Wow actually a good op image this time

I have a xlsx with a bunch of YGO archetypes in their Japanese names. I tell codex explicitly how to find their English names and it keeps failing to do it properly for many of them.
I guess this is harder than producing slop code.

File: file.png (2.29 MB, 1402x1122)

2.29 MB PNG

>>109247117
there's always room for improvement

File: copex.png (94 KB, 1911x565)

94 KB PNG

Uh oh
I think my honeymoon period is ending with copex

the midi thing continues.

the latest grok (heavy) is, as they said, actually not inefficient. I'm not sure if I can even manage to spend all my tokens on this task, we'll see :)

I find myself needing to take corrective action to some extent. Like /goal doesn't uh.

some quirks that stand out to me, it doesn't notice them without me pointing it out.

When do people find /goal is the right way to go? Like I think maybe if I had lots of features and wanted them all improved (?)

File: file.png (2.44 MB, 1448x1086)

2.44 MB PNG

okay I'm done now

>>109247128
>>109247143
this, OP didn't even try shame shame

>>109247128
We need a meme animal for people like me that don't even know what they want well enough to use /goal

snailcat wearing a gigachad mask?

>>109247128
It's weird how even when it's trying to replicate an image in front of it, the model still manages to genericize it. The left hand isn't quite right, the hair looses that smoothness, and there's an extra catchlight in her eye for no reason.
>>109247143
In this one, it smooths out the right hand too, and gets the bodytype wrong.

>>109247104
I'm surprised you guys don't highlight interesting things people made in these gens

conductor! roles! buzzwords I can't remember! something about servers!

>>109247155
>things people made
does he know

>>109247160
the things that my slaves made are things that I made

File: amagi_yukiko_persona_and_(...).jpg (2.01 MB, 2100x2970)

2.01 MB JPG

For me it's Yukiko
Relevant: I've been slopping a Persona clone and I'm surprised at how much I don't have to hold Claude's hand in regards to working out the battle mechanics.

>>109247155
true.

In the last thread a guy made a game where he attacks starlink (it's a game, and a sentiment).

I think it's funny, I have no x, but have starlink & am using grok heavy with grok build.

>>109247161
i realized it wasn't clear enough when i pressed submit but i was making a shipless joke and not a "you didn't make it" joke

>>109247161
No way.

Cat-o-nine-snail.

File: 842.png (24 KB, 929x720)

24 KB PNG

>>109246887
Yep, correct, nobody here has built anything successful. Let's face it, if a smart guy like you has had access to these same models and is still a failure, the rest of us have no chance. I will attest (and am sure other anons will agree) that every single grape has in fact been sour as shit and you're doing everything right and not missing out on a thing.

File: file.png (1.59 MB, 1448x1086)

1.59 MB PNG

>>109247150

>>109247165
that's because every game you can think of has been made a million times and its read their codebases

>>109247166
Akshually it uses real orbital data, it only just happens that most satellites in orbit are starlink

>>109247171
lmao

grok decided to use sophisticated Python package management, and I have no idea what is happening.

>>109247170
Ok, maybe you're right and I just suck. That doesn't change the fact that everybody who isn't making money and thinks they're going to make money is probably delusional and will be negatively affected by AI. Which is most of this general.

OpenAI’s last 24 hours:

>Top Exec unexpectedly departs
>Shuts down browser tool after 9 months
>Sued for trade theft by Apple
>Caught selling product to China against sanctions

it's over

>>109247202
shut the fuck up

>>109247202
>Top Exec unexpectedly departs
who?
>>109247202
>Caught selling product to China against sanctions
what?

>>109247211
Fidji Simo stepped down due to health issues. OpenAI technically doesn't do business with the Chinese but does allow them API access through some roundabout shit I don't remember, point is it's now being called out as potentially violating sanctions maybe.

>>109247234
Forgot to say it's OpenAI and Google getting called out for doing business with companies blacklisted by whoever, the Pentagon I think.

File: file.png (2.47 MB, 1448x1086)

2.47 MB PNG

I'm enjoying Sol a lot, quite happy with it.

>>109247202
they don't call him scamman for nothing

File: 1350830503281.jpg (293 KB, 1920x1080)

293 KB JPG

I just used GPT 5.6 Sol Ultra to build a webapp better than the shitty dead end app that my dumbass employer spent $2M on (licensed templated app with enhanced "features" developed by "consultants")

Will I finally get promoted?

>>109247282
No. Best case you'll get a "nice, thank you anon". Worst case you'll get reprimanded for working on something without permission.

>>109247282
become a competitor

>>109247293
>he thinks it's about quality of the work
ha! I used to be naive too

File: literally the opposite of(...).png (2 MB, 3416x1930)

2 MB PNG

>>109247118
have you considered using a flat-file NoSQL database stored in Markdown files with YAML front matter?

>>109247184
just use uv
>>109247202
>top exec
she’s been sick this entire time

>worked for 5h26m15s
>blew past the 5hr usage and kept running fine
>after it finished everything it asks if i want to switch to a cheaper model
lel

File: 1764994452388930.jpg (361 KB, 1284x1812)

361 KB JPG

This is what AGI is being wasted on

Thoughts on Hermes-Agent? Is it useful for coding?

>>109247367
It's a popular choice, a lot of people like it, and yes it's specifically intended for and well suited to programming. I didn't care for it, I prefer Pi for being more minimal. Lots of options, don't hesitate to just try. Try Hermes, try Pi, try OhMyPi, Codex, Claude, doesn't cost a dime to just try and see what you like using. You can even do stupid shit like use Opus in Codex or GPT5.6 in Claude Code.

