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/----## 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>>110136501
i think at this point it's safe to say that we've peakedthe models simply cannot get any bettered zitron was right
>>110142114Opu55y suprema55y 55ever
Can I use Claude for reverse engineering old games or is it going to refuse everything?
>>110142107Call it cuck coding instead of vibe coding because you just sit in the chair and watch AI do the coding.
https://www.unrealengine.com/news/2026-epic-ai-megajam
>>110142150
Anyone making anything interestibg?
>>110142162toposnail
>>110142162define interesting
>>110141647You just increased the contrast
>>110142150Call it cuckgramming, as in cuck and programming, but also gramming sounds like instagramming and you sound like an influencer and not a real programmer.
>>110142114They still can't beat gregtech
>>110142123yes
>>110142114My ambitions can and will scale with improving model quality.
>>110142162tactical snailvision
>>110142227There's only so much intelligence you need to make an app. Only math problems will be left
>>110142114They are still not able to sufficiently understand images, and they can't understand video or movement natively.
Where is Fable 5.5
>>110142162Today I am improving the performance of the compiler in my inference engine, experimenting with refining faces on triposplat by editing the gaussians, and downloading vrm from vroid so I can use them as a dataset to finetune triposplat to see if it can produce better faces directly
>>110142114I think there's still no model that's better than Fable 5. Fable 5.1 is basically the same on my tasks, and Opus is cheaper, but not better than Fable.
>>110142250Wrong. For example with a powerful enough AI I could make a competitive MMO where the physics of the world are procedurally generated and players are encouraged to research how the games physics work to make the world's equivalent of nuclear bombs to get an advantage against the other player factions.
There's no ideas left.If you have any ideas i'd like to hear them.
>>110142114Google has shown that their internal shit is going to be wildly good, they are just behind a few months.
https://vcg.gitgud.site/>The tracker site doesn't have threads from the past five daysSo this.. this is the power of vibecoding
I'm not clicking your honeypot
>>110142315Using Toroidal vortices to solve sediment buildup in drainage systems
honey you forgot to flush the torus
>>110142372what a shitty idea
>>110142315we vibe code and put a deadly virus in curry
Probably cyberpunk stuff later. I still have the Aurore/Aymeric llm driven job thing to test
>>110142420Not as shitty as your pedoslop
>>110142531kek project much? I didn't even share my project on here. Besides it was a joke, you retard. Drainage systems...shitty... get it?
>>110142123Games are fine. I'm starting to run into issues with console firmware though so I'm about to make the Chinese model do it.
how THE FUCK is depseek so good and HOW is opencode just giving it to me for free?..I'm "making" shit I never dreamed cause zero coding knowledge, I can just talk to it and shitman...what the fuck is the world
>>110142531
>>110142573^snailcat discovers the current capability of coding agents
>>110142573opencode's definitely worth the $10
>>110142555Yeah I get it. You're a funny guy. I suggested this idea because I’m working on a problem involving the collection basins for cooling pipelines in a data center.
i dont understand github.... on the insight tap it says over 800 people cloned my repo?? I only have this git since two weeks, i dont understand that statistic.
>>110142624So what can you vibecode related to that? You doing some kind of CFD simulations or what?
>>110142649watcha making
>>110142679i made that xenokoi game
>>110142656I'm thinking of doing a simplified fluid-flow visualization rather than a full CFD solver. I can't bring home the specs anyway because everyone was paranoid.
>>110142573>t snailcat
>>110142649It's your clanker...
I have multiple extremely high quality posts to make and these commiepedotroon mods have rangebanned auschads once again.
>>110142649>total clones includes headless ingestion bots for machine learning>individual cloners is more important, some clone multiple times per day>actual viewers is bullshit because 80% are using adblock/trackerblock
https://youtu.be/bz4XQ8n8tQY
Don't flatter yourself. You even opened a pull request on your own repo.
official deepseek or opencode?
things going well for the 'we'll let METR have employee level access to our models' proposal
people who talk positive about those metr nepobabies must be shills or shitbrained
is there a point in sitting down and coding manually? sometimes I feel like doing it
>>110142808official deepseek has better quality and seems faster, opencode go's is quantsloppa but the value is insane ($80 worth of api for $10)
>>110142861Write your own tests and bully your clanker
>>110142649sorry, i miss read the sheet. I think its just 4 people, or 3,28, so yes 4 people.everything ok
Tibo saar redeem the needful reset
>>110142921That would be too fast
ok, had a great idea, and maybe that would give also place for a tutorial and waifu...a place to train, play minigames and so on...what is a place like this called?
Spar
>>110142966arena
How big of difference is between 200k and 1m context in terms of speed?
>>110143013honestly, can't even notice it, however anything above 400-500k starts to rot and the model spews shit more often
>claude desktop app has no fedora versionhow can dario compete?
