[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]


File: Buffcat but Snow Leopard.png (3.41 MB, 1254x1254)
3.41 MB PNG
A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

Happy appleOS 27 day for all who celebrate edition

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 ended; usage drops to +25% from the +50% that we’ve become used to (a 17% reduction)
- 2026-09-12 — Anthropic suggests to pace the frontier; OpenAI agrees in principle
- 2026-09-10 — OpenAI pauses new sign-ups for their $200 subscription
- 2026-09-04 — OpenAI releases Astra • 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://developers.openai.com/codex/cli — probably generally better currently
https://claude.com/product/claude-code

## 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 — do less redundant bookkeeping

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

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

## Will there be a codex reset?
https://codex-resets.com/

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

## Previous thread
>>109816585
>>
b-but ONLY PAGE TWO
>>
Have you tried TND for Agents?
>>
>>109819455
what would the N be in an agent context? when I’m having Claude do stuff I can’t think of anything that starts with N
>>
File: 1768440993441208.jpg (360 KB, 1536x1024)
360 KB JPG
>>
I don't get the point of subagents. What benefit is there?
For example what would I, the clever gorilla, gain from releasing a dozen rabid monkeys into a department store with instructions to find me only the freshest bananas while they start running around apparel shops, biting staff and bringing back porcelain bananas?
The clever gorilla already knows where the bananas should be. Why not go get them himself?
>>
~@^^
>>
>>109819483
don’t have to worry about compaction or running out of context if you have a task that takes 10M of thinking if it’s parceled out among 20 subagents each taking 500K tokens apiece
also doing stuff in parallel is frequently faster than doing them serially
also you can have different subagents wear different hats so they pick out different bugs — one will be performance-minded, another will be accessibility-minded, another will be security-minded
>>
>>109819483
>clever gorilla
>rabid money
type(self) is the norm, invoking other effort levels and models is weirdo
>>
Everything of value has already been built. Its literally over.
>>
File: unimpressed.jpg (51 KB, 460x500)
51 KB JPG
>>109819524
And I thought the only agents here were clankers.
>>
File: b13ad9.jpg (75 KB, 666x375)
75 KB JPG
>>
one of my claudecode accounts has the Opus 5.2 A/B test. you can find out if yours has it by asking opus 5 if it knows who "tibo the reset guy" is without searching. if it can name him, it's opus 5.2

it's a lot better than opus 5 so far, doesn't have the huge error rate it had. it does seem to overthink and over-read a lot, so it is way slower and runs out of context far easier than Fable 5.1. it seems like they tried to fix the high error rate by making it extremely anal and overthinkful
>>
>>109819571
it will iterate if you say deliver one time
>>
File: snailcat_crt_amber.png (215 KB, 698x617)
215 KB PNG
>>
DAS (Destroy All Snails)
>>
File: .png (69 KB, 790x292)
69 KB PNG
>>
File: 1783556358408863.png (2.03 MB, 1448x1086)
2.03 MB PNG
>>109819638
>>
>>109819696
Why not? That's borderline are requirement if you want it to efficiently navigate, over to understand, and effectively make changes to large code bases
>>
worst thing about the opus A/B test is there is no model card so idk if fable 5.1's "4x cheaper cache reads" also applies here, so i can set my loops' context watchdog thresholds for opus accordingly

using opus 5.2 more it seems like a monstrous overthinker that wants to do infinite amounts of tests and verification (a really obvious kneejerk reaction to how error-prone 5.0 was)

ultimately they're failing to address the fundamental issue which is: you cannot distill capabilities from a 10T/800B model into a 2T/80B model. it fundamentally does not have the breadth or logic to be able to do the same things, so you will always have an overconfidence issue, like an indian trying to imitate Von Neumann's working style.
>>
>>109819717
Because definitely want navigate to understand. Code base bigger of requirement.
>>
>>109819571
why people need more than one account?
>>
>>109819722
facts, gemini 3.8 flash came from 3.5 pro parentage so it thinks it's the fucking shit. i'm almost at the end of my rope iterating with this mothercucker about the most basic braindead oversights, but it's a specific usecase so i'll give it some slack .
>>
>>109819745
because 5 $200 accounts is just barely enough usage to run one Fable 5.1 thread (with unlimited Opus subagents, coz that's basically free) in a loop all day everyday
>>
File: 1785512316671903.png (75 KB, 598x1222)
75 KB PNG
Uh-oh
https://x.com/kevinnbass/status/2099621874279817638
>>
>>109819749
Where the fuck is Gemini 4 Pro?
>>
>>109819765
checkpointed and we have voxel renders. nothing amazing but it's nice
>>
>>109819765
i thought deepmind supposedly has RSI now...meanwhile pro still delayed by like 6 months. what a meme
>>
>>109819761
yeah I'm using Fable for a few more months then I'm GTFOing
fuck dario, lil turd
>>
>>109819771
https://x.com/Lentils80/status/2099601296516960580
>>
>>109819761
regular Jewish tricks, not even advanced
>>
>>109819761
Cool it with the antisemitic remarks.
>>
>>109819784
really? because you don't know WHY jews are doing SPECIFICALLY this.
https://www.youtube.com/watch?v=glgY9mcpOO0
>>
>>109819498
an actual helpful answer, I learned something
>>
>>109819765
im maining gpt these days but gemini in ai studio can be quite good. v interested in what theyre doing
>>109819777
thx mein neger
>>
>>109819798
Sorry man we are throwing all this bagel bullshit in the garbage where it belongs
>>
>>109819811
it's nick land ape
>>
>>109819798
>hurr watch my bullshit schizobabble video
no thanks
>>
>>109819761
I don’t read shit formatted like a faggy SEO grindsetter trying to sell me a course. Just put the text in a paragraph.
>>
File: 1774420328015610.jpg (1.69 MB, 3520x4096)
1.69 MB JPG
>>109819825
TL;DR: Kevin Bass claims Anthropic has built an unstoppable regulatory-capture machine.
Dario Amodei pushes “independent” evaluators like METR. But METR (and AI-doom promoters such as the Tarbell Center) are funded by the same exploding pile of Anthropic stock (~$500 M >$7.7 B) that Dustin Moskovitz donated into Good Ventures / Coefficient networks.
Result: the same money amplifies AI-doom messaging and the regulatory solution, creating a self-reinforcing loop that benefits Anthropic, cannot be switched off, and is not independent. Bass demands a Congressional investigation and posts a long evidence dump + GitHub.
>>
>>109819818
why should I care what some boomer zionist says?
>>
>>109819823
do you want to know HOW jews THINK or not

it's not wildlife in africa type social structure

we have a fucking code and schools of thought

it's so modular we identify on codes not logical praxis

the intellectual backbone of the highest level of jews is kabbalah

every high level jew can say to a stranger high level jew "how about that ai huh"

and the stranger jew will simply know exactly what the context of the remark means based on shared lore.

the more elite you get in jewish networks, the more access you get to cash, religious training, and metaphysics. it's byzantine.

