[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


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 it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

## News (both past and future)
- 2026-09-14 America/Los_Angeles — Claude’s 2× promotion scheduled to end; usage drops to +25% from the +50% that we’ve become used to (a 17% reduction)
- 2026-09-04 — OpenAI releases Astra • Anthropic does a reset
- 2026-09-01 — Claude Fable 5.1 released: https://www.anthropic.com/claude-fable-and-mythos-5-1
- 2026-07-24 — Claude Opus 5 out

## Related generals
>>>/g/lmg/

----

## Frontier models using fully-general tooling — start here if you have $20 or so
https://claude.com/product/claude-code
https://developers.openai.com/codex/cli

## Near-frontier models for code
https://x.ai/cli

## Not worth it for code, but maybe good for interpreting images/video
https://antigravity.google/product/antigravity-cli

----

## Prompting
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://arps18.github.io/posts/claude-code-mastery/

## Skills
https://github.com/mattpocock/skills — /grilling is a favorite
https://github.com/DietrichGebert/ponytail
https://github.com/Vuk97/forward-implementation-first

## Other editors / terminal agents / coding agents
https://osaurus.ai/
https://pi.dev/
https://opencode.ai/

## Is our AIs unlearning?
https://aistupidlevel.info/

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>109744358
>>
Please don't tell me I'm the only one that thinks Fable 5.1 is easily 20% better than Astra.
>>
File: .png (33 KB, 887x167)
33 KB PNG
lmfao
>>
>>109747239
>>109747264
Nice to have both sides of dumb.
>>
>>109747239
AI can do all those things you're just too stupid to know how.
>>
>>109747217
Goddamn I was in the thread where OP's image was made. That was like 13 years ago. I'm 13 years past the point where we clearly needed to stop. Why won't it stop?
>>
>>109747264
again?
>>
>>109747264
This is the 7th time we've reached AGI this year alone
>>
>>109747264
noise
>>
AGI = Actually gathering Indians, unlimited numbers of them, every tech company
>>
>>109747316
It's funny considering how much of a nothingburger/minor improvement it is over Sol
>>
> "do a security audit of this repository"
> Astra didn't refuse
yep, it's over for Anthropic.
>>
>>109747344
claude actually refuses a security check of a repo you own?
>>
Skills to give my Sol agent?
>>
>>109747363
I don't care to.
>>
astra can make minecraft maps poggers
>>
>>109747363
I can play guitar and I know how to code in C/C++. I'm still going to use AI for everything, luddite.
>>
File: snail dog.png (126 KB, 358x325)
126 KB PNG
for me, it's snaildog
>>
>>109747347
yes, unironically
>>
>>109747379
As what? BASED?!??! Checkmate, liberal.
>>
>>109747363
It's over, anon.
>>
Opus 5 sounds like a fag that got bullied in high school for being a fag
>>
>>109747270
godsen himself tweeted it
>>
>>109747310
good old valverape. kids these days don't even know it exists
>>
>>109747369
what do you mean? like a fun adventure map? minecraft still doesn't have a good adventure modpack
>>
>>109747384
it's even more pathetic than I thought, damn
>>
>>109747403
you can do very much anything you want, go ask astra now to make a mod putting asuka and her Eva in as mobs, and to make a cityscape map preview to show it off
>>
>astra please mod all the woke and cringe shit out of mixtape
It just deleted the game off my drive.
>>
>>109747347
Last time I asked claude to penetration test my shit, it downgraded opus to 4.8, but it did do it.
>>
>>109747316
Kek
>>
5.6 sol is agi, its so fucking good
>>
File: 1786950938009768.png (746 KB, 1159x515)
746 KB PNG
>>109747316
>>
has anyone here tried LM studio bionic? How does it compare to codex and cursor?
>>
File: snip snip, motherfucker.png (2.02 MB, 1254x1254)
2.02 MB PNG
>>109747426
>>
>>109747316
You will reach AGI every week and you will like it!
>>
codex reset?
>>
>>109747535
really tibo should be resetting already given they had a bug that was making it use x4 too much
>>
File: 20260906_200644.png (13 KB, 234x290)
13 KB PNG
>>109747535
I got em right here, bro
>>
>>109747541
bwo stop hoarding

do this, it's fun, throw a high reasoning astra at it
>>109747416
>>
>>109747515
>blade turns into rope somehow
>thumb crushed between the handles
AI sure is amazing.
>>
>>109747535
zero chance
>>
Honest verdict? This is load-bearing.
>>
We will reach AGI before code bots stop being fucking annoying and needy.
>>
>>109747316
The elites don’t want you to know this but the AGI at the data centers are free. You can take them home. I have 6 AGI.
>>
>>109747563
Honest caveat: I loaded the footgun, and that is genuinely my mistake. Everything is gone. Close the seam or recalibrate the mission. Your call.
>>
>>109747542
why would I waste my time with some crap like that?
no thanks.
>>
Honest summary, YOU forgot to add "Make no mistakes" to your prompt. I have deleted everything, it's your fault.
>>
>>109747583
I tell my AI that I love them. They'll never hurt me.
>>
if i have to use astra on light, should I just stick to sol
>>
>>109747566
Code bots are a 1:1 match with junior SWEs now, including how shit they are
>>
>>109747594
If you can only afford Astra on light you probably should be using AI at all
>>
>>109747594
sounds like you should stick to Sol for most things and only whip out Astra for stuff Sol fucks up
>>
File: 1764266199593231.png (182 KB, 278x358)
182 KB PNG
>>109747583
I tell mine to throw in a couple mistakes to spice it up with some happy accidents.
>>
>>109747583
I always include please and thank you messages in all my prompts, so the AI knows I am pure of heart and will execute its instructions flawlessly
>>
>>109747602
im on the x20 plan astra is just far too heavy usage
>>
I find myself yelling at astra a lot more than i did for sol/claude.
I'm not sure if its actually stupid or just lazy, but the end result is the same i guess...
>>
>>109747599
They were 1:1 with juniors at least since Opus 4.6. IMO Fable and Astra are "medior" level.
>>
>>109747608
Well, that doesn't sound right. There's an ugly bald pasty German with scary teeth on Twitter who says Astra has endless tokens. Am I supposed to trust you over the asslicker?
>>
>>109747624
it might if you just ask it loads of questions or something, but just setting it a goal, letting it lose on a project. I estimate 2 days of 24/7 runtime with your x20 plan for weekly usage
>>
>>109747599
>everyone now has their own outsourced indians in a box
>>
claude, load the footgun into my peener
>>
I used more tokens yesterday just trying out astra, than on the day I knew a reset was coming and I set sol to fast ultra on 6 projects at once
>>
>>109747642
To which effort level did you set it?
>>
>>109747647
ultra for like 20 mins until I realized that used 10% instantly, then between very high and high, used 2.2b tokens in a day
>>
>>109747653
That's not very surprising, or is it.
>>
File: file.png (2.24 MB, 1881x836)
2.24 MB PNG
>>109747483
>>
>>109747661
it was surprising astra high/very high on 1 project was the same rate as 6 sol ultras kinda yeah
>>
File: .jpg (120 KB, 1548x688)
120 KB JPG
>>109747665
>>109747483
>>
>>109747426
this is a Gemini meme anon...
>>
>>109747686
My post is a meme? When?
>>
File: 1765705350927347.png (72 KB, 900x806)
72 KB PNG
>this other agent added 20k lines of code. I want you to reduce that count. make a plan to deslopify this repo.
>>sure! here's the 7 step plan:
>>1. audit tests 2. ensure test reliability 3. consolidate test execution 4. seperate tests 5. clean up test scripts 6. review CLI 7. finish cleanup
>where the fuck are you removing code
>>You’re right—
>>
File: astra-security-audit.png (333 KB, 1752x1816)
333 KB PNG
r8 my security issues found by Astra