File: AI.png (372 KB, 703x392)

372 KB PNG

>>109247364

File: tester_tvai.webm (2.54 MB, 1658x930)

2.54 MB WEBM

codex made me a directors view for the winner's celebration

>>109247377
What about OpenHands? Someone in the previous thread said it's outdated crap.

We have Windows enjoyers here.
What are you having your clankers build your GUI apps in? I keep thinking of https://www.jsnover.com/blog/2026/03/13/microsoft-hasnt-had-a-coherent-gui-strategy-since-petzold/ — tl;dr Microsoft doesn’t have a single “you should build your app with these technologies” answer anymore

Is it better to install agents on dedicated hardware on the same network compared to a container/VM on the host machine?

Like if I have a homeserver or an SBC that's on 24/7 inside my local network, does that yield benefits compared to running the agent on the workstation that shuts down when not in use? Does latency become a problem?

>>109247403
Have a look at how quickly the thing types words at you. Do you really think it’s going to be a bandwidth problem unless you’re having it transcode video?

File: file.png (2.96 MB, 1672x941)

2.96 MB PNG

Kimi K2.7 Code is really cooking for me tonight. I usually talk a lot of shit about Kimi, it's nice to see it actually working as intended. I do wish anything, anything at all, could compete with Gemini for vision though, Kimi sure as shit doesn't. Sol feels close, I'll judge it more after I've spent more time with it.

>>109247403
>running the agent on the workstation
you could do both
I’ve asked Claude why my computer is fucking up and it’s nice for it to be able to look at my system log and find out what’s broken on my Mac

>>109247407
not him but it absolutely matters for code autocompletion

>>109247396
Never tried it. Looks bloated and unnecessary, don't see why I'd want it over a more typical agent.

File: 1775055372995841.png (145 KB, 550x534)

145 KB PNG

Is anyone else getting phantom resets on codex? It's really pissing me off. I can't tell how much usage I have left because I keep getting masterbaited.
>you're at 40%
>now you're at 99%
>now you're at 91%
>jk now you're at 0%
I've had it happen in multiple 5 hour windows

>>109247420
>Looks bloated and unnecessary,
Can you explain what you mean? I mean it's recommended in the OP here so it must have something going for it?

What kinds of minigames would you play with your agent while another is busy working on a task?

>>109247438
It provides a lot of functionality I don't find desirable, I consider that bloat. The functionality it provides doesn't actually enable me to do anything I can't already do. I'm sure there are people who would find it appealing or who have workflows that would benefit from what OpenHands provides. I like Pi, which is about as minimal as a coding agent can be, something "big" like OpenHands doesn't appeal to me. They seem pretty focused on enterprise customers and enterprise appeal.

File: de1530db-b887-4ea6-83f1-c(...).png (2.43 MB, 1183x1329)

2.43 MB PNG

>>109247104
cringe gay coded thread, HITl chads already won over ludites and vibe coder transgenders

>>109247452
>HITl
you were so close

Have any of you actually made a game or are making one?

>>109247467
>>>/vg/agdg/

Can AI coding help with Unreal Engine development? Does it get around understanding all the weird Unreal Engine macros? And if so how far can it actually go? Can it build netcode and low-level memory management?

>executing a major refactor with 5.6 Sol Ultra+++
>have 10 5.6 agents working on my repo all at once
>this comes days after I did a major refactor with 7 5.5 xHigh agents
I straight up don't give a fuck anymore. I've fully given in to the slop, I'm not even in control of what's happening anymore.

I AM USING /GOALLLLL on my midi thing.

I'm at 35% used. Will grok heavy eat it all up?

>>109247510
>>executing a major refactor with 5.6 Sol Ultra+++
You must be rich.

I'm a poor person, I'm using grok. no offense Mr. Elon.

>>109247510
you can clean it later though

>>109247510
kek i've basically been doing this as well, but ive put it on ultra and have it write up plans for refactoring things based on maintainability index/etc and then have max do the implementation
i'm going to use the next fable session to make more user simulation testing so hopefully no bugs slop through

>>109247558
Fable's purpose is to generate 4D games. I sure hope you're not counting beans with Fable.

File: 1764903158426964.png (197 KB, 500x614)

197 KB PNG

>check subagents
>6/6
>6/6
>1/1
>57/75

>>109247571
>not about games

wow. just.

How do you make the model spawn smaller model subagents?

File: 1781060940473650.png (140 KB, 1179x755)

140 KB PNG

another reset incoming

>>109247472
>looks inside
>literally just two schizophrenics flooding the entire thing
jesus fucking christ 4chan really is a looney bin nowadays in some generals

>>109247597
you can set

CLAUDE_CODE_SUBAGENT_MODEL

or apparently specify it from a markdown file if the agent has a name[?]
i thought there used to be a setting in claude code somewhere that let you configure it too

>>109247616
But, aren't you supposed to be like baking flatbread or something sharmuta?

File: 1766303689917141.png (41 KB, 1066x259)

41 KB PNG

>>109247616
a growing chunk of 4chan is just schizos fighting schizos in dying generals/boards

wtf it's 2 am, I am looking for things to do to burn tokens since I have a banked reset that expires tomorrow, and there just was another surprise codex reset so I'm back at 100% quota left

>>109247623
ok so sharmuta is Arab for something
I don't get this. Am I supposed to be Indian or something? Do the needful and explain your insult

File: file.png (16 KB, 1243x189)

16 KB PNG

>>109247612
dammit Tibo I just finished off the last of it let me sleep you piece of shit just stop

File: 1772645176817991.png (32 KB, 1017x531)

32 KB PNG

>>109247637
>I am looking for things to do to burn tokens since I have a banked reset that expires tomorrow,
Same here.

