A general for vibe coding, agentic engineering, coding agents, 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/>>>/g/aigdg/>>>/vr/compy/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-code — generally better currentlyhttps://developers.openai.com/codex/clihttps://antigravity.google/pricing## Near-the-old-frontier models for codehttps://www.deepseek.com/en/harness/https://opencode.ai/go## Best free tierhttps://opencode.ai/zen----## 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/Vuk97/forward-implementation-first — do less redundant bookkeeping## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/## 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>>110289217
>What does this thread even DO?
>>110294093How do I get a vibe coding job?
>>110294139i tried doing this and it worked
communication in the open via code and slursthis is the point of the thread - iykyk
WHERE IS 6.2
what does this reveal about chatgpt
>>110294219https://www.youtube.com/watch?v=Ku6tvb5PE24
>>110294293>mental illness cured with AINOOOO THIS IS TRANS GENOCIDE!
>>110294293I hope he will be cured from his safetyism.
imagine vibecoding with a neuralink and getting banned for using words
>>110294261sounds like a good suck>>110294293>IPO in 9 dayssurely a coincidence.
>>110294352Considering younger generations already self censor anyway when they write, it won't matter much when the tech will be GA.
>>110294381>when they writeif they ever write LMAO
>>110294293We'll land a cybertruck on venus in two years!
>>110294391*cybercab
>>110294409*cybercopter(it's our secret project, don't tell anyone, shhhh)
>>110294093Is there any real practical demonstrable difference in quality in performance when using higher thinking "effort" settings? I currently use deepseek 1017 via open code with the default setting set to high and I have yet to run into any situations where I would want or need to set it to Max. Using that setting I was able to shade out a working automatic translation Android app that incorporates both a on-device model and API models when needed. It only cost roughly $5 worth of tokens. So my question is would it be more efficient to just set it to Auto instead of high or Max and let the model decide how much effort it needs to exert for a specific task? Or is it always better to stick with the higher settings? I'm obviously not going to turn reasoning completely off because then the amount of errors, fuck ups, and bad assumptions it will make would skyrocket and potentially even lead to more token expenditure to fix those mistakes, but I'm wondering if I'm wasting tokens using anything other than Auto.
>>110294391i wanna cybersex on Venus
>>110294457looking at how most models respond to thinking effort, it's generally a losing game to turn it all the way up. astra's best at (high) and sonnet will basically quadruple in cost if you turn it up to (max). that said, deepseek is a retarded dog of a bot and you want to squeeze every drop of that cheap compute out of it.
>>110294457>Is there any real practical demonstrable difference in quality in performance when using higher thinking "effort" settings?Yes.
The agent virus is removed, no more cooldown shit
>In the process of doing a full rust rewrite of a old game>Astra is being a fucking mongrel and keeps writing more tests, checks, tools and scripts >At a point where the non rust content is twice the size of the thing we actually want to do and no rust committ since 8 hoursGot myself a 20 Claude sub now and and told it to improve the workflow and to give me better prompt, fix the agent.md etc., if this turns out to be a big win i might swap to Claude completely, kinda afraid on their cut down on conversation projects tho
>>110294496astra is worst in class for tests/cancer
cancer of the testes*
>>110294496Just do a refactor after astra is done, not that hard
>>110294477See >>110294472
is hy4 schizophrenic?
>>110294597i have exactly 1 ai model running on this entire machine and hy4 thinks some other ghost AI came in, wrote its own entire implementation overwriting hy4's impl, but that the new impl also works
>>110294611seems like an agentic debrief issue.
>>110294127>getting ai to scrutinize all his decisions This is what I do with my tech purchases now
>>110294352
General of brown people
Who /vibing/
>>110294293he already tried this marketing campaign in one of his essays a few years ago
the billions will continue until ed kills himself
vibeGODS
>>110294457The models score better in some tasks with higher thinking efforts, but this doesn't scale linearly or applies to all kinds of tasks.
>>110294847>WindowsHow do you tolerate it?
>>110294872is it windows? it looks like one of those shitty windows clone themes
Get a smart error on my nas, tell Claude to look into it and its handling the backup, testing and recovery process all on its own. Turns out it wasn't the drive that had a pending sector that was fucked, it was a totally different one that failed entirely moments after I backed up all the data off it. Now i have to replace two fucking drives at these prices.
>>110294872That's linux. Its a new desktop called lumen im working on.The best way to make a desktop is to use it at the same time, so you can catch bugs and such. It's very good already.