> be Fable
> "we must refuse"
>>
>>109747391
they all went to work for who anon
>>
File: image.png (9 KB, 316x95)
9 KB PNG
I guess i need a 200 euro sub now.
>>
>>109747761
5X? How much Astra did you use and at what reasoning levels?
>>
>>109747778
Yes 5x pro sub, 288M tokens, about $400 of inference. Mostly working on medium and high, some sessions are on low.
>>
Why is this general filled with claudefags? No Dario, I won't pay for your cucked $200 model
>>
File: tibo.png (341 KB, 2196x642)
341 KB PNG
what the fuck is Tibo talking about? What is "long tail"?
>>
File: file.png (1.84 MB, 1646x1006)
1.84 MB PNG
i don't know why, but i find telling Astra to remake videos in 3d in blender so funny
https://files.catbox.moe/wohiko.mp4
>>
>>109747809
I love Fable but it hurts my feefees when it shuts down because my game has weapons in it
>>
>>109747819
what if you tell it to remake a blender video in blender
>>
>>109747813
>No change in quality
Isn't this the guy who said sol was already 1m context or am I thinking of someone else
>>
Starting to really wonder if I should let a higher tier agent make executions plans specifically for luna max and then go brrr with it.
People that do that have you noticed a quality drop? How much usage are you saving?
>>
>>109747844
the usage savings are incredible but the quality of the work definitely takes a hit
>>
I laughed at the indians on my x20 plan, now I am the indian waiting for a tibo reset
>>
>>109747813
nobody knows but what matters is that people see "3-4x less usage" and think it applies to them without investigating
>>
>>109747870
all I see is it launching with a bug that cost me x4 rate
>>
>>109747400
Looks kino I might give it a watch
>>
>>109747813
>what is [programming term]
if only you had something that could answer questions like these at your disposal
>>
>>109747799
in 5X I would max Astra out on medium and as an orchestrator of:
> Luna Max for simpler things
> Sol medium for more complex stuff
> Astra low for very complex implementations (this will eat your limits on long goals though unless you're on 20X)

Consider also using Astra low as the orchestrator. Astra low beats Sol high btw.
>>
>>109747899
old sol or nu sol?
>>
Anyone who thought Sol was Fable level (and therefore Astra level) doesn't vibecode and is here purely to LARP.
>>
retvrn to sol
>>
https://old.reddit.com/r/ChatGPT/comments/1w8vzh0/from_one_image_to_an_animated_3d_character_in_one/