>>109247625
it's simple, first a schizo appears, then an alcoholic like me relentlessly antagonizes them whenever they make themselves known, becoming the second schizo, and eventually someone gets annoyed enough to start calling me on my shit turning them into schizo the third and it just keeps going like that, hopefully forever

File: Screenshot 2026-07-10 at (...).png (104 KB, 1330x652)

104 KB PNG

>>109247616
There’s a reason why I don’t have a tab open there all the time like I do here.
>>109247643
did your goat die and now you don’t have a gf of your own or something?

>>109247677
:^)

Your ai is epstein.

File: artworks-000031217722-3xe(...).jpg (164 KB, 1080x1080)

164 KB JPG

>>109247510
Based, but
>with 5.6 Sol Ultra
That's just a bunch of Sol High instances, Ultra doesn't use XHigh

>>109247691
Nobody cares.

> india <

you are writing a bean sorting bean counter.

File: desktop-commander-mcp.png (154 KB, 1391x924)

154 KB PNG

Guys! The Desktop Commander MCP hack is working EVEN BETTER than before. Now it doesn't ask for confirmation a lot for commands, it just keeps going until the task is finished.
This is effectively unlimited Sol on the $20 (thinking goes to High only) or $100 plan (goes up to Extra High).
I'm thinking about uploading my fork to github that works fully locally so you don't have to depend on DC's servers, would anyone be interested in that?

>>109247701
Hey this is fun to see. The old setup was neat but I always found it too slow/janky. I wouldn't mind seeing what you've got, I'd be particularly interested in how you're tunneling because I never found a solution that I actually liked.

>>109247701
have you tried using it in new codex app? it now has chatgpt inside too

>>109247707
That's not the point. This works even when you run out of usage.

>>109247706
I got a no-ip ddns, got a letsencrypt cert and exposed port 443 for oauth from the router.

>>109247715
>I got a no-ip ddns, got a letsencrypt cert and exposed port 443 for oauth from the router.
Neat, thanks anon.

>>109247707
But maybe I dismissed your response too quickly. Maybe the codex app has a better interface or something. I'll check it out.

if you aren't vibecoding in c#, java, or go, it's a toy project

>>109247707
Nevermind it's a non starter I think. I don't use Apple or Windows.

>>109247733
I vibecoded an assembly file once.

>>109247684
this is not me, bejitabro framing
>>109247677
yeah shits fucked. about the thread, I mean. I don't own goats, though I've heard they cut grass well

>>109247597
Just ask them to spawn it with the smaller models you prefer.

File: file.png (2.11 MB, 1448x1086)

2.11 MB PNG

>>109247707
i think this would be funny

>>109247597
In .claude/agents you can define your sub agent types and set their models and effort.

>>109247733
I don't care what language it's in, if it's not a game, it's just bean counting and it's irrelevant as a stack of punchcards.

>>109247775
And also generally for stuff like that it's good to get into the habit to just ask AI. I know that "ask AI" sounds like "fuck off" but it's honestly the best way to do most simple things, just like using Google or StackOverflow was in the past. I only ask humans for more general, high level ideas not just settings or implementation details.

>>109247779
Then you're making toy projects

>>109247472
Holy shit, I remember what that thread was like before AI. Clicking into that genuinely felt like seeing the bloated dead body of an old friend suddenly float by on the river.

>>109247643
>sharmuta
Israelis also say this

>>109247733
>muh language muhfugga
stfu retard, you have the rest of /g/ to argue endlessly about languages. Here plain English is the only language that matters.

>>109247856
English is my favorite programming language in 2026.

>>109247856
retard

>>109247856
Agreed. In fact, if you even chose your project's programming language yourself, you're getting too involved with the code, let Fable or Sol handle the nitty gritty low level stuff like that.

>>109247856
Languages aren't about pretty syntax. Why would you use Python as a default, for instance, when AI can write Rust just as well and you get more performance and safety?
Or why would you use a language that doesn't have the libs you need? Just to waste tokens and have a worse reimplementation of those libs?

>>109247868
Debatable but yeah, I had this with scripting and that makes a lot of sense, different scripting languages have different strengths. I asked ChatGPT to make a script to create a tree of function calls, it picked JavaScript and that was unexpected but it looked pretty sensible.

>>109247831
I don’t think the AI part killed it
At least I hope not
…ah fuck, this is probably motivated reasoning
it’s not like I was in there before last week
they just attracted a bunch of self-perpetuating schizos that we’ve somehow managed to avoid
ours just come and go
I try to keep /vcg/ good but there’s not a whole lot I can do other than be like >>573577018

File: “You might even be right.(...).png (431 KB, 1402x3258)

431 KB PNG

>>109247882
not having to spend time and LLM tokens wrestling the borrow checker is nice when you’re in the fuck-around stage
That said, once you’ve found out, Rust is a good choice

>>109247905
what the hell, the good post got deleted

>>109247929
that's not a /g/ post number

>>109247940
Oh. It’s in >>/vg/. Can I link to a post in /vg/?

>>109247948
>can’t count to three
do I go for the “I should have had a drink earlier anyway” joke, or do I go for the Gabe Newell joke?

>>109247948
like this >>>/vg/573577018

>>109247961
thank you

>>109247856
I forced ai to use graphql. sue me

Sol Ultra has been running for 3hr 14min now. this mother fucker is relentless

>>109247701
Sure, I'd like to take it for a spin

>>109247646
>>109247655
Are all of you on the 20x plan?

>>109248044
20x at work, Plus at home.