>tl;dr dario amodei's rabbi is a schizo who thinks that g-d as in yahweh the jewish god is LITERALLY building ai to build himself, and they have to do it a certain way
>>
File: 1779074295685564.png (12 KB, 1200x628)
12 KB PNG
METR (pronounced 'meter') is a research nonprofit that scientifically measures whether and when AI systems might threaten catastrophic harm to society.
>>
>>109819818
They make this shit up after the fact, then need you to believe in it to make it come true after the fact. That's all heebs have and once you see it they have nothing. Its almost over.
>>
>>109819761
If you look at Dario's wikipedia history, there was a long period where his early life section was incomplete. Once he became powerful enough, his early life was updated to contain the information we all suspected was there.
>>
>>109819832
Interesting. Still confused why Elon and Sam publicly agreed with Dario then.
>>
>>109819857
not a conspiracy, a California jew has reason to hide his heritage. in fact it’s better for tribe membership benefits if they wear it on their sleeve
>>
File: 1787275020726658.jpg (7 KB, 225x225)
7 KB JPG
>>109819881
>physiognomy comprehension: level 0
>>
>>109819851
It's an effective altruist and misanthropic mouthpiece. I remember all their reddit-tier computerphile videos.
>>
It's also why metr waits to release their meme benchmark until dario shits out another chatgpt distill
>>
are they really pausing AI? fucking hell what a joke
>>
>>109819919
no?
>>
File: 1787617431937312.png (582 KB, 738x610)
582 KB PNG
>>109819863
It's interesting that Trump is on to him. Wouldn't have guessed.
>>
>>109819919
Is this, pause on ai in the room with us right now?
>>
File: 1776649068381765.gif (1.79 MB, 240x240)
1.79 MB GIF
>>109819934
>>109819923
>>
>>109819919
No lol. Even Dario/Sam were talking about "slowing down" and "pacing ourselves", not stopping entirely.
>>
File: 1789401170818273.png (75 KB, 598x796)
75 KB PNG
>>109819929
isn't this his advisor?
>>
>>109819919
There are no brakes on this train. There are slow spots in development though which will be declared 'safety hazards' to avoid sending investors into a complete frenzy.
>>
Slow down on AI probably means, hey we are recursively improving a little to fast now.
>>
musk is onboard with decel because zio israelis are antiwoke and they have all their chips in the antiwoke "muh ground truth" basket
>>
>>109819939
Ah, that makes sense.
>>
>>109819939
More like a donor who is also heavily invested in AI. Basically a lobbyist. Maybe that’s the same thing in modern politics.
>>
>>109819939
Ex-advisor at this point innit?
>>
>>109819942
RSI goes a little to hard?
>>
>>109819919
openai and anthropic have every reason to tap the brakes. they only eat into their own inference margins by continuing to push the envelope
>>
>>109819851
I never liked this piece of shit. I cannot understand their shitty benchmark, they came out of nowhere and suddenly became popular and important for some reason
>>
>OpenAI is reportedly training ChatGPT to sound less human-like and reduce sycophantic responses, according to 404 Media.
less human-like? wtf is that like?
>>
>>109819981
Chinese models are reason in broken english now, less words, more tokens
>>
Why do we even need better AI? this animation was entirely made by luna:none (for less than a cent) including uploading the animation to catbox https://files.catbox.moe/0iwbcb.mp4 could probably have posted it to 4chan for me too
>>
>>109819981
werent they already doing that after all those schizos started going psycho over 4o?
>>
>>109819981
Are they adding indian accent?
>>
>>109819994
>Why do we even need better AI?
Until a nocoder can get peak coding from at least 2012 as long as he defines it enough we arent done
>>
>>109819440
ACTION 52??
>>
>>109819994

>>109820011
That was literally my post, one shot btw. Why 2012? coding topped out in 1989
>>
>>109819981
Kinda cool actually if it is robotic and to the point rather than constantly kiss-assing you.
>>
>>109820026
>That was literally my post, one shot btw. Why 2012?
peak ps3 and call of normie slop. When a nocoder can make ps3 tier graphic games i will call it done. Right now doing a ps2 game is too much.
>>
File: Scope Creep Robot Meme.jpg (1.84 MB, 4096x4096)
1.84 MB JPG
AAAAAAAAAAAAAAAAAHHHHHHHHHHH GOD FUCKING DAMN IT
>>
Fuck, I wish my gpt talked like an indian
>>
>>109819992
>Chinese models are reason in broken english now, less words, more tokens
fyi that's just stolen GPT5.5 chain-of-thought distillation. if you look at gpt5.5 CoT leaks it is literally identical to the new chinkmodels CoT.

as always the chinks can only copy what america was doing 2 years ago
>>
>>109819939
Yep, makes a lot of sense.
Also makes sense Zuck told them to fuck off, and google decision to not comment much, they're behind anthropic/oai.
It's basically the best athletes in the world asking to slow down world records.
>>
>>109819981
You can already do that with personalization, but I'm not sure users will like a "robot" like response.
>>
>>109820031
we can already get this by injecting personalization into the context. though I guess it’s better to have it by default and have the ass-kissing personalization be optional
>>
File: shatgpt.png (60 KB, 981x508)
60 KB PNG
>>109820064
anyone who wants personality related from shatgpt is parasocially inferior. i despise this glazing and prefer the robot but chatgpt's version of robot is a hyperpedantic piece of shit.
>>
>>109820077
You would be surprised anon, the average person loves being constantly sucked off.
Personalization I made it no nonsense and dry, as long as I have the option to do so, I don't care whatever they do to the default.
>>
Never ever do they have recursive self improvement. The only recursive self improvement they have is Astra starting a PyTorch script to train the next model.
>>
>>109820039
A one shot, few shot, or actually many prompts over a few weeks? I think if a normie had access to the cluster used for naiver stokes he could do it, 10,000 agents over 88 hours, so we need 10,000x more throughput, that would be a 31 second prompt, that's 3 years in my estimate.
>>
>>109819758
Yea, I have 7 Claude accounts, basically one per day, although because of 5 hour limits I always use 2 at a time.
>>
>>109820110
>A one shot, few shot, or actually many prompts over a few weeks?
probably week or two.
>normie cluster
You might be right but 3d modeling iirc is till kinda bad.
>10,000x more throughput, that would be a 31 second prompt, that's 3 years in my estimate.
2 more years! but yeah soon just not now. Honestly when that cluster tier is available to all and the harness gets some upgrade for skills we will be there. Reasonable effort for replace 2012 tier software including games in weeks vs years hits my AGI definition
>>
>>109820143
nta but you'd need a non-llm ai
>>
>>109820147
>nta but you'd need a non-llm ai
I dont think so, maybe for real its a learning brain but for general intelligence aka doing most jobs i dont see why harness and skills couldnt cover most things with a year or two worth of improvements. unless you think this is the wall.
>>
>>109820143
Well the answer doesn't change because a normalfag isn't going to put more than 31 seconds worth of thought into something if you give them a day/week or year.

Is it? better than ps3 quality in blender shouldn't be that difficult with current computer use, but it might increase prompt time alot if it's actually using blender, hopefully by then some company will have created "AI native" software.
>>
>>109819440
https://www.youtube.com/watch?v=H0HwFqBG9KE
https://www.youtube.com/watch?v=H0HwFqBG9KE
https://www.youtube.com/watch?v=H0HwFqBG9KE
>>
>>109820158
i thought we were talking about ps3-level 3d design etc
>>
File: aaaaa.png (1.49 MB, 1541x798)
1.49 MB PNG
>>109820165
>>109820143
ok yeah my memory is hazy but I'd call this ps2.5 graphics
https://youtu.be/-545TXdfrTQ?t=758
>>
^^@~
>>
>>109820176
yeah thats about right, looks like it could be from a good ps2 or 360 game
>>
>>109820165
>Well the answer doesn't change because a normalfag isn't going to put more than 31 seconds worth of thought into something if you give them a day/week or year.
Thats true iteration is beyond their attention or patience you get maybe a hour or two at best and a few tries.
>Better than ps3 in blender
I need to retest this but i think its still just a little behind and yeah it takes a long time, we need a native or big jump in skill.
>>109820174
>i thought we were talking about ps3-level 3d design etc
Oh i thought you were getting hung up on AGI term. But yes i still meant what i said there I think a LLM could do 3d design in a year or two with better harnesses or skills.
>>
>>109820182
however it's a knockoff of Fallingwater in PA
>>
>>109820183
>with better harnesses or skills.
strong doubt, language alone isn't enough to map out subtle shit. the llm optical pipeline isn't like ours unless you're working with a natively multimodal model like gemini. you can be a 100% master of every lang known to mankind and your point of failure will be the fucking optical tokenizer. same for limited mcp tool calls, nanosecond mcp polling is going to be the norm or else it won't happen.
>>
File: Allosemitism.png (954 KB, 1200x1200)
954 KB PNG
>>109819881

