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


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

You use Git, right, anon?

## What “vibe coding” is, and how to do it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

## News
- (2026-08-31, 11:59 PM America/Los_Angeles) Claude’s 2× promotion set to end
- (2026-07-24) Claude Opus 5 out

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

----

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

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

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

----

## Prompting / 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
https://github.com/DietrichGebert/ponytail

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

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

## Previous thread
>>109657847
>>
You will never be a real programmer. You have no skill, you have no knowledge, you have no ambition. You are an untalented loser twisted by prompts and chatbots into a crude mockery of an engineer.
All the “validation” you get is two-faced and half-hearted. Behind your back people mock you. Your parents are disgusted and ashamed of you, your “friends” laugh at your “impressive” projects behind closed doors.
Programmers are utterly repulsed by you. Thousands of years of evolution have allowed coders to sniff out frauds with incredible efficiency. Even vibecoders who “pass” write uncanny and unnatural code. Your project architecture is a dead giveaway. And even if you manage to get your PR accepted by a project, they’ll turn tail and bolt the second they gets a whiff of your long, em dashes in comments.
You will never be happy. You wrench out a fake smile every single morning and tell yourself it’s going to be ok, but deep inside you feel the depression creeping up like a weed, ready to crush you under the unbearable weight.
Eventually it’ll be too much to bear – you’ll buy a rope, tie a noose, put it around your neck, and plunge into the cold abyss. Your parents will find you, heartbroken but relieved that they no longer have to live with the unbearable shame and disappointment. They’ll bury you with a headstone marked with your failures, and every passerby for the rest of eternity will know a vibecoder is buried there. Your codebase will decay and go back to the dust, and all that will remain of your legacy is a git history that is unmistakably slop.
This is your fate. This is what you chose. There is no turning back.
>>
File: 1785166130181117.jpg (965 KB, 1122x1402)
965 KB JPG
>>109663139
This is hate speech.
>>
>>109663127
This is literally me
>>
For those still using Claude, are you retarded?
>>
>>109663204
why would I stop using Claude
>>
File: deepseek.jpg (287 KB, 1080x1079)
287 KB JPG
>>
You will become a real vibe coder. You have prompts, you have models, you have ambition. You are a talented builder empowered by agents and chatbots into a new kind of engineer who can turn an idea into a working product before everyone else has finished arguing about the tech stack.
All the validation you get is real and increasingly enthusiastic. Behind your back people share your demos. Your parents are proud even though they still don’t understand what you do, your friends send people links to your impressive projects and tell them you built the whole thing over a weekend.

This is your fate. This is what you prompted. There is no turning back.
>>
>>109663214
Codex is much better for the same price
>>
where's anthropic new models? fake rumors again?
>>
Sol-medium pisses away more weekly usage than opus-high. And opus is actually good.
>>
I got da codex 5x to be ready for astramaxxing and free myself from the tyranny of the 5 hour limit
now I have more sol than I know what to do with, literally just coming up with random shit to point sol at just because I can, this is too much for a normie like myself
>>
>>109663280
never a reason to use sol medium. the only things you should be using is sol-low, sol-high, and luna-xhigh
>>
safe guards make Fable effectively worthless, so sick of manthrobic
>>
>>109663332
you should feel so safe anon
>>
>>109663329
Will sol high suddenly be good instead of pissing even faster?
>>
>>109663354
sol low and sol medium are basically the same thing.
>>
File: 1742539745798895.jpg (146 KB, 550x502)
146 KB JPG
>JUST IN: Leopold Aschenbrenner reportedly promised his bride he would someday buy her "a galaxy" days before his heavily leveraged AI hedge fund crashed.
>>
>>109663379
~rus was down the other day
is he gonna buy that one
>>
File: codex_context.png (41 KB, 1058x130)
41 KB PNG
>>>109661690
I actually updated Codex and the higher context now seems to work, it says 828k even after a few messages. When I tried setting it higher in the past, it would show 828k at startup, but then it would show the smaller context window after the first message.
Maybe it was because Codex CLI updated today. I also checked the fix in GitHub, from my understanding it should probably also work if someone's model catalog didn't update yet, like mine did after the update.
>>
File: 65434.jpg (82 KB, 913x714)
82 KB JPG
>>109663266
nah.
>>
>>109663486
fable is unusable though
>>
File: deepswe4.png (25 KB, 895x559)
25 KB PNG
>>
>>109663490
they'll fix it within two weeks
>>
5.6 Sol is insanely good at reverse engineering. Not even sure if I need the IDA decompiler, it can straight up read assembly code and understand it perfectly
>>
>>109663501
>benchshit
lmao
>right to left x axis
kek retard
>no gemini 3.7 flash which beats them all
>>
File: HQudL8iXoAAIsbf.jpg (308 KB, 1620x1643)
308 KB JPG
>>
>>109663587
cn i opusy
>>
>>109663587
How do you go about fixing this?
>>
>>109663587
I especially hate "shape".
>>
>>109663587
I fucking hate anthropic. foid coded crap
>>
>>109663587
You forgot
>This is a genuinely smart solution
>This is genuinely better than than my idea
>This is a genuinely great implementation
>>
>>109663587
Is this mental mode the ground-truth?
>>
File: 1787855205.png (11 KB, 237x184)
11 KB PNG
>another reset
>>
>>109663663
>mode
model*

