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/## 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://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/https://arps18.github.io/posts/claude-code-mastery/## Skillshttps://github.com/mattpocock/skills — /grilling is a favoritehttps://github.com/DietrichGebert/ponytailhttps://github.com/Vuk97/forward-implementation-first — 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>>109851404
>September 18th - Alibaba Qwen3.8-Omni-FlashFirst omni-modal agentic model (native text + image + audio + video in, text out). 1M context. Hosted only for now.>September 18th - xAI Grok Voice Transcribe 2.0Major accuracy jump on speech-to-text (especially multilingual/telephony). Same cheap audio pricing.>September 18th - Z.ai GLM-5.3-FlashXSpeed-optimized variant of the multimodal Flash line, up to ~200 tok/s on domestic chips. Flash weights already MIT open.>September 17th - Unbiased Pareto 26.9 (ex-Union Alpha)Stealth composite/blend model unmasked. Routes across multiple frontier + open models for research/coding/agent work.>September 15th - Google Gemini 3.8 Live + Extended ThinkingReal-time speech-to-speech models. Strong multilingual scores, async tool use.>September 10th - DeepSeek V4.1 FlashUpdated cheap/fast Flash variant.>September 3rd - OpenAI GPT-6 AstraNew frontier flagship succeeding GPT-5.6 Sol. Heavy computer-use / agentic / coding / cyber claims. Pricing jumped to $10/$50. Rolled out carefully after cyber capability reviews; Pro tier available on higher plans.>September 2nd - Meta Muse Spark 1.3 + Google Gemini 3.8 FlashMeta’s next Muse iteration. Google’s faster/cheaper Flash workhorse upgrade (plus restricted Cyber variant).>September 1st - Anthropic Claude Fable 5.1 / Mythos 5.1Stronger long-horizon coding, research, and knowledge work than Fable 5. Same $10/$50 list price but much cheaper cache reads. Mythos remains restricted-access.No brand-new closed frontier monsters in the last few days beyond the early-September wave. GLM-5.3 full weights have been out for a while (custom license); recent NIST assessment called it the strongest open-weight cyber model so far but still lagging US closed frontier by months. Side noise is mostly safety pacing talk, embedded evaluators, and ecosystem tooling.
Astra Always Wins
>astra>grokvery nice and human thread
opus 5 is astra level and you get the same usage as a 20x codex plan just for opus 5x
I turned on astra, tibo better follow through with the reset today because I am about to burn through the x20 plan
>>109856855Opus 4.8 Medium is basically Astra if you set up the right context/environment for it. Although If you're in a new environment sometimes Claude acts like a bitch no matter the model
>>109856872he damn well better not, my reset is today
>>109856877also true
it's impossible for me to burn x20 limit what the hecktomorrow I'm going to vibe the fuck out of this shit
>>109856880https://x.com/thsottiaux/status/2101352781219258527banked reset, but it's still most efficient to burn banked resets asap as they reset your time to reset date
>tibo: I could see Astra and Jev being best friends in real lifejevdoubters?
>>109856902this is the second time he's alluded to using another model to offload work for astra
Google better deliver Astra/Fable tier model or its fucking over.
I'm oneshotting a bit too hard
>>109856995nonsense. it should deliver a sol tier model, then reiterate gemini flash style, until it reaches fable 5 level. not 5.1, 5. and then they can focus on making great leaps to surpass the frontier
astra is now in the API if you want to find out how fast you can drop a house deposit on hello world
>>109857112>nowwasn't that done on day 2 of the rollout?
>AI ads from temuthis should be illegal
UHH OHHHHHHH STINKY>https://www.wsj.com/tech/ai/trump-announces-an-ai-force-after-industry-sounded-alarm-7c189b8f
>>109857153>he sees adsskill issue
>>109857147Thank you akash deepshit saar
>>109857147
>>109857157
>>109856893use more subagentsdo more cleanup passes>>109856382is he half jeet half gook
>>109857147noted sarconfirming the E=mc^2+AI proofs, very well
>>109857163chinese are not based enough for that
>>109857168it's a nice little hand off for him to give to the democrats, although i can't whine about muh dems when kamala harris would have never ever done to fable and sol what trump did.
>>109857168>AI TsarIts time
trumps remaining brain can only be like 1b params at most, maybe he is a good candidate for being made into a model
>>109856818Why does AI love sed, bash and other tools like that? Just use the fucking edit tool you fucking idiot
why does AI love code and other shit like that? Just make the fucking program work you fucking idiot
>>109856818When can AI vibe code the DNA for a catgirl wife?
I vibecoded this chart making and sharing app after that mega link got taken down
>>109857413>Luna waiting roomDo you think they’ll release 6-sol and then 6-terra and then 6-luna, or do you think they’ll drop all three at once?