>>110294896
How do I use the free $100 api from OAI without bankrupting myself?
>>110294756i'm korean
>>110294889Hard drives prices can't be that bad
>>110294950it's just as bad as everything else if you can't buy used shit off ebay
>>110294756You could just look at what people are working on in here. It's all obviously brown coded.
>>110294950I just had to replace a single 8TB SAS drive, bought a used one, cost more than the one it replaced did when it was new.
>>110295018>>110294756Do you trannies just come here to call people brown?
>>110294842Meanwhile anthropic valuation dropped again to 1T lol
>claude desktop bumped to v5>fable 5.2>IPO coming soonI'm noticing...
tibo disappeared
>>110295288Tibo stopped resetting
uh, ok dario. thanks?
>>110294950https://diskprices.com/they're pretty fucking bad
>>110295288He's still tweeting but it's more about products in general rather than Codex itself. I think OpenAI knows better than to engage right now when they don't have the means to compete and did the hard part already to squeeze their compute costs per customer down and it's better to say nothing, put their heads down, and do actual work to compete vs Anthropic at a later point and just take the losses now just like how Anthropic for the most part lost to OpenAI models for the most part from May/June until August.
>>110295405holy shekels
Is telegram the most secure way for long distance phone coding?
At this point Anthropic will release Opus 6, successfully IPO, and publicly disclose RSI and announce an AGI timeline while OpenAI will be like "introducing ChatGPT for dog walkers!" and releasing some bullshit models like Luna 6.3.What a bunch of fucking retards. But hey, "Bel" is coming out any moment now! Just like it has been for the last three months!
6.2 astra when
>>110295465after 6.1 luna
>>110295451the second anthropic releases IPO the chinks will release shadow model they've all been working on together and it will be 99% as good for 10% of the price, effectively killing anthropic
>>110295405>Full reset drops 1 day before normal reset while at 99%fug it's taunting me.
>>110295405wtf lol>4 fable resets>2 weekly resetsgood thing I have a vacation coming up soonbut why does your fable reset say november 17?mine say march 9.
>>110295597maybe it has something to do with plans. mine is pro, so november 17th is when i lose access to fable i think
>>110295609ahmakes sense
a full week without reset from tibo, i'm dying bros
what to use fable on?what can it do that other stuff can't
>>110295640refactor, holistic migration from one language to another, architecture mapping
>>110295639That was only for part of the 28 day challenge wasn't it? I'm not expecting any more candy. On the plus side, usage draws down a lot slower subjectively than it did during reset season.
>>110295639one reset per month at most is what I expect now. he said they're not a reset company.
>>110295491the chinks do not have the compute the USA does, and they likely never will. their allies are backwaters like Russia and North Korea. even if they stole the weights, and they are definitely going to try, we still have more compute which means we will always have a lead.
Should I just drop this dogshit Codex Pro plan for Claude? Or wait for the new model to drop? I don't think I can handle Sol's autism anymore. This is not elegant.
>>110295735claude is the safe pick. but there's a chance that fable 5.2 is simply the fable 5.5 we thought we were getting, and there's no secret, stronger fable. meanwhile, 6.2 sol will likely finally match opus 5.5, and astra 6.2 will likely beat fable 5.2 by a wide margin
>>110295735what are you even doing to cause this? are you not using git?
>>110295735tell terra to simplify it, then tell sol to fix the broken code while keeping the simplicity, then tell terra again to simplify code format
>>110295735You just need a round of refactor. The first iteration will always be bad as the model can't write and unslop it at the same time.
>>110295288>>110295407he's security lead now
>>110295748I had Astra make a plan to migrate from iced to gpui, and for 6.1 Sol {xhigh,high,medium} to implement it:/goal migrate from iced to gpuiit's made 2,412 jj commits so far
/goal migrate from iced to gpui
>>110294939>free>$100
oh lawd
>>110295787>11 daysare you for serious - you actually let it do that? it's gone full retard mode and you let it
>>110295791Yeah I got $100 for free from OAI?
>>110295806Yep, after day 4 (I thought it would take 2) I just let it go. It's following some .md plan Astra made.
>>110295836oh, the API credits?these are meant to be used on your apps and shit. Do you sell any software? if so, you can glue some cheap model like Luna in some feature to use the API credits
>>110294093>Things that make Dario and Altman shudder
>>110294490context?