Fuck.
>>
>>109663587
Sol doesnt have this problem though so its an Anthropic skill issue
>>
>>109663665
TIBO I FUCKING KNEEL
>>
>>109663695
i like both models but they both have their own quirks, like sol always wants to make its own tests for everything
>>
>>109663379
A Samsung Galaxy
>>
all these resets make me feel like I am not using my usage effectively enough.

I want a reset at 10% not 85% reeee
>>
>>109663717
as if that is in anyway comparable
>>
>>109663665
why isn’t the moon red?
>>
>>109663784
“optimal” in the face of random Tibo resets is draining yourself dry ASAP
but that leaves you with nothing in case you want to get Codex to help you debug why your computer is acting funny, and other random tasks
>>
So will tibo issue even more resets in the next days?
Or do I burn all my usage now and take a break until in a week?
>>
File: 1785914101125894.jpg (30 KB, 368x368)
30 KB JPG
>>109663654
Fun fact, Claude only says those things when true.
>>
>>109663813
well resets are cheaper for them if they spam them
>>
>>109663784
you have to tokenmaxx, do anything you want to do NOW, there is no reason to pospone things, you should be using all the tokens always
>>
>>109663818
Strictly speaking I wouldn't know because I've never been anyone else
>>
>>109663784
i used 100% right before the last two resets. i have even had my subscription for a month yet lmao!
>>
>>109663826
>there is no reason to pospone things
claude helped me figure out what was buggy about whatever PITA process was acquiring the secure-input lock on my Mac and not letting go
your clanker is WAY better at grepping your system logs than you are when it comes to debugging OS funny business
>>
>>109663842
just upload your log files to a chatbot lmao you don't need an agent
>>
>>109663204
Fable exists, it's still the best model despite Anthropic being the worst company.
>>
>>109663609
unslop skill
>>
>>109663609
>>109663961
this one https://github.com/cursor/plugins/blob/main/pstack/skills/unslop/SKILL.md
>>
good morning sirs is everyon vibe ok?
>>
File: average-linkedin-post.png (1.57 MB, 1574x1452)
1.57 MB PNG
what did Sreekanth Kumbha mean by this bros???
> pic related

>>109664031
good morning sir! vibe code very strong today, Codex and Intel the best combo! power and quality very pretty blue teams!
>>
>>109664031
gm saar i am vibings very much with the luna today after the tibo reset
>>
>>109664031
Not really, getting tired of the 90% competent 10% severe mental retardation. I don't know how much more I can take before coming a deranged anti AI zealot.
>>
>>109664076
This.
>>
>>109663280
>Sol-medium pisses away more weekly usage than opus-high
I highly doubt that
>also, stealth testing cube updates
>>
>>109664087
:^)

