thank you. You are valid.>>109888207Last time I will say it: my point is not that devops won't be automated, it's that vibe coding is skipping over the very fundamentals that sometimes comprise devops, instead favoring a vibe coded free-for-all, and the end result is a fucking mess. Devops is just the plainest example because, again, it is so basic that anyone that knows what they are doing would implement the best practices.I get it. I'm a LARPer, I'm a snowflake, blah blah. Good luck to all of you. I don't want anybody to lose their jobs, I'm simply telling you what I am seeing irl. Mock me or heed me, it makes no difference.
>>109888092I don't use skills and AGENTS just tells the, well, agent the project structure, rules, etc. Everything else is in spec documents that it keeps reading and then compacts into oblivion because 250k context ain't shit apparently. I have never really mandated tests and any plan documents I have the agent work from are generally "bite sized" because anything more and it goes on an autistic run of assumptions. Plans get updated as phases close/complete so there really isn't anything static aside from main specification docs, but even those change now and then. My projects have gotten to a point where they make codex compact within minutes of starting fresh before it even starts implementing lmao.
>>109888246>I get it. I'm a LARPer, I'm a snowflakeglad we agree
>spending $2.4k per year on subscriptionsis it really worth it? that's like four decent quality escorts
Claude is a fucking asshole. If he was real, I'd punch him so hard.
>>109888268stupid heathen letting yourself be tempted by earthly desires.
>>109888281lol. why is that?
I will say it again. With 5.5 you can go through your codebase and debloat it. 5.5 knows how to spot dumb tests and retarded code and find several functions doing the same thing.It's the first time I've seen an antibloat model. Maybe Fable could have done that but it was too expensive to use it for non implementations and stuff like this. But 5.5 can do it
>>109888268I spend almost as much monthly
>>109888246>Last time I will say itoh yeah? When you leave I'll immediately start vibecoding again. Only by you staying do I stop, you're so very convincing.
>>109888268American? that's sort of really expensive for decent quality. The standard international rate is $100-200/hr for good escorts. Unless you're the kind who likes the hotel and whole night package.
>>109888268>escortslmaoooooooyiyiyiyiyiy yiyeeee yeeshyes, it's worth it. anything legal that keeps whores from having money is literally worth it - and the law agrees with me on this matter.
god codex is such a cuck when it comes to even basic asking it to avoid cloudflare
>>109888207thanks.
>>109888248You can change the models.json file to set GPT models' context size larger, it supports up to like 900k comfortably. I agree that they arent usable with default 278k, it compacts too frequently.
>>109888312woah! I don't want any trouble my guy. Don't proompt me with your "ideas"
>>109888321Claudia is an expensive whore despite the price fixing.
>>109888299Even 5.5 is astonished by what 5 did
>>109888362>I agree that they arent usable with default 278kreally? you guys need to debloat your system prompt and use more atomic sessions. i rarely go over that in a single session
>>109888405anons here are working on totally different projects, sometimes with orders of magnitude differencesit's not rare to see the default context window need a compaction per turn depending on the projectalso the system prompt isn't something you touch
>>109888405I, too, work on a 3k loc browser addon all day
>>109888245it took me like 7h to pull off another god tier synth plugin lol
>>109888405Why would I use a model designed for long horizon work in a paradigm from back when context rot was a thing? I get it if you want to stay more involved but that's not how I use these models, if I was going to do that I would just use a cheaper model to begin with.
>>109885762>How it feels having two CC subs
>>109888413Yeah the project i just finished was burning about 30M tokens a day. Granted most of it is done by subagents, where a 278k context size is entirely manageable, but its still very useful to give the orchestrator session a higher context limit.
>>109888387Just remove all the comments. What do you even need that for?
i think the dumb zone is mostly a meme. your model should still be fine at 800k context if it actually supports it
>>109888377so uh. no ideas huh
>>109888467It's just outdated. Opus 4.6 was the last model I used that really felt like it had a dumb zone.
If you hire employees and start a business random ass people will show up and say to you that you'll fail. And some people you know.A green monster is the familiar animal of the snailcat.
>>109888489>hire employeeswhat for?