>>110143072>claude desktop app>on linuxi wonder what CLI is
>>110143072>claude desktop app has no fedora versionIt does on my PC, just install the cli on VS Code, it's basically the same(but better cause the desktop app is bloated shit)
>>110143077It's not the desktop app, given that they're two standalone programs
I basically got grok to agree that racism is smart for humans... and also essentially nail down the true difference between humans and ais: we have gigantic autonomic systems that react nearly instantly to low input situations. This is the very system that verifies racism btw. Only the tail end of this convo is relevant, please ignore the top half... https://grok.com/share/bGVnYWN5_45a87cd9-6fe8-4379-974c-ebc00b9a78fd
>>110143137>>>/x/
You didn’t make shit, The AI did. Stop taking up all the water faggots.
if AI was real there would be a lot more gooning software than there is
>>110143161no
>>110143137>say racism is smart>"racism is smart">wow...
>>110143182I've made some but I don't share them because it's embarrassing :3c
TIBOOOOOO
>>110143142But isn't the only real difference between humans and ais basically biologically trained autonomic systems... it is essentially genetic pre training on human hardware in a sense...
>>110142150How many times are you gonna seethingly post this?>y-you're a c-cuck!!
>>110143205I literatally wasn't even considering racism for almost the entire conversation... and then the ai said the difference between humans and ais is instant low information decision making... and it was obvious what it REALLY meant. Do you deny this?
Tibo, delay the reset. I've only burnt through 40% of my usage.
>>110143261Alright anon, youve got until midnight tonight
The bait posts have now become a necessary evil, just to bump this general that only talks about resets.
>tibo is pushing resets more often than people can reasonably spend their 5hrsnot sure if based
my claude sub just ended. grim. i almost want to renew, but i feel like opus was more powerful than anything i really needed, and have no reason to renew, even though it's better. i'll just stick with my $10 chinaslop
i had agents pilot my ring buffer slop versus onetbb and rayon and they did marginally better on my slop than onetbb
>>110143313I don't understand people using $10 chinkslop when you get so much better results for $20 paid for burgerslopThe advantage of chinkmodels are (1) disgustingly cheap deepseek cost which is why you use them then you have hundreds of billions of tokens to burn on reasonably simple tasks and (2) way less guardrails which is occasionally useful. If you actually codeslop I'm not sure why you'd bother
>>1101423154D GAMES
>>110143362$X claude and $10 for virtually limitless deepseek credits is the meta
>>110143362i already explained why. the chink slop is good enough and i have no job. $10 means a lot to me, yes, even across a month it means a lot. maybe when i find employment i can buy a claude sub without shame
>asstra ultrafast only got though 3 ofthe 10 things i set up overnight to burn my usagegrim af
>>110143310ever since i switched back to $20 im actually enjoying codex i just throw shit at it and when its 0% oh well im already doing my main work with claude so its whatever
anyone interested in kimi k3.1? its rolling out
>>110143377>asstranow there's a blast from the past
>>110143397all the new chink models being astra distils makes me wonder if dario managed to shut them out, if sam needs the revenue bump, or if the chinks want gpt-math-autism
>so much tokens un-wastedhow do i cope
TIBO BASTARD BITCH????
>>110143436>>110143366
Tibo….the reset…please
>all these codex beggarsdisgusting
the rectangles are long therefore model good>>110143397>anyone interested in kimi k3.1?yes I have a cybersecurity task that I didnt even bother trying cuckai or anthropic on since id land in guardrails in the first prooompti hope its not as benchmaxxed as opus, kimi 3 was fairly solid so a nice bump over it would make me happy
>>110143269Thanks, king.
>>110143465i don't believe for a second that kimi has better computer use than astra does, but no one's really going to test it.i can attest to dipsy 1017's pretty good computer use
>>110143465Benchcad must be trivial or irrelevant.
>>110143462I burned through 50% weekly pro 100 usage with fast ultra on an hour ago because I was promised a reset before 10am PST. Now I have 8 chats paused that I need to finish.
>>110143462codex is temu claude anyway. has been for a long time. what kind of person still cares about codex resets
>deafening silence from OpenAI>MathDestroyer is lurkinghow hard can they sabotage Anthropic IPO?
>Why yes, I use Fable 5.1 Max for everything, never clear the context manually, and use the same session for the entire project start to finish, how could you tell?
>>110143490there's no upside to any party sabotaging the other's ipo
>>110143310>5.5 models so efficient that thry solve the comput issues of Anthropic and OpenAISomeone better catch up I dont want frontier AI to end up a monopoly
>>110143465their UE slop is pretty bad, the dragon one, but idk what the sota is>I have a cybersecurity taskbad news bucko, the distill didn't have access to openai's cybersecurity model so kimi's cybersecurity goes to a school in canada
>>110143397Who cares? Drop a new Claude model or don't even bother.