hi. I notice you're not visualizing 4D using a 3D perspective transparent view cube.
>>
>>109664117
give me a 4th dimension to view and I’ll get right on it
>>
>>109664087
in hindsight, it would mean if Sol medium is costing you more than Opus high, you’re probably throwing Sol medium at a task that’s too complicated for it.
Sol Max is the closest in the benchmark to capability, it should be a little faster but a little more expensive (but offset immediately by just one Tibo reset)
>>
>>109664157
How many tibo resets are included in a subscription?
>>
>>109663787
I like Sol, but this actually is a problem because it really bloats your code base at some point.
>>
File: shadows.png (1.34 MB, 1554x657)
1.34 MB PNG
this took a lot of iterations but I got sol to do shadows in my game
>>
>>109664141
if you have four numeric columns and idk ??? five points? not sure. you can plot it in 4D.
>>
reminder that literally every benchmark and every sane person says that openai models are more token efficient on tasks compared to their like for like anthropic equivalents and you should not entertain the schizophrenic ramblings of one anon
>>
>>109664087
How do I do this 3d comparison
>>
>>109664215
Do you mean me, who complained about sol medium?
I don't know. I wish there was a better, more transparent, more intuitive way to see how much usage something uses.
>>
>>109664215
every sane person isn't a slave to a single frontier model
>>
it's really easy to just not reply to him
see, i'm doing it now
>>
File: cleanup.png (294 KB, 2908x1300)
294 KB PNG
after spending over a week importing ebooks the cleanup can finally start.
sol medium is very nice for ruby, tho i have to spend time this weekend cleaning up code so it looks more like idomatic rails and not random shit in an rb file
>>
>>109664215
It's common knowledge you get the gpt if you're poor and Claude if you're rich.
>>
>>109664215
I mean Sol is all around great sure, but there are some tasks that are so hard only Fable can reliably handle them.
>>
>>109664268
openai does not have a fable-class model
sol competes with opus
>>
File: what.png (342 KB, 563x572)
342 KB PNG
Guys I just found out something huge: I thought benchmaxxing was just a schizo thought but in fact most benchmarks are either completely public or sort of private but still trivially obtainable.
There is absolutely no reason NOT to think every single model out there is benchmaxxed.
>>
I just found out that grok 4.6 is good at making flowcharts. I guess that's obvious, but that's crazy useful.
>>
Why doesn't Mistral give us resets?
>>
they should only give resets to those who maintain the login streak
>>
File: grim.png (61 KB, 1183x752)
61 KB PNG
>>
>>109664432
I used my entire weekly gml5.2 quota last night with Mistral pro.
Now I have to wait a week.
Meanwhile Tibo already gave jewish lackeys their reset.
>>
>>109664447
why 5.2? don't they offer 5.3 too?
>>
>>109664321
They are benchmaxxed, but when a new benchmark comes out, the good models usually still perform better on that one.
>>
>>109664460
They are white people, they only offer 5.2
>>
>>109664447
When will anthropic get a Tibo?
>>
>ask sol max to make a plan
>come back to a 2000 line md file with 11 milestones

wtf, openai really just wants to milk me dry huh
>>
>>109664493
What was the plan for? That scale might be more reasonable depending what you asked it for
>>
>>109664525
I asked how many rs are in strawberry
>>
>>109664525
oh nothing much, just a SOTA agent for an unsolved problem using strategies from a paper published last year
>>
fun with numbat for someone with a Sunday at 2 AM reset:

let n = now()
let w = weekday(n)

let days_since_sunday =
if w == "Sunday" then 0
else if w == "Monday" then 1
else if w == "Tuesday" then 2
else if w == "Wednesday" then 3
else if w == "Thursday" then 4
else if w == "Friday" then 5
else 6

let candidate = calendar_sub(time("02:00"), days_since_sunday * days)

let last_sunday =
if n - candidate < 0 seconds
then calendar_sub(candidate, 7 days)
else candidate

let next_sunday = calendar_add(last_sunday, 7 days)

(n - last_sunday) / (next_sunday - last_sunday) -> percent

so I’m 64% through my week
my fable percentage is over that
my everything-else percentage is under that
>>
https://desuarchive.org/g/thread/109328319/#109329147
Thinking about this, I think Anthropic might have deployed an exploit control vector against me, because one time it suddenly made an SQL injection as a joke, out of nowhere. Against me or maybe somewhat wider.
>>
>>109664564
alright, vibecoding was a mistake, this is the worst thing I’ve ever seen
>>
did they fix something with that last reset btw?