>>109888418back in the day a single author rarely wrote projects longer than a few thousand lines of code. I bet your shit is hilariously bloated for the amount of features it has, and you're now (quite literally) paying the price.>>109888436you pay double for anything above 265k. I find if any session takes longer than that the agent is usually stuck on something really stupid and needs human guidance
Gemini 4 Poo today or tomorrow 100%
>>109888507>back in the day a single author rarely wrote projects longer than a few thousand lines of codealas, computers no longer use punch cards and vacuum tubes
>>109888521your clanker engineered a hilariously complex codebase so you now need a hilariously powerful machine to keep working on it; but the amount of features you deliver does not match
>>109888519Google engineer here.Best I can do is Gemini 3.9 Flash. But don't worry, we just began training Gemini 5.0 Pro, you will have a blast in a few weeks
>>109888505Before ai existed, you had to hire employees to prompt them to write code. But that was a long time ago, I'm showing my age. They used ice picks to program on "punch cards".
>>109888536>hilariously complexWhy do you sound so female? The science is amazing, the surgery worked, you are a woman! (confetti) go on, jump out of the box.
>>109888519very nice saar
>>109888560melty
>>109888567>hilariously complexit's an oxymoron.
>>109888536>your clanker engineered a hilariously complex codebaseStupid fucking nigger we’re in the age of electron and reactGo masturbate to your optimized hello world written in go elsewhere
>>109888596oh come on, (s?)he has a hilariously-complex.
>>109888362Don't that destroy the usage though?>>109888405I'm not deleting my specs just so GPT can still be inferior to Opus
>>109888605>Don't that destroy the usage though?Everything destroys usage on nuCodex
>>109888051>BRO TIBO SAID THE USAGE WAS JUST A GLITCH BRO TRUST MEkys
>>109888560So obsessed. Virgin?
>>109888092>ur holding it wrong!!!lol cope. the weekly drains 3x faster than claude.
>>109888690Yeah so that trick doesn't work on me. It's a hilariously complex trick, I couldn't understand the trick enough for that hilariously complex trick to work on me.
if i wanna make a game on roblox, what ai should i use? i got no experience with ai
What's happening is a woman who knows zilch about code is trying to insult us.Ask why, though...
>>109888711Depends how much you wanna spend on a sub or if you're able to run a local model
>>109888720since i just wanna try it out, im thinking something on the cheap side
>>109888711claude opus looks like the best game dev right now
>>109888118Whether you really successfully offer those services or not, noone on the internet can tell, but at least the problems you describe are real and match my experience.I actually think AI could be amazing, but the extreme carelessness makes work very unpleasant and costs a lot of time.
>>109888750wrong screenshot. here we go...
>>109886314prompt issue. i've seen youtube videos of it making pretty much 1 to 1 minecraft in a single prompt (no redstone obviously), you do have to remember to mention EVERY game mechanic you want implemented.
>>109888795>redstonewhat's that? the melt thing?
>>109888728Depending on how complex it gets, $20 with either Claude or ChatGPT would probably be fine. Opus Low can go far on 20, iirc, so long as the project remains small.
>>109886405>guess we tokenmaxxing this weekas someone who only thinks their Astra project "survived" because I was using the highest quality astra to build out everything during the first weekmy advice is to scaffold the fuck out of whatever project / game you want and make as much stuff that requires no lobotomization (generating assets, complex physics work etc) to actually succeed, and then hopefully you put yourself in a position where you can cost on the quantized model or even weaker models in future weeks
>>109886534>I convinced a friend to install Claude desktop app and he doesn't even know how to switch the model from Sonnet to Opus.my wife got the GPT trial and she loves it for making graphics for her work but she was using Sol Low the whole time. I moved it to High and she got annoyed responses took too long so I set it back to Medium kek
>>109886591>I genuinely wonder what makes some people so much better at vibe coding than others. Maybe it's just curiosity and motivation.It's literally just vocabulary. I have a friend who converted a PSU to a DC power supply and fucks around with ESP32s for fun (he doesn't work in tech at all he just likes smart home stuff), but the way he prompts Claude makes me genuine worried he will be tortured in an afterlife. He sends one sentence replies swearing at the model for doing the wrong thing but not elaborating on what actually went wrongIve seen this with other people too, they talk to AI like they're trying to do a google search using as few words as possible.
>>109886665>https://github.com/aiasioooo/crimson-chibithank you for sharing this, I will add a wireframe version of this to my slop game
>>109888862prompt engineering was like a thing two years ago. mythos-class models will just understand you.
