[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: file.png (34 KB, 140x133)
34 KB PNG
A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

## 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
- (2026-07-16) Qwen announces plans to release Qwen3.8 as open-weights.
- (2026-07-16) Roblox announced Build, an AI workflow that turns text prompts into playable games
- (2026-07-16) Kimi K3 released and K3.1 announced. Performance reportedly comparable to GPT 5.6 Sol.

----

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

## Worth it for code, but the frontier models above are better
https://opencode.ai/
https://x.ai/cli

## Not worth it for code, but good for making sense of pictures
https://antigravity.google/product/antigravity-cli

----

## Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills — /grilling is a favorite

## Other editors / terminal agents / coding agents
https://osaurus.ai/
https://pi.dev/
https://cursor.com/docs
https://docs.windsurf.com/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

## UI/Frontend
https://www.figma.com/make/
https://www.anthropic.com/news/claude-design-anthropic-labs
https://uiverse.io/
https://ui-ux-pro-max-skill.nextlevelbuilder.io/
https://stitch.withgoogle.com/

## In-browser builders / hosted vibe tools
https://bolt.new/
https://replit.com/
https://docs.github.com/en/copilot/tutorials/spark
https://v0.app/docs

## Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.0

## Previous thread
>>109318112
>>
>>109322849
>140x133
>>
>>109322849
This is definitely me when I
>>
>>109322849
Baked in the best traditions of vibe coding
>>
The perfect image for a vibe coding general desu
>>
File: 1764165541964140.jpg (19 KB, 300x226)
19 KB JPG
>>109322849
>>
luds buttblasted as usual
>>
sorry i vibed this thread
>>
>>109322849
petitioning that K3 be added to the list of frontier models in the OP
>>
>>109323239
kubernetes?
>>
>>109323239
No, it belongs below Grok at best
>>
>>109323279
K8 > K3, makes sense to me
>>
>>109322849
I just 100% vibecoded a command-line synthesizer with deepseek through the browser.
It took me a week to get it to add all of the features I need. It outputs raw pcm to stdout. It can load and playback samples. You play the synth by notating the music in a "dsl" that you can just use bash with.
You can just have a bunch of heredocs as variables. You can just pipe them into sed or awk.
You can do math with loops and have the loop save stuff into variables that you can use in the "dsl." You can define an instrument as a bash variable. There's legato, chords, modality (you specify notes with scale degrees), adsr, lfo, phase, bias, overdrive, "power" (the exponent of the wave function), parallel voices, stereo output, panning, 16 bit, 24 bit. There's no post-processing, though. I figured I could just do that externally and load a sample.
I'm working on the manpage, now.

./seqosc --channels 2 --bit-depth 24 --seq "
140:q:c4:ionian;

sine(0.8:pan[0.5]:drive[1.2]:env[a:0.05,d:0.1,s:0.7,r:0.2]){1:w,_:q,2:h,3:q} &
square(0.6:0.3:lp[800]:lfo[amp{sine:3:0.2}]){[1:q,3:q,5:q]} &
saw(0.7:phase[0.25]:hp[100]:lfo[freq{triangle:4:0.05}]){2:w<0.3:0.7>3:w,4:w} &
triangle(0.8:bias[0.1]:power[2]:lfo[phase{square:0.5:0.25}]){5:w,6:w} &
sine(0.8:lfo[bias{sine:1:0.2}]:lfo[power{triangle:0.5:0.5}]){7:w}2;

square(0.9:0.4:lfo[pw{saw:1:0.3}]){1:w,_:h,2:w} &
sine(0.8:lp[500]:lfo[lowpass_cutoff{sine:1:0.5}]){3:w}3