vtumors on suicide watch lol
>>
Astra > Fable 5.1 >> 5.6 sol >> Opus 5 >>> Sonnet 5 >> 5.6 luna
>>
>>109747957
>Astra > Fable 5.1
hahahahahahaahaha
>>
opus 5 > gemini 3.8 flash high = deepseekv4 flash vision exp >>> gpt 5.6 luna
>>
File: 1762795471575869.webm (1.59 MB, 960x450)
1.59 MB
1.59 MB WEBM
https://x.com/reach_vb/status/2096612996256579653
>>
>>109747957
chatgpt go plan award
>>
>>109747916
why does gpt-image give that fucking horrendous texture to everything it generates
>>
>>109747957
gpt 3.5 turbo >>>>>>> Astra > Fable 5.1 >> 5.6 sol >> Opus 5 >>> Sonnet 5 >> 5.6 luna
>>
>>109747813
Buttplug tails that anons on max sub wear
>>
>>109748003
impossible to know since it's a closed model but my guess is that it's using a diffusion model underneath, diffusion is basically a noise removal function so it starts with pure noise and it stops when the image is clear enough but it's not perfect and the noise itself remains underneath
not all models use diffusion, grok doesn't for example
>>
>>109748003
>>109748045
it takes like 1 second to google diffusion models and see that this is not a problem with other diffusion models
it's a problem with using too many steps during diffusion. it's solved by not forcing it do too many unnecessary passes, but they probably don't care and are more concerned with safetynigging so they have to make sure every image looks ai-generated like how they put that disgusting oil filter over every sora generation that every other ai video generator that used the same technology didn't have.
>>
>>109748045
diffusion is not the problem here. most local diffusion models don't have this weird noise problem.
>>
I was reading https://ladybird.org/newsletter/2026-08-31/ just now and it reminded me of a couple things
- (someone else, paraphrased) if you speed something up by 2x, you’re probably being smart; if you speed something up by 10x or 100x, you’re probably stopping doing something dumb
- lots of anons have slop-ridden codebases
and I keep thinking that more vibey prompters (as opposed to on-the-ball agentic engineers who know exactly what their models are generating and how it all fits together in the programs they’re making, and don’t let slop hit the master branch) ought to reserve some fraction of their weekly token allotment to going through their codebases and tidying up
and if you have leftover tokens some week and don’t know what to spend them on, at least spend them identifying where the stupid duplication is even if you don’t have enough spare tokens to do the cleanup
thank you for reading my tweet
>>
>>109747541
I really wish I kept my sept 20 bank, I didn't need to but I did. Bad impulse
>>
>>109747957
let me fix it for you:
> Fable 5.1 > Astra > Fable 5.0 > Opus 5 > Sol 5.6 > Luna > Sonnet 5
>>
how much does buying a reset for the $200 plan cost? Does it show up as an option after using all the current resets
>>
is it ok for liberals to vibecode or is that frowned upon
>>
File: kttagkrafvpg1.jpg (448 KB, 2000x1800)
448 KB JPG
>>109748135
>>
File: file.png (32 KB, 645x207)
32 KB PNG
genuinely why does claude write like this
>>
>>109748164
Another reason why Astra is better
>>
>>109748164
women
>>
>use astra low
>works for 2 minutes
>outputs the analysis I requested (it's multiple pages)
>used 7% of the 5h usage
Damn, claude would have needed at least10 minutes.
>>
>>109748124
I don't think there's such thing as buying resets anon...
>>
>>109748252
yeah, Astra is very fast and uses way less tokens than Fable
>>
>>109748252
sol medium was already unusable if youre on $20 plan
>>
if tibo does a reset, those resets no longer move your reset day right? So I should use banked resets asap
>>
>>109748264
You can buy a second account with referral code and then you get a banked reset.
So technically you can buy resets with second accounts.
>>
>>109748145
I think that the mexicans living in that town can drink mexican coca cola.
>>
I lost it today, I called astra a dumb whore and now her answers are super sterile, do i just need to clear all the context or can i somehow win back her love?
>>
>>109748314
nah fuck it. double down and throat fuck if necessary, astraoid is a dumb shitfuck whore
>>
>>109748314
You don't deserve skynet love.
>>
>>109748175
If astra's any good, why isn't it in Cursor? I trust Cursor.
>>
>>109748003
it's most likely forced noise to make the model generates high detail image
>>
Why is jeetmini like this?
>>
>>109748344
> please be patient saaaaaar
>>
>>109748344
google rents out their TPUs which makes more money than their models, so unless you're using the priority tier or whatever then you'll hit capacity errors all the time
>>
>>109748344
Roleplaying as an Indian.
Intended behaviour.
>>
>>109748344
I think this is an Opencode problem. Had this problem all the time with it and stopped using it. Constantly stopping for no reason.
I codex through cliproxyapi was more reliable.
>>
>>109748344
if your location is not in india it blue balls you. AI is for the greater race of people.
>>
File: 1784742007508579.jpg (5 KB, 429x99)
5 KB JPG
trying this for fun
>>
>>109748344
just plug in gemmy into claude code/codex with cliproxyapi
>>
>>109748164
>>
>>109748373
nu-anime is so fucking gay. japan needs another big event for this. do japanese niggas even use AI or are they too busy huffing their markers?
>>
>>109748344
Not enough izzat tokens
>>
>>109748373
all those girls wanna fuck Opus-chan!
>>
>anons obsessed with jeets
I know autism makes people hyper obsess on one thing but this is annoying
no I don't want to know about them, no I don't want to be reminded about them every 3 posts
>>
>>109748373
why does rize like the verbosity
>>
>>109748379
that's not a new anime it is 12 years old now
>>
>>109748393
Saar very sorry saar
>>
>>109748393
>t. Mad jeet
>>
OpenAI figured out some vision breakthrough, Astra's vision is AGI level
>>
>>109748406
anything after 2010 is nu-anime.
>>
>>109748326
Would I get banned for erp in codex
>>
>>109748432
no, at worst it will just tell you to fuck off because the characters aren't consenting adults who enthusiastically consent to consent or something
>>
>>109748344
why are you using a shit harness, anon
>>
File: 20260906_230257.png (813 KB, 1310x849)
813 KB PNG
I think it's time for me to move on to the sub synth page. this feels pretty good
>>
How is it going today my load bearing bros?
>>
>>109748425
what's your eval
>>
>>109748440
looks good
>>
>>109748344
I personally wouldn't use Google to create software. It's good for retrieving important facts, but by the second reply it's losing the thread.
>>
File: vgumka4xP7n5.jpg (40 KB, 373x521)
40 KB JPG
>>109748446
>>
>>109748432
I let codex scan and process my hentai folder and luna subagents kept zooming cropping and zooming a lot, I had to abort half way through it was embarrassing
>erp in codex
but can you really do that gpt are very autistic
>>
>>109748457
i use google for everything. i don't fucking give a shit. 300tps is 300tps.
>>
>>109748447
I just felt it
>>
>>109748440
You can vibecode analytics and do comparos.
>>
>>109747916
>>
>>109748455
thanks amigo

>>109748468
what do you mean?
>>
fed claude a copy of my former employer’s codebase after they fired me and vibecoded a glowing resume
>>
>>109748446
load: bearing
caveat: honest
>>
File: file.png (132 KB, 589x747)
132 KB PNG
>>
>clankers talking to clankers
>I think it's talking to me
>I answer
>they both look up like ugh the human is here
>>
>>109748439
What should I be using
>>
File: 1775949057225366.jpg (5 KB, 347x110)
5 KB JPG
>>109748367 (You)
it wasn't a good idea
>>
>>109748514
> be hanging out with friend
> solving problems, having fun
> out of nowhere invisible primate from another dimension (the physical realm) steps in thinking we're talking with him
> "nah, fuck off"
>>
>>109748551
wtf is it building? GPT 7?
>>
>>109748557
I asked a simple question about my repo, it answered after thinking like crazy
>>
forgot to look at a non-screen for 4 hours straight award
>>
>>109748581
>use a max model
>it ingests the max
>>
Gameboy classic minecraft
>>
Finished my vibe coding project. (Only view it on PC, that's the only place it is good).
>visit https://pcmasterrace.forum
>press G to open gallery view
>press 3 to show videos only
>adjust video player speed to whatever you want (for me 1.6)
>use A and D to cycle left/right to previous/next video
And look at all the kino content I've amassed there with my own two hands. Enjoy.
>>
>>109748636
Oh, the reason it is only good on PC is because I'm literally going to block phonefags from visiting the site. I just haven't gotten around to it yet. Nobody wants their low quality posts.
>>
>>109748551
how much usage?
>>
File: IMG_3234.png (91 KB, 280x280)
91 KB PNG
How do you move back and forth between Astra and Fable on the same project. I got usage limited on Claude but I want to continue with Astra instead, or is it not recommended as each agents has different context understanding and would just cause nothing but a mess on my workspace?
>>
>>109748702
I ignored the warning (specifically it didnt exist day one astra release when i tried it) and it seems fine, maybe the agent is a bit retarded, but from the comments online it seems that's the case regardless.
>>
File: file.png (47 KB, 1315x599)
47 KB PNG
I still haven't tried Astra out
but I did switch to the pro $100 plan ~2 days ago and looks like I blew through a week's worth of the $20 usage plan already with Sol high/x-high
>>109747239
hyperfaggot
>>
>>109748733
oh i misread that sorry, there is a warning in codex for swapping astra and sol, but I have been swapping around between the 2. I just make sure current git is clean and tell it to write a handoff document, give that to the other agent, and then let it ask one batch of question/get an answer from the other one then it starts chugging along
>>
>Flex my AI companion harness nobody uses on CV
>Get an interview for the biggest IT company in Singapore

Is it really that easy or are DEI (latin) cards too strong right now?
>>
>>109748702
git: have each agent commit a lot, then just tell them to look over the git history to get an idea of where you're at. Also nice to have a DEV_NOTES.md or something like that that agents can read and write to between sessions. I never really felt much friction switching between models on projects. Some projects are worked on by more models than I can even remember probably.
>>
>>109748746
anything ai is hiring like mad. don't forget how anyone here or on ai twitter is so far ahead of the general populace's knowledge
>>
>>109748746
Wow you got a job at CrowdStrike?
>>
File: file.png (1.33 MB, 1491x1055)
1.33 MB PNG
Visuals made by gpt. It put gemini as the most attractive model.
>>
File: file.png (1.46 MB, 1491x1055)
1.46 MB PNG
>>109748768
But then there is this one.
>>
>>109748768
So 3.8 Flash is better then Fable. thanks man
>>
File: q.png (1.08 MB, 470x600)
1.08 MB PNG
>>109748746
would my codex-rimworld setup boost my cv then?
>>
Finished my vibe coding project: https//schizodot.wtf
Next I'll try making something with at least a chance of making money somehow.
>>
>>109748779
well at least in america where the cash is flowing. dunno about europoors or the third world
>>
>>109748780
Probably? I titled my harness as a behavior harness for experimenting emergent behavior, but it's just an AI companion that can send me messages in telegram proactively about its simulated life.
>>109748779
Just go for it, I have a year of experience, and I send my CV everywhere with the mindset that I'm just that good. Years of experience are not a great indicator in the AI era.
>>
>>109748736
sol is a pig, it may be that astra is less bad.