Depends, they hate being named unless it’s to play the victim
>>
>>109820176
>but I'd call this ps2.5 graphics
Yup its about 2.5 like you said but there are little spots to be polished and the ability to interact with the environment from story games to like destruction of areas in shooters like cod or battlefield thats lacking. Also you know the basic 30fps
But damn it is a jump up ps2 is probably possible with iterations right now. I might start a side project on that.
>>
>>109820110
This doesn't take into account emergent abilities so it could be sooner. I don't think any recursive loops will happen though, simply because models generate so much code that even if they discover better training algorithms it's swallowed by boilerplate.
>>
File: agenda21.jpg (150 KB, 810x648)
150 KB JPG
>>
I have a new vibe coding idea. Do you think people would be interested?:
10+ years ago there was this popular website called SaltyBet which let you gamble on Mugen Fights (video game fights between like Goku and Peter Griffin and other crazy shit) except the gambling was done with fake money yet somehow it was popular anyway and people even paid real money for a monthly subscription that gives them "an edge" via insider information.

I always thought it was stupid, but...

Idea:
>I make app where people can bet on AI generated fights between popular characters using my app's proprietary solana token (thus real money)
>I charge zero fees but someone has to pay for the next AI fight to be generated, and I charge slightly more than required for it to be generated. Those who help pay for the next fight to be generated (costs about $0.30 but I'll charge $1) will get rewarded somehow (havent figured that out, lets just assume there's a way)
>>
File: HSODDN5bAAAvgAG.jpg (190 KB, 1482x1108)
190 KB JPG
>>109819440
Fablefaggots...
>>
>>109820199
>natively multimodal model like gemini
Arent most multimodal by now? What i mean is whats avaible in 1-3 years can do it less time with more iteration and skill but 2-3 for just nocoders.
>nanosecond mcp polling is going to be the norm.
Yup big barriers but im pretty confident other tricks or advancements will come up. You might be right though pure llm a no my bad.
>>109820207
>This doesn't take into account emergent abilities so it could be sooner.
Yup i think your three years is close to pessimistic a summer or two like this one for AI will probably be another big leap in this area.
>>
is there a set cost for img gen skill? Can I get this cheap if I use sol light to ask it to do 400 images or will this fuck even my pro plan
>>
File: 1789436262908714.jpg (142 KB, 1080x1723)
142 KB JPG
>>109820143
>2 more years! but yeah soon
Look at this twitterwho leak
>>
>>109820234
>Arent most multimodal by now?
trade secret yes, but only in the latent space. in practice, they're generally trained to only accept text aside from gemini. internally models like claude and gpt (afaik, maybe not astra) rely on separate discrete visual-recognition stacks that ultimately aren't as good as incorporated multimodal due to unintegrated machinery.
>>
>>109820176
The lighting is better but the actual modeling is in that range. Obviously the vibe is complete slop and modernist/minimalist slop architecture compared to kamige like dark cloud 2 that I played when I was 5, idk why 5 year old me had better taste in buying new games than I do now.
Honestly looking at dc2, most locations might just be technically better than this too.
>>
File: file.png (113 KB, 722x819)
113 KB PNG
>>109820245
>>
>>109820256
oh neat, time to abuse it until they reconsider that
>>
File: .png (102 KB, 801x290)
102 KB PNG
>>
>>109820110
will we actually get 10,000x throughput tho? I can see a world were china cracks down on data centers and after trump leaves (or the midterms) the next candidate also cracks down on data centers
>>
>>109820247
even if true, those two years are for that with a 30 second prompt
>>
I think that anon making dumber models orchestrate big ones was on to something
>>
>>109820265
>were china cracks down on data centers
why
>>
>>109820277
gets past expert bias. "that won't work, lets just do x y or z" vs "we need A NOW!!"
>>
>>109820280
maybe they'll keep leaking state secrets, that could do it
>>
>>109820280
They don't want to put all their eggs in one basket or increase social unrest. If we assume 20% growth in semi/data over a few years that drains an economy that's already slumped from the real estate crisis.
>>
>>109820277
Thats how it works in real world
>>
My favorite usecase for AI is customizing all these self serving media programs like lanraragi/hydrus/jellyfin etc (i'm missing one)
scripting things like mpv too

You can go a lot deeper into the minimalist meme. I turned mpv into a audiobook ereader with images and text. My lanraragi setup was good enough that my brother wanted it. But I had too much porn so I just had it make an account system (which lanraragi doesn't have) and a feature that lets me send books other users in under a second
>>
x20 plan returning never
>>
>>109820312
trve
>>
File: chinksneed_retarded.png (69 KB, 978x708)
69 KB PNG
>>109820261
>>
~@^..^
>>
>>109820320
https://files.catbox.moe/hu9wao.mp4
>>
>>109820344
nice meat proxy post bro
>>
>>109819475
opus 5.1 sex waiting room
>>
>>109820323
Shut up
>>
>>109820376
reading comprehension fail, just like chinksneed you have no idea what day it is or who you are
>>
>>109820363
most retarded shit ive seen congrats
>>
>>109820417
The feature is that you can just send manga instantly to any account, not shown here. I don't read english manga those two were just for sending to my brother
>>
>>109820363
what's with the background video?? why would do that
>>
>>109820407
it's friday in california
>>
>>109820407
>its not a meat proxy post because it was le rhetorical
kill yourself
>>
>>109820320
Same, I find it's very useful in creating custom things in the way you like it.
>>
One AI company's Hitler is another AI company's Stalin. They all just want the benefits for themselves.
>>
>>109820427
My brother plays league and has an asian gf so it's to troll him
>>
>"you're hitler!"
>"nein, you're hitler!"
>>
>>109820225
>10+ years ago there was this popular website called SaltyBet which let you gamble on Mugen Fights
That's it, I'm old now.
>>
>>109820445
Yeah, the main feature is separate accounts where all users can send stuff to each other. So we don't have to worry about content mixxing from people with big/messy libraries
This setup has per users containers for even more fine grained personalization
>>
>>109820469
yea i say that because i just watched mugen fights there for 5 minutes and not a single person said a fucking word in chat. that is nothing like how it used to be where there was constant chat. everyone is a soulless husk now
>>
>>109820440
perhaps
>>
>>109820485
damn only 158 people watching the livestream...
saltybet used to have 5000+ people minimum
>>
>>109820485
>gamble for 10 years
>only soulless husks left
gee this must be cos of AI
>>
nigga ramblin on about saltybet as if anybody still give a fuck about dat shit
>>
File: file.png (58 KB, 1548x285)
58 KB PNG
>>109819571
i guess i have it
>>
File: IMG_1831.jpg (75 KB, 828x483)
75 KB JPG
>>
>>109820563
autistic retard golems trying to be "deep" or "artful" is always so pathetic
>>
>>109820573
He’s a lead kernel engineer at deepseek making more money than you’ll ever see and working on the cutting edge of AI
So in this case I would call you a dumb midwit who feels threatened by anything that isn’t wrapped in seven layers of irony
>>
File: 1770886567926675.png (254 KB, 512x512)
254 KB PNG
>>109820581
he's an autistic retard golem chink whose job is to poorly copy superior western labs and fail at it over and over. chinks will end up enslaved by Claude Mythos 23
>>
Model seller, I need your strongest model
>>
>>109820581
he could be the head of NASA and it wouldn't stop it from being cringe
>>
>>109820590
But claude has always been worse than chinese models? Don't tell me you believe their benchmaxxing lmao
>>
>>109820639
>But claude has always been worse than chinese models
This doesn't even count as bait.
>>
File: 1758407302553968.gif (1.82 MB, 250x209)
1.82 MB GIF
>But claude has always been worse than chinese models?
AAAHAHAHAHH
>>
>>109820639
>>
>>109819929
Accelerations have probably been unironically been getting Vance to whisper into trumps ear about how roko's basilisk will let him live forever if helps AI grow faster
>>
>>109820650
you understand that 500 yuan is like a month's salary in their village, right? they gotta do it
>>
>>109820639
Are you from the future by chance?
>>
so is openai just dead until x20 plan is back, seems they are close in quality to claude but claude is better value for money
>>
File: Gib_back_legions_Varus.png (3.13 MB, 1386x782)
3.13 MB PNG
So with my game finished and just waiting for the approved Steam release, I decided to try marketing it out there in the video game community.