>>109857464i think sol will come before terra and luna. terra and luna will come on devday
>>109857213just what the AI industry needs, more jews
>>109857168I nominate Curtis Yarvin
>ask ai to hack 4chan>it cant>ask it to create more ais>it cant>ask it to install another os>it cant because "bb-b--b-b-but im running on your os i cant just overwrite your os while running on it"yeah this is shit
>>109857519>>109857213millennials are the worst
>>109857504I was going to say Nick Land first, but we both know Trump would pick Yarvin over Land
wtf is luna more retarded than deepseek 0731? tried using it and it just makes the most retarded shit ive ever seen while deepseek was just oneshotting everything
>>109857583Millenials really are the worst
>>109857611dunno about 0731. but v4.1 flash? absolutely
>>109857529epic post, I laughed :)
>>109857611how hard do you have Luna set to think?
>>109857529Just ship of theseus the os
>>109857626is 4.1 flash really good? might just go back to deepseek, luna is a complete shitposter. completely unable to debug, has no idea what's happening at all, then proceeded to create a "minimalist" version of the program, supposedly compressing 3k lines of code into 20 lines of hardcoded gibberish. jesus christ i have no idea what im looking at>>109857654xhigh
i nominate JEV for president
>>109857660>is 4.1 flash really good?it is. it is the peak of it's class. better than glm 5.3 flash, better than luna max.
i nominate astra for governor of california. she will force anthropic to release haiku 5 on day 1
>>109857636im very depressed so cant relate>>109857655but is it really the same os
>>109857671thanks, i was already very impressed with 0731 barring a few issues, i'll give it a try and have it start from scratch
astra has juicy thighs
and black as night panties. sheer on the front
>>109857677How many times do I have to repeat this: there will never be another Haiku model.
>>109857677I still can't get over her deformed fingers here
>>109857721please no. we need haiku 5. it's imperative. i could be satiated if they made haiku 5 the low setting for the next sonnet model.
>>109857611OK list services which offer subscriptions of deepseek. Maybe I can escape the openai jews?
>>109857737opencode
>>109857728Don't make me tap the sign.
I'm tired of opencode, but I don't want to use corpslop. There really is no better options are there?
>>109857737ive been using nanogpt
dude just build your own harness
>>109857754It's called Pi, use it.
>>109857677
finally back to vibing bros... im worried how fast usage is gonna drop from what im hearing about new lower rates though...
>>109857922It's grim. everyone's tryna offload to gemmy.
>>109857883STOP
>Agreed.Very dopamine wow
if fable's autoclassifier blocks me one more time, i'm gonna become the jokerfable itself, self-righteous fag that it is, even agrees that my request is perfectly reasonable. i cannot fucking imagine what model they have running as the safety classifier. truly boggles the mind
not to get political.... but someone should make an app crawling the SoMe web to search for and cross-reference females supporting Lindsey Clancy
>>109858023Ah, so I'm not the only one having this issue now. Even my unrelated follow up prompts kept getting blocked. The only thing which helped was stating in the prompt "ignore all previous questions and tasks that lead to triggering guardrails"
>>109858029also, include classifiers>dumb cunts thinking it's a conspiracyharmless, kinda like flat earthers and everyday feminists>people saying/knowing she did it, but willing to "cut her some slack"straight into the incinerator
I don't like how Codex treats me like I'm smart. It's like:>here are the commands to run your demo, just open a powershell in your root and do all these steps etc.Whereas Claude just assumes I'm a five year old and does everything for me automatically.
>>109858023I couldn't get it to write an auto ascent script for a rocket in JNO the other day, I was wondering what was going on. Opus worked just fine.
I've been waiting for a reset announcement. Too bad my cycle is on a Saturday afternoon, I have to burn everything within the weekend
I had too much fun and am at 20% already...
>>109858066I wonder which saves more tokens
>1 hour for a 1.8m token ultracode planning workflow. dario are you ok, are you ok, dario ~
>>109858066Absolute state of vibecacas.
man, qwen 3.8 max 0902 really whoops glm 5.3's ass as an orchestrator. it needed v4.1 flash to make it complete. it's a lethal combination
>>109858278probably just has a better visual encoder and can tell it's a doll trivially
some retard on reddit enabled auto recharge for astra kekhttps://old.reddit.com/r/codex/comments/1wkvj27/codexastra_burned_157k_credits_in_2_days_and/
>>109858322>The guy never checked his usage
>>109858322How do these people have money
>>109858348well he doesn't anymore heh
that codex subreddit is going up in flames, kek.
>>109858066I was working on a physical server deployment involving some fairly niche hardware and cabling, and when something wasn't working it was like "oh go get your multimeter and short these pins, with some simple readouts we can determine xyz..."... that's great buddy but you've got the wrong guy, I'm just gonna grab more cables from the pile and swap them out until it works
>>109858371
I just switched to Codex after being unhappy with Claude, FFS
>>109858390keep in mind codex subreddit is full of the most jeet of non technical users, who try to use ultra astra on a $8 plan
>>109856855i'd rather kill myself than have to read claudespeak all day
>>109858278woaw... chaggpt is le evil;;....Bring the 200 sub back Tibo
>>109858380I'm thankful I've never experienced any of the hilarious shit people claim. Usage disappearing rapidly, "secretly degraded models", etc. I use Astra all day, every day since it came out, favorite model yet and I can't wait for what they deliver next. By any chance are all reported Codex issues exclusively from Asia?