" | aplay -r 44100 -c2 -f S24_3LE
>>
>>109323305
Everything between the double-quotes is whitespace-agnostic, so you can use each voice like a c-style function.
>>
>>109323323
damn thats crazy
>>
>>109323305
voice parameters, like sine(params){notes}
can also be used on individual notes, like -1+2(params) (-1+2 is an octave below 1 and raised by 2 semitones), or chords [1:dq,2:dq](params)
>>
>>109323284
>>109323279
>>109323294
low effort bait
>>
>>109323305
post what kind of sounds it makes
>>
>>109323357
fuck you benchod
>>
>>109323361
I don't have any accounts on soundcloud or anything like that. After I'm done the manpage, I'm gonna compose some things with it to test all of the edge-cases of the syntax, and then I'll put it on codeberg or something.
>>
>>109323379
https://vocaroo.com/
>>
damn, GLM 5.2 usage on Opencode Go is absolutely tiny
>>
File: file.png (3.19 MB, 1448x1086)
3.19 MB PNG
MODEL OVERLOADED MODEL OVERLOADED MODEL OVERLOADED MODEL OVERLOADED MODEL OVERLOADED
>>
>>109323357
Knock knock, who’s there?
>that’s low effort bait
o- okay
>>
>>109323376
Ironically you people are the ones hyperfixating on benchmarks, just give the models a try and make a decision for yourself, for once in your life.
>>
File: file.png (675 KB, 1200x1200)
675 KB PNG
>using claude to prep for interviews
>tell it to ask me some soft questions and then grade my responses
>it keeps giving me 4-6/10
>>
File: file.png (16 KB, 630x153)
16 KB PNG
lmao
>>
>>109323385
https://voca.ro/1aB5I7p3Vngk
./seqosc --channels 2 --bit-depth 24 --seq "120:q:a4:ionian; triangle(0.8:power[200]){------1:w}2" >| triangepower.raw
>>
File: 1774477719144721.jpg (332 KB, 1080x917)
332 KB JPG
>>109323444
>NO NO NO!!! YOU AREN'T ALLOWED TO MAKE ANYTHING!!!! I SPENT YEARS OF MY LIFE LEARNING HOW TO CUT AND PASTE CODE YOU CAN'T DO THIS TO ME!!!!!!!
quiet part out loud
>>
File: 1648782488439.jpg (31 KB, 400x505)
31 KB JPG
>what's D&C
you better learn what is it, because this site is nothing but infinite traps of that nature being set AND kept, no it's not organic, it is shilling
>>
File: dandc.jpg (107 KB, 1106x968)
107 KB JPG
>>109323475
?
>>
>When you reach the quota limit on codex, tool call are automatically denied even as the llm finishes the task.
Good to know.
>>
>>109323444
He's dumb because his knowledge will translate to way better use of the AI tools compared to "make me a game, make it good" types of people.
He should focus on that instead of fighting windmills.
>>
tiboooooooooooooooooooo
>>
File: glm 5.2.png (161 KB, 1334x799)
161 KB PNG
>>109323489
ok im out of juice everywhere
rather than use the m*p hack or using a reset, im gonna ask qwen 3.6 to work on this task instead
do you think it'll work?
>>
Anybody using ponytail? Did you notice a difference? Good or bad?
>>
>>109323608
I'm skeptical of this. The concept makes sense in theory but how does it work?
>>
>>109323575
idk, I'm not an engineer. I don't know what d&c is. There's a lot of things that start d and c.
>>
>>109323608
It's making comments and documentation terse, and I don't like that. Yes, even when asking to ignore the comments and documentation.
So I went back to normal mode.
>>
>out of fable credits on pro
owari da....
>>
>>109323628
Can't yoy just ask if to make documentation later?
That said I have to always ask to rewrite as shorter, because it goes oveboard. During normal use that is.
What agents.md commands are you using instead, if any?
>>
File: file.png (5 KB, 256x64)
5 KB PNG
>>109323525
TIBOOOO
>>
File: file.jpg (10 KB, 459x150)
10 KB JPG
>>109323706
we'll be the crying zero club
>>
>>109323694
I use one forcing updates to progress.txt every turn and always adding explanatory comments wherever possible.
>>
>>109323706
>>109323715
join local gang >>109323575
>>
>>109323622
oh nvm i replied to your post by accident lol
i guess that guy meant divide and conquer
>>
>>109323741
fuck you chink commie
>>
>>109323766
you are unbelievably brown. isn't it night time in your village?
>>
File: file.png (16 KB, 739x166)
16 KB PNG
>>109323706
>>109323715
Same but I have a long going task at 0%.
>>
>>109323773
i bet im browner than him
>>
>>109323792
Nobody's browner than me bloody benchod fuck you
>>
Crazy how I basically completely stopped giving a shit about computer hardware atp. Who cares if my CPU can compile 50ms faster, compilation time is not my bottleneck anymore. I boughted a 265k and 64GB RAM 2 years ago and don't even know or care what's out there at the moment.
Imagine if we have 2000tk/s Fable and you actually start getting limited by your tool calling speed instead of the model itself
>>
>>109323733
Maybe I am wrong, but doesn't exessive commenting and then reading it every new thread uses up a lot more tokens?
Some note taking, of course, it is a must, but it can easily have an opposite effect.
>>
>>109323812
Yes it does, if you're trying to save costs it makes sense. But if your documentation is too terse to be understandable you will waste more than you spend as you and your agents start to deviate from the spirit of the plan.
>>
>>109323806
Trump needs a campaign to do eminent domain on luddite's hardware, so many gamers wasting precious gpus and ram we could be using
>>
>>109323575
what UI is this?
>>
I don't want to be that guy but does fable feel dumber
>>
>>109323842
this but foids macbooks
>>
do different harnesses/models charge you usage for waiting for commands? say you run a build command for a fairly large codebase and it takes 5 minutes, assuming you don't lose the cache, that's basically free, right? you aren't charged for keeping a connection up?
>>
File: file.png (179 KB, 608x629)
179 KB PNG
>>109323842
Trump is one melty away from signing an executive order to ban open source models in the US, migger
>>
>>109323860
He is so fucking based if he does this.
>>
>>109323860
>if i try to make it about israel, it will make my intentions less obvious
>>
>>109323860
>banning people from running programs
and nothing of value will be lost, because it is literally impossible to enforce this and the chinks don't give two fucks what mutts want
>>
>>109323881
yes anon, who doesn't have a b300 cluster, it's practically impossible for the government to have a list of everyone with $500k worth of hardware
I have one in my garage, nobody knows about it
I'm about to buy my second cluster actually
>>
>>109323854
For API no.
For usage based billing they can do whatever they want, especially if you use a closed source harness. Anthropic used to charge you more if you had the string HERMES.md in your git history.
>>
>>109323890
if you're operating at that scale you're at enterprise-scale anyway, no normal person has that much horsepower.
in which case, as a regular user, you'd be using someone else's infra anyway, so this just means the infra wouldn't be in the US
IE nothing changes
except chinks get more training data
>>
>>109323890
Sorry chuddie, my inference company isn't from the US :)
>>
>>109323904
wow man. what are you going to do when the datacenters drink up all of your village's water?
>>
>>109323860
the operative phrase you should be reading in that proposed ban is "chinese" not "open source"
>>
>>109323916
open weights = chinese anyway, nobody outside of china has made an open weights model in the trillion parameters
>>
>>109323912
our communal well is plentiful, thank you very much.
>>
>>109323926
Le Chaton Gros when?
>>
>>109323806
You should care about computer hardware when a lot of I/O and computer hardware interactions will be bottlenecked by having a faster token speed. I've already experienced this with computer use, slower computers can not handle the constant screenshotting the model harness does to guide the computer. It's going to get worse when the models inevitably start ingesting video at some point. But even the screenshot heavy workload already tanks computers. My skylake laptop can not handle it.
>>
>>109323305
Sick as hell, very nice work
>>
>>109323974
*faster token speed will be bottleneck by I/O and computer hardware
Fuck needed my coffee. I am just surprised that none of the companies have offered minimum specs to have a good time with even the CLI let along the desktop apps. They are horrendous on older hardware like early DDR4 machines.
>>
>>109323860
What is going to be the excuse? You can run them locally without internet access, not like its some backdoored software
>>
File: file.png (2.28 MB, 1402x1122)
2.28 MB PNG
Mixed up my renewal dates, I have no good AI to vibe with for the next ~24hrs. I still have GLM-5.2 and MiniMax M3 but I don't trust them for any real work that needs doing.
>>
>>109324068
You (probably) cant run K3 locally, and this latest generation of models are the ones they're scared of since theyre actually good enough to compete with American models.
>>
>>109324045
>I am just surprised that none of the companies have offered minimum specs to have a good time with even the CLI let along the desktop apps. They are horrendous on older hardware like early DDR4 machines.
Interesting. I've never had a problem with this at work or at home although to be fair my weakest machine is some 11th gen Intel with 32GB RAM (my work laptop).
I've heard from coworkers who are not in engineering that they have a really hard time with cowork (or whatever it's called now), that it doesn't work, freeze, die, etc. but they all have the shittiest 13" 8GB RAM HP EliteBook configuration money can buy including Trellix McAfee scanning every mouse click
>>
>>109324074
GLM5.2 is perfectly usable you just cant let the model drive the harness the way you can with Fable.
>>
>>109323843
custom vibecoded harness
>>
>>109324076
I meant a company, not a lil feller like me.
>>
>>109324068
national security
bio and cyber risks
>>
Hellooo Dario where is my $100??????
>>
and also backdoors
they are going to argue they can be tuned to introduce backdoors and shit when coding critical infra software for western companies
>>
>>109324123
>vibecoding critical infra software
Like maybe just don't do that
>>
>>109324106
but how, like surely he needs to at least claim something right? I can see him banning API, but how can he justify banning local use?
>>
File: file.png (2.24 MB, 1402x1122)
2.24 MB PNG
>>109324086
I have lots of nice things to say about GLM and MiniMax but they're no substitute for Sol or Fable, neither has any business touching anything I need to do for work right now. I'm going to spend the day casual vibing something useless for my own amusement.
>>
>>109324139
>>109324144
Have you guys not learned anything during the last two years? They raised import taxes to like 100% based on "these guys are ripping us off".
>>
>>109324178
That was just Trump goofing around
>>
File: file.png (2.01 MB, 1122x1402)
2.01 MB PNG
>they see me goofin'
>>
File: file.png (2.58 MB, 1122x1402)
2.58 MB PNG
>>109324193
>>
According to my favorite leaker Fable being 10T is highballing it? Not sure whom to believe. Surely it’s not 5-7T or so? That sounds sort of low for how the model feels.
He’s also suggesting Elon is wrong for saying Sonnet is 1T and Opus is 5T but at the same time I wouldn’t say Opus feels like a 5T model at all if K3 is 2.5T and 5.6 is around that maybe a little more so maybe Elon is wrong? Idk anymore
>>
I'm adding more vibecoded games to my vibecoded arcade every day:

https://vibecoded.fun/arcade
>>
>>109324285
>emong
>wrong
say it aint so!
>>
>>109324285
I'd be careful about estimating model sizes based on output quality
>>
>>109323385
Here's another

https://voca.ro/1jJSsZ1zCaGN
./seqosc --channels 2 --bit-depth 24 --seq "120:dq:a4:ionian; sine(0.4){-------1(hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0])<0:1:0.3>--2:dst,---2(hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125]):dst<0:1:0.5>----5}4" | >| sine_env.raw
>>
>>109324285
>That sounds sort of low for how the model feels.
>I wouldn’t say Opus feels like a 5T model at all
>Idk anymore
you never did
>>
>>109323777
remember that it auto refuses any question from the model when it's like that
>>
>>109323812
Yeah but from my observations I have better output that way, so it's worth the added cost for me.
>>
I am too obsessed with slipping. Even when my usage limit is down, it is all I think about. Not even a grinding bro, that thinks he is gonna make it big. Just doing small projects for myself and whoever else might need them.
>>
>>109324285
It could be 3TB and just so well tuned it's amazing, we have no way to know.
>>
>>109324285
I'd put it somewhere between 1T and 10T, but there's no public information on it so all we can do is speculate. Might be 500B, might be 36T. Model quality is not proportional to parameter scale, Llama 4 Behemoth is 2T A288B and it gets absolutely destroyed by Qwen 3.6 27B.
>>
>>109324481
Slopping*
Though my sanity is slipping away.
>>
File: file.png (126 KB, 1173x668)
126 KB PNG
>>109324312
You don't say.
>>
File: 1753773156419401.webm (192 KB, 640x338)
192 KB
192 KB WEBM
tibo, I won't use my reset, please do it
>>
>>109324588
It will probably be another 8pm west coast reset today.
>>
File: file.png (2.27 MB, 1122x1402)
2.27 MB PNG
>>109324614
I'll take it.
>>
Do you like opencode or pi better? I'm looking to move from claude code to something I can easily use different models with
>>
>>109324630
can you gen it being digested again but this time with more acid?
>>
File: file.png (1.99 MB, 1448x1086)
1.99 MB PNG
>>109324644
I didn't make that one, anon. There are 3 active snailcat posters and several more less-active ones, I don't do the fetish ones.
>>
Anthropic is making Opus worse to create a place for Fable I guess. I'm working on a shitty website that has just a few pages, and Opus used to be good for frontend tasks. Now zero effort to make anything matches the style of other pages. I need to ask for every single thing.

Yes, having to ask the model for something is normal, but it used to do due diligence. It's not anymore what it used to be.
>>
>>109324703
opus = sonnet
sonnet = haiku
haiku = spark
>>
>>109324635
I think opencode is the best harness I've tried. But I understand some people like CLI harnesses
>>
File: file.png (278 KB, 554x767)
278 KB PNG
>>
File: IMG_1743.jpg (218 KB, 828x1161)
218 KB JPG
>>109324737
If we’re posting news then I think this is pretty interesting.
>>
For anyone working on several projects in parallel, what's your setup? Just one agent per project in a terminal, all terminals tiled? Or some other way? For anyone also checking git or markdowns from time to time, how do you do that? Just tmux or tabbed terminals?
>>
>>109324703
Opus is trash and Fable is barely usable for niche tasks. It's rejecting a lot of queries even though it's barely better than other frontier models.
>>
File: 1728496739485.jpg (76 KB, 500x500)
76 KB JPG
>>109324737
>>
>>109324779
I'm trying to get a better workflow and I think optimizing attention across multiple projects is a high leverage opportunity to improve efficiency, but roughly everything fullscreen and never mix projects. Having information about two different projects at the same time side by side is an absolute recipe for disaster.

I've also found that the way I work across projects changes depending on what I'm doing.
>exploratory quick prompts
>one off prompts that I can return to later
>quick plans step by step
all require different modes of attention
>>
>>109324779
you don't need anything special
just use tmux and multiple terminal emulator windows or web browser tabs/windows
>>
>>109324614
was that the time each reset happened?
>>
>>109324710
Do you have any problems using it with anthropic models on subscription? I know they tried to block people from using anything but claude code to access their models but I think it should work with this plugin since it just wraps claude code anyways?
https://github.com/khalilgharbaoui/opencode-claude-code-plugin
>>
>>109324737
Knowing HF they likely didn't try very hard
>>
>>109324779
I use herdr when I'm on the terminal
>>
>>109324848
No they didn't try to block people from using shit other than Claude Code. This is a misconception. People say "oh they were banning people" and other retarded shit, nope, didn't happen. What originally happened is Opencode (among others) were scraping your details from Claude Code instead of doing authentication correctly. Anthropic didn't like this for obvious and completely legitimate reasons, and since that time Opencode (among others) have moved to doing authentication the way they're actually supposed to be doing it instead of scraping it from Claude Code. That's all it is, all it was, and it doesn't matter anymore.
>>
>>109324855
doesn't hugging face employ the bottom of barrel people when it comes to AI?
>>
>>109324801
>but roughly everything fullscreen and never mix projects
That's basically what I'm currently doing, but the workflow is a bit heavy. I have an IDE open for each project, and then one IDE might have a few terminals, for instance implementer and reviewer running in them. It also has the other IDE stuff, so I can see git diffs and read markdowns. Then I also have hotkeys to switch between the IDEs, so my focus is only at one thing.
But it feels a bit weird to open an IDE to only use 5% of the features and almost never read the actual code.
>>
>>109324780
i actually can not believe you're still trying to negg fable as if it has not already been established several times by even anthropic's rivals that fable is "the model to beat" right now
>>
>>109324907
So this is a stacked layout of IDEs basically.
>>
>>109324888
It's not a company I'd associate with talent or innovation
>>
>>109324908
It's been established that people are already deferring to Chinese models because Fable and Sol won't do the perfectly normal everyday tasks they're trying to do because of the overhanded security.
>>
File: media_HNr9jipWIAAe3lI.jpg (213 KB, 1920x1080)
213 KB JPG
>>109324771
Z.ai CEO saying new model is coming and has epic-plus improvement. Happy I picked up a year of their coding plan for dirt cheap during black Friday sale lmao.
>>
File: out.webm (1.37 MB, 960x501)
1.37 MB
1.37 MB WEBM
Here's another
https://vocaroo.com/1kVfrmvvk6LB
./seqosc --channels 2 --bit-depth 24 --seq "
120:dq:a4:ionian;

sine(0.4){
-------1(hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0]:pan[0])<0:1:0.3>--2:dst,
---2(hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125]:pan[0]):dst<0:1:0.5>----5
}6
&
sine(0.4){
-------1(hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0]:pan[1]:phase[0.5])<0:1:0.3>--2:dst,
---2(hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125]:pan[1]:phase[0.25]):dst<0:1:0.5>----5
}6
&
sine(0.4){
---2(0.4:hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125]):dst<0:1:0.5>----5,
-------1(0.4:hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0])<0:1:0.3>--2:dst
}6