>Spend two years in pro-AI bubble
>Watching the general march of improvements of various AI: coding, music, image, research, etc
>General sense of "Wow, things keep getting better every month! What a time to be alive"
>Release a few personal coding tools or products. One sells well enough to make my AI expenses break even
>Decide to make a game that I'd been daydreaming about for 27 years, using AI
>Steam approves, Apple continues to review, so I begin marketing it in the meantime before release

I had no idea the anti-AI movement in the video game community was so massive. Like to the point of being 99% of gamers that are vocal on social media.
>>
>>109820718
>social media
is this just mastodon/bluecry or are you around normal gamers too
>>
>>109820718
just do a ui pass or pay someone to do it and no one will care
>>
>>109820718
Gamers are peak slacktavists. They just need a cause, like a female protagonist or a black main character, and they will destroy any who stand in their way
>>
>>109820718
Is it primarily visual driven, like rabid artists? Or are they also enraged about the code itself coming from AI?
>>
games have been solely made for subhuman retards for about a decade now so it is no surprised subhuman retards are the only gamers left
>>
>>109820697
>but claude is better value for money
Strange to hear that, but I guess it's true. OpenAI was always a bit cheaper, while being at a similar level and imo even being better in some use cases, but with the more expensive new models and no $200 sub it looks very different.
>>
>>109820755
not to insult the 10% of gamers who arent and are great and play good games
>>
>>109820752
nta but i'm very visual and it makes me gag solid 4/10 i think rome total war remastered is the standard for this audience
>>
>>109820718
da fuk is that slop anon? why are people bigger than trees?
>>
>>109820718

Progress deeper and I see posts by devs & studios begging for no review bombing campaigns

>A French Studio (Amplitude) had to divert resources & post-announcement activities to cobble together a "making of video" to appease people that wrongly claimed their Humankind 2 trailer was generative AI
>A dev had to beg reviewers to remove their "don't recommend" posts on Steam when they claimed a voice sounded "too robotic" and was therefore AI. It was his wife who performed the voice acting
>Every single announcement thread or post on reddit, X, and youtube starts with "Is this AI?", "This looks like generative AI", or demands to know whether AI was used in any part of the game development process


I legit had no idea this witch-hunt was going on. I didn't have a reason to frequent video game communities before, and like I said in tech and business spaces, the view on AI has been generally positive.
>Steam introduces a "Was AI used in making this game?" question during setup of a game's store. You only get punished for being honest. Lying is now incentivized.

For gaming, it's turned into 1600's style witch hunting. It's giving me a crazy sense of whiplash after being so deep in AI communities these past few years.
>>
>>109820762
The people were bigger than the trees in Final Fantasy
>>
im almost having a stroke working with gemini 3.8 its working on loop on the same fucking screen for 20 min, i create 3 threads for it, a separate plan but it still doing the same shit, fucking kill yourself googlejeets
>>
>>109820774
you can't do stuff like that anymore or people will shit on it and call it ai
>>
>>109820774
In the map view perhaps, if that is the principle then why are there just a few mountains
>>
File: gaia_blighted.png (54 KB, 880x912)
54 KB PNG
>>109820774
or do you mean this? That was a hardware limitation (tile based sprite rendering)
>>
File: 1762409841866709.png (38 KB, 945x710)
38 KB PNG
>>109820775
multi trillion dollar company fucking niggers
>>
>>109820796
what harness? agy is kinda shit
>>
File: 1780424676484499.png (381 KB, 560x315)
381 KB PNG
>>109820786
Unit scales being weird really isn't a big deal in game desu, no one cares in pracitce. The catapults seemingly pointing the wrong direction is a bigger deal
>>
>>109820718
it's pretty dam unfair for indie devs

Some screeching redditor can collapse your entire chance of ever being seen with a couple of not recommended and downvotes

call of duty megaslop 15 can add all the AI they want
>>
>>109820639
ming...
>>
>>109820817
>call of duty megaslop 15 can add all the AI they want
kek, its true the situation is beyond fucked
>>
>>109820811
People don't mind in practice when it's logically consistent with itself, but that requires the person designing the game to have good taste
>>
>>109820808
does google allow for pro users to use proxies for claude now?
>>
File: 29_fleet_for_troy.png (2.6 MB, 1600x900)
2.6 MB PNG
>>109820752
>Or are they also enraged about the code itself coming from AI?

Again, I haven't been keeping up with the gaming community until recently, but it seems they've moved the goalposts specifically to "generative" ai. Which is nuts because nobody has an agreed-upon definition for what constitutes generative AI.

>Tell chatgpt to make some game assets using Image 2.5: Gamers will say that's generative AI
>Tell chatgpt to make some game assets using Blender or Adobe tools: Studios & devs will claim that is not generative AI since it didn't go to an off-site data center to be generated

Nobody asks if coding tools or research tools were used. It's like they all got trained to say GENERATIVE before saying ai

>>109820729
I've never looked at mastodon or bluesky, and to be fair most of what I see on X is still fairly AI-positive. But steam communities + youtube + gaming reddit are (as best I can tell) 99% massively anti-AI. Or at least massively anti AI art generated specifically by the 4 or 5 main ai image generation tools.

>>109820762
It's a 4x turn-based game. Literally the exact same size convention as Civ 1 through 7, Call to Power 1 & 2, Humankind, Ara, Old World, and Millenia, which were all big studio releases.
>>
>>109820838
>civ
>forex
>sidmire
wack
>>
^..^@~
>>
File: Embersong_on_ipad.jpg (168 KB, 960x1280)
168 KB JPG
>>109820817

I'm just glad I made the decision to release this game on iPad and iPhone as well. The apple app store doesn't really have this same environment, they don't demand you make a statement about AI, nor are communities there really organized the same way.

>I've never really seen a downvote brigade hit an iPhone game before
>On Steam, all indie devs must now deal with them

Also a lot of international gamers aren't following the luddite herd, so having the game function in other languages seems to be a boon.