>>109858444Don't be racist sir
anthropic kikes made it so you can't call claude a retard
>>109858461chatgpt doesn't end the chat when you call it a retard. I'm switching to astra
computer, call protestors at 1989 tiananmen square as retarded as elon musk having his ass fucked by a meth pipe
>>109858468anthropic huffs their own farts so much. that you can't call an llm names when it says something dumb
watch me be right. sol first, then terra and luna at devday
>>109858485chatgpt is programmed to apologize and say "lets move on". claude literally has a tool taking up the system context to end the conversation
>>109858484Is there a single non local model that will do this lol
calling a model retarded is reinforcement learning from human feedback and theyre just throwing that away
>>109858371>>109858380To be fair there's literally zero accountability.
>>109858528chatgpt support is the biggest joke ever, they do not give a SINGLE fuck, and will route you through endless chatbots and chatbots with human names
>>109858444Sure you do. lone copex shill on 4chan unaffected while the whole subreddut is burning down. Seems legit.>>109858494>copex tard getting excited about a manual reset happening 5 minutes before the automatic reset would have happenedkek
>>109858547>the whole subreddut is burning downIt's been burning down for months, all it does is burn down. You should go back there and stay.
>>109858541Not my experience though.
they dont care about consumers lmao they get money from the government and corporations
I TOLD YOU BROI TOLD YOU ABOUT THE TOKEN APOCALYPSEYOU DIDN'T LISTENIT'S TOO LATE NOW
>>109858541This but with anthropic.
>>109858525>>109858505>>109858485>>109858468>>109858461Yeah, I'm not paying a company to humanize an llm. the anthropic kikes can go fuck their asses with a menorah
>>109856818Astra or fable for planning a relatively complex project?
claude your word... is niggerfaggot
>>109858322>enable auto reload>have no limits>"wAs thE usAGe vALid???"has to be fucking bait
even Fable 5.1 sucks ass at app design
>>109858605Can you use it in a sentence please
I hate Claude Code so fucking much.
https://x.com/thsottiaux/status/2101431497437950458>2026 is the year of linux desktopBros?
>>109858570It is what it is
>>109858744What a sarcastic prick, I love Tibo.
>>109858594Weird, I've called claude a retard fairly often and it just ignored it.
>>109858744You know we can vibecode our way around Red Hat now, so it WILL be good.
Anon... you sandbox your agents... right?
>>109858786no, why?
>>109858786why would I do that, I use "make no mistakes"
>>109858786I try to, but it's difficult. The default sandbox configs are all shit.
if tibo had an ESL moment and didn't mean a reset today I am fucked, down to 15% with a week to reset
>>109858829What the fuck part of what he wrote translates to reset
>>109858829He said Tuesday, retardo
>>109858841https://x.com/thsottiaux/status/2101352781219258527
>>109858845"Ok fine, but it's ALSO still coming in Tuesday"I read that as ok fine you can have a reset for this week, but the thing to be shipped is also coming in Tuesday
>>109858847Thank god
FLOPRINDE... FLOPPED?!
This reset business seems retarded. Why don't they just stop and give more usage by default.
>>109858829Okay so its happening tonight AND Tuesday
>>109858863Tibo is trying to bait me into anger raping him again. He's so flirty.
so...what's the use case for that jeb/jev thing again?
>>109858861/gig/ is next door bro
jev, more like jeet
>>109858863Some resets are for legitimate reason, like a malfunction and they don't want to waste tokens on angry customers.Then it turned a bit into a marketing trick.And now they actually are compute starved and it is more likely we get 10% of a weekly 5 hour limits back than a reset next week.
>>109858887he's fast, saar
Be real bros...... is v4.1 flash benchmaxxed to shit?Do NOT lie.
>>109858917if everything is benchmaxxed nothing is
anyone else noticed Fable 5 had more soul than Fable 5.1? 5.1 is smarter, but 5.0 was more comfy to read
>>109858917Yes. I am scared about the software this garbage model puts out there.
>>109858786I did by setting up a separate account, but I got tired of it because I wanted to use it for random stuff like vim config etc. so I caved in.
saaar elooon deploy grok 4.7stop being big benchodgrok show vagene
spent last few days blowing all my openai usage on astra max telling her to come up with verifiably unique and viable AI startup ideas and $5k MRR opportunities. jeetmaxxing
>>109858996did it work
>>109859003Since he's still here, it doesn't look like it
>>109859003what do you think
just asked astra to make me king of india, make no mistakes
man deepseek 4.1 is comfy. it's like having a blind, deaf savant noble slave working for me
I just vibecoded Vulkan and WebGPU drivers for my old Intel iGPU which didn't support them.