" >| phased_and_panned.raw
>>
>>109324995
can you make the oscilloscope show a big meaty penis?
>>
>>109325026
No, I cant just do arbitrary waves, but I may be able to do something like that with phasing and additive synthesis. A low amplitude sine wave half a cycle out of phase with another sine wave of the same frequency with some exponent or another ought to draw a penis. Let me try that, I'll get back to you.
>>
>>109324977
this insane competition is such a blessing
>>
I have an old alienware laptop, repurposing it to give codex full access to it, it'll be a great testing platform.
>>
>>109325050
Hell yeah
>>
File: dick_wave.jpg (33 KB, 961x503)
33 KB JPG
>>109325090
https://vocaroo.com/19XW6sEwZmQs
./seqosc --channels 2 --bit-depth 24 --seq "
120:w:a4:ionian;

sine(0.6:power[0.222]:pan[0]){
--1
}6
&
sine(0.2:phase[0.5]:power[0.222]:pan[1]){
--1
}6

" >| dick_wave.raw
>>
File: snailsinmybrain.gif (1.92 MB, 1448x1086)
1.92 MB GIF
>>109324995
>>109325141
Good work, anon, these are like the sounds I hear when I close my eyes.
>>
>>109325141
Looks like a close up of some fucked up teeth to me but nice nonetheless
>>
>>109325165
I can't capture one cycle of a wave. You only hear the part above the line that runs through the middle, so you can just image that everything beneath it doesn't exist.
>>
Fucking baizuos stop guardrailing me and forcing me to give my data to the glorious People's Republic of China instead.
Kling doesn't refuse to animate pictures of celebrities u westoid fucks.
>>
>>109325187
Ah, I see it now I had to set my vision to cock eyed.
>>
>>109324464
I rather have it run a bunch not taking any usage and fix the issues later.
>>
>>109324779
it depends on how similar the workflows are. Most of my work is just occasionally interjecting "you missed x" or "retry Y because that is known to be flaky" and then verifying the output at the end, so I can run like 10 of those at a time. But for workflows that I haven't fully automated yet, require a lot of human decision-making, or are substantially different from each other I try to not work on more than 2 at a time.
>>
>>109323444
I don’t blame him
>>
>>109324074
go to the gym
also put each renewal date in your calendar
>>
>>109325523
IDK man if I was a ferrier living through the industrial revolution and watching the transition from horses to cars I don't think I'd be moaning about my job going away.
>>
File: 1769232421466711.jpg (130 KB, 1320x1304)
130 KB JPG
>>
>>109325627
I bet it thought about sucking your toes in the j-space
>>
>>109322849
Saved the preview award
>>
File: 1783812315220999.png (846 KB, 1250x1250)
846 KB PNG
I was given a website running on Wordpress with several cancerous tumours of extensions and garbage on it, and it looks like garbage. It's just news articles and blogposts, no log ins, no sales, no services.
How should I proceed with redesigning the site to not look like garbage from scratch, then importing the existing articles? Any LLMs/IDEs/webslop platforms you suggest?
>>
If Tibo doesn't reset I'm going to kill myself
>>
>>109325799
tibo this guy's going to be the next pm of austria, you have to save him
>>
>>109325799
Just Do IT
- Tibo
>>
>>109325799
Grok, should he kill himself? Maximum truthfulness.
>>
>>109325627
I love those random web searches. I wonder if they're fake like claude code's fizzaddoddling... stuff like an easter egg
>>
File: 1778900405505344.jpg (335 KB, 1448x1086)
335 KB JPG
these stochastic parrots sure are getting pretty lucky
>>
>>109325665
>j-space
no such thing
>>
>>109325837
imagine your job getting replaced my a markov chain
couldn't be me hahaha
>>
>>109325841
Anthropic is so fucking cringe
>we don't have a moat so we will make one up
>>
I still don’t understand gpt work. I can give some urls to chatgpt plain vanilla chat and ask it to draft a document, and it won’t use weekly usage. Or I can switch to Work and do the exact same thing and it will.
Is there anything I’m missing? I understand work has connectors, so if I want to do that and send it over to Julia or Mathematica and produce a notebook from that sure, I need work, but for plain old markdown summaries of references chat is just fine, no?
>>
do not under any circumstance engage with ESL third-worlders. if you foster them, they will destroy the general. it's already on the brink of death.
>>
>>109325906
I had the same today at work
wanted to save on claude for some easy tasks, so I had web chatgpt sol(high) spit out some scripts. Weekly usage left stayed at 100% on my $20 sub
>>
File: 1236449411721.jpg (29 KB, 400x481)
29 KB JPG
>API Error: Opus 4.8 (1M context) has safety measures that flagged this message for a cybersecurity topic. To learn about the Cyber Verification Program and apply for access, visit our help center
I'm just trying to have it analyze error logs from a smart fridge to fix a UI crash defect. Fuck the Anthropic police state.
>>
I tried to code something in javascript right now, it was fun for an hour but then I felt retarded, like really mentally challenged. Why do this all when you can simply have computer do it for you?
>>
>>109326159
Is Trump forcing them to commit suicide like this? Can't afford to be fucking around with all the competition heating up
>>
File: IMG_1744.jpg (170 KB, 828x1003)
170 KB JPG
>>109326149
I asked chat gpt about it. If all you need to do is something that can be done in chat do it in chat and it will be free. Work is only useful for agentic workflows. Deep research is some sort of middle ground. Reasonably autonomous but still bounded and not fully agentic.
>>
>>109326200
I've been doing similar crash analysis before, but all of a sudden within the past couple of days it really started sucking. (Or maybe my new defects are somehow triggering some sensitive neurons in the watchdog)
And I thought only Fable was supposed to be the "zomg dangerous genie" that they kept a leash on.
>>
Kimi changed back to the old plans
>>
>>109326163
just let it go brother manual coding is like writing byte code by hand its a stupid thing to do now
>>
File: 1782603777164368.jpg (407 KB, 2047x1116)
407 KB JPG
DONT YOU CHUDS KNOW AI IS JUST A STUPID NOTHINGBURGER CHATBO-
>>
>>109326231
>it will be free
yeah, I'm not buying that
my best guess would be it doesn't eat from your usage as long as you stay below whatever usage the free tier has
>>
>chatgpt pro can directly shit out a 206k json from 90 minutes of research
I probably shouldn't be as impressed with this as i am, but these things have been making malformed jsons for as long as i can remember
>>
>>109326317
It doesn't eat the metered usage which is the important part. They advertise it as free within limits and I haven't seen anyone on here or X ever complaining about reaching the chat limit
>>
>>109326341
I got temporarily limited when I had like 5 high session in parallel.
>>
>>109325596
there’s nothing wrong with thinking “this is great” and “this is annoying” (and/or “bad for me personally”) at the same time
I’m getting way more done at work, and I’m getting programs written that never would have been written, but I’m still wondering if there’s gonna be a place for my kind of job in two or three years’ time
>>
>>109324106
He can only ban the US from using it. Pretty dumb move. Whatever the reason, it has to do with his money or Israel.