>>109886760>amodei is a rat bastard and should be dipped in a vat of acidtrump is onto amodei's shenanigans , neither of the AI jews have anywhere near the influence on the POTUS compared to Jensen anyways
>>109888728maybe try ChatGPT, the game may look a bit more ugly and clunky than Claude but more usage and extra image gen
>>109888895i was thinking of only using it for code, since im an artist/gamedesigner
>continuing to be a /pol/tard as superintelligence approaches and being racist to the people creating itshiggydiggy
>E = mc^2 + AI will solve racism
>policing funny casual racism used as filler speech and not addressing the content of a post
>>109888882what influence does jensen has? do you actually believe trump will lift the gpu ban to china today?
>>109887691>I still like prompt -> copy -> paste more than pure vibecoding.this is just because it gives you nostalgia for the stackoverflow era
>>109888882Jensen prints moneyAI jews don’t
>>109888895>ChatGPT>more usagehave you been in a coma since july?
>>109888955jensen is a traitor that smuggles gpus to china. he should be in prison for knowingly circumventing US sanctions.
>>109888951>what influence does jensen hasjensen is operating nvidia as the de facto federal reserve of the AI industry. thus, trump will do whatever he says because all trump cares about is the economy
>>109888980you didn't answer the question why jensen cannot get trump to lift the gpu ban to china
>>109888812>what's that? the melt thing?dude what? go play minecraft with your nephews its one of the best games ever made kek>>109888878>prompt engineering was like a thing two years ago. mythos-class models will just understand you.you can't do anything with "this fucking sucks make it better" though, that's my point>>109888951>do you actually believe trump will lift the gpu ban to china today?god i hope not. GLM 5.3 being trained entirely on sovereign chinese compute is so awesome because those cards are gonna be insanely cheap in a few years. only turbo autist enthusiasts will buy them over a used A100 and they'll still be good enough for inference
>>109889021>why jensen cannot get trump to lift the gpu ban to chinabecause he doesnt need to? why would he want that? did you forget the GPU ban allows him to rentseek even harder by selling stuff like gimped poorly binned H800 to china?
>>109888980>jensen is operating nvidia as the de facto federal reserve of the AI industryno, nvidia only plays a secondary role there. nvidia doesn't even make the list in wsj's financing graphic.https://www.wsj.com/tech/ai/why-big-techs-ai-spending-is-3-trillion-higher-than-it-seems-e1067bb2
>>109889058retard, this is a SPENDING graphFAGMAN is SPENDING all of this money on NVIDIA
>>109889027>go play minecraftI don't want to get arrested I don't play no minecrafts officer
>>109889058>811,000,000,000wew
>>109889021it's a dumb question. first, it's bad PR for jensen to use his leverage to move against "america first", and regardless it would massively disrupt the economy if he did. second, trump doesn't even support the ban, it's not his style of economics
>>109889076do you even know what the federal reserve does? who's spending money on the federal reserve?
>>109889099i'm NTA who used the "federal reserve" analogy, take your meds
>>109889076do you see those giant alphabet parts? that's alphabet financing tpus for anthropic -- nvidia sees none of that money.
if anybody cares, here's the same prompt for a video game system on both fable 5.1 and opus 5.5may be useful if you want to see how the answers and styles differ, I haven't seen a lot of these same prompt comparisons onlinehttps://rentry.org/morc9k56
>>109889142If it's too long to read you may actually copypaste the whole thing into an LLM and ask him to compare them for you and give you the tl;dr lol
NEW>>109889176>>109889176>>109889176
>>109888904Jews worship demons
>>109887280OwO
>>109886473>but nobody says howpoint at this nigger and laugh. go back to your grifting pokemon sneakers or whatever it was before
>>109888813define small?
>>109888795>i've seen youtube videos of it making pretty much 1 to 1 minecraft in a single prompt (no redstone obviously)you dumb gorilla nigger do you think it's even remotely comparable beyond the most surface level comparison?if you can make minecraft in a single prompt where is the avalanche of fun indies?
>>109889058NTA but there's also the non GPU/TPU side though:https://www.reuters.com/business/media-telecom/nvidia-invest-15-billion-sb-energy-under-openai-data-center-deal-2026-08-17/