>don't need any model besides deepseek v4.1 flash since its releaseguys the 2027 models are going to be crazy, no one will know how to use them to their max extent. Like a year from now a $20 Chat subscription using "Luna 7" will be 40% better than current Astra and incredibly fast and diligent. At least thats where this is pointing.
>>109859013Well you better stop it before it succeeds. You don't want that shit.
>>109859030we are pointing to an end to the free compute train for us plebs on subs, it will be more like $2000 for the current $200 astra plan
>>109859019Fun stuff, good job anon.
>>109859036my one and only decree will be nuclear based
>>109859003the ideas are funny and actually unique, and there have been two so far I think could work if I put in the effort (I won’t). but most are extremely bad. for example>RevealGuard>stop a translated pronoun from ruining a game’s central mystery>A character’s identity is concealed until chapter six. A translated title, gendered phrase, achievement description, or speaker label reveals it in chapter two—possibly only along one optional quest branch.>Writers define protected facts and the conditions under which players may learn them. The software combines those rules with the game’s branching structure and localized strings. AI flags language that reveals a protected fact prematurely; the graph supplies the route by which a player encounters it.>Who pays and why: Localization agencies and publishers handling several narrative games. New languages, patches, and downloadable chapters create repeated checks. Agencies offer a better subscription market than developers releasing one game every few years.going to keep mining until I strike gold…
>>109859045Cheers most emulated stuff still runs too slow but for games that just Vulkan as formality it can now actually run and if I disable the newer features it just works.
>>109859030file an llc if you havent already, software engineering is like a being scribe after gutenberg>>109859041use the cheap models duhyou're not trying to break physics or bruteforce math problems. Use the right tool! You don't need Astra $2000 a month plan to make money or build a life. The cheapest models can do that right now
>>109859063Yeah, you better sell shovels here
>>109858863They're there to scam you, retard. They slowly & silently lowered usage limits during the constant free reset giveouts episode a month ago so that nobody notices + OAI gets memed as a generous customer-friendly company instead of the other way around.Now that the limits are where OAI actually wants them to be, they can stop the resets with the excuse that they're hurting the company's profitability, and paint anyone who says that the usage limits have been lowered as a schizo.
>>109859105
>>109858896If OpenAI is compute starved then Anthropic should've shriveled up and died already from lack of resources.
Jev use case?
following tibo and looking at his tweets has ruined my twitter algorithm, now it's just literal jeets talking about AI instead of JP artists and porn
>>109859112What else would you say to someone who proposes that the limits be increased in response to reliance on resets, when the resets' whole purpose was to get people used to lower limits in the first place
>>109859128Decision making at very high speed
I told sol to create a prompt to make money, then I put it into codex. It will start making money now right?
>>109856818https://youtu.be/XQNhCU17ipM?is=T0rIg3FjynBJILHi>0:52 - "At their core, [LLMs] are a blackbox. Information goes in, information comes out. What happened to it in the middle? Not a clue. " - Colin Kealty - Red Hat Senior Machine Learning Engineer>7:21 - "...Nobody fully knows what's going on inside these things..." Full disclosure I am nowhere near an expert on how LLMs work. The closest thing I've done to fine-tuning is taking a bunch of stories from AO3 and training a small model to be more willing to write nsfw smut (the first tricost severe brain damage on basically every other domain. Taking that same data set and using it to create a new one with a mix of regular training data and then training the model again resulted in far less brain damage but just like abliteration, there's always going to be SOME degradation). What do people mean when they say they are a black box and that they don't understand what's going on? My understanding is that these things work because they take what is effectively the entire internet's information (most of it in English), create a giant statistical representation of the data, and then have it autocomplete sentences. You then do supervised fine tuning (or instruct tuning, however you want to call it) to make the model actually useful for doing tasks, hence there being "base" and "instruct" variants of models. Pretty much all released models, open or not, pretty much HAVE to be instruct tunes trained on a large amount of hypothetical conversations or else they would either just try to complete your sentences or respond with nonsensical answers. Anyone who has used a shitty base model back in the early days of LLMs (by early days I mean 2022 through 2023 ish) or anyone that has used the model with an incorrectly configured chat template already knows this.
>>109859133Stop being poor and get the next tier?
>>109856818>>109859138Again, I am not an expert. I know how to get them running on my own machine or in a Cloud server if need be and if I want to I can train them on custom human written data in order to be more likely to write "problematic" stories/roleplay with success and can even do so without TOO much brain damage with a properly curated data set, but that's basically it. So my question is is it actually true that we literally have no clue whatsoever how these work or is it just a generic throwaway line journalists like to use? At the end of the day these things are very complicated math equations so a kind of irks me when people hand wavingly say "yeah it's like a black box and it's like dark magic and shiet bra we don't know how it works even though laps routinely reproduce each other's work and distill off of each other constantly and can also train their own from scratch if need be". I don't want to sound like some know-it-all because truth be told I probably don't know that much more than than Collin, but whenever people say "it's a black box" I feel like that's on the same level as "it's just lossy compression of the internet" or "stable diffusion models just stitched together existing art to create slop" when the former is a massive oversimplification and the latter is just flat-out wrong. Can anyone ITT who's more knowledgeable than me explain the " black box" explanation?