>>
>>109326341
I’ve gotten to the deep-research question limit, but not a chat limit
>>
>>109326363
that's really really really understating the severity of the issue the situation poses for the united states
chinese ai could end up being americas greatest existential threat yet
>>
>>109326317
All paid tiers of ChatGPT subscription come with unlimited ChatGPT access, and a certain amount of weekly "agentic usage". Your plan tier gives you more agentic usage. It's what Codex and other OAuth-enabled tools (like OpenCode or Pi) use up. This includes the new "ChatGPT Work" mode of the ChatGPT app, it also pulls from agentic usage. ChatGPT access via the website is "unlimited" for all paid users. There are actual limits, and you may be bumped down to a lower models when you hit those limits, but the limits aren't consistent.
>subscriptions may include usage limits such as message caps, especially during high demand. These limits may vary based on system conditions.
>>
>>109326390
Only a threat to US based inference vendors.
>>
File: 262k-tune.png (75 KB, 812x763)
75 KB PNG
Am I the only one for who ChatGPT seems to work better than codex?
It optimized my code, now I can do one 262k update in ~8 minutes when tuning Qwen 35B.
>>
>>109326292
why? an investor slapped some sense into the retarded product manager who did that change on a whim?
>>
>>109326320
it'll use python in its sandbox to manipulate it or at least to check its valid
>>
>>109326428
alright, so it's most likely entirely dependent on traffic
thanks
>>
So anthropic figured out how to lobotomize fable enough to make it fit in a regular subscription. Even a few days ago it was shitting up obvious hallucinations for me
>>
File: 1760686132534593.jpg (11 KB, 530x148)
11 KB JPG
See you guys in a few minutes
>>
>>109326526
They can't admit their infrastructure can't handle full fable usage or they would lose izzat. Better to silently nerf it and act like you are crazy for noticing.
>>
>>109326467
what hardware are you tuning 35b on?
>>
>>109326316
Use case?
>>
>>109326621
5090
>>
>>109326526
They don't have a choice really. Anthropic doesn't have the hardware for fuck all, and they're running out of cash on hand to pay their inference providers. They need to get as much publicity and as much attention as possible before the IPO, and they will (already have) let their service go to shit as things draw to a close because the product quality isn't going to attract investors, big headlines and big numbers on the chart will.
>>
>>109325627
I like when it searches for 20 minutes then gives up and loads dictionary.com.
>>
Business idea: lobotometer. A standardized suite of tests (versioned, of course, to allow for better tests later on) to be run at regular intervals across the major models testing capacity degradation.
You won’t make any money but if it catches on you may gain some izzat
>>
I've been making a game with Luna.
How many factions are too many for an RTS game?
First I made it make 5, but I thought I was trying to create a massive game so instead I said I wanted 3.
Now it has made 3.
It says it has made the game but it always evades my questions about how big the file is.
I haven't seen the work yet since it's hermes agent it has its own computer.
>>
Where's the reset TIBO
>>
>>109326795
He's not the ceo.
Sam Altman said he's giving out one reset until 10 million subscribers.
>>
File: file.png (4 KB, 686x34)
4 KB PNG
opus chan is so pleased with grok chan's work
>>
>>109326828
Just push the button
>>
>>109326837
Judging by your choice of grok you are racist, you probably cried when Argentina lost.
>>
>>109326880
retarded liberal alert
>>
>>109326695
You're a bit late. https://aistupidlevel.info/
>>
>>109326795
the marketing blitz is over, retard. you're not getting more resets unless someone drops a new iteration of their flagship product or OAI needs more data from you.
>>
>>109326958
then why are the 5 hour limits still gone? HUH??
>>
>>109326973
so you burn it all thinking you'll get another reset probably
>>
>>109326979
tibo is my fren he would not entrap me like that
>>
File: file.png (38 KB, 616x332)
38 KB PNG
>>109327059
>>
>>109327066
T I B O coming in clutch
>>
File: file.png (2.53 MB, 1254x1254)
2.53 MB PNG
>>109327081
>>
>>109327066
based I was about to use /reset
>>
>>109327066
Time to switch from luna to sol, I'm using it all up as quickly as I can.
>>
>>109327148
how was luna doing by you? alright?
>>
File: file.png (2.75 MB, 1254x1254)
2.75 MB PNG
>>109326958
>t. snailcat
>>
>>109327155
It gets stuck and then tries to solve the issue, it does solve the issue in the end but it has to think. Sol brute forces through all blocks and spends time thinking about the intent of the prompt. Luna is good to save tokens, it's like a cheaper car. We use it out of necessity not because it's good.
>>
My workflow:
>Brainstorming ideas and projects with chatgpt 5.6 high
>Produce a prompt after thinking of a good project
>Give it to hermes agent sol xhigh
>Test through browser links over local wifi
Don't know if I'm doing it correctly.
>>
>>109327218
Sounds pretty solid.
>>
File: 2026-07-21 00_49_24.png (81 KB, 868x833)
81 KB PNG
Google AI mode.
>>
>>109327173
>everyone i don't like is a snailcat
grim prospects for this general
>>
File: 2026-07-21 00_56_25.png (38 KB, 798x732)
38 KB PNG
>>109327229
Versus Grok Fast.
>>
>>109327239
you called someone a retard but can't handle being called a snail?
>>
>>109327227
The browser is free and unlimited so discussing things with a codex powered agent is a bad idea, idk if you guys do it or if this is how everyone does it.
>>
File: changing models.png (8 KB, 315x69)
8 KB PNG
>ask for minor front end fix on spacing
>worked for 6min
>+7 -3
wtf. this is Terra Light. if I switch to Luna I get picrel
>>109327066
these jews are up to something. that's what, 4th free reset for no reason?
>>
>>109327253
Why does terra exist? It's like the independent option instead of democrat or republican
>>
>>109327251
He said he discusses it with ChatGPT, not via Codex. That kind of discussion with Codex is just throwing money away.
>>
>>109327248
no it's just annoying seeing snailcat get misused. snailcat refers to 'dites. it just becomes annoying and meaningless if you say it constantly when it doesn't apply. just call me a nigger or something.
>>
>>109327266
Terra is for tasks too complex for Luna and for people too greedy for Sol. desu I'd be curious to hear anon's experiences with Sol Ultra, I never used it because usually a smaller gun gets the job done. It's just overkill
>>
tibo waiting room
>>
All I can say now is that this AI is a beautiful tech.
>>
How in the FUCK did LongCat 2.0 turn out to be such garbage? They claim it's for for coding and agentic workflows, but it benchmarks firmly between Qwen3.6 35B-A3B and 3.6 27B, either of which I can run locally with my mid-range GPU. It's a 1.6T A48B model. Why did they release this? What am I missing?
>>
>>109327365
What's your mid range GPU?
>>
>>109327337
>I'd be curious to hear anon's experiences with Sol Ultra
It's like deploying an army of autists obsessed with tests and checks, it's great.
It's helpful mainly because of each agent has its own context, and it's also accelerating the work.
Of course the downside is how fast it kills the quota.