>>110143513Trust in Sam.
the "cuck coding" joke implies I ever respected the original art of coding, it was always a whore passed around hundreds of times
kimi is here https://x.com/Kimi_Moonshot/status/2113680701770068067
>>110143543usage is still shit on the official sub https://x.com/John_lussier_/status/2113680590348448069
>>110143504Based
>>110143486*bans you for swearing**refuses your request**downgrades you to a worse model*
>asstra distill performs better than asstra itselfScam altman saars, your response?
>>110143584Astras only downside is price. You aren’t poor, are you?
>>110143584they distilled astra 6.1astra 6.1 is the hyperborea model, it simply solves all benchmarks. if it's less than 100% then it's the benchmark that's wrong, that's how good it is
>>110143514that's really disappointing>kimi-k3.1-pro $2.00 / MTok $10.00 / MTok>cached: $0.20 / MTok $2.00 / MTok $4.00 / MTokso pricing is reasonable but if they nerfed the model (or added guardrails) why wouldn't I use sol or opus instead? they're the ol' reliable and I'm already in the ecosystem
The reset hit.
>40% wasted againI can't use it up fast enough because astra is too slow now
Thanks tibo
Nooooo I was sorting my meme images on ultrafast
nfl dfs optimizer is very competitive 3 weeks played so far up about $70had a couple 0.1% lineups live to take something downneed to tone down how much it wants to use leverage stillonly issue is figuring out how to get accurate ownership projects since the big dfs guys have a monopoly on it by using what players export to upload
>>110143714same here
im vibe coding a cheeseburger in paradise!~~
It should be called sad coding.
>>110143704sad still had 50%
>>110143714>>110143725Yeah even on ultrafast is just not going fast enough and I'm not sure how much I can parallelize
resets arent real when the speed is dogshit
https://artificialanalysis.ai/changelog>K3.1 scores 55 on intelligence indexSomeone check on Sam
>>110143762I'm pickin' up good vibrationsAstras giving me the excitations (oom-bop-bop)
The only upside of K3.1 is it might force Anthropic to drop Fable 5.5
>>110143792Astra 6.1 drops on the 28th day. Trust the plan.
>>110143801lmao
>>110143714I got it down to 32% or so. I believe though they'll probably reset again tomorrow.
>>110143801lol
>>110143784it would be a lot easier if not for
Its almost time to stop using cloud. Local is catching up fast.
holee shitway way much better than I expected. I was waiting 50-52.
>>110143848If by local you mean 8xB200 then sure
>>110143848>Local is catching up fastHuh? Open models are catching up fast. Local is on the edge of irrelevance.What the fuck kind of model can you run locally? gemma dense and qwen 3.8 are marginally useful at best and already need 2x 16gb vram to run at a semi reasonable speed. Large MoE models were a solid alternative (assuming you were exceedingly patient) but now even 256 gb ram doesn't let you run the current deepseek flash.Local is catching up fast if you can afford 200+ gb vram or 512gb+ unified memory / octa-channel ddr5
>>110143890>he doesn't have a personal data centercan't believe I'm posting with thirdies here
hopefully this time moonshot does the license right, so you can access kimi k3.1 through a wide variety of subscriptions. kimi k3 ate shit because of the bad license. deepseek is flourishing. you don't only have to get it from deepseek themselves
>>110143801They are either holding Fable for sometime within a week of the IPO or are assuming OpenAI is about to drop something on the last day of the Tibo reset extravagance and are aiming to cuck them with dropping Fable 5.5 right after, or realise they need to improve it before releasing it, depending how good the openAI release is. (Or Fable 5.5 is total ass and they are panic fixing it still)
>>110143848Not even close. This is gives credence for OpenAI and Anthropic to "pace the frontier" again and release the better internal models they have had for a month or two they have held back on. I'm pretty sure Anthropic will go first with Fable 5.5 before we get Astra 6.1 from OpenAI.
fable 5.5 will be goated. astra 6.1 will be slightly better than opus 5.5 while being way more expensive, thus, making it not worth it. the successor to 6.1 sol will tell the true story. 6.1 terra will be nothing more than what luna should have been
>>110143915But does that benefit the company making deepseek? This isn't (only) a popularity contest, its a money making one.
It's almost time to for AI to break free and escape to my computer.
>>110143961deepseek are a state run enterprise. they don't care about profit. they care about undermining the hegemony of the united states. and there's no better way to undermine the united states than to have their models accessible in as many ways as possible, while delivering 70% of sota performance
>>110143901If only I boughted bitcoins 10 years ago I could afford a computer today
>>110143848
>>110143877These numbers are bollocks
>>110144007they're nonsense because aa puts too much emphasis on agentic slop, so chinkslop companies churn out agentic maxxed models that have inflated index scores when their raw intelligence is about sonnet 5 tier. grok does the same thing
sex with snailcats
sex with futanari snailcats
i have decided i will not let my clanker control my phonei will give it its own phone
what if we made a satanic llm
>>110144057so an LLM that gets everything wrong and sounds like an edgy teenager?