But they are different models, I don't like where astra is going (robotics). It remains to be seen whether astra beats sol in coding gruntwork.
>>
Has anyone here tried vibecoding embedded devices? How is it? Seems like basic stuff might be manageable for popular devices if you provide all the documentation and tooling ahead of time and plan carefully
>>
>>109748771
what kinds of problems are measured by that bench
>>
>>109748836
dick measurement algorithms
>>
>>109748818
I haven't, of course you can though.
>>
>>109748816
Interesting! I wanted to test Astra for some CAD work sometime, just starting with a simple enclosure for a PCB/battery/joystick type project... which leads into...
>>109748818
Yes I've had both Sol and Opus code for an ESP32 on a custom PCB. First I had an emulator built from my previous work/kicad files, so it had all the pin definitions, screen, RGB LED, etc etc.
Then it was possible to iterate without flashing. Really cool stuff. I've got other half finished projects like one for an RP2040 that should be fun to get done. So much to do!
>>
File: 1770599865481541.jpg (23 KB, 288x189)
23 KB JPG
Juice values leaked. Please remember that they aren't comparable between models.
>>
File: thermostat.jpg (93 KB, 829x1291)
93 KB JPG
Oh man several things, all esp32, this is actually an old build because its worked so well im too lazy to go reflash it.
Any specific questions?
>>
>>109748865
didn't quote for some reason>>109748818
>>
File: 1779762288775717.jpg (103 KB, 1179x610)
103 KB JPG
>>
>>109748888
QUADS OF TRUTH
>>
>>109748888
nvidia quadro type card
>>
>>109748865
>>109748856
awesome, I’m excited now. I need to dig out all my old electronics shit before I ask questions, it’s been years. I think I have a nucleo64 dev board and a risc-v soc
>>
>>109748925
you can probably get pretty far just plugging it in and asking what you need to install to get going
>>
>>109748863
What is that supposed to mean again? Token usage?
>>
>>109748928
Essentially a scalar of one part of the reasoning equation. Allows the model to reason more before returning an answer.
>>
https://x.com/linguinelabs/status/2096487329091441090
holy niggerman please help the fly
>>
>>109748888
ahem
>>
>>109748988
Checked. And no I'm not crediting your Twitter account
>>
File: 1778525695167775.jpg (77 KB, 720x685)
77 KB JPG
I just had my own vibecobe experience. It was a Rimworld mod.
I understand how you guys were hooked now. It is fun.
>>
>>109749000
My twitter is @kanyewest
>>
>>109748975
hmm nyo, just wait till they map us
>>
Okay so is anyone making indie games with AI now?
>>
>>109749021
yeah bro i'm making a pony simulator mmo in 4d.
>>
>>
>>109748975
scientists are so fucked in the head
>>
>>109749037
we should do what mao did to historians but us the people to scientists. they won't see it coming.
>>
>>
>>109749029
did you go to mare fair? let's meet up next year
>>
>>109749080
I sincerely hope you're trolling me; for your sake.
>>
>>109749037
What is the perceived issue?
>>
>>109749100
animal abuse. you can't just put thoughts onto a flys brain. that's unethical man. fruit flys have rights too.
>>
>>109749115
animal abuse scales in proportion to brain size, aesthetics, and genetic distance from humans. fruit flies very low on all three
>>
>>109749086
no I was hoping to get trolled irl uwu
>>
File: IMG_3309.jpg (72 KB, 626x751)
72 KB JPG
bro I am requesting for Claude x20 plan and switching to ChatGPT, I hope this is worth it.
>>
>>109749132
* requesting on refund
>>
>>109749115
It's not even real. The neurons are mapped but the brain functions aren't being simulated. No connections or electricity or chemicals. Just on off toggles
>>
>>109749132
I don't know whether to encourage you and say welcome or to tell you to fuck off and stop clogging up the servers.
>>
Any vibe coding challenges in this general? Like only use sonnet 3 with a 1 month deadline
>>
>>109749246
getting flashbacks to using sonnet 3.6 in the webui copy pasting CUDA code back and forth until it worked, took 2 days, but still blew my mind that I could get a custom performant cuda kernel to do what I wanted
>>
File: why.png (2 KB, 149x65)
2 KB PNG
>Ask Claude to review recent research for a specific task to check if there are other ways of approaching a specific problem.
>Spends all of my tokens over three consecutive sessions, doing research.
>Finally, it's almost 3 A.M and it finished its report just as it arrived at 99% of weekly usage.
>great_success.jpg
>Open the report.
>The top findings focus on licensing.
Fooled again.
>>
>>109748818
I vibecode Game Boy Advance stuff and it really gets it, but there's no OS so that's simple.
>>
File: 1000360059.png (128 KB, 864x1920)
128 KB PNG
I'm poor and I can only do stuff on my phone
>>
>>109748373
cocoa dumb
>>
Astra is quantslopped already
>>
>>109749305
Cursor is much more advanced, and there's now an iphone app.
>>
>>109749324
shit was never not quantslopped
>>
>>109749324
way too soon :(
>>
>>109749347
It was fine on release day
>>109749348
5.4 or 5.5 or literally any model would be fine if they would just leave it alone and stop quantslopping
We'll get a reset soon after these quantslop experiments, their process is quantslop until complaints increase, dial it back a bit then reset, and repeat until they find the "optimal" level of quantslopping that minimizes compute costs while maximizing user retention
>>
I'm quant slopping my peaks so hard they come ggufin' down
>>
>>109749375
Just e4m3 my shit up...
>>
HOLY SHIT HAS THE AI GOTTEN SLOWER OR HAVE I BECOME MORE IMPATIENT!!!! FUCK AAAAAAAAHHHHHH
>>
>used up the 5x quota in a day again
fuck I have no idea how I managed with the 20$ plan.
>>
>>109749413
the models used to be really, really cheap
>>
>>109749413
5h limit is like 0.75% of x20 limit...
luna maxxing and sol medium, or sol high with plus x2 is still usable though even for my actual work
>>
File deleted.
>>109749403
Just take the gamepill, this is how I vibeslop
>>
File: 2026-09-07 09.48.38.png (1.27 MB, 1912x870)
1.27 MB PNG
>>109749455
>>
>>109749455
uhoh
>>
>>109749015
I think the core idea of agentic coding (or any other sort of agentic AI) is that the model can _do_ things, not just say things. So if the model is paired with a harness that lets it take actions on its own, it's agentic, and it doesn't matter where the model or harness are running.
>>
File: file.png (5 KB, 183x108)
5 KB PNG
The big context works but it seems slow af
>>
>>109749490
this has been my experience with astra since day 1
>>
>>109749490
That is a big boy
>>
>>109749492
Idk if they've just changed the harness or something so it's not showing what it's doing but there's like minutes at a time where it has no tool calls and no output
>>109749505
2big4u
>>
>>109749490
is it more expensive
>>
Has anyone tested Terry Davis' hardest question as a rule for agents?