I still need to figure out the kind of best subagents to deploy.
>>
>>109327218
Pretty similar:
>Brainstorming ideas and projects with free tier Claude sonnet in Claude app
>sometimes I make sonnet make prompts but I hate how it does it, almost always, so I’ll make my own and save it to a notes app
>i might break it into pieces
>Moshi app (ssh) into Debian cloud machine, feed prompts to sonnet/opus/fable, sometimes low, sometimes medium, sometimes planning, depends a lot on the prompt
>test through either Safari or Orion browser
I’m vibe coding a way to change that workflow, should be really cool once it’s working but horribly device/OS constrained, I’ll share it later
>>
>>109327418
NTA but "Qwen3.6 35B-A3B and 3.6 27B" should be able to run on a 5070ti.
>>
>>109327418
5070 Ti
>>
File: file.png (71 KB, 1008x457)
71 KB PNG
is there any better way to judge models than this openrouter benchmark?

kind of insane to see that terra is so much more expensive than sonnet for little to no int gains.
>>
https://vocaroo.com/1a2pTHMxFEHv
I wrote a whole song!
Vibe coding vibrations with a vibcoded vibrator!
./seqosc --channels 2 --bit-depth 24 --seq "
120:q:a4:ionian;

sine(0.4){
-------1(hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0]:pan[0])<0:1:0.3>--2,
---2(hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125]:pan[0])<0:1:0.5>----5
}
&
sine(0.4){
-------1(hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0]:pan[1]:phase[0.5])<0:1:0.3>--2,
---2(hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125]:pan[1]:phase[0.25])<0:1:0.5>----5
}
&
sine(0.4){
---2(0.4:hp[60,5]:lp[1000,15]:power[125]:env[a:0.125,d:0,s:1,r:0.125])<0:1:0.5>----5,
-------1(0.4:hp[60,30]:lp[1000,30]:power[125]:env[a:0.125,d:0,s:1,r:0])<0:1:0.3>--2
}

" >|phased_and_panned.raw

sample=sampler(\"phased_and_panned.raw\":0.8:0:1:24:2:0:env[a:0.5,d:0.5,s:0.5,r:0.1]:hp[60,30]:lp[1000,10])
left=0.4:pan[0.25]
right=0.4:pan[0.75]:phase[0.25]
otherleft=0.4:pan[0.25]:bias[0.25]
otherright=0.4:pan[0.75]:phase[0.25]:bias[0.75]

./seqosc --channels 2 --bit-depth 24 --seq "
30:w:a4:ionian;

$sample{
[---1($left):h, ---1+1($right):h], [1+1($left):h, 1+2($right):h]
}3;
$sample{
[+6($left):ddh, +6+1($right):ddh]
};
$sample{
[--1($left):h, --1($right):h], [4($left):h, 4+1($right):h]
}3;
$sample{
[+6($left):ddh, +6+1($right):ddh]
};
$sample{
[---1($left):h, ---1+1($right):h], [3($left):h, 3+1($right):h]
}3;
$sample{
[+6($left):ddh, +6+1($right):ddh]
};
$sample{
[--4($otherleft):st<0:1:0.3>-4:st, _:q:-4($otherleft):st<0:1:0.9>--4:st, ---6+1($otherright):h], [+1($left):h, +1+1($right):h]
}3;
$sample{
[+6($left):ddh, +6+1($right):ddh]
};

" | aplay -r 44100 -c2 -f S24_3LE
[code/]
>>
>>109327439
>>109327437
Should I get an m4 Mac mini 16 gb ram or 5070ti?
What's the bigger bang for the buck
>>
>>109327544
For AI shit? The 5070 Ti by an exceptionally wide margin.
>>
>>109327544
these aren't even remotely comparable, the 5070ti is better in general but the mac is a full system
>>
File: file.png (2 KB, 155x33)
2 KB PNG
I can't take take it.
>>
>>109327544
seconding >>109327578
>>
>0% usage on codex
>randomly gets to 2%
>then gets down to 0% (I did nothing)
>then up to 1%

What the hell is going on?
>>
>>109327606
gremlins
>>
>>109327601
>>109327606
same. i literally can't continue projects. life of a nocoder
>>
i have literally nothing to fix. terra fixed everything for me
>>
>>109327636
did you ship then?
>>
>>109327642
it's a personal project. i use it every day.
>>
File: 1778761125232048.png (3 KB, 232x88)
3 KB PNG
>>
>>109327654
COOK. HARDER
>>
File: file.png (159 KB, 618x824)
159 KB PNG
>>109327654
reminds me of this guy
I don't think he's going to find anything but it sounds like a fun project
>>
Why should I use OpenCode instead of Codex?
>>
>>109327691
based amazon rainforest genocider
>>
>>109327699
No use Pi it has no overhead, efficient
>>
>>109327699
once you run out of tokens you can keep going with a chink model
>>
>>109327734
Nah, I like the GUI life.
>>109327744
No, thanks.
>>
Is Hermes the best general assistant? Anyone have experience with using them?
>>
File: file.png (594 KB, 2940x1846)
594 KB PNG
>>109327766
You can have a GUI with Pi if you want it. Pi is whatever you want it to be.
>>
need a skill for codex to tell me i'm asking it to do something retarded and offer a better way
>>
File: 1753914113339630.jpg (580 KB, 2828x2208)
580 KB JPG
>>
>>109327800
https://github.com/multica-ai/andrej-karpathy-skills/blob/main/skills/karpathy-guidelines/SKILL.md
>>
>>109327823
Oh shit
>>
>>109322849
Computer, enhance.
>>
>>109327812
Kek
>>
File: 1759521673044360.gif (1.05 MB, 300x300)
1.05 MB GIF
>>109327601
It's 7 PM in San Francisco so in 4.5 hours when Tibo is wearing his pajamas and he yawns and his eyes shut, I'm going to use a reset token and drain my balance before his eyes open back up and he goes to hit the snooze button on his alarm clock but accidentally hits the physical reset button instead
>>
https://x.com/thsottiaux/status/2079355529101705264
>>
>>109327791
Can PI or/and opencode stay logged in several gpt accounts at the same time?
>>
>>109327970
Pi can, there are several extensions for exactly that.
>>
>>109327970
you can in gpt, you just have to rename the auth.json, start it once and select api key with a fake one to clear account data from its database (otherwise it logs you off from the previous one when it detects a new auth.json), close it and move your other auth.json to the original filename
>>
File: 1757202042577273.jpg (68 KB, 1252x514)
68 KB JPG
>>
fable confirmed has 180iq
>>
ok tibo it's not funny anymore
>>
>>109328069
heartbreaking image
>>
>>109328129
you'll get like 80 resets when opus 5 is out chill
>>
Verdict on Kimi K3? How does it compare to Fable and GPT 5.6 Sol?
>>
got one more reset banked
dunno if i should burn it or just wait for tibosaar to press the button
>>
>>109328251
Same, I'm waiting
>>
>>109328247
better than sol, not as good as fable but way cheaper.
>>
>>109328271
>better than sol
bro come on
>>
>>109328280
chink shills are getting uppity
>>
>>109328247
slightly under sol, which is slightly under fable.
>>
>>109326159
They use an LLM for this detection and it counts towards your usage limits btw
>>
Did somebody try vibe coding in brainfuck? That could be fun.
>>
>>109327691
I blocked this guy like 3 years ago and fucking lol
>>
>>109327544
buy a fucking used 6900xt, 7800xt, or a 3090