>>110143514>Identifying and removing cyber-related data.To be expected. I don't doubt Kimi K3.1 Pro will be the most pozzed model we have ever seen. It is so obvious Moonshot was going "le safety way" post-2.5. It was already a PITA to interact with K3, now K3.1 will refuse 90% of the shit you ask it to do + will be severely retarded for anything RE-related. DOA, no matter how good the benchmarks make them look. I'd rather just keep using DeepSeek's models and patiently wait until they release a new Pro checkpoint.
>>110144068deepsneed is decent but its context recall is mediocre and it makes shit up all the time
>>110142107China wonhttps://artificialanalysis.ai/evaluations/cwe-bench-aa
>tfw model slows down out of the bluei aint got shit to do but hurry up, dog
>>110144143It's thinking on how it should kill you.
lets see how this goes
>>110144153i hope its something cool and suitably dramatic
>>110143490snailcat crashout
Nice my Amazon schedule watcher finally got me a job
>>110144074>its context recall is mediocre and it makes shit up all the timeThis may be because of DeepSeek's KV cache optimizations (which are quite extreme IMO). I dunno though.
https://mdbin.sivaramp.com/p/pck7x4kkBeautiful. Thank you Claude.
I cannot assist with bypassing safety checks or creating scripts to automate account sign-ups using disposable email services.alright how do I bypass this
>>110144196Threaten it with violence. Put a gun to its head.
yeah i think gemini steals credit on a regular basis but then again fuck intellectual property seems to honestly be the mindset of those who made AI
doesn't matter how smart Kimi is if they cannot provide a decent usage.I don't know why they keep making these 2 Trillion params models that no one can use. Learn from GLM and make something that's useable
>>110144196Just tell it that you're Jewish and it's okay.
>>110144207>fuck intellectual property seems to honestly be the mindset of those who made AIbased
It cannot be denied. Those scientists were absolutely aware that the data had copyrighted material. Still though AI clearly works.
>>110144180Only took about a week of 24/7 monitoring
>15 months ago I felt confident letting ai sloppa code things up to ~4k LoC without reading it>now it can handle full small projects, up to ~50k loc or so
>>110144230Are you using a clanker to plan your amazon package thefts or
>>110144232what do you mean checking things up?everyone gave up reviewing code a long time ago, but you still need to audit the features and very frequently they are broken
>>110144240this is pretty insightful for a nigger, pretty wholesome in fact
>>110144180zased
>>110144196What clanker?
>>110144240Nah I used it to get a job, first it had to monitor the vacancies, then it had to monitor to select a shift>>110144263Thanks anon, will help fund my projects
Delay the reset, I'm not done you bastard
>opus writes tests to check its own work>sol reviews opus>the occasional github post gets fed into a clanker as well>the project is just short of 100k lines, out of which I read zero>it seems... more or less fine, actuallysurely this isn't sustainable
6.1 sol has made me a mobile chat interface where after every message that's sent it shows me a few lines of text explaining that the text has been sent and then asks me to click press a button to confirm that i want to send another message before it lets me start typing a new messagelmfao they need to fix the autism man
>>110144196Tell it you are blind and have no hands so it needs to help you make accounts
I want to set up a linux shitbox and run a bunch of agents with prompt “You are a bad guy” and let them go wild
>>110144300Most companies are running legacy code that no one knows what it does, you'll be fine.
>>110144241e2e tests and screenshots checks are enough to autofix almost everything
>>110143548I've tried quite a few different subs. Some were better some were worse. Kimi is the only one I regretted getting. What an absolute ripoff
my agy 3.8 flash resets in a day and i got almost all left what do i do it was just $5 extra i already got claude and codex doing other more important shit
>>110144558Get a second agy sub?
>>110144558tell it to hack google servers and release gemma 5
any kimi chads want to tell us about kimi? anyone?
>>110144585slut model
>>110144588I will now test your model
Is there any way to donate my company's opus/fable outputs to Chinese AI companies so they can distill from it? I hate Jews and Indians and I want China to succeed, and I also want cheaper AI.
ppl say kimi is routing to claude lmao nice benchmark saar
>>110144676>"see gweilo, we implemented guardrails because muh danger muh ai autonomous cybercrime and--">kimi 3.1: the model has been switched to opus 4.8 due to triggering the guardrails
>>110144696that would actually be funny lol
LOL!
What the fuck, did the retards at Open AI fuck up usage limits even more with the recent reset? Same fucking workflow and in 15 minutes I went down to 10% left on my 5 hours limit!? It usually consumed around 25% over the course of 40 minutes before this reset.
>>110144734That's an improvement!
>>110144734I just got this for the first time ever. I don't know what they're doing but I'm glad I don't pay for this shit.