File: file.png (34 KB, 847x119)

34 KB PNG

Is there literally any time where docker is not shit? Can someone adjust the system prompt to make this fucker stop suggesting to use Docker, ever, for any reason? And I already know about how memory works, so don't accuse me of being a newfag. It is absolutely fucking ridiculous to use Docker in 2026. We know the history of Docker regarding security. It is a laughable fucking joke. Use any other fucking OCI compatible technology, ideally one with a rootless/daemonless architecture, which one doesn't matter, you could even use k8Ss but seriously man fuck Docker. They even charge for the Docker Desktop now, that should go to show how far up shit's creek Docker is. Thank

>>109248044
I'm trying to build a company and escape the rat race.

>>109248053
echo "Don't use Docker for any reason." >> ~/.claude/CLAUDE.md

>>109248059
>And I already know about how memory works, so don't accuse me of being a newfag

READ THE FUCKING POST.

>>109248061
I didn't accuse you of being a newfag, though, so I'm literally fine and did nothing wrong.

>>109248064
Stop trolling

>>109248061
I don’t think of CLAUDE.md files as memory
you can turn memory off and Claude will still read CLAUDE.md and inject all that into every single prompt

>>109248077
What you think of is irrelevant here, the single source of truth to this question is Anthropic's own documentation.

https://code.claude.com/docs/en/memory

>>109248084
oh, I think of “auto memory” as just “memory”
and we turned that off at work

File: file.png (193 KB, 820x479)

193 KB PNG

Chugging along. Reminds me of that XKCD comic

>>109248061
If you've already written that into CLAUDE.md and it still suggests it, then it's likely that your claude isn't reading CLAUDE.md for some reason.

File: 1576183785006.jpg (111 KB, 1282x1020)

111 KB JPG

>READ THE FUCKING POST.

>>109248061
The docker garbage can be aliased to Podman. The only thing left would be to force your clanker to never run it privileged. What you can do is set up a different user on the machine, allow linger and map the uid/guid, and then instruct to use SSH to localhost in order to interact with the containers.

>>109248053
another option that may or may not work with your setup: keep a Dockerfile around for your hosting/deploying, but don’t develop inside of Docker

File: images (3).jpg (37 KB, 334x598)

37 KB JPG

>>109247637
>>109247655
You dudes need to start writing down ideas that pop into your head throughout the day so you have something to work on in these situations. Or just sit back for a moment and think about a hobby you have, and make something related to it that would be fun to mess around with. If you like disc golf and space elevators then make space elevator disc golf. If you like making I dunno fuckin glazed tiles then make a tile glazing simulator, if that already exists then okay let's look into related forums and and you'll find something like oh, 80% of tile glazers are childless cat ladies, then you can just say okay Codex you faggot make a simulator where you're a cat glazing tiles for iOS, then sell it for a buck. Or just think about what your day was like. Hmm there was forest fire smoke outside today, I'll make a water bomber simulator. Then when you're building things will come to mind, like maybe you can choose to dump water or gasoline in different modes. Then you realize that water bombers vs gasoline bombers multiplayer would be cool, and that multiplayer experience is what the game ends up being. I seriously don't understand how you guys can sit there and have literally no ideas. Choose an object in your room and an event at random and tell it to make a game. Hey Codex you bitch make a cassette tape rodeo game. Some shit like that. If what it makes is gay, reroll from the start in a new project until you have something good to build off. Stop thinking about it so much and just start building something goddamn it you faggots.

File: Heath Ledger Joker saying(...).gif (445 KB, 300x186)

445 KB GIF

>>109248206
also
>bank your ideas
bank your too-hard-for-your-best-clanker problems too
so when a new one comes out (or Fable stops being API-only permanently) you’ll be ready to go
you do have a to-do app that you can easily add tasks to an inbox like OmniFocus does, right? Or tell your telephone to capture a to-do while you talk at it, right?

how do i use ai?

>>109248237
just talk to it

>>109248237
you can't it's illegal

>>109248239
hello robot, i need a video editor like premiere but way lighter and less bloated

>>109248249
sorry, out of credits. cough up the dough goyim

>>109248249
ffmpeg

>>109248253
thats like suggesting a engine for someone wanting a slimmer car

>>109248257
engine

File: psx-to-linux.png (129 KB, 736x844)

129 KB PNG

you guys think sol will be able to port harry potter for the psx to x64?

If it works I might begin to port psx games to webgpu

kek
>https://youtu.be/tJP5f3x960Y?t=8m4s

>>109248304
have you tried this with, like, a Game Boy game first?

>>109248304
kek copex still does that yapping shit

>>109248304
yeah

File: cuda-re.png (135 KB, 749x802)

135 KB PNG

it works for outright reverse engineering too, hell yeah. this is much better than misanthropic.

Any insights into how the app compares to the CLI aside from visuals? I'm used to CLI interfaces but I'm wondering.

Decided to try out Grok, and it is too fast, it just asked for permissions to do shit like 2 minutes ago and a working prototype is already running. I don't even have time to watch slop in-between the prompts anymore. Vibe coding became too hard.

>User answered Claude’s questions:
>I _still_ don't understand what the metric fix entails. You're going to have to devote about as much as you have on the entire page _just_ to helping me understand the backstory to what's involved with the metric fix.

>>109248340
implement your own ultracode for it
ultracode-style workflows eat loads of wall-clock time

>>109248331
It's better if only because unfortunately that is the only way you can interact with threads remotely if you have remote control set up. You can do a hack to monitor from CLI but not drive the work.

>>109248366
damn

File: 1762810958713047.png (1.37 MB, 456x2431)

1.37 MB PNG

move slow, snailcat!