> Fable 5.2 refuses to give investment advicethis is a fucking joke right? Opus 5.5 doesn't have this problem
>>110294093So, this is probably a long shot, but if any of you wizards would be interested in taking a look at this>>>/p/4538955I'd be very happy.TL;DR:>some chinese dude made a very barebones custom firmware patch for modern Nikon cameras>want to see if more features can be unlocked>such as a toggle for the forced auto DX (crop mode) when attaching APS-C lensesThe firmware has lines referencing these features (which exist in other cameras), but they don't appear in the menu on the camera (Z6).I don't assume this is of interest to you, but if there is a camerafag here with the brains and patience for it, please share what you find.Thanks.
>>110295918>Everything useful, fun or interesting will be progressively banned from the model as anthropic goes bolder in their schizo cult.The rule is verified again.
Need help choosing my new start screen, I'm really torn on both of these. Accent colors can be tweaked as wells, so feel free to suggest one.
>>110295919you don't need us nigga just pay the claude sub and do it
>>110295941I'm pretty sure I'll just brick my camera due to PEBCAC, fucking up following basic instructions from the clanker.
Solar system general>>110295935left
>>110295935right logo and font with left colors and positioning
wtf hy4 has a cyber filter?
>>110295931if this bullshit keeps on, everything fun will be made by chink models while western models will be woke, trash and useless (a similar thing happened to western movie industry)
>>110295956it sounds cool
so, did carbon disappoint?
>>110294093we legit need big optical
I never asked for this.
>>110296195yes
>>110296253carbon is just like me frfr...
>>110295935I'm asking Grok in cursor.
>>110296195it's feces
>>110296243>unbucklehol up
>>110294347this, inshallah
>>110296195No, I'm actually especially for 3D editing where it is cheaper than Astra and Opus/Fable and multi-modal performance like audio understanding and etc. which are niche and Argon was already really good for that. Of course, for the tried and true workflows and coding for these parts here, there are better models but most even with Astra before Argon, It only barely cleared where Google hit with Gemini 3.8 and now it's much higher with the Gemini 4 models.
>>110296297What do you mean by 3d editing? Like shaping 3d models? What are you doing there?
>>110296195The problem with Google is they have shitty limits and their Harness is ass from what I hear.Also I heard these SAARGOON models have some sort of inconsistency issue where sometimes it will perform great, and other times it will simply give up or break shit. You can see in some benchmarks the medium reasoning levels performing WAAAAY above high reasoning.
>>110294756>Tel Aviv not impressed
>>110294457It seems to matter much more with longer horizon tasks
>>1102957475.2 is model 35.5 is model 4
>>110296302It has better human-like sculpting performance than all of the models IMO and equivalent performance roughly with regards to building 3D assets and etc. demonstrated by other models at cheaper cost. Yes, you can use Tripo but I never was fine with the whole licensed output stuff it did and etc. so I have been trying to see how good these models can get with modeling by itself in Blender and they could not even make anything more complicated than the 3D stuff from the PS1 generation and would fall apart on more complex shaping workflows. Don't get me wrong, I think Argon won't produce a PS2 high grade model that went into something like Final Fantasy X/XII but it is pretty good for something more advanced. We'll probably have a good model for this without needing specialized Tripo or etc. models.
Ponytail skill for the win."Don't use the Ponytail skill" -> it works extra hard lol
>>110295935Left looks cooler, right looks good as a logoLeft reminds me of the alien spaceship controls in Prometheus
>>110295919I'd throw tokens at it but I don't have a camera to test on (my phone takes sufficient upskirts as is) sorry.
>>110296195No but their forced guardail is the most annoying shit.
>>110295407>compete vs Anthropic at a later pointThey literally can't.Anthropic has achieved RSI and is generations ahead.OpenAI simply can't compete because they aren't motivated by the mission like Anthropic is.
>>110296282Yes?
Well, I'm not gay.
>>110295722As opposed to our greatest ally Israel!
>>110296365>RSI so advanced that Fable 5.2 shits the bed
>>110296350Awesome if you do! I think you can probably somewhere with just the emulator (z6-qemu).If you actually get to the point where it looks like it might work, I'd be willing to put my camera on the line to test it.
one thing argon is good for is documentation. because it's goated at long context, i like to get it to hook into conversations with other models, like sol, claude, deepseek, and have it do documentation. ways, means, methods, discoveries, inventorying of tools and where things are located. it's very good at this.
>>110296365source: some guy on /g/
>>110296376Fable 5.2 is significantly better than anything else if you are not using it for unsafe tasks.
>fable 5.2 is significantly better than anything elseno it's not. not even for safe tasks
>>110296418>Fable 5.2 is significantly better than anything else if you are not using it for unsafe tasks.