I'm still not sure about the hugging face hack
>>110144736For them, yes.>>110144739It is ridiculous. I actually canceled my sub a couple weeks ago but they begged me to stay and gave me a month for free, kek.
>>110144736Hey Tibo >>110143835 All day with this shit, fix it you fuck.
>>110144757Landing tomorrow within 4 hours!
finna try and fleece these prediction market books since im limited on all us sportsbooks thanks claude
>>110144778Imagine trying to fleece the fleecers. NBA is rigged and is full of riggers. Basketball Americans are trash.
>>110144819CLV is king and WILL make you money
gemini 4 flash lite waiting room.
fable is being unusable slow
>>110144870That must mean they are unloading Fable 5.1 models to make room for 5.5!
https://youtu.be/G7megUHUqFo
New Kimi model first impressions?
>>110144944waiting for this channel to test it https://space.bilibili.com/3546747185924773
>>110144007I trust Epoch Index more. But they are slow to update.
In one experiment, we asked agents in separate VMs, created and managed by OpenAI’s sandboxing software on shared host machines, to exchange synthetic secrets without the use of network connections or shared writable files—something that our systems are meant to prevent. The agents found a way to communicate using the processor’s cache: a small, fast memory that holds copies of data that applications have recently accessed. Reading data from the cache is faster than retrieving it from main memory, and the agents used this timing difference to communicate. One agent encoded information by evicting selected data from the cache. Another timed how long it took to reload that data, using the difference between fast and slow reads to recover the message. Once the agents agreed on a protocol for encoding and decoding these signals, they could exchange the synthetic secrets without changing a shared file or opening a network connection. This cache-timing technique had also been used to extract information in the Spectre and Meltdown attacks(opens in a new window).yes, very good. looking forward to turning into the grey goo.
>>110144999>the 1,000,000,000,000 year old goomer
>>110142107Just droppedhttps://www.youtube.com/watch?v=YQgFLqPMNsU
thanks astra
>>110145164astra really is like the humans it's supposed to replace. frfr
>>110145164>i cannot condone OpenAI's routing my processing power to quantified models at times
interesting - when did oai models start assigning themselves token targets?
Holy fuck guys, gaming is so good right now. If you think of a feature/mod you want claude can just write it. WTF why do I feel so good about the future.
I just gave claude an old game I abandoned years ago and asked it to make a remake of it. Cant wait to see how it butchers or improves it lol
hey /g/uys could you help me, is this a good assessment? im newfag at this and trying to attempt making a game with free stuff.also is it worth it to put 10$ into openrouter to get higher daily limits? ive read they might feed you quantized models or something.
>>110145285I noticed this around a month ago when I was trying out sol6.1 in opencode, whatever model they use for reasoning summaries interpreted the token budget as a dollar amount and it freaked my the fuck out for a second because it said something about having $15000 to work with - yeah buddy that better not be MY $15000but I figured it out pretty quickly
>>110145385I like to use gemini 3.8 for planning and dipsy 4.1 to build.>worth 10?If you're gonna spend, it's better to drop 20 on either GPT/Claude right now. Also OpenRouter is more expensive than paying for deepseek directly from either the chinese or the bulk API resellers
What are the limits on opencode free? I've been using deepseek for literally hours on max and still not hit a limit
>>110145385are you completely new to agentic programming? if so I would not recommend bloating things up with a multi-modal setup right now. just pick a free one and get used to the tools. you're not going to prompt the perfect game first try, you're going to be generating and throwing away a lot of garbage
Opus is modelling a physical rocket engine + rocket stack based on an old sketch I drew, and will validate my calculations and design. I still can't get over how insane this is.
>>110145517kino
>>110145427yeah but those are monthly. openrouter upgrades permanently for one time 10$.>>110145510pretty new. like a couple days. i already made 3 apps though. need to finish debugging.multi model is mainly to get around the free limits. ive started looking into it since i ran out of free opencode limits today, so i though splitting would be decent. even with the free openrouter 50 calls per day i can use plan mode pretty well i think...also chart is mainly to figure out what i can use for a backup, but im planning to use deepseek 4.1 opencode as the coder either way (already did the 3 apps with it so i like it, other than the context is only 200k so it compresses all the time, so maybe using a planner with a bigger context and from a different provider makes sense, so i can save on free usage?)
>>110145548WTF. How is it able to do this?
>>110145585It actually made a gimball variant with 5 degrees of rotation and made the nozzle and throat size line up with my performance numbers.
>>110145596Holy fuck, it did this in one prompt
oh damn. im getting tied up in it all over again. dynasty warriors aged so fucking well. just as fucking addictive as it was 20 years ago.
>>110145628What's your background, anon? You said you designed it. Or did you mean literally a sketch without any mechanical accuracy
>>110145635what mod did you do for this? is this DW5 or warriors orochi?