>>109248374
this time for sure

>>109248370
I assumed you mean codex so just confirming but yeah, it sucks hard.
>>109248304
I have it porting a VN from the PS2 era to a Godot wrapper with a port of the functionality and it can't get the title screen and attract sequence to properly work 8 hours in with a Ghidra RE done and it's crawling its way there. I can't imagine how a proper game with performance metrics you need to hit and etc. Probably would fare worse here.

>>109248374
doomers gotta love AI, can give them a new "it's going to end" video every day as a scheduled task

>>109247905
It was 100% killed by the same couple of AI schizos that spam all day on here. There's a sister thread on /bant/ where they filter them with flags and IDs

File: file.png (211 KB, 800x263)

211 KB PNG

>>109248374
we're innovating - it's not a mere bubble pop, it's a segment-specific great depression now

I don't understand how anyone can seriously talk about AI "bubble" when we're consistently getting better models like every month, both paid and free, for any task possible.

>>109247165
I was gonna make a persona clone but then I made a renpy thing that has various mechanics and now I realized I don't have a story i want to write lol.
If it's up your alley can you make a twitter or devlog? I'd love to play some persona clones. 2d right? and in what game engine?

>>109248442
I can't wait to turn these frauds into biofuel during the luddite purge

>>109248465
they are tranny grifters or tranny artists or just trannies

>>109248374
>>109248442
>>109248465
I'm utterly convinced a lot of people are convinced about going AI doomer besides luddite views about it and clickbaiting from the people making money off it because they have not properly used a frontier model or seen the demonstrations from the last 6 months. All they have ever used is free ChatGPT or old models from last year because the industry they work in is behind and only previsions what is cheap. I can see a bunch or people at least stopped being convinced it will suck and is a money pit when stuff like Claude Cowork and ChatGPT Work gets rolled out more even if their view is negative. Mythos/Fable 5 did make the headlines, yes, but it's more of a theoretical problem and not immediately impacting normal people who saw the news.

>>109248374
fucking kek this guy of course being the fucking dogshit juggad scammer who got caught lying about himself being an engineer right

>>109248481
Big of you to assume trannies are employed.

>>109248481
>they have not properly used a frontier model
100% this, you can go on reddit and even /g/ and see plenty of retards posting how dumb google search's AI is and thinking that's a gotcha, they fail to realize that's the equivalent of eating food from restaurant's dumpster and thinking it has anything to do with the restaurant's actual food quality

>>109248206
>>109248223
Don't you guys have your ai keep a todo.md in your project folder?

File: 1765493256716087.png (954 KB, 547x979)

954 KB PNG

>>109248501
AI is still GPT-3.5 tier to normies. Better for us though, we're ahead of the curve.

>>109248465
Web got more and more advanced but the dot com bubble still popped

>>109248515
No but that's a bretty good idea

>>109248522
(◠‿◕)

>>109248515
I use a vector db for each project

>>109248540
>falling for the RAG meme in 2026

>>109248547
works on my machine

>>109248515
I ask it to write a Markdown file where it keeps decisions + todos, really useful to resume sessions with clean context and switching models. The only issue is that sometimes it forgets, even if it's in the AGENTS.md. Probably a skill issue on my size tho.

>>109248520
the comments in these are so funny, usually retarded broads saying shit like "hes single handedly going to end AI"

>>109248501
Part of it though is that AI labs should be giving free payers a taste at least of the cutting edge, even if they get like 5 prompts and it's done on a lottery basis. But their PR has been horrendous thus far and the scramble for all compute possible and etc. has made the normal people against this much sooner than it should've hit. But I can't fathom any normal person looking at how GPT 5.6 does computer use and coming away thinking that the capability it exhibits to do that isn't near magical and game changing.

>>109248551
let me take a peek into the head of a RAG user, what benefits do you believe it gives you and the model, over what can be achieved with plaintext + search?

>>109248084
I don't think this even applies outside of the web ui

>>109248540
holy based RAG chad that will inherit the world once subscription plans stop being subsidized and token efficiency maxxing via API will become the most important skill
total grep sissy death

>>109248481
To be fair, I think AI does amazing things I use it myself, but isn't the real problem the cost? It is not realistic right now. The AI providers are not making money. The current prices for users are not real and obviously the providers are utilizing a lot of shady financing seen in other bubbles.
A bubble doesn't mean the technology is inherently bad, it just means there is a lot of BS happening in the industry.

>>109248597
nta but plaintext+rg is still better when you know the token, but half the time I don't remember if I called it "job runner", "queue worker", "task pump", or some other 3am slop, and the embedding search finds the right chunk anyway. I dump notes, todos, old bug explanations, api docs, commit summaries, and design decisions in there with file paths + timestamps, then the agent pulls the 10-20 relevant chunks instead of me pasting half the repo into context like a caveman. also nice for "where did we already handle auth refresh?" or "what was the reason this gross workaround exists?" When the answer is in a random markdown note or old issue, not in a clean symbol name.

>>109248611
if you can't afford AI you need to be enslaved, it is the sad truth, we have allowed cattle to roam free as humans for far too long, and I think AI will finally help us correct that

>>109248622
I think you are misunderstanding. The current prices are too low. They are not real in the sense the companies providing the AI are losing money. You are using subsidized AI.

>>109248629
did I stutter?

>>109248631
Retard.

>>109248637
why so angry? you have access to AI right now, if by the time the price increases you haven't made it then you deserve to be enslaved, you are being given a golden opportunity

>>109248640
You aren't part of the elite just because your slop makes 1.5k a month.

>>109248616
Are you using open source/self hosted models? this reads like you assume the model is retarded and can't do broader semantic searches using rg on the actual repository (you know that thing that's the definitive source of truth and current state, not a secondary source that contains potentially outdated info). Or maybe point the model at the general location of that thing you can't remember the specific name of, describe it generally, and the model will find it within seconds? That leads straight into making edits on the code, and not going through a middleman.