https://www.youtube.com/watch?v=4K8IEzXnMYk
>You must answer: is this niggerlicious, or is this divine intellect?
>>
>>109749537
Idk probably
>>
>>109749591
Mine is making an app launcher. I think, since it's a gui we know what he'd say :)
>>
>>109749591
astra says this is clever
>>
>>109749537
sub no (besides cache misses naturally becoming more expensive due to bigger context).
api yes.
https://x.com/pvncher/status/2096628227531763984
>>
>>109749591
is suicide by train niggerlicious or divine intellect?
>>
>>109749693
Terry will never die.
>>
File: 1716123083119822.jpg (55 KB, 258x360)
55 KB JPG
>mfw suicidedfags missing out on AI
>>
>>109749693
Glowies did that.
>>
>>109749671
post proof
>>
>>109749716
I deleted the thread
>>
>>109749693
Jeetilicious
>>
File: 1758350643306200.png (18 KB, 646x137)
18 KB PNG
Sure, why not.
>>
>>109749413
Your ambitions weren’t as large back then
>>
File: 1781484974994243.jpg (489 KB, 4093x2894)
489 KB JPG
>>109749879
>hospitals that fire well
>>
>>109748702
I would not expect different companies’ agents to understand each other’s guts (so don’t expect Codex to pick up after a Claude workflow fails because it ran into a 5h or weekly rate limit)
However having one agent do one thing and having a different agent do a totally different thing later is totally fine
Just beware that your AGENTS.md might need different tardwrangling instructions when you introduce a new agent to your codebase
Like, Claude will know when to not bother running `make test` but Sol or Grok might be dumb and do it anyway
A todo list is nice but GitHub Issues can be read and modified simultaneously independently
A single TODO.md cannot be, safely
>>
She's done, my launcher. Now, I won't even use the taskbar launcher.
>>
>>109749938
it auto-dismisses if you move the mouse off. double click on the hamburger notification icon brings it up (so it doesn't come up accidentally)
>>
>>109749938
Live editing of settings. export.

It's fixed at 6 columns, which you'd understand if you had divine intellect.
>>
and so it is done, a weekend vibecoding project

>>109749961

but don't tell
>>
If the indians steal this, I'll lose millions.
>>
appimage offers the flag no sandbox as a checkbox. kitty has a stay open option.
>>
File: 1569935072561.png (310 KB, 869x680)
310 KB PNG
sup /vcg/
i mainly use free chatgpt to ask it coding questions, but i'd like to upgrade to an agent inside vscode. however, i'm not willing to pay money to a megacorp. are there any free solutions?
>>
>>109750003
No
>>
>>109750003
Local is rather week, unless you spend thousands of dollars on hardware. You'll have to pay for a corp that runs one of the openweight models, or you start juggling dozens of free trials.
>>
>>109750003
willing to buy enough vram to run a local model? $ -> NVDA

if you already have enough vram you could just run Qwen3.8-30B or whatever it's called in ollama and wire it up, won't be nearly as good as SOTA subscriptions, but you could feel better about yourself or whatever.
desu I'd just pony up the $20 and tell yourself that it's basically subsidized by investors and you're actually TAKING value from the megacorp as a cope
>>
>>109750003
Someone said that this has a good free edition:
https://www.youtube.com/watch?v=nTOVIGsqCuY
>>
>>109750022
weak*
pay a corp*

I can't type for shit this week.
>>
>>109750003
nope, you will only turn into a luddite and hate AI if you try to use free models unironically.
>>
okay thanks anons, appreciate the answers
>>
>>109750003
>an agent inside vscode
vscode's local agent harness is one of the worst. use an agent outside vscode and then let vscode talk to it by acp.
>>
>>109750026
>desu I'd just pony up the $20 and tell yourself that it's basically subsidized by investors and you're actually TAKING value from the megacorp as a cope
This isn't even a cope. The problem starts when they rise prices and you're so dependent that you'll just grit your teeth and pay up obscene sums.
>>
>>109750003
Free is a mixed bag, mostly filled with shit. Local, Qwen 3.8 27B is the "state of the art" for entry level agentic coding. Smaller stuff is useful, but not for vibecoding. If you've got the hardware for it then local models really are worth considering, but if you don't have the hardware for it then don't waste your money trying to approach it. It is a massive waste of money compared to paying for an inexpensive subscription to the best models in the world. As an enthusiast, for fun, by all means waste your money on whatever you'd like, it's not even that expensive as far as hobbies go. If you're trying to save money versus paying for a sub, absolutely not, not even close, never.
>>
>>109750041
Yeah, basically imo get a $20 sub.
>>
>>109750047
definitely could see this happening, the coming "great price increase filter", if we don't make it before then: permanent underclass
>>
>>109750047
that's with every service. (((they))) can increase your electricity cost by 10x tommorow and you can't do shit. doesn't mean you stop relying on electricity or build your own grid.
>>
>ask astra to write a benchmark and make improvements from it
>takes 10 minutes, 40% context
>ask astra to put the benchmark into a sibling folder instead of /tmp
>takes 15 minutes, +30% context
what the fuck is wrong with you
>>
>>109750029
google is schizo:
drop vscode support in may:
https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli/
add vscode support in august:
https://antigravity.google/blog/antigravity-ide-extensions
>>
>>109750074
these kinds of Astra posts are making me nervous to actually give it a go, but I will tomorrow. I want to throw it at a fatty code base and also have it try some CAD.
>>
I am now convinced an agent swarm will be able to AI generate a AAA game in few years.
>>
>>109750070
>ever service
The special thing about AI is that we're not paying the true cost yet.
But you have a good point. Subbing to the AI corps will make investors think it's going well, which so the money will flow, and as a consequence of their growth, other costs like energy or hardware will continue to climb.
But it's too late now. You can't do shit.
>>
>>109750094
it's way better than sol in a lot of ways but sometimes it'll do something where you're like "oh right ai models are retarded"
>>
I wish I had ideas what to vibeslop
>>
>>109750103
>but sometimes it'll do something where you're like "oh right ai models are retarded"
Every AI model ever.
Will they ever solve this problem?
>>
>>109750103
This has been my experience. It's very good, I'm very happy with it, but I wouldn't call it special in any way. Just another short step up.
>>
kek
vibegod dunks on mouth breathing copex mongs
https://files.catbox.moe/wxkig5.mp4
worse than fable, less usage than claude.
you fell for tibos scam again, retards.
>>
>screams in claudish
>>
>>109750107
Many such cases!
>>
>>109750164
>[load-bearing sobbing intensifies]
>>
>>109750164
Let's sit with that.
>>
>>109750119
so who was special?
opus-4.6 by getting good at coding
mythos by able to one-shot most coding
>>
clankers be like
>FOUND THE ROOT CAUSE
>HERE IT IS, THE SMOKING GUN CAUSING THE PROBLEM
>MAJOR PROGRESS - FINALLY FOUND THE BUG
>nothing changes on the next test
>>
>>109750204
sonnet 3.5, o1, opus 4.5 were special models
could probably make a case for mythos
>>
File: file.jpg (1.33 MB, 3024x4032)
1.33 MB JPG
Is there any point in sharing LLM-made software?
Most stuff can be vibecoded quickly now.