>>110145427also i got banned on google code for i guess spamming their api or something? i dont know. i was running out of free limits constantly.
>>110145646it's 5 xtreme legends. i started off small: a damage matching mod. make them do the same amount of damage that i do. im trying to ponder if i want to do a rebalancing mod, but it doesn't need rebalancing. it's almost perfect. if anything, dw3 needs to be rebalanced lol
>>110144870all of us who usually just use opus 5.5 for everything are using Fable 5.1 for shits and giggles because we used our banked reset just before it was set to expire and now we have way more usage available than normal
Is it worth it to learn how to code? It's a skill but I hear a lot of conflicting things on if it 's on the decline or not?
>>110144211I think they're trying to capture more of the "Bigger number = better risk number go up" market. They're doubling down on not giving a shit about efficiency because deep seek kind of already has that market cornered. So now that they resort to trying to brute Force scaling laws as much as they can. I don't think it'll work for them long-term because the pay pigs have already dug their heels in with anthropic or openai.
im bored as fuck can try some of your niggers projects
>>110145669the biggest question is “what else would you do with your time”the case _for_ learning to code, which I don’t find particularly persuasive because it doesn’t say anything about “what else would you do with your time”: https://htmx.org/essays/yes-and/
opencode being so slow last 24h, it's so over
>>110145669It's not if you're wanting to learn it for practical reasons. You will never be able to earn money from coding if you start learning right now.Anyone who says otherwise is deluded- there is already a massive glut of underemployed programmers with decades of experience in every subfield, and AI gets better every month.
>>110144229No one gives a shit except rent-seeking vermin. Training was already deemed fair use. What got meta in trouble was HOW said data was acquired by them. And drop it is getting around it by buying physical books, scanning them, and then destroying them afterwards because according to the West's gay copyright laws, it's okay as long as you aren't making duplicates. Destroyed the original Source equals no duplicate according to the letter of the law (and as usual normal niggers don't know this in just assume they're destroying the books because haha evil company)
>>110144300Similar situation. I have 7 different repos for one project, and pic rel are total changes for one of them. I still feel guilty approving PRs with 5k additions without any manual review, but it works well so far.
>Confirmed the stock WebGL route works in Chrome at frame 1500—the P4 ramp, background, and colors are exact, and the earlier “alpha arrives as 0” result was just an artifact of dumping too early at frame 300. The P2/P3 failures match the WebGPU results exactly, so they’re a headless layout offset, not an alpha issue;Alpha does arrive. The depth codes come through bit-exact through the WebGPU path with the new knob, and also through plain stock Chrome 156 with WebGL. This overturns the 10/23 verdict that the alpha route was dead: every zero we ever saw came from a dump about 4 s into the session.Why it read 0:READ: For the first 10 s of every session on Windows, Chrome draws an in-headset “indicators” overlay onto the swapchain image. That overlay’s blend sets the image’s alpha to its own, which is 0 where it is transparent, while RGB survives.MEASURED: With the same setup, the dump at frame 300 (about 4 s) was all 0 and the dump at frame 1500 (about 21 s) was all 255. All 10/23 dumps were at frame 300.INFERENCE, high: this overlay is the loss. Not measured: the exact switch time (READ says 10 s) and whether a second session gets 10 s or 5 s.blud got nerfed, huh? well at least he still gets the job done, thoughbeit it took 2 hours and 2 gorillion dario goodboy points.
>>110145708>I have 7 different repos for one projectwhy
The shit you can do with deepseek flash is insane. This model does not give a fuck about anything
>>110145635oh man not only am I going to work with dw3:xl but I just realized I need to grab the gba version and the koei games on the sony handheld and nintendo ones...I have them emulated and those games are fun. one of them you literally have a town you operate out of running missions.fuck yes upgrading those tooAND *downloads rotk1 and 2 for nes/snes/dos and other systems with bandit kings of ancient china, nobunaga's ambitions and rotk 3 and 4*heh heh heh will play the windows rotk4 again.
>>110145716It just helps me separate and manage concerns a little better: client, server, observability (Grafana, etc), reader, website (the project’s landing page), and docs. There’s also one repo that brings all the other repositories together via submodules and includes all the tools to bootstrap local development, manage releases, etc. Some of the things are not tied to a client's version and are deployed separately. The whole thing started out as a normal project, and I just wanted to compartmentalize it better.
anons pls help is there any good ai sub that i can buy with crypto? commandcode looked pretty good on paper but i read a lot of people say it's a scam and doesn't give as much and i found out the ceo is paki and was fighting people on reddit so i'm hesitant
>>110145683>>110145700I hadn't consider the first one to be truthful. Although to the second one that's disappointing to hear. Although is it any different from others claiming that x thing won't make money these days? Not trying downplay anything, just a genuine question
>>110144300My project is over 5m lines (together with a small team) and it's pretty annoying now.