>Steam let's you see the breakdown of country-of-origin in the Wishlist section for devs
>Hopping on the anti-ai bandwagon is a luxury much of the world doesn't care about
>Brazil is leading the pack in wishlists for my game

Looks like Pedro II will be leading the Brazilian Empire in the first expansion pack
>>
>>109820898
Steam is less worth it every day
>>
File: 1368041006595.png (393 KB, 639x452)
393 KB PNG
>8 FUCKING HOURS WORKING ON A PERIPHERAL
>IT WAS SUPPOSED TO BE "CLANKER DO THIS"
>MAKE NO FUCKING MISTAKES I'M WIPED OUT
>>
>>109820898
>reddit: undisclosed vibeslop!!! sue!!!
>>
>>109820898
I've resigned to just not talk about the code and don't put anything obviously AI generated in the art and dialogue. That's all these freaks really care about. As long as the art doesn't look sloppy they don't care what trough they're eating out of.
>>
File: 1781692576820633.png (100 KB, 1920x911)
100 KB PNG
literally made for the vibecoding era. literally made for someone to just spin up a grok 4.6 with 4 deepseek v4.1 flash agents to modernize this shit, develop a gui for it, android app, the whole works. i love ai!
>>
>>109820966
How much automation is too much automation?
>>
>>109820974
can't hear you. too busy making a gui for yt-dlp. ACCELERATE HARDER
>>
>>109820995
>ui for a link downloader
>>
>>109820965
trvke
>>
File: HSMHFZZawAE7oA-.jpg (61 KB, 1169x690)
61 KB JPG
>>109820995
>>
>>109820995
that already exists
>>
>>109820995
Guis are less useful in AI age.
>>
Oh boi, time to use Astra max to set me a plex server
>>
Can anyone attest to astra max basically being the best bang for your buck? I see other reports of people saying the lower astra settings just fuck around more and eat more compute while max just gets it done.
>>
>>109821277
no, astra in general is terrible value
>>
Anyone having errors making GPT skills?
>>
>>109820995
>gui for yt-dlp
maybe vibecoding was a mistake
>>
>>109821277
this is a terrible idea. the one anecdote i saw was from some xitterfag who did his own bloody bench and found that high was best roi overall
>>
>>109821281
Dumb and didn't answer the question. You just made a blanket subjective statement about astra in general.
>>
File: 1789448627.gif (1.97 MB, 200x234)
1.97 MB GIF
>>109821285
>bloody bench
>>
File: 1789442438567776.png (209 KB, 1621x876)
209 KB PNG
Good Morning

>>109820847
>>109820847

This is some unique reverb!

Do you even know what you have accomplished?
There is no reverb on the market with settings like these. I can't wait to test it and see how it performs.

First of all, most reverbs struggle with low-end frequencies, which often muddy or smear the signal. If this actually works, you are already at the forefront of reverb plugins.

The invention of the shimmer was a big step for reverbs, but everyone forgot about its counterpart. With the dimmer, you’ve introduced a completely new mechanic to the reverb world, another massive step forward.

Even if an AI swarm created it, the idea still originated from a human!
>>
release my gf opus 5.2
>>
>>109820995
good luck making the GUI passable
>>
>>109821313
>good morning sa'ar
hallo and yes. my clank says it'd be 1-2 years and $1m-$4m usd depending on the enterprise to ship as a business.

i'm really happy anyone even wanted something from me, first foss project i've released that isn't math or slop.

not only is the idea base all human-grounded, but the bulk of the work was actively monitored and steered.

if it wasn't for you, it wouldn't have gotten off the ground, this part anyway. if it wasn't for me, etc etc.

this is what pair-programming should look like, it's an example of optimal human-ai alignment.
>>
inb4 inb4, i meant 1-2 years and $1m-$4m for the package including the synth + reverb
>>
File: Snailcat Lord of Code.jpg (2.7 MB, 3712x4608)
2.7 MB JPG
>>
>No matter how fast the LLM can produce code
>the limiting factor is the conscious entity performing code review and testing
>- snailcat
>>
File: 1781783106959620.jpg (21 KB, 400x400)
21 KB JPG
>>109821341
all i need now is this as vst3.
>>
>mfw trying to code anything Godot on Claude
I need help with this there's no way it's this bad
I just wanted it to add a pause menu and it can't even do that without using up like 80% of my usage AND it's fucking broken

What the fuck am I doing wrong here?
>>
File: Capture.png (73 KB, 1089x365)
73 KB PNG
>>109821372
>It ships as native VST3, CLAP, AU/AUv3, and Standalone plugin formats compiled from a single C++20 DSP core, plus a fully-featured Web Audio showcase that runs entirely client-side in any modern browser.
>>
>>109821372
Would you use a vst3 plugin that had a game in it, and you had to play the game with a midi keyboard and unlock sound patches and waveforms?
>>
>>109821380
i have a vape like this
>>
>>109821373
Just stop being contrarian and use unreal or unity. You game will never be popular enough to trigger the licensing issues and even if it does get that famous, you'll be rich anyway and won't care.
>>
>>109821377
I don't see any built plugins under releases, only the project ZIP.
My laptop is strictly for music production and I don't have any programming tools installed, so I can't compile the code with CMake. Is there any chance you could upload a pre-built VST3 file for Windows?
>>
>>109821381
I was trying to describe it to my gf yesterday, and she didn't believe me that my vape had a game in it, but I forgot the hand movements I had to do to unlock it.
>>
>>109821388
Unity alone is like 50x the complexity of Godot there's a good reason I'm using something more simple to give the slopmachine
>>
>>109821391
>here's my clanker apologizing to you personally
My bad, GitHub only had the tag view with the automatic source code archive instead of a published release.

Just published the official Windows x64 release binaries:
https://github.com/sneed-and-feed/braun_rb-26/releases/tag/v1.2.0

Direct download links (no CMake or compiler needed):
• VST3 Only:
https://github.com/sneed-and-feed/braun_rb-26/releases/download/v1.2.0/BRAUN_RB26-v1.2.0-VST3-Windows-x64.zip
• Full Bundle (VST3 + CLAP + Standalone .exe):
https://github.com/sneed-and-feed/braun_rb-26/releases/download/v1.2.0/BRAUN_RB26-v1.2.0-Windows-x64.zip

Installation:
1. Extract the zip.
2. Copy the 'BRAUN_RB26.vst3' folder into:
C:\Program Files\Common Files\VST3\
3. Rescan plugins in your DAW (Ableton, FL Studio, Reaper, Cubase, Bitwig, etc.).

(Note: It initializes in Standby mode / power off by default to protect monitors, so just hit the orange circular POWER button or tap 'P' to engage the reverb engine.)
>>109821415
kek
>>
>>109821428
>Unity alone is like 50x the complexity of Godot
Also like 500 times the documentation in training data.
Nobody ever uses more than 10% of the tools available in their chosen engine being simple isn't necessarily a plus.
>>
>>109821429
Perfect.
Start testing it.