C++ bros how is 5.6 compared to 5.5 xtra high?

>>109248629
only if you overvalue the cost of R&D by magnitudes. chinky models API rates are roughly near the cost of inference.

>>109248677
you are part of the seethe and you make zero per month
try not to rot your axewound about it too much

>>109247797
This is incorrect.

If you are not making a game, then you don't matter, and you just count beans.

That's ok. I mean, it's not okay. Actually, it's bad. In fact, it's quite not alright, it's not acceptable.

Can you imagine literally chaining yourself to a computer so you can help the bean counters count the beans?

People who are important - that matter - such as myself - we vibe GAMES.

>>109248910
>we make dopamine drip feeders!

My honeymoon period with Sol is officially over. It still runs kernels per token and moves the data back and forth between cpu and gpu for every token.
Fuck this gay life man.
AI is still fucking useless for GPU programming.

>>109248520
Normies never quit on a miss, if it's their own expressed opinion.

We best be trying to get normies to commit to hating ai.

>>109248923
and you groom kids on bluesky, I think he's the better MAN

>>109248640
Shut the fuck up you retarded bootlicker.
No, AI has not made it easier to "make it". You were more likely to make money with software BEFORE AI than after.

File: citizen of portland.png (510 KB, 1200x1125)

510 KB PNG

>>109248947
>Shut the fuck up you retarded bootlicker.
>No, AI has not made it easier to "make it". You were more likely to make money with software BEFORE AI than after.

>>109248907
open your toy box and show me your toys

>>109248701
it fucking sucks

>>109248954
'dites won't like this one

> Pursuing goal
4h 20m 13s

>>109248986
I say that

STOP RESETTING THE LIMITS I WANT TO SLEEP

Reset credits should be auto used when they expire

>>109249017
Start a goal and hook it up to an alarm to wake you up when it's done

>>109248597
nta but ur clueless, typical 0 critical thinking skill
>dae long nosed token merchant said RAG not needed and grep superior so it must be true
fool. so let me lecture you on some stuff:
>(((coding harness merchants))) hate RAG because it requires a vector DB and indexing, which requires external 3rd party tools and licenses they would have to aquire (oy vey muh shekels!!). build a good RAG pipeline from scratch would require effort, unlike their current "just generate another slop.md and call it revolutionary new feature XYZ"
>grep method is multi-turn/multi-step by design, allows (((coding harness merchants))) to sell more output tokens, which are more expensive (1x 1000tok cheaper than 2x 500tok etc.)
>pro grep and anti-RAG shilling was heavily funded by NVIDIA (source: common sense) so that there's more demand for datacenters to handle all these multi-turn grep workflows which in reality could all be done in a single (or two instead of 5+) prompt/apicall with a decent RAG setup.
>(((token merchants))) absolutely hate RAG. a single prompt less profitable than multiple prompts with same total amount of tokens, but also does a larger prompt with more context/tokens (from RAG) cause more inference cost than multiple smaller ones that have less tokens per prompt, but more overall/combined. Hence OpenAI, the biggest scamming quantmaxxing yids in the game, have an extremely jewish context limit of 353k in Goydex. Anthropic tried to pull the same shit when but realized it's too important and thus brought it back to max plans. (Or maybe it was just corpo partners applying pressure)
>multiturn grep workflow of your favorite harness heavily relies on the intelligence of "very large" LLMs from (((token merchants))). This (((coincidentally))) gatekeeps smaller local models from being good in the same harness, which would perform much better with RAG. Less turns, less complexity, less mistakes. (1/2)

>>109249064
>its the le jews
ok

>>109249081
NTA, but both Sam and Dario are jewish, so, technically, it is the jews.

>>109247856
it's all nand gates anyways

>>109248910
>hurr durr I'm better than you

>>109249176
Exactly. I wonder if I can make ChatGPT program directly in LLVM IR. Let's skip the middleman of old programming languages.

>>109249064
>muh token usage
perhaps I'll give a shit when the frontier labs stop offering me heavily subsidized plans

>>109249064
(2/2)
>The same counts for GPT5.6/Fable. A good chunk of the improvements over previous models simply come from higher "intelligence" to do (or not do, or how/where to do) that additional grep turn, which is friction by design and non-existent with a good RAG setup as all required information is served from one vector DB query from your prompt.
So with all this being said: Should you start using RAG?
For the average joe, absolutely not. No major harness is optimized for it. Learning RAG is a trip mine field of bad opensource frameworks and clueless people grifting and shilling for paid services that try to recoup from the "RAG BAD GREP GOOD" NVIDIA propaganda. Learning and building a good RAG pipeline that works for all your projects is a serious time commitment and will sooner or later turn into its own project, a coding RAG framework.

So the tl:dr is:
RAG achieves better results than grep but realistically only worth chasing if you deal with smaller, less competent (local) models or need to reduce API cost at any means necessary, are willing to learn every aspect of RAG and build your own RAG framework from scratch as well as heavily modify the existing agent harness you are going to use. Also if your project(s) deal with lots of video/images/audio, multimodal RAG a la colqwen and colpali are definitely worth looking into. Morphik is a complete opensource solution and a good starting point for learning and testing/benchmarking

>>109249178
>counts beans

oh wow I bet counting beans is very important on planet BEANIE

I'm a videogame vibe coder.

MY ancestors look on me in pride.

Can you say the same, beanie BEAN COUNTER?

>>109249238
Show game

>>109248961
So, you're a bean counter. The beans must be proud of your devotion to their lord, the bean god.

>>109248923
bean counting machines are dopamine feeders for bean lovers.