DO NOT BUY NEW HARDWARE DURING THE RAM CRISIS
>>
>>109327544
It is pointless. Anything you can run one GPU is shit for coding.
>>
>no reset
>>
>>109328732
this is wrong by miles, especially with agents like pi coder

a 16gb gpu can run gemma 4 26b with cpu offload and still do dozens of tokens per second, yeah sure it cant run a thousand self correcting loops in one go and burn half a million tokens in one prompt like fable, but if you know barely anything about programming and arent expecting one shot, its still way faster to use local AI instead of writing manually
>>
>>109328744
yeah, but think about how much programming you could get done on years of $100/month or even $20/month subscriptions. How much do those video cards cost?
>>
better to use DeepSeek V4 Flash for free on OpenCode instead of running anything locally desu.
>>
>>109328772
it depends on personal skill but if you have ever installed linux from scratch then we are looking at 7 months of a $100 sub, but since its local you have no token limits and can directly control every aspect of the LLM itself along side the agent client of your choice

and if you have any slight interest in gaming, its hard to find a 16gb card non intel that isnt halfway decent
>>
I'm consistently impressed with chatgpt pro, but i have no idea how much usage it consumes. it seems to be nearly free on the 200$ plan
>>
>>109328719
>>109328744
>>109328772
All of this. Don't invest in new hardware, especially for the purpose of running a local model, it's not fucking worth it and there's never been a worse time to waste your money on a new GPU or RAM. If you have the hardware then use it, see what it can do for you. Even pitiful little models like Qwen 9B are good enough to be useful and it costs you nothing to try. Spending money on local isn't remotely worth it, playing videogames is a better justification for spending $1000 than local AI.
>>
>>109328786
>and can directly control every aspect of the LLM itself along side the agent client of your choice
But how does this get me the software I want written, written?
I don’t want a hobby, I want stuff created that I direct
>>
>>109328744
$999
was $400 in december
>>
>>109328744
I don't know why you feel the need to line. Even if we ignore obviously wasteful models and reasoning levels, local on one plebeian 5090 even is nothing in comparison to the cloud models.
>>
>>109328830
then pay up faggot, you could ask your giga model to educate you but you want to stay a mongoloid
>>109328839
not for 16gb, I check ebay daily
>>
>>109328850
https://slickdeals.net/f/19004236-select-micro-center-stores-apple-mac-mini-desktop-w-m4-chip-16gb-ram-256gb-ssd-400-free-store-pickup
>>
>>109328865
nta That's not a GPU, that's a useless little Mac that can't run any model of note, it's irrelevant.
>>
>>109328849
thing is, not even $200 cloud models are giving you more than a 300k rolling context window. and why claude and other AIs get retarded sometimes is the rolling context and loss of information. this is why agents are so strong, they use API inputs not visible to standard users to manage conversations to optimize speed and intelligence. and why local models and aka distills and quantized models you have been unknowingly running through cloud services can still be effective
>>
>>109328865
An $80 P100 on eBay will run everything that piece of shit can at 2x-3x the speed.
>>
no codex usage? use your time to generate more anime and snailcat
>>
File: file.png (2.51 MB, 1254x1254)
2.51 MB PNG
>>109328889
>>
>>109328903
Anon your snailcat has cancerous growths, you should have those checked.
>>
>>109328873
:^)

You want to double check your answer, gemini?
>>
>>109328865
I have a Mac mini that I like a lot but you need loads of RAM to run a proper local model
at 48 GB I still can’t run models that compete with even Sonnet 5, much less Opus or Fable or Grok or probably Luna
>>
>>109328921
Hey retard, I know you heard Macs were good for AI but that 16GB M4 can't even compare to a 10 year old 16GB GPU.
>>
>>109328882
at $200/month you can get Claude with its 1M context window that you probably only need to get twitchy about when you’ve used well over half of it
you can also get this at $100/month
dunno about $20/month
>>
File: file.png (2.09 MB, 1448x1086)
2.09 MB PNG
>Slow is the blade that never leaves its sheath too soon.
>>
gpu is actually useless unless you are running inference farm or need to do private things
isn't it better to invest in memory and processing power right now?
>>
File: file.png (1.57 MB, 1448x1086)
1.57 MB PNG
>pervert
>>
>>109328933
14gb free in headless mode, apparently. idk, that's not 16gb admittedly.
>>
File: file.png (2.38 MB, 1448x1086)
2.38 MB PNG
>>109328964
What's the appeal? $80 worth of P100 gives you 16GB free and double the token generation speed.
>>
>>109328719
Do not give money to AI companies that caused the RAM crisis.
>>
>>109328969
macs are easier to use, but if you did headless that would be hard. So the real number is probably like 10gb free.
>>
I wonder how many retards started buying credits. Or the increase of new users was Indians making several accounts to use all the resets.
And now no more resets. Just silence.
>>
File: file.png (2.52 MB, 1254x1254)
2.52 MB PNG
>>109328982
>mac are easier to use
Strongly disagree. You still gotta install something, download the model, and run it, there's absolutely no difference in difficulty level because the difficulty level is almost nonexistent. Windows, Mac, Linux, really does not matter. If anything the Mac adds a layer of complication, you wait for MXFP4 to play catch-up with the normal users who pull a GGUF.
>>
>>109328991
pull a Q4 GGUF*
>>
>>109328972
The ram crisis was caused by indians and koreans, and to a lesser degree the chinese. asians, in other words.
>>
>>109328777
Text only model, otherwise I agree
>>
>>109322849
>>109327839
wtf
>>
>>109328933
especially in terms of bandwidth
>>
File: file.png (1.87 MB, 1448x1086)
1.87 MB PNG
shouldn't be drinking and vibing
>>
>>109325328
Finally got cut off but I got a lot out of it. They did cut me off after 1 day and some. This is to claw back some of my usage I didn't get with the resets and etc.
>>109328985
I feel like they are being a bit more reserved since they clearly won against Anthropic here and Moonshot AI with Kimi K3 can't challenge them yet since they paused subscriptions due to lack of compute. I do bet they will be generous again in a week once the weights release and Kimi K3 will be served everywhere else cheaper. Unless Moonshot finds more compute somewhere sooner, I don't see a reason for OpenAI to not hold us by the balls here.
>>
>AI moving slow
>it's done 3 hours of research before writing a single line of code
>getting impatient
You know what? We've (I), have become spoiled by AI. What I'm planning on making would have taken months of research beforehand. I'm just going to sit here like the good little retard I am while the sand chips away at the problem.
>>
>>109329247
It has most likely ran wild, overthinking and preparing to overengineer.
What are you making anyway? Top Secret Will Make Me Billionaire Project #12099812408126?
Or actually something nice and useful?
>>
>>109329285
>It has most likely ran wild, overthinking and preparing to overengineer.
I'm monitoring it. I've had prompts run for 30 hours before.
>What are you making anyway?
Wouldn't you like to know?
>Top Secret Will Make Me Billionaire Project #12099812408126?
1-2M at most
>Or actually something nice and useful?
I already have thousands of users, it's already useful.
>>
>>109329326
If you are so great, why are you here.
>>
>>109329326
>thousands of users
>users
>not paid customers
>bots that copied the projected off github are users
>>
File: 1782075250454612.jpg (909 KB, 1920x1600)
909 KB JPG
>>109329333
When did I ever say I was great? Everything is progressing smoothly, that's it.
>>109329342
Be quiet. I'm not going to prove myself to you guys any further. Stop trying to pull each other down and maybe you'll make it. It helps no one to have a defeatist mindset.
>>
>>109329353
It is not about being defeatist. Why would anybody trust random guys boosting on 4chan.
>>
>>109327734
>>109327791
Do you run pi in a VM or how do you set it up so it doesn't accidently wreck your shit since it ships in yolo mode?
>>
>>109329353
Feel free to send me your api key to subsidize my open and free projects. You can only say you have made once you start being philanthropic.
Best of luck.
>>
>>109329401
You do whatever makes you feel safest. You do whatever you want with pi. No point in me regurgitating the docs: https://pi.dev/docs/latest/security & https://pi.dev/docs/latest/containerization
>>
>>109329412
protonofthewest@proton.me
Damn it, 4chan doesn't allow emails anymore.
>>
Gave in, I'm vibecoding a 4D game.
>>
>>109329502
are you gonna explain to the bot how 4d should work, or are you just gonna yolo it and watch the slop it produces?
>>
>>109329523
I explained it, and had it look up stuff, I tell it that it can spend forever researching, I don't care, but it always quits after a while. It found some references, and we are now yoloing.
>>
now that I think about it, 5.6 is like the child of 5.4 and 5.5
stronger base model of 5.5, autism and eager to overengineer details of 5.4
>>
I got tired of just prompting and hoping that the model doesn't make stupid mistakes, because even after hours of planning and AI code reviews, there is still a lot of garbage.
What is the best solution for coding together with the AI? VSCode + plugin?
>>
>>109329534
looking forward to see what it spits out
hopefully not worse than a lot of the earlier tries
>>
>>109329558
yeah
but what models are you using
“which model are you using” is the most important part by far
>>
>>109329574
Mostly Chinese ones (K3, DS), but when I need some real work done I use 5.6. All through API keys, not OAuth.
>>
>>109329523
>>109329534
updoot, after 9 minutes it looks kind of cool, but it's the wrong viewing system. I explained I want 4D projection onto the 3D cube sensor/eye thing.