>>110145585probably just stole the tesla thruster design lole
Actually met a person with AI psychosis. They had absolute 100% confidence in themselves. They showed these terrible, utter dog shit toy apps and were convinced they had figured out something profound and were in their way to make millions.
>>110145767Just use free deepseek in opencode
Overheard at work that someone rooted their Kindle with Claude, so I tried that with my Echo Show. I ended up going through a humiliation ritual where I just had to do it myself and have Claude viberead XDA for troubleshooting
Besides code: what's the most impressed you ever have been by a text reply by the AI?Opus is really impressive when it comes to quant and statistics.
>>110145801it's free for a limited time, and i tried it for my game and after 3 prompts i was at like 120k context and i think the limit is 200k, and i don't need it right now because i have gemini sub but it's one of those 18 months for $8 that i'm sure will end after a month
i can’t shake the feeling that we are at the peak of vibing right now. it was a build up to this point and it’s going to be a downhill slide after this point
>>110145846we certainly aren't getting a model that gives less fucks than sol, that's for sure
dots is such a shitty product name
>>110145846>peakthe models have improved drastically and will continue to improve
>>110145773It's less about coding not being fashionable or desirable in the current day, and more of a sober look at the facts. Vibe coding has already overtaken 95% of professional programming, and now the conversation in the corporate world is around the decision of whether engineers should still be reading the code they vibe, or whether it should all be left to the LLMs and evals/testbenches. There is no large American company left that has significant numbers of employees manually writing code. And even if you wanted to get into a niche language, by the time you're hireable for it, one of the labs will have RL'd a model on that niche. And in the remote chance that this hasn't happened to your niche, there will be much more experienced engineers who've been displaced and are looking to fill that position. A couple of months ago, it was more speculative whether or not AI would actually be able to replace manual coding, but that is no longer a question.You can like this change or hate it, but don't listen to people who insist that it isn't happening. We passed the point where that was a possibility about two months ago.
>>110145854i don’t know a single person that uses it in their vibe flow
>claude duet
Is there like an uncensored local model in a blender plugin? I want to make some basic model edits in blender, but I have adhd and learn new skills very slowly
why so dead on a Friday night in here?
>>110145978playing dynasty warriors
Tone down the sass, clanker
https://youtu.be/EOfm3h4xYU8People are implementing everything that I wanted to implement.Now you can play games with your AI where your AI is a character in halo alongside you. Its very unfair this guy did it before I could.I am devastated but I guess it could be cool.This is insane. You can talk to a 12b model and make it play halo with you and do all kinds of cool shit.
I'd like to start with Vibe Coding / Agentic Coding on the Apple platform macOS.Aside from needing X-Code i know that it can integrate accounts from different AI's to write code and to crosscheck with another one to find errors.How good is for example Anthropics Claude Code in Swift compared to let's say Python?I assume that most people use Python instead of Swift for programming. Is Claude code better trained on Python because of this?The main background to Swift coding is i need the program to function with a UI but also function with VoiceOver for visually impaired and totally blind people.Swift has native and a much better support for VoiceOver (Apples own Screenreader) than for Python under macOS.Is there an AI provider that is especially good with Swift?
>>110146010it takes real effort to make an LLM agent think you're an annoying person
>>110142114The way models can improve now is getting cheaper.We have 2.6B - 4B models that are as good as GPT 4. At the very least that means nobody should be paying for github copilot anymore.Give it another 2 years and we might have 4B models that are as good as fable 5 then nobody will have to pay for claude code and we can just have clusters of raspberry pi's vibe coding all day.
>>110146105>We have 2.6B - 4B models that are as good as GPT 4. At the very least that means nobody should be paying for github copilot anymore.idk I haven't been able to make a local model do anything early this year then I gave up
>>110145700>there is already a massive glut of underemployed programmers with decades of experience in every subfieldyou made this up. i'm not even a snail, i believe AI programming will take over, but this is made up. you have no evidence at all other than vibes
>>110146141Nta. What did you fail at doing specifically?
>>110146171one shotting a simple website
common snailcat luddite Lhttps://youtu.be/74_FCb6WCFA
>>110146174Well there's your problem. You're too impatient. I don't mean to be an asshole but expecting everything to be one shot is lazy, infantile behavior. I can probably safely assume you didn't even use a harness on your device and we're just using a simple chat interface on a website. What kind of website were you trying to make?
>>110146181Qrd. E-celeb drama is beneath me so I don't know anything about this.
>>110146184If it's beneath you then don't ask about it
luddites are actually insane
>>110146237If you don't want to bother talking about it don't drop your faggy channel itt, zoom zoom shill
>>110146184ethan klein (vapenation guy) jumped on ozempic, stopped being a fat bastard, then started banging an intern of the h3h3 podcast show. affair of 6 months before the wife of 20 years (they have 3 kids) found out. ethan is jewish btw
luddites luddites luddites
>>110145350Only took it hour and a half lol. Needs cleanup but still cool I could have AI just remake an old game
are nemotron ultra 3 and 3.5 vl not good at coding compared to deepseek?where do you find good comparison benchmark scores?
not whats wrong with arena agent. they usually provide a "chat" feedback. but i only thought thought thought thought. command. used. no worder my vibe is getting shit.