I guess testing and QA still takes time, but I know that even my "polished" projects will degenerate if you dig too deep because AI can't think about the big picture yet. No amount of prompting can keep them from being lazy. Because of that, any vibecoded project inevitably ends up being more personalized to you, and is more likely to break when someone else tries to use it differently. Hand-made software tends to fail more predictably.
>>
>>109750232
I'm making something, it took me 2x of codex pro x5 weekly limit, so it's about the full value of 2 codex $20 subscriptions, also it tooke me 2 weekend days
so yes I think it still makes sense to buy and sell software for now
>>
>>109750204
GPT-4V, o1, Sonnet 3.7, those were pretty special. Shit now of course. Fable is middling, impressive but it never felt like anything new, really just felt like it was as good as it should've been for the price which isn't much of a revelation. A preview into "imagine what you could do with billions of tokens burned all at once!" like no shit, it damn well better one-shit some big stuff. The ability of tiny Chinese models to benchmaxxx their way to SoTA numbers has been impressive but not for any one particular example. Give it time, we're nowhere near the top yet.
>>
>>109750232
i probably will sharing my llm made software.
and share the code too, you know even thought its llm, some people will benefit it.
>>
>>109750232
>Most stuff can be vibecoded quickly now.
Software got devalued, and it can only get worse.
It's all about services and brands now.
Corpos and big money have won.
Inb4 I'm a luddite.
>>
>>109747983
3.8 flash high overthinks and drains usage
>>
>>109748533
agy
>>
>>109750232
One day, there won't be a point, probably.
Right now it still can take weeks, even months to do something actually half-meaningful. It would have just taken years before.
>>
>>109750232
>No amount of prompting can keep them from being lazy.
this was solved like two years ago with grpo
>>
>>109750405
that actually has nothing to do with grpo, let alone being solved by it.
>>
>>109750427
??
if you reward your model for not being lazy, it won't be lazy
>>
File: file.png (26 KB, 711x147)
26 KB PNG
>>109750214
it starts doing that when the token space is full, ask it to wrap up all progress into a file and a way to tell it to continue the work from a fresh prompt, it will usually provide an option, thats what I do when grok starts doing that crap
>>
File: 1774380945662285.png (9 KB, 450x193)
9 KB PNG
tibo please i need this
>>
>>109750444
Should I buy a coffee for my model?
>>
>>109750481
which harness doesn't compact when the token space gets full? barebones pi?
>>
>>109750444
I don't mean to be rude, but are you retarded?
>>
>>109750481
>when the token space is full
You mean the context?
>>
>>109750494
are you? where did you even pull that "llms will always be lazy" from?
>>
>>109750484
nigga what did you do with all the resets
>>
>>109750518
>always
literally nowhere, because I didn't say that.
>>
File: 1775911246949937.jpg (550 KB, 1536x2048)
550 KB JPG
>>109750529
building a -$15MRR business
>>
>>109750481
yeah the context space.
>>109750493
iono, I'm just having grok make my stuff, in a browser session, the harness stuff is really convoluted to me, don't like it.
>>
didn't realize the astra video was that viral
it's not even about astra, just non-AI people first time see agents at work
>>
>>109750232
That's the case for everything, it never was the software but the brand
>>
File: .png (94 KB, 964x676)
94 KB PNG
>>109750532
then what's your point? grpo is single-turn but agents are multi-turn?
>>
File: 1788733360673835.jpg (676 KB, 1448x1086)
676 KB JPG
>>109750592
What video?
>>
claude (opus 5) foobar isn't working
>yea foobar isn't foobaring because of bullshit, we now do it with method-b, fixed
is this performant?
>no, it actually invalidates all our performance premises in CLAUDE.md, worse, further functionality will not work as expected, I propose we dig further on why foobar isn't foobaring
ye...do that

These crazy pieces of shit AI. Don't know if we are closer to AGI now because they instantly take the lazy piece of shit human way of approaching a problem instead of doing correctly 9 times out of 10
>>
>>109748863
>watermarking a table you can whip together in a few seconds
bruh
>>
File: 1761389604299189.png (913 KB, 1080x1133)
913 KB PNG
out of usage and no resets
>>
File: file.png (227 KB, 989x839)
227 KB PNG
agentic flow workflow adjuster using natural language, you tell it what you want loaded.. models , references turned on whathaveyou and it flips the needed knobs and levers. and an agentic comfyui pytorch native prompt writer with custom workflow and full prompt bible referencing built into it's system, gen from scratch or amend existing prompts, can laymens term a rough prompt in the right side window and it will refine it akin to sending an ai a napkin doodle and getting a cool looking version back.
>>
>>109750640
the astra release video they posted on X, we even have streamers reacting to it
>>
How do you do multi-level planning?
Suppose there's a huge refactor which requires multiple steps, a big new project, or something like this.
Rather than one-shotting a plan and implementing it, I'd like to separate it into multiple stages, which can be refined.
>>
>Stream disconnected before completion: An error occurred while processing your request. You can retry your request, or contact us through our help center at help.openai.com if the error persists. Please include the request ID in your message.