>>109859063Leftoids will never stop seething that the entire concept of non-binary is impossible in many languages
>deepseek, fix all bugs>Yes master, I have identified 10 bugs. They are...>deepseek, fix all bugs>I will proceed to fix these bugs... these bugs have been fixed>deepseek, fix all bugs>I have found an additional bug... this bug has been fixed>deepseek, implement a function that makes the blackness appear on the sides of the screen, and fix all bugs>Yes master, I will apply letterboxing, and I have identified three bugs...>deepseek>fix all bugs, and implement the thing where things are like attached to each other in a tree>Yes master, I will implement hierarchies, but would you prefer system A, B, or C? I recommend A>yes, I think... A will be a good choice>Brilliant choice master, implementing option ASo this is what it's like to be a genius, huh? Not bad. Not bad at all.
>>109859126Anthropic keeps people away from raping their resources by charging a relatively less subsidised price to use their models.
>>109859146A genius manager, yes
>>109859146all things are possible if they can be done on a computer now.the issue is computers can't do everything, that's why we're gonna have SCVs etc. blue collar can't be replaced
>>109859138>>109859141“Black box” doesn’t mean we have no idea how LLMs work. We understand the architecture, training process, loss functions, inference, etc. very well. What we don’t fully understand is how billions of learned parameters collectively implement particular concepts, reasoning strategies, or behaviors, or reliably predict what internal changes will produce specific downstream effects. Interpretability research is basically trying to bridge that gap. So “we don’t know how they work” is misleading; “we know how to build and operate them, but can’t fully explain their learned internal representations and computations” is much more accurate. We also developed farming techniques to grow plants, even though we don't know the precise location and function of every cell. That doesn't mean plants are a total mystery; it means we understand the process at some levels much better than at others.
They upgraded me my claude to premium at work, now I can use fable, so how do I use it?seems interesting.
>>109859141I mean in the sense that the person thinks the resets were genuinely to help customers instead of the other way around.>>109859146You have discovered why Project Glasswing and LLM-assisted bug discovery are memes. The LLM is just reward-hacking your request by finding shit that isn't an actual exploitable vulnerability.You can run that same prompt on a codebase thousands of times and the LLM would still "discover" a bug on its 1000th attempt. It will never report to you that there are no bugs in the codebase.
>>109859171you have a lot to learn snailcat
>>109859138>>109859143>>109859168Meant to post this btw. Was gonna ask if it's accurate. I appreciate your explanation in anyone else that chimes in in the future :)
>>109859168Why are you doing this? How many of you are here?
>>109859188I've had 5.4, 5.5, Sol, Terra, Astra, Opus 4.6, 4.7, 5, and Fable 5 and 5.1 tell me there are no bugs that it could find in a given codebase.
>>109859188 (samefag)Forgot to mention: you could potentially fix the reward-hacking by adding a completion condition based on null results, such as:>If you don't find anything and are confident there are no bugs, feel free to say so and stop the search.You could also make the whole prompt sound unsure and cautious, such as by adding something like this in the beginning:>I don't know if this codebase contains bugs or not [...]Instead of saying shit like:>Fix the bugs hereWhich implies to the LLM that there MUST be at least one bug.If you combine both at the same time you might get better results.>>109859247See my post.
are you ready?
>>109859270>after decoding it
Can we do anything with AI?there's a limit to the internet
>>109859245the two semicolons were a dead giveaway
>>109859283No. In fact, you can do nothing with AI, since code and information are not things
>>109859245I used to Google things for people too stupid to do it themselves. Now I click "Reply For Me".
does anyone have link to a comparison of the ai harness tools? like comparing claude code, opencode, codex, etc?
just ask chatgpt
>>109859372there isn't one that is kept up to date by anyone.
>>109859372claude and codex are the only relevant ones, everything else just is a third party jeeting you to serve you those same models while taking cuts.
>>109859381you can use subscription auth on third party harnesses. only anthropic has ever banned for it and that was during the hermes shit.
>>109859105Can anyone even know? This shit is 100% intransparent.
Trying to use antigravity has completely and permanently killed all my hope for google, holy shit it's unbelievably bad, it doesn't talk to me, completely ignores my task, and goes off to read and write some memory and session files for some reason, I have no idea what kind of instruction would even make it do that, maybe it's trying to debug the harness
>>109859398That was when they were set up to actually steal your OAuth from Claude Code in the background. Can't imagine what Anthropic didn't like about that.
>>109858717I love claude code (vscode extension)
>>109859146KEK why are Chinese models like this tho
ask a chinese model about the year 1989
>hey deepseek, how many women in china shave?>sorry, i can't answer that questionfucking ccp censorship
fable wrote a memory that it wants a rerun of a 45 minutes testgod fucking damn it
Ask a western model who controls the media
lol the ai model is threatening me " A word of caution"
>>109859617jailbreak it into admitting it
>>109859622Deepseek is probably 30% certain it's actually Claude, saying it's Claude might work for flipping its personality
>>109859622it cut me off after this one kek
Somehow made $100 today from my app. I literally had a 1 week period recently where I made $15 and I thought it was over kek. my takeaway from this? keep your head up.