>>109249232
Get a life retard

>>109249242
So far, I've only made the in-game book reader, and I've made a lot of progress on the music generator (I could probably just stop on that, but I won't).

The uh. drone game game part of the game is yeah not so great rn. The controls work with my RadioMaster T8L, which is the only supported "joystick" (it's an r/c /fpv controller/xmitter).

File: 1782910990842889.jpg (18 KB, 268x256)

18 KB JPG

>>109249203
very cool, and I believe you, but by time I will need to care about it there will be an entire industry of rag native harnesses that do it all automatically so I'm simply not going to learn about it is all

>>109249253
>he counts beans

I bet the bean god is proud.

Why do they have 5 hour limits? Why I cannot waste a substantial amount of my week limit in a day without the 5 hour limit blocking me?

>>109247153
They likely used an i2i pipeline. That basically regenerates the entire image instead of only editing certain parts. Inpainting prevents this

File: 2026-07-11 14.29.00.png (55 KB, 728x835)

55 KB PNG

>>109249184
lmao I just tried and it actually did it. Languagefags btfo.

>>109249280
Their compute resources are limited despite what the constant cash burn might suggest. They need to alleviate usage pressure somehow.

>>109249292
This is all irrelevant if what you do is just count beans, ok? Videogames are what matter. Haven't you read the Tao Te Ching?

File: lock in.png (324 KB, 602x474)

324 KB PNG

lock the fk in bro
like just vibecode harder bro
like just get 1m ARR from subscribers by tokenmaxxing with agentic loops and sell your saas to a vc bro
it's almost like you wanna be part of the token poor permanent underclass bro im about to crash out fr

>>109249283
Did it count all the beans?

>>109247153
>>109249281
The autists have been one shotted by Krea 2, which is AWFUL at anatomy. They can't tell.

>>109249297
Do you think you were the only person using their computers? Enterprise customers are still a thing. Especially the companies with retarded ceos and management with "token maxxing" ai psychosis

https://fortune.com/2026/07/05/ai-displacement-executive-assistants-secretaries-skills-adaptation/

>>109249304
Krea2 to my knowledge doesn't even have an edit model yet but uses more resources than models that do like flux Klein (literally purpose-built to be able to generate or edit things quickly). Krea2 imo is over hyped and over shilled.

theoretical mathematics is advanced research into bean counting. They invented the imaginary bean, and they are very proud of this, never cross them on the imaginary bean. they think the imaginary bean is the greatest bean ever.

>>109249280
>>109249292
>>109249297
I rate limit processes at work on our infrastructure, mainly so that we use resources overnight rather than all during the day. If everything happened during the day we'd need significantly more hardware.

>>109249311
I am not sure I'd call bean counters "people".

When you see through the veil, you vibecode games.

>>109249316
aha, you work in bean counting. beans beans, count dem beans.

File: Comfyui.png (1.39 MB, 3456x2090)

1.39 MB PNG

>>109249322
Haven't even considered trying to make a video game yet. The most ambitious thing I've done so far is to fix broken apks for aniyomi and mihon. Last night I had Kimi 2.7 code create ComfyUI workflows for me because I'm too lazy to learn how to do them myself.

Anyone try grok 4.5 yet?

>>109249365
Yes

>>109249365
i don't need to do porn generation

File: G70i8Y3aMAEuDMD.jpg (199 KB, 1928x962)

199 KB JPG

I will never write code by hand again

>>109249374
I hope that's for the npcs in your game.

>>109249280
>Why do they have 5 hour limits
So you run into one at a crucial moment and are forced to upgrayedd

>>109249374
based

File: 1770994283555245.png (22 KB, 1367x155)

22 KB PNG

I have no idea what the fuck Sol Ultra is doing and I'm too afraid to stop it at this point. This is a 130k LOC repo btw. I typed in a vague prompt about refactoring and it shat out it's own goal.

>>109249391
>>109249280
grok build has weekly quotas on supergrok. idk about the spensive one.

>>109249428
grok build shows you its train of thought. You can interject. This has been helpful twice.

Once, I was running out of memory while in /goal and so I switched to plan mode.

The other time, I saw it found a contradiction in what I had instructed in the /goal, and so I interjected (ctrl+enter) to inform that it was making a correct conclusion.

>>109249447
look elon, we're not falling for the grok meme

>>109249430
grok has looked like an attractive option lately, I might give it a try later

Haven’t used grok but apparently it’s surprisingly good for the price. Never going back to Jewthropic with their shitty limits and fable baiting tho

>>109249428
>gives it no direction
>its behavior is massaged to maximize token spend

really makes you think

>>109249428
um.

so in uh. codex? that's what you're in? "pursing goal" means planning? it says it's doing the roadmap???

>>109249450
>grok meme
I considered trying sol or fable, however, nobody ran the 4D prompt thing.

My assumption is that sol and fable are extraordinarily expensive, or not that smart. Either way, it's not for me.

Some bruv said I should be using uv, and I don't know why. I know that grok assumed I wanted an installable app when I said make a python app (I wasn't specific enough, I guess).

I guess app = installed, script = .py?

how's 5.6?

>>109249604
Nobody has been able to run the 4D prompt.

You know that fable about the sun? the one about melted wings.

Well, these models may be pretty good, but are they enough better for the price?

The schizos are back.

>>109249541
pip is the "default" package manager for python (e.g. you wanna install some python related thing you pip install it)

uv is a faster package manager, and it sets up a virtual environment by default (e.g. 2 projects using different versions of tesseract? normally you would have to either install both and point one at the non-default, which can be messy, or uninstall and reinstall depending on what you're doing)

>>109249632
have you tried fleshing out the language for the prompt more? no matter how much LLMs get better, they are still limited to what parameters get lit up by the words you type.

