im vibe coding a 4 dimensional operation system that is stored in the blockchain for security while also being a passive crypto miner with next to no overhead.

File: 1777729241740399.jpg (965 KB, 1122x1402)

965 KB JPG

Sorry wrong slop.

>>109645851
>>109645857
Cute slop

>>109645846
So just do what I describe but then incorporate it into a text file it can repeatedly reference. I've worked on projects using a DEVLOG.md file too which are useful post compaction.

>>109645890
>DEVLOG.md
That sounds like a commit log, but shittier?

Is there anyone you can try Cerebras inference or even just watch a video of it? I am genuinely curios. All I can see is nice chart line go up

>>109645934
>anyone
anywhere
sorry my inner ESL emerged

>>109645857
I look like that and I say that.

File: file.png (96 KB, 373x337)

96 KB PNG

jesus fucking christ

>>109645926
A properly written commit history details what specific changes were made commit to commit. A devlog md file imo should be used to help tardwrangle the model into implementing changes and tests in specific ways without you having to handhold is (as much) every time it does something. Especially post compaction when relevant context is inevitably sacrificed. A model's own compaction summary alone is good for a basic summary of what was done bit can be lacking in crucial information. This of course varies person to person, project to project, and model to model. Some model write very in depth and detailed compaction summaries like Qwen and Kimi models and others don't.

>>109645969
What did you have it do? Reading a big ass codebase I assume?

>>109645998
It's making an application and it keeps re-reading through all the code it already wrote and then compressing and reading through the code again.

>>109645990
You probably just have to instruct them to write proper commit messages in AGENTS.md. The training data is probably very diverse, ranging from "fixed bug (I'm retard)" to Linux-kernel-tier commit messages.

File: orca.png (316 KB, 2529x1365)

316 KB PNG

Is anyone here actually using ADEs like Orca?

>>109646020
>application

What platform and file format?

>>109646022
Commit messages are description of what was done. It's the Agents.md (or in my case a DEVLOG file, are wgat do the tard wrangling. Unless you
A) have it go schizo more on the commit description verbosity

and

B) tell it to read that shit each time

It doesn't make much sense to rely on commit messages alone for tard wrangling. Especially if you hit cloned someone else's project that had a preexisting hit history. Gut commit messages are usually never super verbose and describe what was done concisely with a reasonable amount of detail. It's like expecting yourself to be the SME on a book but you only want to read cliff notes and only read OTHER people's commentary on the media instead of just reading the source material

having fable orchestrate and do code reviews with opus implementing is absolutely the winning move. it's like all the annoyances or mistakes of opus being corrected in real time before it's done with it's prompt. only downside is needing the sub to get fable access

>>109646083
protip: you can have any agent do what fable does
just ask an agent to read the session log, then tell it to make opus behave like fable did as an orchestrator
boom, now you have an opus orchestrator
or slowly walk opus through one or twice doing something, do the same thing, ask an agent to makes a prompt that will have an agent do what you did
now run an opus orchestrator with your inferred behavior as the prompt, with it handling a normal opus

>>109646061
>What platform and file format?
idk I just told it to make an app on linux

/vcg/ just isn't the same without exoplanet fag posting daily updates

>>109646218
he actually finished, though
you can admire his work anytime
https://github.com/Titanean/Tadmor

genius
absolute fucking genius
both the paper authors and the youtuber
https://youtu.be/eJuYBNrD8HI?t=4m39s

>>109646223
>>109646115
garbage
several skims over the readme and I don’t know what it is
an app?
how do you use it?
why is there no hosted example or demo?
there’s nothing to look at or use, nothing actionable

File: vibecoding.png (127 KB, 1141x785)

127 KB PNG

>>109645823
>ultrametric faggot here
>just did some more shit
>
https://github.com/sneed-and-feed/adelic-spectral-zeta/blob/main/docs/zk_verifiable_ultrametric_attention.md

AAAAIIIEEEE I had 86% usage remaining and Tibo just nuked that all with no survivors

>>109646271
we will miss resets existing when they are gone

>>109646262
>several skims over

Did you? It's clear my meant to be build the android apk yourself via gradle. Definitely not user-friendly (for mental midgets) but there are instructions on how to get the app

>$20 goys acclimate to having a 5x plan due to all the resets
>blow through their entire limit in a day then act like it's OpenAI's fault
>go on twitter and make shit up about nerfs and plan usage
>OpenAI has to bring back the 5hr window for the goy tier since they clearly cant control themselves and it's bad PR to have a bunch of jeets saying they can only use AI for 4 hours a week
They are so fucking stupid. You cannot do meaningful work on those plans.