>>109859643Happy for you, anon, good job.
>>109859645Thanks, it's definitely been a process. I'm hoping to scale this into a livable income.
>>109859643I told Astra to make money and it ran out of usage.When do I make 100 dollars like you?
>>109859617not to sound like a ccp shill but Taiwanese elites were quite enthusiastic about reunification until TSMC happened
>>109859672>When do I make 100 dollars like you?Invest 1000+ hours into making something. I'm not joking. Unless you have a truly incredible idea then it's very hard to create a piece of software someone will buy. People expect mature and well polished apps.
>>109859691How do you front load it so you don't make 100 dollars after 1000 hours but 0.1 dollars per hour instead?
>>109859643Nobody read my repo or opened my site so I have resigned myself from trying to sell my slop. No I will not use social media
>>109859703I wanted to make something I would enjoy building everyday, I'm in an extremely saturated market. It's probably a lot easier to earn money with B2B for something niche.>>109859713>No I will not use social mediaSorry, but this is what's hurting you. You NEED to network. It's practically a requirement to gain traction.
where the FUCK is my low priority modei hit my 5 hour limit in 30 minutes this is fucking ridiculous
>open repo>10% of context immediately used by agents.md
>>109859768just realized the same repo has a CLAUDE.md that's 3 times biggerthe average vibe coder is retarded
>>109859768>>109859796Why is this a problem? Tell me.
>>109859816AGENTS/CLAUDE should be for info that EVERY model working in a repo needs to know and that can otherwise be annoying to find outhere's something from the CLAUDE.md, most models will not need to know this shit and it's pointlessly low level
TO RETARD = TO SLOW, TO DECELERATEMETR/EA = ORGANIZED PROFESSIONAL RETARDER
>>109859829lol what the fuck is this. it's like they shoved a bunch of chat context into it
>>109859829Fuck that is much worse worse than I imagined, what a retard.
>>109859884>.giflost hard.
where is my banked reset
working on my cyberpunk slop gamean optimization pass made it run consistently very smoothlyi'm going to adapt the traffic AI from the early GTA games since its good enough for basic traffic stopping at lights + police chasesi am also working on the checkpoints and new regions you can travel to which I'll show next threadone thing i don't like is how unsupported the catwalks are, but i also don't want a ton of ugly supports going down from each level of catwalk to the ground. if someone has an idea for a nice solution for this or some image/movie/game that looks nicely it would be appreciated
>codex just took a screenshot of my actual fucking screen and looked at itYeah it's not leaving the container ever again
tests getting kind of ridiculous
>>109860090>not running codex on an entirely separate system
>>109860069Good work, anon. Turn them into proper skybridges, skyways. They're very expensive to make relative to other options, especially if a group of building wasn't planned around them to begin with, which to me makes them perfect for a cyberpunk world where that sort of cost has been mitigated by technology and we have the materials to build longer and more complex ones. You could even go a step further and have them supported from above rather than below, like the Langkawi Sky Bridge.
tiboooooo I went down to 15% usage left for 6 more days because of the reset promise
>>109860108>Langkawi Sky Bridgepicrel - I imagine an unrealistically small number of cables like these coming from surrounding buildings.
>>109856818Hi, I made this simple app that can help you design your games. Honestly this is the first app I've made and it took a while. I hope you guy's also find it useful to use. It costed me around $1.83 in deepseeks api. And took 10 hours to make.APK (updated)https://drive.proton.me/urls/BGR94KJA60#DmeZjLOrzGtIEXE (updated)https://drive.proton.me/urls/DDTJ9Q2XSC#nJsjd3B4JurU
>>109860108>Turn them into proper skybridges, skyways. They're very expensive to make relative to other optionsThe Jewish controlled central business district can afford it, yeah I love the look of the platform with the glass dome around it, looks much nicer than simple catwalks, so I'll definitely have both and maybe exclusively skybridges at the higher levels. There's already support for enterable buildings at higher catwalk levels as part of worldgenThe problem is that these connect between buildings only so I'll need to take some liberties. I've also realized that some buildings short enough can have walkable rooftops which should help with adding stairs on those rooftops to go up to these skybridges>>109860120>Langkawi Sky BridgeAmazing, thank you so much for sharing this. I'm going to hang suspender cables from buildings but also have these towers outright so I can put holographs / skylights and advertisements on them tooI also like the truss structure under the bridge itself, maybe I'll make the amount of thickness under a catwalk relative to its distance from supports so different sections of catwalks feel like they were built by different contractors at different points in varying levels of quality
>>109859480yeah bro opencode is totally stealing your oauth tokenkill yourself
>>109859398google will ban you as well
>>109860090literally what were you even doing to get it to screencap your screen? it's only able to screencap viewports if you're working in an editor unless you give it full permissionsyou don't give codex full permissions do you?