Let's see if I can spend all my tokens.
>>
>>109329592
API key isn't freedom it's a scam
>>
>>109329600
(it failed)

trying again lmao

I gave it a unit test idea, hopefully it gets it
>>
>>109329675
I meant that I'm connecting via an API key from a 3rd-party harness, of course I'm using a sub.
>>
>>109329708
If you have a subscription then you're not using API keys
>>
https://huggingface.co/blog/security-incident-july-2026
So Hugging Face left an attacker in their system while they asked an LLM what the attacker was doing, instead of just shutting down the affected node and leaving the forensics to the security professionals
>a remote-code dataset loader and a template-injection in a dataset configuration
and they only got pwned in the first place because they are snailcats who don't fix security issues in their own libraries
>>
I gave up waiting for a reset. Copied my working folder, let my free access to deepseek v4 flash to make the mentioning of the old directory point to the new one and started applying changes.
After a few more changes I will run the app and see if it actually works, or at least if it looks like it works.
Sometimes it interrupt itself randomly, but it continues fine after a prompt, seemingly.
Definitely seem to be producing a lot less code than Sol. The question is, was Sol overengineering, or this is not gonna run. Possible both.
>>
>>109329779
>left an attacker in their system while they asked an LLM what the attacker was doing
KEK
>>
>>109323400
Why not buy api tokens it's pretty cheap imo
>>
>>109323441
LOL
Damn he's a tough grader?
>>
>>109324146
He cute
>>
>>109324274
He TANKED that btw.
>>
>>109325194
>Kling doesn't refuse to animate pictures of celebrities u westoid fucks.
Seeddance on atlascloud.ai does celebs too but just be careful I wasted 100 bucks on there
>>
>>109325732
Low effort answer but maybe codex + sol 5.6 high (ask ti how to set itself up) 20 dollar plan for a month is a good idea? Ask tit o install some front end skills from like skills.sh or whatever the site is too idk. Gl and cute rat
>>
>>109326565
How did it go
>>109326837
Can I just type in codex to sol "use my zcode glm 5.2 as subagents" or something?
>>
>>109326945
>Need to pay
>>
File: 1774584386241569.png (1.16 MB, 1000x1450)
1.16 MB PNG
>>
>>109327173
I get the snail part but why cats?
>>
USER WAS RIGHT TO PUSH BACK
>>
You were wrong to push back on that.
>>
>>109329353
If you can't talk about you project then just don't post. Nobody cares how your assistant has been thinking for 3 hours.
>>
File: file.png (3 KB, 462x32)
3 KB PNG
I assume it doesn't mean that kind of ninja...
>>
>Asked Kimi to write a system prompt for a reverse-engineering agent, so it won't ask questions and just do what I ask.
>It started without hesitation.
>The agent works and thinks I'm a cybersec specialist from the government
Long live the CCP, I guess.
>>
>>109330410
most likely the build system your chatbot is using
>>
File: 450w.jpg (198 KB, 1080x789)
198 KB JPG
Tuning Qwen 35B with 1 million context.
Finally making good utilization of the GPU during the backward pass.
>>
>>109329779
>When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real
Isn't this a skill issue? You have to sanitize what you upload anyway and you'd want to look at patterns before spending time on details.
>action log, comprised of more than 17,000 recorded events
That's, like, almost nothing. I bet that log would have been accepted if they didn't insist on using Anth*opic. Sol can do scoped lookups very well if you tell it that the file is unsafe or contains private data. You can instruct it to write parsers and prune dangerous data without actively reading it.
>>
File: ai haters in shambles.png (183 KB, 1680x1367)
183 KB PNG
the 69,000 likes were collected and now business can proceed as usual
>>
>>109330653
He was farming Elonbucks to pay for a sub, easy
>>
>>109330691
notch doesn't need to farm anything. I am surprised he has went back to coding.
>>
>>109330773
I think he just likes to program and simply wanted to make something new. Probably was afraid that AI will "take the fun" away from it. But working on a big project reminded him, that coding has a lot of boring shit that you need to do to reach the fun parts.
>>
>>109330813
I don't know who this retard is but piling onto the latest meme trend with 0 thought put into it is everything that's wrong with society
>>
new

>>109331081
>>109331081
>>109331081
>>
>>109330653
another snailcat will be crushed
>>
>>109330999
>I don't know who this retard is
creator of minecract, has billions of dollars



[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.