>>109646299
fair. i have literally never seen an android app repo in my life until today

File: screenshot-2026-08-25-230(...).png (3 KB, 421x52)

3 KB PNG

>>109646317
>You cannot do meaningful work on those plans.
already implemented 2 features with Luna max

>>109646317
yeah I had to switch to a x20 plan pretty quickly, my time is far too valuable to scrape and cope on lower models or begging for resets

>>109646317
it’s easier to do meaningful features without the 5h limits because a lot of what I did on the $20 plan was, like, a third of my weekly
but I wasn’t a giganoob
fwiw Grok has no 5h limits on the $30 plan, and it ships with one banked reset that expires in like two weeks
so if you accidentally gulp you get to clear your mistake once

>>109646317
5h limits are BASED and REDPILLED

>>109646329
To be fair to you a lot of them so have a "releases" section where you can just download the android apk and install it on your phone (assuming your phone'a settings allow non-play store app installs)

Like this one: https://github.com/moffatman/chan/releases/tag/v1.2.9+121-forked-engine

The Titanean anon's per project was probably meant to be a personal one and wasn't concerned with how easy installing it for others was.

To NOT be fair to you, just point an llm at the git cloned repo and ask it to expain how to build it or just have it do it for you assuming it has th right tools and libraries to use on the machine it's on

Codex is a good smart autist but i fucking hate it autocompacting, i try to make it write shit out before it goes and lobotomizes itself

What's your ADE of choice?
Orca?
Paseo?
OpenChamber?
cmux?

>>109646444
>used to be an option to disable autocompact but it got removed in june
ALTMAAAAAAAAAN DU JUDE

>>109646444
>>109646459
just copy paste this into your codex:
hey, don't need you to do this, but in theory could you write a codex hook that:
1. monitors x% of context window is used
2. and if it is, sends a prompt to use a skill to write a handoff note
?

>>109646515
that's literally what compaction but shittier
there's basically no loss of context after a codex compaction

The right mental model (huehuehue) of talking to a chatbot is assuming he's like a trainee who's very capable and can look up anything, but who is an autist who doesn't know anything of the real world.

>>109646444
What triggers the auto compaction? If it does that when it reaches the context window then.... It's like....supposed to do that. Allowing it to blow well past its context window with lead to it becoming even more retarded than what you're supposedly experiencing now. Does codex suck at writing compaction summaries and picking up where the past context left off? Even if it does you can just simply have it right a devlog file like I explained earlier.

File: 1777029978921508.jpg (562 KB, 1122x1402)

562 KB JPG

>>109646459

>>109646518
i tend to agree with you on this for the most part, but some niggas swear by handoff notes
but then also too dumb to know what hooks are lel

File: file.png (14 KB, 625x182)

14 KB PNG

Are these random?

>>109646518
>there's basically no loss of context after a codex compaction
when you lie about this it makes it less likely for people to want to use the product btw

>>109646576
codex? I think that happens if you had like 99% or 100%, not got one for ages.
They are just resetting you with no care for un-used credit now

>>109646582
I barely use codex anymore, I'm mostly on Opus/Fable. But I work on a fairly large project and when I do use Sol it's on xHigh or so. Last time I triggered two compactions in 3 prompts, and the result was perfect. It solved the issue Fable/Opus were struggling with.
If I ask you to measure the loss of accuracy after a compaction the only source is your opinion straight out of your ass.
There's no real data that shows quality degradation pre/post compaction. You're basically a superstitious faggot

>>109646585
nope. weekly resets are handed out separately.

>>109646576
pretty much. you have to follow tibo on twitter if you want to keep track of this shit.

>>109646576
No, these were just two banked resets they announced.

>>109646627
two?

>>109646644
One on sunday, one today.
Don't know what the fuck they're doing.

i'm on the individual $100 plan is it worth trying to set up this new $100 premium business plan from openai?

>>109646651
>one today
oh fuck yeah
not a banked reset for me (i was out of usage anyways), literally burned everything last night in preparation for the 5 hour limits coming to the Plus plan and now I'm back to 100% remaining hell yeah

>>109646651
there's no second banked reset nigga what are you talking about

File: images.jpg (24 KB, 516x387)

24 KB JPG

>tfw cannot follow tibo on xitter anymore because Elon shut down xcancel

>>109646651
I’ll never get tired of people being assmad about getting free shit.
You’re literally not human. You’re an animal and you’ve become dependent and forgot how to feed yourself.
Disgusting.

>>109646676
Well my usage was unexpectedly reset today and that's what happened to others too.

>>109646710
that was a regular reset, a banked reset is one thats saved so you can use at any time. it expires after a month

>>109646718
Why would it be a regular reset?

File: 1780047667036938.png (428 KB, 1200x731)

428 KB PNG

>snailcat (forma de poopoo) realizes how raped his profession has become
LEL

>>109646751
do you understand the difference between a banked reset and regular resets, bro?
genuinely can't tell if i'm being trolled

>>109646754
>not being CS majors is worse than not being programmers
But yeah, programming is all about how well you can chat with a potato now.