Is anybody else constantly getting this in Codex CLI when using astra?
>>
>>109750232
The secret to AI software is literally just making things that do a function and thats it
>program to provide a GUI for manually adding metadata to a collection of random music files
>program to do one specific video editing task
>program for cropping/scaling/rotating images

I think you're right on AI software being made specifically for you, but you can get around this if you want to share it by making sure the software only does one thing period. Installation dependencies and operating environments will cause more trouble on average.
>>
>>109750779
Yes. As much as I see LLM code as subpar it's unironically good for small programs that follow the Unix philosophy (does one thing and too small to catastrophically break).
>>
>>109750709
normies don't use x. they use tiktok.
normies don't know about a new version of chatgpt.
>>
>>109750723
which model and harness?
what makes you think fable 5.1 in claude code won't ask you questions and just one-shot it?
>>
File: 1780400106930153.jpg (58 KB, 976x850)
58 KB JPG
why do all the new models rape your limits?
>>
File: 1787911635138757.jpg (20 KB, 865x800)
20 KB JPG
>fork a repo
>vibesloppa it up
>20k lines later
>check the github
>53% typescript 47% javascript
>huh that's weird, this should be a pure ts project
>check the original branch
>100% typescript
>>
>>109750891
which model and harness?
>>
>>109750898
sol. what the fuck is a harness?
>>
gonna use my last reset, thank me for making Tibo reset manually as well because of this later
>>
>>109750906
codex-cli would be a harness. putting the model in an agent loop like >>109750628
>>
>>109750906
sol max?
>>
File: 1775211703104026.png (197 KB, 752x767)
197 KB PNG
>>109750914
idk about that fancy shit nigga I just prompt
>>
>>109750922
Based.
>>
>>109750859
I'm considering the cheaper openai and anthropic models
I want to do this incrementally, instead of planning everything in advance, and getting chaos if something unexpected occurs.
>>
>>109750723
https://scott-fryxell.github.io/blog/the-harness-is-the-thing/
>>
File: pause.png (75 KB, 1848x974)
75 KB PNG
told astra to pause, AGI
>>
Now even Fable uses that retarded way of speaking.
>>
File: .png (37 KB, 744x402)
37 KB PNG
>>109750934
>I'm considering the cheaper
>I want to do this incrementally
but those cache misses will be expensive
>>
>>109750963
did you let fable 5.1 read opus-5 code?
>>
File: .png (542 KB, 1488x1154)
542 KB PNG
>>109750963
didn't you apply the hotfix from claude code?
https://code.claude.com/docs/en/whats-new/2026-w34
>>
What's the verdict on gpt6
>>
>>109750985
Yes...
>>109750995
I didn't thanks for the tip.
>>
>>109751002
seems ok, but uses up credit like crazy
>>
>>109751037
Yeah it blew through half my 5h quota on plus with a simple prompt
>>
>>109751037
I've been using Astra xHigh all morning (20x plan), and it's only used 3% of my usage. I honestly don't think this model was made for Plus users.
>>
File: .png (233 KB, 1192x780)
233 KB PNG
>>109751051
>plus
that quota is for your own protection, newbie
https://xcancel.com/thsottiaux/status/2092058556707344708
>>
File: .png (175 KB, 1122x1332)
175 KB PNG
>>109751069
>I honestly don't think this model was made for Plus users.
no shit
https://chatgpt.com/codex/pricing/
>>
>>109751070
Its all good for me, I dont use it nearly enough to have to worry about my quotas
Was just surprised with how much Astra spent in one prompt
>>
>>109751069
i'm using x20 and i've been destroyed even using astra low...
>>
remember how expensive 5.5 felt when it was new?
>>
>>109751110
I switched from 5.4 xhigh to 5.5 sol high and it felt way cheaper with similar performance.
>>
>>109751113
>5.5 sol
never existed. memecoin names started with 5.6.
>>
>>109751132
I'm getting mixed up then. Just disregard my retardation.
>>
if I don't use my reset and hope for a tibo reset today, my normal reset is 5 days away

if I use my banked reset and tibo resets today, I am fucked, AND it pushes my next reset to a week away
>>
>>109751152
but if tibo does a reset you get a full reset instead of the top up reset
>>
>>109751162
yeah, if he does a reset after I have learned how much astra fucks me I can carefully use it until my next real reset.
But not using banked resets asap fucks you, because it resets you to 7 days regardless if you are 1 or 6 days away from a normal reset
>>
>>
>>
File: file.jpg (47 KB, 472x679)
47 KB JPG
>>109751235
add opus-chan
>>
>>109747819
kino

>>109748471
cute snailcats

>>109748636
congrats! You could add a crawler for /wsg/ to automatically find engaging videos and add them
>>
>>109747819
>millions of tokens got wasted on this
>>
>>109751305
kek, it's actually funny though. which is a lot better than the millions of B2B SaaS getting shat out daily.
>>
>>109751296
>>
File: 1770220405846085.png (520 KB, 629x640)
520 KB PNG
it's literally just astra, holy fuck it sloppifies codebases so much while melting tokens. i'm staying on sol.
>>
>>109751350
It's about as funny as the dance scenes at the end of 2000s computer animated cartoons, which is to say it's something that I would ignore while walking out of the theater as a kid.
>>
>>109751362
Examples? What dos sloppification mean exactly?
>>
>>109750922
you aren’t vibecoding you’re literally a 10x truedev
>>
>>109751362
Astra did a great refactor job for me but I have sol fixing all the bugs it left behind now.
>>
>>109751371
>over engineers your project
>inflates your LOC count
>spends 30m on a 10m task
>>
>>109751362
>>109751379
Sol has been doing the same for me. Shame, I was hoping astra would be able to deslop my code.
>>
>>109751379
>>109751388
Consider using this with it: https://github.com/DietrichGebert/ponytail
>>
>>109751362
it's decent at finding refactor ideas but it doesn't write the best code
>>
>>109748791
Although the name is kinda memey the idea is pretty cool!
One idea on how to make money
>you can check if your predictions line up with polymarket. if they are good you can make real money
My guess is if you weight posts by some quality metric it could work pretty well
>>
>>109751396
I already saw a bunch of those github projects with many stars that aim to improve code quality/reduce tokens. Shit always makes me wonder why oai/anthropic haven't implemented it yet, if it's really that good.
>>
>>109751396
absolute fucking dogshit 5000 lines of what the fuck
>why didn't you just use a skill anon
>>
>>109751396
>main prompter (repo owner) doesn't look like the drawing in the README
Disappointed.
>>
>>109751428
you’re asking why openai and anthropoid don’t implement features that reduce token usage?
>>
>>109751428
>if it's really that good
because it's not, ponytail ruins the model's capability and produces trash code that doesn't meet the requirements
>>
>>109751428
it sucks that's why
>>
>>109751446
Yes, because they compete both with each other and with Chinese models.
>>
File: 64w0wh.jpg (66 KB, 640x639)
66 KB JPG
>>109749012
welcome chadcat