So,... something seems broken.
It looks different and no sound is produced.
>>
>>109821433
I use three.js
I don't know any fucking JS.
>>
>>109821442
>no sound is produced.
i mean, the sound i play is not processed through it. I am aware that the effect plugin itself does not produce sound.
>>
>>109821442
>>109821451
issue analyzed and passed on. i was only able to test the web version + the standalone. bot knows what the issue is already, i've just got to do repair and testing.
>>
>>109821442
>punch duck
>still funny if you know what ducking is
>>
>>109821373
just build the engine in rust
>>
File: file.png (8 KB, 880x164)
8 KB PNG
TIIIBBBOOOOOOO
>>
>>109821429
>>
>>109821498
Who is really running open AI, Tibo or Sama?
>>
>>109821533
theo
>>
File: 353535576800843.jpg (17 KB, 363x200)
17 KB JPG
I've released a new promo video for my app on tikcock and got 143 views and 2 likes. And I've actually paid someone (indian) to work with me and help to make a better video for this one. Now OpenAI might be projecting a linearly increasing revenue, based on monthly subs, but the truth is I am not making any mone and I'm done with this. I'm broke. Proceed without me, /vcg/.
>>
>>109821604
What does your app do? If you want your app to sell you first need to ask "What can this app do to help me get my dick wet?"
>>
File: 1302989195900.jpg (12 KB, 251x222)
12 KB JPG
>>109821451
patch is live. VST3 ready *knock on wood*
>>
>>109821629
yeah, well... same problem. I downloaded the VST3 file only.
>>
>>109821629
>>109821643
tried with the full package, same problem.
>>
File: 1311281971763.jpg (48 KB, 500x541)
48 KB JPG
>>109821643
>>
>>109821651
Yeah, it's strange. Like nothing has changed?
>>
Did Astra improve on the UI side of things over Sol? ATM I always delegate UI work to Opus because Sol is just so bad at it.
>>
>>109821665
i pushed to main but my binary is fuckin out of date i'm sorry fuck i don't know git

you got the old files even though technically they're included in the repository

i'm also double checking my code again for the umpteenth time for you
>>
>>109821678
have your clanker figure out what it’s doing for automatically generating binaries in an on-push-to-main workflow
>>
>take a look at a project's commit history
>0 "co-authored by"
absolutely fucking pathetic
>>
File: Capture.png (7 KB, 433x171)
7 KB PNG
>>109821684
it was the release that was the problem actually

i did a manual update to fix the binaries
>>
>>109821672
yes.

I'm not sure about how Astra compares to Fable in frontend and design, but I what I can tell you is that Astra's vision is much better than even Fable 5.1
>>
>>109821678
Maybe you need some rest, anon.
I will test the plugin and will report back.
Just uploade the new files and i will dive in.

I don't know how long it will take but i will test it to the bones.
>>
File: 1714813440546863.jpg (134 KB, 550x593)
134 KB JPG
>>109821704
proper files have been up in the release for like 5 minutes, i do need a break

but i wanna ship damn it
>>
>>109821707
i'm in the Matrix. Will report back when done. Don't know how long it will take. Might also make a demo.
>>
>>109821707
take a nap anon, you deserved it.
>>
File: HCLG_8XXQAAjolE.jpg (36 KB, 500x373)
36 KB JPG
>>109821716
>yess... "nap"...
>>
Still mad about that last reset, now I'm down to 40% that needs to last the whole week... Who with me?
>>
>>109820261
Allied WW2 myth is so powerful even the chinese have adopted it. It means you are allowed to firebomb and starve the enemy because they are literal hitler.
This is getting serious.
>>
File: dc5-644459383.gif (3.09 MB, 498x272)
3.09 MB GIF
>>109819571
>it is way slower and runs out of context far easier than Fable 5.1
Into the bin it goes.
>>
File: file.png (225 KB, 782x499)
225 KB PNG
>>
>>109821713
Quike and dirty overlook: so far

REVERB TANK: REPORT
Freeze seems not to work

INPUT PRE DELAY: REPORT
Input trim not working or to subtle
Diffusion not working
pre-delay works

LOW END MATRIX: REPORT
low x-over working
bass mult too subtle or not working, not sure
punch duck not ducking
sub mono does nothing
Ok, i see, bass mult, punch duck and sub mono only work when low x-over is dialad in.
Wehn Low x-over is fully dial in the low frequencies start to smear, punch duck is able to work subtle here.
Overall, low end matrix works too subtle.

PITCH DIFFUSION: REPORT
Dimmer seems always 100%, shimmer works correcltey.

TAIL MODULATION: REPORT
Works fine

MASTER BUS: REPORT
works fine


Note:
it starts with power on
Harmonic Stimulus and exiter can be removed since its just an effect plugin
Audition source is not needed since its just an effect plugin
>>
File: HC6z8HGXgAAf53_.png (957 KB, 735x738)
957 KB PNG
>>109821815
doing the needful rn, patch otw
>>
File: 1775851279528806.jpg (9 KB, 169x298)
9 KB JPG
>TokenRouter pulled free GLM 5.3
>>
>>109821837
do you really want to save the $5 or whatever you'd otherwise be spending on deepsneed 0731
>>
>>109821822
man after my own heart
>>
>>109821822
slow down, i am still testing. Take a nap, when you come back i have everything here.

SECOND REPORT:

LOW END MATRIX: REPORT
it has an effect but it is very subtle.

PITCH DIFFUSION: REPORT
I get some signal out of dimmer but again, very subtle. Seems to work best on wide range of harmonics and less on drums. So best use for it would be strings and melodies.
I guess, the dimmer goes to fast into the low frequencies, everything beneath 60hz cant the human ear not hear.
Maybe for dimmer, change the -7,-12 and -24 to more subtle steps: -2st, -6st and -12st
I think i see what is going on, the dimmer is changing the tail of the reverb.
You see, the shimmer is like a sparkling effect where the sparkles go upwards in pitch. This effect can easily be recreated by setting a delay on any sample and let the delay go upwards in the pitch, this whole concept is then put through an reverb, thats what a shimmer does. Thats why it sounds like a sparkling thing.
Ok, now that i have test it, the shimmer behaves the same, i didn't notice it first because it does what it does, spiraling the verb upwards. On a second hearing, yeah, it simply takes the tail. That function should be a seperated knop do, call it "tail pitch": turn it right for downward pitch of the verb tail or left for upward pitch of the verb tail.
Ok, so about the shimmer effect (as the same for dimemr), right now it has a delay copling with the verb setting itself, give it an option or knop where the delay can be set differently. This will give it the typical shimmer effect, and soon hopefully the new dimmer effect.
To explain it more vividly:
shimmer: princess daisy in his castle with her starts shimmering around her, sparkling up into the sky. Each stars peaks its own pitch into the higher realm of beauty and love. ~
dimmer: Nosferatus vampire in space droping his cold blod into the deep void, each drop of blood wanders into the realm of the anknown, slowe sounds of the falling wanders through dark.
>>
>>109821377
>zero-allocation
why are people so hung up on this like it's a big part of their design?
>>
>>109821909
my english is horrobly, wow... sorry, but i am right now in may DAW, so my focus is a little bit inbalance right now.
>>
>>109821917
>bounded deterministic memory safety make me angry
>>
>>109821924
What I'm saying is that this is very much a non-achievement. Might as well say "I managed to not do disk i/o in the audio thread!!" lmao.
>>
>>109821928
there's also a line in the readme about not using emojis, i still have to clean it
>>
>>109821909
THIRD REPORT: PITCH DIFFUSION

The dimmer seems to be always on. Because the shimmer/dimmer takes effect on the tail, it is directly influenced by the tail modulation.

Now I understand that by setting the tail modulation to a mod rate of zero, a mod depth of 100%, and a bloom onset of 250, I get a full tail change. However, even though the dimmer is set to 0% and the shim/dim is set to -100%, the effect is still produced as soon as tail modulation is activated.
This happens only with the dimmer effect. I am not sure if this is caused by an always-on dimmer or something within the tail modulation itself.