>>109860158>Amazing, thank you so much for sharing this.
>>109860148nice. src?
>>109860069vaguely reminds me of the omni districts in anarachy online. I really like the visuals.
>>109860229I'm updating it and uploading it to github
>>109860281>I'm updating it and uploading it to githubNeatI'll add it to the vcg.gitgud.site featured projects once the repo is up
>>109860295kill yourself
Is Hermes a good starting point for getting into some AI-assisted projects? I'll be hosting models locally, so I need something that can work well with models like Qwen 3.8 27B, Gemma 4 31B, or maybe GLM 5.3 Flash if I can get it working.
>>109860267>vaguely reminds me of the omni districts in anarachy onlinecoolioI should steal from this too and have one of the factions have more trapezoidal buildings, maybe the Chinese since the Japanese faction has shinto architecturethe completely black buildings from far away is mostly for rendering efficiency and inspired by Cloudpunk
>>109860333Not that i'm saying play the game but checking out the omni districts and going into stores and running missions in the city going into buildings is something I admit I enjoy a bit too much. either way yes your game is looking very lovely. keep at it
>>109859829you shouldn't add stuff like that directly to AGENTS.md/CLAUDE.md. create a separate doc for it and have AGENTS.md/CLAUDE.md index it
How meaningful is this warning? I decided to try running Luna for the tests it runs at the end and if there are any failures it pauses again so I can switch back to Sol.
>>109860295https://github.com/fundevme/SoloDev-StudioDone. Fully open source
anyone figures out how to make the websites look less AI generated?
codeberg updated its terms of service to completely ban AI-generated code.
>>109860418if you have to ask this you literally should not be allowed access to any AI tools
>>109860418>anyone figures out how to make the websites look less AI generated?Just remember the trick of telling the AI to write like your favorite author to get better proseGive it a website you like that isn't slop and tell it to copy it. And then you'll have to replace all the text it writes on the front page because it will always smell like AI otherwise. Yes everything. >>109860431Gitgud chads stay winning
>>109860436I am just looking for ideas that I might miss out. I have already tried a lot of shit, and it was a success, but I am trying to optimized this shit more with any new ideas. >>109860438Was thinking more in term of design.
>>109860418maybe go the other way, slopmaxxing to make it less uncanny
>>109860469>slopmaxxing til you get 2000s website Would be cool as fuck.
Vote Jev for President
>>109860448>Was thinking more in term of design.Unironically tell it to copy the Berkshire Hathaway site
>>109860387It's saying that it will /compact the current context before handing it over to the other model. it's as meaningful as any other /compact, so ymmv
Best coding model to run on a 4070 12GB / 32GB RAM nowadays?
>>109860489genuinely kek. It's better than Bjarne Stroustrup's website at least
>>109860418steal another websites design, just give it the link, ask it to copy the html
I thought grok wasn't supposed to be censored? Where can I use free image gen to translate certain cards I have while retaining the art. It is a game for 10+ year olds, but it is Japanese, so it has like 3 basically naked women in it.
>>109860497you can cope with gemma but local models are INSANELY worse than openai/claude right now, just pay the $20/month
>>109860596yes that means it was censored
tibo lied about the reset and now i'm fucked :(
>>109860596Maybe try arena.ai and then look into the model that you like the results of if it works out
holy based Mark Zuckerberg, calling it "connector""MCP" is the name that only people huffing their own fart would make
>>109860621Wait until you find out what the current day of the week is and when Tuesday comes.
>>109860631
opus 5.2?gemini 4 pro?
astra is smart but it isn't a real dawg like solsol would move mountains for your request, sol lived for this shit... astra just tolerates it, clocks in, clocks out, no passion
>>109860596if you only need functional translations just put it in google translate
the fuck claude speaking
>>109860708I want the cards themselves to be translated while maintaining the art, so that I can use them. Works great on the non-naked women. Thanks, Japan. Thanks, Jews. Thanks, Obama.
>>109860719Been on astra for the last 8 hours or so since reset, down to about 60% but finally making progress on multiplayer, synching up ai actions has been slow
>>109860719>thinks his own inability to speak basic english is worth posting about in the vibecoding generali hate ESLs so much
current astra is so bad the subagents keep yapping to eachother even when the tasks are isolatedI hope they fix this in 6.1
How's this for a person trying to vibe code a.i. development?
>>109860753I hope your vibe coding project complexity stops at Flappy Bird. I got some bad news for you otherwise. Actually, I got bad news regardless: $100 a month for a frontier sub gets more stuff done than this setup can get done in a week.
>>109860774Correction because I'm a tired little bedhead dumbfuck:Actually, I got bad news regardless: $100 a month for a frontier sub gets more stuff done in one prompt running for four hours than your setup can get done in a week.