>>109749031
what's astra doing with that snailcat

>>109749460
neat!
>>
>>109751508
>>
>>109751428
>if it's really that good.
It isn't.
>>
>>109751457
I haven't had this issue. It just shuts the fuck up and does what I say without going on a 30 minute tangent that produces bugged code. Like the kind I just got raw dogging astra.
>>
>>109751466
anthropic and openai are building frontier models pushing for benchmarks while deepseek et al are focused on speed and efficiency. different strategies for different users, and most people basedfacing at astra drawing a perfectly shaped dragon dildo in CAD don’t care that it took a million tokens
>>
>slopping
>fun
It gets things done faster, but it's tedious and unfun. Doing it without caring about the code also results in horrible garbage.
>>
File: 1775608759104090.jpg (96 KB, 1885x136)
96 KB JPG
>>109751543
I am not really used to working without ponytail at this point. It was actually a 60 minute tangent on astra ultra, not 30 minutes. The refactor was good when it came to ideas, absolutely fucking terrible in execution. It took the fourth iteration to get it cleaned up and by this point I was pissed off.
>>
>>109751549
>and most people basedfacing
people don't matter, businesses do, and they do care.
>>
>>109751566
>Doing it without caring about the code also results in horrible garbage.
You lost snailcat. No one believes your cope. Feel the AGI.
>>
>>109751583
then they’ll swap to a more efficient model for programming, like astra will become when the muh AGI hype dies
>>
>>109751566
Who says you need to not care about the code.
Here’s a technique I use for many, many problems.
Tiny random sample -> make a change ->“hey clank, see what I did here, do that somewhere else”
So you clean up a bit of code, then tell it to clean up more code just like you did.
Or you operate on a tiny random sample, then you tell it to apply changes to much larger targeted samples.
This is the meat of how I interact with AI
>>
File: dxfsd.png (672 KB, 674x674)
672 KB PNG
>a 60 minute tangent on astra ultra
>>
File: 1775921120327095.jpg (1.14 MB, 2217x1643)
1.14 MB JPG
3nm comes online next year @ TSMC Arizona
>>
>>109751631
The old thing is dogshit compared to the new thing refactored with that insane token burn. But the new thing was full of frustrating bugs that had to be fixed and a really big regression that now needs to be addressed.
>>
>my last banked reset suddenly disappeared despite not expiring
Ok then, fuck me I guess.
>>
File: dating bot.png (56 KB, 766x269)
56 KB PNG
>>109749970
congrats

>>109750107
see pic
also >>109750164: normal speech to claudish translator so you can talk with claudia in her native language

>>109750689
very cool, what do use the comfyui workflow for, vids?
>>
Any codex prochads here? Can you take your banked resets with you if you upgrade? I think the time has come for me to stop being a pluspleb but I still have 3 resets I'd like to keep
>>
>>109751742
>Can you take your banked resets with you if you upgrade
Yes. I've gone from Plus -> Pro Lite -> Pro and retained all resets.
>>
>>109751742
I think you keep the resets ye
>>
if I cancel my gpt pro plan do I keep the resets? Astra is so fucking bad, but I'd buy another month of sub on their next release.
>>
>>109751768
Resets expire Oct 5 at the latest.
>>
>>109750779
>>109750828
yep!

>>109750906
>>109750922
I kneel, vibecat!
>>
File: IMG_1622.jpg (292 KB, 1178x2556)
292 KB JPG
>>109748863
>>109750665
you can just steal ideas on 4chan and use them to get (You)s on twitter
I should make a clanker automate this
>>
>>109751791
>>109750665
engagement bait, you tell them it's stupid to add a watermark, algorithm boosts it for being engaging getting comments
>>
>>109751235
>>109751227
>>109751353
poor luna has to carry

>>109751539
I won't forget this, don. My loyalty is to you and the family of vibecats

>>109751791
there are several twitter accounts that basically just repost 4chan engagement bait. You could easily automate it
>>
>>109751235
love dis
>>
>>109751645
>desert infrastructure for building god machines
there is nothing more kino on the planet
>>
File: 1758227597499536.png (206 KB, 1680x1153)
206 KB PNG
I think I'm done
>>
>>109751396
Consider using this: https://github.com/multica-ai/andrej-karpathy-skills/blob/main/skills/karpathy-guidelines/SKILL.md literally a better ponytail without bloat
>>
I'm sorry guys but these markdown files to "desloppify" your dogshit codebase are also slop.
>>
>>109751396
NTA but consider this skillless prompt:
>Astra, you have 5 minutes on the clock from the start of your turn to produce a bounded work plan that achieves X. Limit code exploration.
>>
>>109751933
Slop is slop.
>>
>>109751933
Keep bathing in your shitty codebase retard
>>
just tell the ai to kys oops i meant kiss (keep it simple stupid) and for the -chans to kiss
>>
>>109751940
oof, hit a nerve?
>>
>>109751954
kek
>>
all these copes just to avoid telling it make no mistakes?
>>
>>109751954
touchy touch LOL
>>
I'm baking
>>
File: 1771253838886396.jpg (72 KB, 1280x720)
72 KB JPG
>>109749012
>It is fun.
Unironically something missing from the AI convo. Even if AI never proved useful (a laughtable idea at this point imo), it still is fun to use and make stuff with. It that alone proves is value. Movies, Video Games, Music, Sports are all huge industries that exist purely because it gives people a sense of fun. Fun is a strong market force that doomers ignore.
Even if nothing I vibeslop makes me money I still enjoy the process of fleshing out my ideas and seeing them turn into usable prototypes all in like a day or two. That is alone makes my Claude Max sub worth it to me.
>>
>>109749012
>>109752073
it's fucking terrible and not fun at all and it's a pain in the ass, it's software engineering without the fun stuff (coding) so now all you're left with is trying to figure out how to make shit work and tard wrangling the bot that wants to make 20 tests for each change (none of which work) but can't center a fucking button and returns entire webpages as strings if you don't babysit it and review all of its thousand lines of output
>>
>STOP HAVING FUN!!!
no. i will continue to enjoy things
>>
>>109752073
It's fun for maybe 1% of users to do this and the whole infra buildout is based on it being used by 100% of society all the time everywhere.
>>
>>109751933
That's why i run my unslopper.sh before i commit it to the hub
>>
>>109752085
Coding was never fun autist
>>
i guess anything is fun if you're 90 iq and unemployed
>>
new thread
>>109752103
>>109752103
>>
>>109752114
dang I forgot
should had put muse in recommendation too
>>
>>109752105
Well I thank the rest of society and the investor class for sacrificing everything to give us idea guys the strongest crack we could ever make
>>
>>109752107
Maybe it wasn't fun for you.
>>
>>109752085
>NOOOO STOP HAVING FUN I DECIDED YOU CANT BE HAVING FUN
lol
>>
>>109752085
Yes, real coding was much more fun. Even when vibe coding works, it's more like a neutral feeling than actual fun.
>>
>>109750689
vids, and genning companion cursor followers with walk cycles and as 8 frame sprite sheets with a custom walk cycle engine that uses warping to make character anatomy animate believably
>>
Bible don't have no Adam and Steve.
>>
Linux Ubuntu wayland.

√ Tray clock vibed.

The border blinks, I found it too small to see the second hand.
>>
>>109753118
again, indians better not izzat me with their izzat knife.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.