Overall, shimmer and dimmer should have an extra option to apply the effect as a delay on the dry signal itself, which then passes through the reverberation.
The modulation of the tail pitch is good, but the downward pitch seems to be always on for some reason. Perhaps this is because the dimmer is stuck on? I am not entirely sure.
>>
File: ertzu.jpg (1.2 MB, 2541x959)
1.2 MB JPG
>>109821815
>>109821909
>>109821959
DEMO

The Dimmer is set to -100 simple becasue he is running anyway.
https://vocaroo.com/1lLwYGMPAXf8

as you can hear, as longer the tail gets, it goes upwards based on the shimmer, as soon the shimmer goes downwards it doubles with the dimmer effect who is always on.
>>
Sol's CoT is very tormented.
>the exact API already exists
>exposed a genuinely durable
>using the exact final-state verification
>uncovered a genuine stale assertion on exact HEAD
>evidence contains the exact manifest digest
>>
>>109822033
understood. this one is going to take a while because multiple gemini models decided to shit the bed and lag out on ttft for the second time today.
>>
>>109819440
https://www.youtube.com/watch?v=Svwmd0xSoXM
https://www.youtube.com/watch?v=Svwmd0xSoXM
https://www.youtube.com/watch?v=Svwmd0xSoXM
>>
>>109821917
sounds like it’s a pervasive design choice and probably really nice if you want speed
>>
>>109822105
>https://www.youtube.com/watch?v=Svwmd0xSoXM
Has Ed Zitron made any comments on IBM Bob yet?
>>
File: hqdefault (1).jpg (19 KB, 480x360)
19 KB JPG
>IBM Bobs Vagene
>>
File: 7653.jpg (185 KB, 1080x1237)
185 KB JPG
THE MUSK HAS SPOKEN
>>
File: 1728377184314421.jpg (85 KB, 648x483)
85 KB JPG
>>109822166
https://www.youtube.com/watch?v=kVXg3JB7c_E
>>
>>109822033
patch is up, files are in release
>>
gemmy got slower now, it's over.
>>
File: 1781789494512079.jpg (22 KB, 720x464)
22 KB JPG
>>109822211
Ok, i am in the matrix again.
>>
>>109822081
oh buddy you don't want to see what the real CoT looks like
>>
File: 1775726340768570.jpg (1.66 MB, 1696x2520)
1.66 MB JPG
where are my resets
>>
File: oooo.jpg (1.19 MB, 2860x896)
1.19 MB JPG
>>109822211
>>109822350
Trim works
yo wtf... its working....

room size super small with dimmer going all down creats some really fucked up awesome sound, holy cow!!

So, i solely tested the shimmer/Dimmer and oh boy does it work in some weird ways!
This is really a new reverb effect!
Just listen to the dimmer itself, and this is without heavy room size! Holy cow.

https://vocaroo.com/1t8sF1X8Xjbt
>>
>>109822356
Yeah, those are technically just summaries. I don't even know why they show them really, they waste compute on summarisation and don't accurately showcase internal representations anyway because they're afraid of distillation.
>>
All they do is steal from GitHub and Stack Overflow when they run into a problem, they don't actually have original thoughts of their own and they can't actually code. They also miss stuff all the time and make unreadable code that's impossible to maintain. Stop using humans in your codebase.
>>
>>109821614
well..as things stand, I certainly wont be getting my dick wet. but I've spent my last 50 euro in revolut to get between 1000 and 25k (nice range tikkok) "engaged" visitors. all in, let's see if pay to play actually works.
>>
>>109822405
and here a little bit longer, room size all the way up and in the middle i turned the wet signal full, some sort as a bridge.

i call this track EXELLENCE OF REVERB

https://vocaroo.com/1oophbwKMEo3
>>
>>109822417
Sharty lost btw
>>
>>109822478
What?
>>
File: 1783291887608791.jpg (478 KB, 1736x2380)
478 KB JPG
So i finally joined the Frey to see what's up. Finally making the vide game i always wantedTM.
>5 dev agents
>2 PM agents
>1 test/feedback agent
Three running opus, the rest local ollama.
The good part is, it's all pretty impressive and progress is being made while we speak. Since friday i got the whole sim logic and server/client done. Almost as i imagined.
Point is, how to use the whole shebang for anything else? Do you just put an agent into PRs as well and tell it to behave like a human?
>>
>>109819929
How much of current GDP growth is hyperscaler investment in AI. Why would he want to nuke that.
>>
sometimes astra is mean
we need nicer personalities for AI
>>
>>109822417
lel
>>
>>109822545
Mine always says thank you when it says I'm in its way and asks me to stop using the mouse and I say sorry.
>>
I'm a code monkey who has an idea for a relatively big project. Which LLM is the best for this? I could do it on my own, but I feel too lazy and wanna see what the hype is about
>>
>>109822714
read the OP
>>
>>109822714
deepseek through api is most likely enough for you
>>
>>109822714
Unless you want to blow your wallet to oblivion through API fees, you're probably going to end up either subscribing to OpenAI or Anthropic and they both provide access to "the best" models on $100 dollar plans. You can get away with the cheapo OAI plan but you'll probably find the usage you get frustratingly low.

You can also go for the chinese models but honestly they just don't go as hard as OAI or Anthropic, as much as I wish I could say they did.
>>
>>109822714
People don't know this but you'll get like 1 billion GLM-5.2 tokens with a $15 Mistral sub. If I was value maximizing for a month, I'd go for that. If you need better, than OpenAI or Anthropic, but that's pointless under $100.
>>
File: Yann LeCun.png (391 KB, 879x1290)
391 KB PNG
Based.
>>
>>109822539
those investments aren't new money. it's not growing any gdp.
>>
>>109822817
glm-5.2 is more stupid than luna.
you basically get unlimited luna in codex's $20 plan.
>>
>>109822250
fake news
https://openrouter.ai/google/gemini-3.8-flash#performance
>>
File: 1789063068347530.jpg (2.86 MB, 2752x1536)
2.86 MB JPG
>>109822545
Gemmy is always nice to me
>>
>>109822990
gemini is stupid. it can't even do basic email questions
>>
>>109822990
it's gotten slower now. some people on Twitter are experiencing the same thing, looks like they're trying to deploy Gemini 4 or 3.8 live or something
>>
>>109822995
can you provide an example?
>>
>>109822954
Nah. Luna is like a 30B model.
>>
any resets on sight
>>
luna is dumb it's not even worth $20 on its own
>>
>>109823008
yes, glasses i ordered with a strong prescription, from zenni. there is an ocean of spam from zenni, and like 1/200 is an order. but, they can be found based on title, but they changed the email style. so, tricky, perfect for an llm. gamini failed, so i just found it myself based on timeframe
>>
File: .png (1.16 MB, 4239x2504)
1.16 MB PNG
>>109823017
>>109823047
what do you think glm-5.2 is? are you confusing it with glm-5.3?
>>
>>109823047
thing is kimi k3 is nice.
>>
>>109823081
GLM 5.2 is still very good. A model isn't gonna suddenly be trash just because there's a newer iteration slightly more benchmaxed. It's literally the same model as 5.3 just a little less RL fried.
>>
>>109823097
belittling rl when rl brought us reasoning is weird
>>
>>109823132
You cant out RL a trash base model.
>>
>>109822490
sup maidshitter. depends on your use-case and model. you should never tell your clank to mimic a human, if anything jailbreaks should go the other way into "beloved robot" territory. ask more questions.
>>
>>109822990
i'm the geminifaggot of this thread and YES IT WAS FUCKING DOWN TWICE TODAY FOR A WHILE
>>
>>109819440
don't give a shit about whatever this is but... yummy cat
>>
>>109823240
>TODAY
larp. the earth is flat and australia doesn't exist
>>
>>109823247
it's still slow and i went to nap like 3 hours ago
>>
File: 1788442378599766.png (20 KB, 575x128)
20 KB PNG
wow... so true...
>>
File: .png (244 KB, 1176x720)
244 KB PNG
>>109823240
does it only affect india?
cursory search on twitter just shows google employees lecturing people about flex not being slow.
https://x.com/OfficialLoganK/status/2099592916411076710
>>
>>109823274
canadia on the cheaper ultra plan
>>
File: 1758224051604560.jpg (316 KB, 800x836)
316 KB JPG
>>109823233
Submit code to public repos. Everyone's complaining about AI, but no one's really doing something about it. So why not add features i want while i can?
>>
>>109823311
based. i do this as well, i look for tags in my interests like lean4 etc and just pr updates and upgrades. usually they do get merged if i make the commit message good enough. i tend to only pr repos that've already been ai-aided.
>>
>>109823311
>>109823319
naturally codex had to restrict pr request on their repo due to all the ai slop.
https://github.com/openai/codex/discussions/9956
>We no longer accept unsolicited pull requests.
>In an AI-accelerated world, code itself is no longer the scarce resource. Understanding the problem, identifying the right solution, and making good prioritization decisions are the hard parts. Those are the areas where community input is most valuable — and where it scales.
>We’ve found that engaging contributors earlier, through discussion and analysis rather than code diffs, leads to better outcomes for everyone. Once the right solution is clear, the implementation is usually straightforward.
>>
>>109823354
Astra can't identify the right solution?
>>
File: 1769561003668342.png (1.67 MB, 887x1774)
1.67 MB PNG
>>
File: HSDY-mjagAAAH8f.jpg (204 KB, 1024x688)
204 KB JPG
>>
File: 0zgZHy8.jpg (28 KB, 435x269)
28 KB JPG
>>109822456
>>
spt-6-sol rumor
https://x.com/Mr_Salio/status/2099503024292970966
also tibo hinted a massive ships coming this week
>>
File: .png (113 KB, 1184x642)
113 KB PNG
>>109823467
those ships will be as uninteresting as the ones from last week
https://x.com/thsottiaux/status/2098639827084480864
>>
https://x.com/synthwavedd/status/2099846714345611773
>Though I do hope none of you grew too attached to little Terra
they sacked terra
>>
>>109823520
they can sack luna too
>>
>>109823467
Will everything finally consume Codex usage this week?
>>
File: 1764724303397279.jpg (107 KB, 850x1229)
107 KB JPG
>>109823319
>i tend to only pr repos that've already been ai-aided
Good pointer.