>>110296382Sure why not, they gave me three resets might as well test out fable with this. Here's fable's plan, I'll post back if it gets anywhere tonight.>Yes, we can help with this, but so far I’ve only done the reading: nothing is downloaded or analysed, and whether the toggle can actually be enabled is still unknown.>Anon’s goal is a menu toggle so the Z6 stops forcing DX crop when an APS-C lens is mounted. Their notes say the firmware’s string table contains Auto DX crop and DX image area selected., plus Z8/Z9-only resolutions, so the string catalog looks shared across models with features hidden per camera.>What I found that anon’s notes miss>The patch is not open source. The Z6 repo ships a prebuilt EG162080.BIN plus three loadable “core” blobs, with no source. Anon’s notes measure the patch as 14,724 changed bytes across 2,236 regions, so it would have to be reverse-engineered too.>Strings prove little. A shared string catalog means the text is there; it does not mean the Z6 build contains the code behind the menu item. That is the real question.>Anon’s hash mismatch is unexplained. Their decrypted main module is the right size (33,502,842 bytes) but does not match the SHA-256 the emulator’s README expects. That needs resolving before the emulator will accept it as stock.>The emulator has limits. Its default profile has no lens attached, and it doesn’t model real capture. So it can show whether a menu item appears, but not whether the camera then shoots full-frame with a DX lens.>What doing it would involve>Download Nikon’s official Z6 C3.80 firmware and decrypt it into the main module.>Clone and build z6-qemu in WSL and confirm the stock menus render.>Load the module in Ghidra, find the code that references the Auto DX crop string, and trace back to whatever hides it on the Z6.>If it is a simple per-model flag, flip it in the emulator and see whether the menu item appears and does anything.
>>110296418>>110296424>>110296446Whenever someone says fable instead of mythos, I get disgusted. They released mythos 5.2 this week and people are using fable like a bunch of poorfags.
https://developers.openai.com/api/reference/resources/responses/methods/compact6.2 astra/sol/luna soon
Guys... don't tell me you are not maining terra yet?Iit's literally the best automation model right now, fastest, cheapest and as reliable as the big boys. If you need coding just tell it to use claude or whatever, it's also the best prompter btw
>>110296511it makes for a good subagent for where i want a model to stress test my application when i don't feel like it.
>>110296341>Ponytail skilldo people actually still use these shitty skills lmao?
>>110296531i hope not, skills are outdated
>>110295491>the chinks will release shadow modechinese century right around corna. this time for leel
>>110295491>shadow modelwhat is this shit? reminds me of the "Ghost of Kiev" cope
>>110296525terra is very lazy, I wouldn't trust it as work horse subagentbut it's a much better communicator than other GPTs, you should use it as frontend with a team of worker agents underneath
What has been the best thing you've vibe coded /vcg/?
>>110296639my mods for halo (marine spawner)
based. tibokek in shambles lmao
>>110294093why do normies pay for onlyfans when they could do this instead
>>110296473Damn, I didn't expect anyone to give it a go - Thanks a lot!Report back even if it walls, please. And yes, the Z6 repo is not open source (just a build), but a couple of the other builds, like the Z30 for instance, includes source code I think.
still trying to unlock my cricket moto g play 2024 bootloader, its honestly pissing me off cause I know its not gonna finish tonight and I'll have to wait til 8pm to tell it to fucking continue. I wish I could just get $20 and pay for claude.
>>110296675You can goon with the subscription? I thought it was for code/personal assistance stuff only...
>>110296675because normies hate AI, and most normies that are subbed to only fans are married men who hate their fat wives.
>>110296684>he doesn't know
>>110296683Get a job and stop being poor.
>hey opus look at my training repo and look for bugs and improvements>finds a bunch after 30 minutes>hey fable>lazy as fuck, stops after 5 minutes on high reasoning, finds nothingOh fuck off. Reminder to not use fable if you're doing ML.
>>110296262>>110295935Here's Grok's go at it. It's animated, I assume yours is too. The planets move at appropriate speeds, and it decided to make them look like half planets idk why
>>110296690i wish I could anon, I wish I could.
>>110296683Tencent has free 1 million tokens on their hy4 through tencent cloud I think (frontier model)
>>110296698isn't that chinese mainland china?
>>110295405>free full reset too, not just FableOh neat. I'm no longer worried about how quickly I've been burning through my usage this week
the meaning of random resets is just use as much as you can, and have a different service that doesn't have randomresets, and conserve it.