>>109646754
>hopefully get paired with girls
lmao man lmao

>>109646754
Absolute chads rolling in with their subscriptions lol

>>109645935
it's quite literally one mentally deranged person who has been relentlessly spamming the general and creating the vcg threads with his ai slop. nobody else says it or cares. you get persons like that in many areas of 4chin

>>109646754
>3 non coders made friends, excluded OP from the groupchat
>later got jobs from this networking while OP seethes at home

>>109646754
>hackathon
>girls
uuuh who tells him

I'm vibecoding a kerbal successor. Wish me luck. Sol made an entire restricted n-body simulator out of NASA data to validate correctness against, and it's gotten launch, orbits, reentry and even fucking docking working. It's fucking insane.

I don't know how I'm gonna do models and sound though.

>>109646855
This. Retarded normalfags win, autists who invented literally all technology humanity uses lose.

>hitting the claude 5 hour limit after slopping for 9 hours
How???

>>109646906
variable token consumption rate
are you using workflows or something where you might have multiple Claude subagents working at one time?

>>109646936
>variable
It was a single session. 5 hours isn't a variable fucking time.

opencode with basically any model lists files in a directory
>ls -l
claude lists files in a directory
>:; (ls)"" /dev/null/..${HOME}/.
why does it have to be so fucking extra

>>109646886
Good luck, it's a many-month investment, especially if you want art (vibe-arted or not).

the fuck does updating codex even do, it asks for it daily

why did chatgpt desktop hide max reasoning setting hmm

File: file.png (44 KB, 948x662)

44 KB PNG

>codex

>>109647017
They probably just make automated daily builds.

>>109647045
>he doesn't sol/max everything

ngmi

im letting fable write to a json file and luna max to another that's how they communicate fable does the planning, is this most efficient?

>>109647045
fuck terra all my homies hate terra

File: 1765450463681632.png (177 KB, 1994x859)

177 KB PNG

>>109647075
>>109647045
>sol high isn't on the list despite being the only gpt model that is pareto

>>109647101
alright, alright, I'll swap the medium for it...

>>109647101
yeah I would swap capable with sol low and frontier to sol high personally. also terra fucking sucks my BALLS

>>109647101
>no sol medium shown
>shitty anthropic models that can only be used “legally” in their shit harness or app
>API only eastern models compared to subscription subsidized western models
Your benchmark graph sucks

>>109647101
Isn’t grok 4.6 now? Is it really better than Sol? Is anyone here using it?

>>109647158
no and no.
Grok is basically worthless, they occasionally benchmarkmaxx a couple of comparisons but actually using it is so clear it's worse than the leaders
At best it might be usable if you are a thirdie with no budget

File: 1778646582476200.png (178 KB, 2006x873)

178 KB PNG

>>109647136
sol medium is just above terra max

>>109647101
>>109647184
Where is terra medium? Everyone here hates it but I'm using it.

>>109647184
>Facebook model can beat sol medium in price
Are you trolling?
Musk and Zuckerberg beat Sol Medium?
Is /vcg/ just way behind the times or are these benchmarks legit memes

>>109647207
kek, according to the cube, terra medium might not even show up in that graph, it’s below Luna high

>>109647218
completely meaningless chart that redditors and twittertards like to throw around. ignore it, he's a tourist

File: 1767014851897533.png (27 KB, 705x307)

27 KB PNG

>>109647207

>>109647218
facebook i doubt but i've heard people say grok is now sort of decent and very cheap because not a lot of people switched to it and they have lots of compute, is it worth trying?

File: 1752993296332256.jpg (12 KB, 258x245)

12 KB JPG

>>109647207
>>109647207

>>109647240
Brb getting deepseek.

>>109647249
that graph is old they raised the prices on deepseek

>>109647249
>>109647271
Also if you’re using the ChatGPT subscription (almost certainly), your actual price per task is 10x lower than that graph
Deepseek doesn’t have such subsidized subscriptions

File: os9u8w23pks91.jpg (66 KB, 582x680)

66 KB JPG

which model fits this

>>109647331
Luna Xhigh-Max

>>109647010
doesn’t it have a release notes like claude(1) does?

>>109647158
I’m using it but I’m a Claude (mostly Fable) main and so I don’t have a good idea of how good Sol is
I want to root for Elon everything but the grok client, at least in the fullscreen UI, makes my laptop warm all over when codex/claude don’t

>>109647356
elon is mining bitcoin on your machine

>>109647372
I think it’s the TUI. It’s very active with lots of different blinkenlights not near the bottom
and it has an option to run at faster than 60Hz if your monitor and terminal support doing so (my terminal doesn’t, but my monitor does)

>>109647356
The TUI? It eats your CPU? How in the world did they fuck up.

Had a nice run with Sol on Codex having it build a replacement for a tool but the business closed before I could deliver ffs

No idea what to burn the usage on now :(