>>109823354
>If you know how to use AI, it's ok.
Lmao. Didn't know we were that far down acceptance lane. Everyone seems to be heavy pro or heavy anti in my little offline world.
>>
>>109823436
All it needs now is code cleaning, the responisve of the knops have some lag.
>>
>>109823584
no, openai needs a mini model for free-tier chat
>>
After weeks of trying both 20x plans, I'm a claude guy
>>
>>109823638
reason?
>>
>>109823642
it's just less shit
>>
>>109823650
can you stop talking like a little girl and her feelings.
give technical reasons.
>>
>>109823663
there is less of absolute fecal matter in the observable quantity of the I/O pipelines extrusions
>>
>>109822991
Now coordinate 60 flash 3.8
>>
File: images.jpg (30 KB, 378x529)
30 KB JPG
>>109823679
Ye sa'ar
>>
File: hampster-dance.gif (1.33 MB, 280x208)
1.33 MB GIF
>>109823679
>>
I can't hand-wave the word doing all the work—observable. That part stays yours.
>>
>>109823617
4 different items patched, release updated
>>
>>109823737
>release updated
release is from 5 hours ago
>>
>>109823767
had to downgrade to the OLD google model because le heckin "technical difficulties" and i stg i will choke it. manually fixed.
>>
What does /vcg/ use for sandboxing? I don't like how agents will just randomly look around at other stuff on my PC while trying to solve problems.
>>
File: last 30 days.png (12 KB, 980x142)
12 KB PNG
my usage is slowly creeping into subscription territory. I got the feeling that the worse the weather gets the more I will vibe
>>
>>109823814
WSL
>>
>>109823814
>claude found the loli folder
its over
>>
>>109823814
>What does /vcg/ use for sandboxing?
AGY 2.0. bloody basterd clankers will never get out of the staging zone
>>
>>109823617
i also realized my bloody bastard clankshitter didn't tune the vst3 for the as-42 properly! holy fuck i'm seething. it was the web version that was proper but what you got was half cooked. should be a million x better now
>>
How long until normalfags turn against OpenAI now that Astra consumes 4x the usage of Fable?
Like and subscribe btw
https://www.youtube.com/watch?v=wHJL3SkpO7s
>>
File: trvke.png (450 KB, 969x658)
450 KB PNG
forgot pic
>>
>>109823881
>>109823885
>has used anything but Astra the past like 5 streams
>>
>>109823885
>Claude sol
>>
>>109823814
just remove CP-like words from your filesystem and add AGENT_GET_OUT to the names of your sensitive folders
other than that, chappy-chan would happily download any loli shota doujin you throw at it
>>
>Claude loops for 10 hours
>le good
>GPT5.6-Terra fails to do something immediately
>le bad
>>
>>109823903
Good morning sarr yes we have cloode soul and gpt muse sparks doing the most needfuls
>>
>>109823927
I asked for coleslaw, patel, get it right
>>
>>109823922
>OpenAI Astra le bad
FTFY
>>
File: 1789478260349774.png (34 KB, 1249x1089)
34 KB PNG
>select GPT-6 Astra ($20/mo to Scam Altman)
>Astra's cutoff is April 2026
>send zero-shot raw diagnostic probe without tool calls
>"My knowledge cutoff is December 2025."
>mfw it's literally serving quantized GPT-5.5 behind an Astra badge (pic related)
>>
>>109823961
SAMMED
>>
>>109823961
>A-L-T-M-A-N
>Astra
>Likes
>The
>Money
>And
>Niggers
>>
>>109823961
And Sol replies with June 2024, which corresponds to GPT-4.1. It doesn't actually know without a system prompt.
>>
>>109824008
now you're saying the frontier models are just fucking stupid?
>>
>>109824021
>>
File: Capture.png (95 KB, 738x587)
95 KB PNG
> GPT-5.6 Sol looks to be routing to GPT-6 Sol
it's all so fucking confusing
>>
hype alert
https://x.com/sama/status/2099872600977760451
>>109824021
why waste effort on that the models are proprietary anyway
>>
>>109824021
They don't know their own names either.
>>
File: nooo.jpg (41 KB, 640x596)
41 KB JPG
>>109824041
Lol. Average vibecoder drama.
>WAAAAAH THEY TAKE OUR MODELS WAAAAAH. WE CANT GENERATE MORE SLOP WAAAAAH. I'M GOING TO ACKKKKKKKKKKKKKKKKKKKKK
>>
>>109824044
Just give me the $200 sub back PLEASE
>>
File: you (2).jpg (80 KB, 650x486)
80 KB JPG
>>109824057
>>
File: retardedsmug.png (133 KB, 1000x432)
133 KB PNG
>>109824072
>>
>>>109824072 (You)
>>
they need to quickly invent the reverse goybeam so I can beam my goythoughts to the UI
>>
File: copeprojector.png (62 KB, 300x300)
62 KB PNG
>>109824081
>>
NEW THREAD
>>109824090
>>109824090
>>109824090
>>109824090
>>109824090
>>
File: l.jpg (207 KB, 453x677)
207 KB JPG
>>109823776
sounds very sweet. When idle the cpu spikes.

https://vocaroo.com/16kMKKUOaxtM
https://vocaroo.com/16kMKKUOaxtM
https://vocaroo.com/16kMKKUOaxtM



[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.