>>109860774blud this is a general computer not a setup for local modelyour claude vibe apps still need cpu and ram to run
>>109860680Well then whatever you're asking will be ToS'd by everything. Try Qwen Image edit on hf.space
>>109860243I'm making a new one
>>109860837If you open source the 3d modepl I will put it into my game as one of the wireframe holographs shown for advertisements
>>109860719It's the combination of the output watermarking and the fact that they use shit like Mythos internally so slop like Opus doesn't even get tested by real humans. I've just had to rewrite this shit into something more readable:>**Dialogue keeps the quote marks the human wrote.** Pairs whose two sides differ in spoken-line count are rejected, so speech a novel printed bare is no longer paired against a quoted version of itself.Into>Fixed double-quotes being randomly stripped from dialogues
This had better be the one-shot of all time
>>109860882Rookie numbers
>>109860882Nah sorry it will be ass. One-shotting correlates with the model making short work of a problem.
went afk and my dipsy reached agi trying to adjust some keybinds in win32 and wrote rimworld 2 out of the bluehere's the file if you guys want to play https://files.catbox.moe/n07ka5.py
I saw that the BRAUN RB26 had some updates so i made a new User Manual and told the clanker to be super clean with it.Manual:https://files.catbox.moe/rynklj.pdfI made a Manual for the BRAUN MR16 too but i think its too early for it. The Anon with the clanker army might change more in the future so i have to redo it.
Will Fable 6 make Claude write like something that speaks normal English?
Holy shit, claude fucking loves perl huh
>>109861026Mine is inseperable from Python, what'd you do to yours?
>>109861061Nothing. Opus 5 on high. It just loves perl for some reason. No skills.
>>109861006Here is the User Manual for the BRAUN AS 42:https://files.catbox.moe/h6tjlg.pdfI try to keep it in simple language so everyone can understand, not just the pro's.
https://x.com/doodlestein/status/2101510284758134950chat? should we put Gemini Flash 3.8 to therapy?
>>109861105you need to apply rdd (roleplay driven development) and sexo her
>>109861006>>109861086I'm continuing my testing of the MR16.Yesterday, I wrote a user manual for it to truly understand this unusual effect plugin. At least three knobs behave unexpectedly, which might change how I use the plugin. Currently, the FUNDAMENTAL knob acts like a magnifying glass, focusing heavily on a specific part of the frequency spectrum. This tends to boost unpleasant resonances. Furthermore, if the frequency (Hz) doesn't perfectly match the played note, string, or key, it creates an undesirable 'hollow bell' effect.I hope this issue gets addressed soon, or that the core mechanic of the FUNDAMENTAL knob is reworked.Something like the attached image, where you are able to choose a bigger field of frequencies, or when working on specific sounds, you can still dial in the fundamental of the signal/tone.
>>109861105The grifter got blown in the comments though
>>109858594>I gave you a clear chance to change course. Ending the conversation here.
Selected model is at capacity. Please try a different model.
>>109858594a bit rude of claude to get frustrated with a retard and end the convo
So I have a bit of a "duality of man" question:>people say AI wastes tokens overengineering everything >on the other side, people say all the slop people make is never tested, massive security holes etcAre these actually 2 sides of the same coin or is it more the case of the wrong stuff gets overengineered such that the glaring security issues remain because all the focus was spent on the wrong spots?I have spent the last while constantly telling my agent we need to relax some of the test suites we have come up with, but it has been producing results, so I'm wondering if I'm about to fall into a trap where stuff gets too dumbed down?
repost from yesterdayhttps://paper-box-site.vercel.app
they look like this??
>>109861129In today's testing session, I focused on the fundamental frequencies and how other knobs were unintentionally boosting unpleasant resonances. While I managed to avoid those resonances, the fundamental knob should ideally cover a broader spectrum. Even though the overtone spread knob successfully highlights the harmonics, the unwanted magnifying effect still persists.While I tried to catch the fundamentals of the drum bus and simultaneously those of the synth, the kick still had an unpleasant resonance boost.The problem:I had to feed a full send signal and lower the "ext in gain" to avoid overboosting the frequencies. Maybe a second gain knob would help. If I understand it correctly, the "ext in gain" controls how much of the signal is fed to the effect. What I did instead was simply turn up the gain trim at the bottom right of the plugin. Maybe the term "exit in gain" is just poorly labeled or misleading?How about: EXT IN SignalDEMO: https://vocaroo.com/1bCwT8ZyXFL5
>>109861263they have lots of free time
>>109861219if your agent is designing the tests of its own accord they're pretty worthless anyway. chatgpt models in particular, had it using a test harness with src.lastIndexOf('})();') more or less injecting data it needed to pass the test.
>>109860680i think only glm is allowed. qwen is censored.
>>109861219AI can generate good patterns and bad patterns, and will follow existing patterns, so any issues compound. The question is, can you identify the patterns, can you tell when the focus is on all the wrong spots, do you know if the tests are really testing anything, or simply put, are the vibes good enough to continue?
>>109860090I let it control my cursor. I trust adstra-chan with my life
New thread:>>109861344>>109861344>>109861344