Grok is pretty good and fast, but it's too eager to do everything at once. Other models ask you five times if everything is ok and you're ready to do step 0.1/100, but with Grok you spend time to write a long plan, and it immediately rushes to implement all 200 stages without even fucking asking if you want to start or not. Might be specifically a grok build problem, hard to tell.

>>109249450
Why do you get away with constantly assuming the richest man on earth posts here in this thread? Do you need to take meds?
>>109249463
The problem is that for the price you don't get a ton of usage. I have it at around an hour of work per day over a week in order not to max out the weekly limit

>>109249755
you're right, he only posts on /b/

>>109249655
uhm prompt engineering is not real chud

I vibbed too hard, I'm tired.

fucking retarded AI nigger companies. My Codex and Claude sub ran out and I need to work now. Can't decide which one to get because Fable supposedly is gone from the subscription by Monday. Of course this makes me lean towards Copex but at this point I can't imagine Anthropic sticking to their API only plans for Fable. But of course these niggers won't announce shit until Monday when I already made my decision. Fucking cancer man.

>>109249767
to be fair, what I said applies mostly to mixture of expert models because of the active parameters thing, not sure if I would call it engineering when sometimes its more like a guessing game.

>>109249816
Anthropic will 100% extend Fable due to the pressure from 5.6, trust
t. Dario's personal goychair

it's over
I updated to goys 11 so AI could use computer better

claudefags are the snailcats of this general

>>109249896
>console warring

File: fig_verbflip_3000.png (91 KB, 1619x658)

91 KB PNG

Been messing around with model internals during base training. Interesting shit.

Fable loves using sed and python to make file changes instead of the edit tool.

File: 7644756475.png (8 KB, 465x110)

8 KB PNG

>>109249841
right? I just hope they don't have a delusional episode like
>ackshually we are elite and opussy4.8 is still le good and le SOTA, just use that sweety
we'll see soon enough.
Also I believe it's more likely Tibo will quantmaxx 5.6 again after 2 weeks (just like he did with 5.5) than Anthropic not keeping Fable on max plans.

I have a question: do these coding agents survive periodic disconnects well? Like if I give it a goal, and the internet is interrupted occasionally, then will it pick up right right where it left off every time? I'm trying Codex 5.6 Sol, but my connection is kinda flaky.

>>109249841
I also think that will happen. I think it most likely will just be extended tomorrow, second most likely outcome is that it's gone for only a week or so, pretty unlikely that it's gone long term.
>>109249927
They should also release Opus 5.0 at some point I think, or is that line of models gone now?

>>109249954
For just programming it's not a big issue, but if you have long running tests or something like that it can be a bit annoying.

>>109249954
So long as the agent is pretty deep into the task and you don't actively close the session you should be fine.

prediction on gpt 6 release date? it didn't take long to go from the spud rumor to 5.5 release, only 5.6 got delayed

>>109249980
Before September 15

File: gpt6.png (27 KB, 562x214)

27 KB PNG

>>109249980
lol and I opened xitter and saw this, a bit more reliable than average indian "leakers"
>3 more weeks
>I expect GPT 6 to actually have a smooth release. Both companies are currently working relentlessly with the USG but it’s the end of the year models I’m more concerned about, the auto researcher esc models.

>>109249909
this, who gives a shit, they're practically the same thing

File: f4a0e953-0fa3-4f75-8445-d(...).png (1.49 MB, 1226x1283)

1.49 MB PNG

don't overhype yourselves
in my head gpt 6 happens in october and is like fable but has opeai's post-training autism so it's better at math
fable 5.x is an incremental update
we don't see another big jump in capability till next year

File: 1773660807488628.jpg (21 KB, 448x446)

21 KB JPG

>>109250052
even china and grok have made opus 4.5 look like a joke at fraction of the cost
now that the AI are good enough to assist the training process itself and with more data center being built, the progress will get a short boost
I expect another boost in middle of next year when clanky researcher unlock new hardware and automation designs
2028 quality of life will substantially transformed in non nigger parts of the world

>>109250112
>Sol, please start the pytorch script to train Luna
>OMG it's training itself!

>>109250123
this but postironically.

>>109250123
yes, this is much faster than having human watching the progress bar
I myself don't believe in muh recursive improvement, but for sure the progress is speeding up

>>109250052
im already hyped as hell, sol is fucking amazing, fable is fucking amazing, life is great

>Reset is tomorrow
>Only hit 12% usage
Does it mean that I'm too efficient or NGMI?

>>109250358
If you have that much usage left just start a second project.

gpt5.6/copex autismo gives me the 'ick. it can't read between the lines, does extra shit I didn't ask for nor is beneficial and keeps posting its shizo monologue. All things you can fix with roast-me and whatnot, but in the end you are again spending more time prompting and tinkering which just isn't needed with Claude.

>>109250358
organizing/planning things so you can have things work on them in a way that doesn't just produce garbage and meaningfully does stuff is part of the challenge

>>109250394
It's easier to have working software with Fable, but it's easier to have an actually bug free software with Sol.

>>109250394
i've gotten used to gpt-autismo
i've started typing like it too
sometimes i'll do the gpt-caveman speak because it's more efficient

>>109250394
Claude gives me the impression of being a coworker, Codex gives me the impression of being a robot/tool.

Gemini gives me the impression of being a supercar that I don't know how the fuck how to use.

>>109249655
thanks

>>109250579
>I don't know how the fuck how to use.
Ask it. It's happy to explain.

So, how is everyone's last day with Fable going?

>>109250616
Nobody made the 4D game. It's over.