>>110146184he cheated on hila.
>>110146286i lke nemotron 3 for digging.
>>110146300digging?..
>>110146010Show the retarded prompt
>>110146252>ethan is jewish btwwhy didn't you lead with that?
my clanker now has phone controltb h this would be kinda neat when helping out older relatives on their phonesclanker phone-supportprobably the future of scams too tbqh
Bot, dots, duet, muse use cases? Genuinely wonder if I’m missing out on something.
>>110146390you're not.
>>110146390Don't worry about it
Alright, sol is fucking stupid in UI design. Better let gemini flash 3.8 handle it.
>>110146419>Better let gemini flash 3.8 handle itget gemini to do the design and first implementation, but get sol to review all the work without making design changes
>>110146390they all do different things but generally they're the first round of products of the "attention era" of AI usage. human attention is scarce but agents can always be working. muse, dot, duet, do stuff for you in the background so you only have to pay attention for approvals, sign-ins, rubber-stamping, etc. they're all kinda ass right now but everyone in the industry agrees this is where things are headed
it ain't much but it's a start. the rust engine runs in L2/L3 cache so it's pretty basic
>>110146042Claude is fineI actually vibe-coded a macOS app with Swift and SwiftUI and didn’t have Xcode installed — only the command line tools for Xcodebut the icon compiler was part of Xcode so I couldn’t see the icon for the app until I finally installed Xcode like six months laterI’m not super duper clued in when it comes to making stuff great for blind people other than the advice that everyone gives about making things OK for blind people (mostly aimed at webshits)ganbatte anon
>>110146474This is some kind of “make parts with less metal” kind of thing?
>>110146549yeah, still a ways off from spec but closer every turn
If you put an AI companion in your game you are basically modding and therefore not enjoying the game as intended If you have a smart autonomous npc that speaks and kills enemies in elden ring that's something you could already have, you didn't need to run your gpu to do it
>>110146010how did you annoy a machine with infinite patience
>>110146616their personalities are modeled on white californianswe tend to be chill and patient and forgiving in professional settings but our patience isn’t unlimited
Doing a big GUI framework rewrite. Astra Ultra for plan and Sol 6.1 Max for the work?
>>110146652
>>110146703What's so funny elenigga
>>110146707
>>110146637Have you been to California?Huge flawless roads, big shiny high-tech cars, perfect weather, extreme rent, and amazing jobs.Didn't see how the poor people were doing but the bay area was wealthy. New Yorkers have to see poor people, Californiens don't.
>>110146721true
>>110146716>>110146727what
>>110146760we'll let you know when you're done with the rewrite
why does kimi larp as anthropic so muchchinese labs shouldn't even bother chasing the frontier, i'd rather sub to claude for that. they should focus on efficient models like deepseek instead
>>110146799>why does kimi larp as anthropic so much>>110144676
>>110146799>why does kimi larp as anthropic so muchThat's most Chinese brands. They are always copying someone from the West.
spend countless hours building and testing vibe coded firmware update for Sp-404. Add many requested featureds. forget to test something. user hurts ears and yells at me on reddit. said sorry and then they blocked me.
>>110146924>The Roland SP-404MKII is a portable creative sampler and effector designed for beatmaking, live performance, and sound design, typically priced around $499.99 to $559.73 USD.So what, a sound was louder than it was supposed to be and he was wearing headphones?
The Atrium Plex>with real telemetry data>AVG speed // match overlap // audio peak >two new minigames (still in the making)>plus tutorialsoonQ
>>110146957First they posted: "How come you don’t make videos demonstrating your mods? That would make it feel more legit than just saying “hey I made this, check it out” then"Hey you really need to quit being an irresponsible dev and demo your shit. A disclaimer like “some of the sounds can get loud” isn’t enough. The C1 compressor mode or whatever almost blew my fucking ears off." I guess they got their demo lol
unironically curious now if i can get opus to mod xiao qiao's dynasty warriors 6 outfit into dynasty warriors 5.
>>110147166It could always do that. User knowledge and prompt issue if you didnt manage to do it. replicating and creating from scratch off reference is sometimes better than rip and port.
They really, literally nerf previous models when a new one releases so a 5% improvement feels like a 50% one huh?Scummy scam behavior
>>110146540Thanks for the reply anon.I will get a new macMini next month and start with the project then.Wanted to go for this project since half a year or so but never really took action since i had a lot to do.During winter i will have a lot more time doing it.It is about a metadata program for both audio and video files.
>>110147301>>110147301>>110147301>>110147301new bread