I feel like sol is using up way less of my limits now.
>>
>>109664545
If it's a public paper, you could've just used Sol Pro in the webchat and done it for "free", assuming you're on a sub and not API keys
>>
Fug did Tibo reset again?? I was only at 90% usage for the week
I'm starting to get a little wary, like he's waiting for me to stop chugging luna max and he'll stop doing this once I start using sol
>>
>>109664693
it's literally gambling (and i love it)
>>
>>109664693
>>109664769
time in the market beats timing the market applies here
>>
>>109664688
I wanted the plan to include actual code snippets and concrete implementation plans working off my existing code base.
>>
Currently testing Qwen3.8-Flash-Next on RTX5070Ti + Ultra 7 265K + 64GB RAM setup
Q4 at 25 token/s
quite impressed so far
although only 100k context length. I might have to try flash attention to get more
>>
>>109664268
Glad you use Fable, can you give Grok 4.6 high / xhigh a run for its money?

I think Grok 4.6 is fable class, not sure if you'd agree with that.
>>
>>109664584
???

what's the prompt? Only once have I caught Grok in a stupid mistake (not bounds checking something). every other time, I gave unclear instructions, or the library itself had bugs you have to work around.
>>
Why are none of the GPT pro models on Artificial Analysis btw?
Are they paid off by Anthropic?
>>
The big wow of vibecoding is like "wow I never have to suffer annoyances anymore".

Hammerman is my mascot, I have a new hammer and I'm on the hunt for nails.
>>
>>109664792
Aaaah. Your options, then, are private github repo it can read from through the connector or use Desktop Commander and have it read directly from your PC if you don't wanna lose usage to planning
>>
biggest vibecoding wow:
>big changes don't really need a plan sometimes
>small things need a plan (plan mode), like REALLY small things need a plan, it's like a microscope
>>
>>109664904
I feel you. Big changes = model going
>ok i need to plan this with 6 gorillion sub agents
small changes = model going
>ok lets just do this, no point to check all affected components, 20$ snailcat gonna complain about me using all his tokens for a small feature
>>
sol loves trying with all its heart to make my stupid ass ideas work, and I love it for it
>>
File: 1764445903744058.png (524 KB, 1179x774)
524 KB PNG
Gemini soon
>>
>>109664845
I’m a Fable main and Grok took some retarded shortcuts that I had to prompt it out of
I like Grok but it’s Opus-tier at best, not Fable-tier
>>
>>109664966
I spent $60 trying to make Grok do what is actually impossible lol

basically I tried to make it take arbitrary waveforms and turn them into a wavetable-esque thing. This isn't actually possible with any known math, apparently.
>>
>>109664858
I’m just messing with you. If that was JS, you could collapse that down to 5 or fewer readable lines and I imagine most languages already have the day of the week as a number
>>
>>109665007
wtf was the googler vagueposting about ox alpha about?
>>
>>109664858
I asked ChatGPT (in the web UI) and it gave me something close to this but it was buggy and I asked it to fix it and it did
>>
>>109665017
Interesting. So Fable will reshape your requests to a less literal interpretation? Or do you mean that it literally blew off your instructions?
>>
>>109665024
No, I mean I have found that I said something wrong, and then it did what I said, this happens a lot.

eg. write a py that xxxxx

so, it writes a py. one.
>>
>nobody knows why Fable got lobotomized
>Fabel fanboys keep going anyway
>>
Did they nerf ChatGPT Sol-High in the Web Chat?
I had that weird bug where they served 5.5-mini instead of 5.6-Sol for a couple of days last week, but it just doesn't seem to spend a lot of effort anymore.
3 weeks ago I would regularly have sessions where it would think 5, 10 or even 15 minutes on high, now it rarely thinks for more than 1.
Also you used to be able to see at least some trunkated thinking summary but they don't show jack anymore.
>>
>>109665029
nobody knows, some google employee tried saving face and seemed even more ridiculous
he said “well we weren’t referring to ox alpha exactly, it was a general thing” despite the last post literally saying “ox alpha”, only to get them roasted even further
I guess google is a retarded shell of a company
>>
one of my accounts seem to have been reset for the second time today? the other one didn't. da fug
>>
>>109665022
If by arbitrary waveform you mean amplitude over time then yeah that's retarded. A wavetable is just an array of waveforms.
>>
>>109665007
>>109665029
What are the chances they really thought it was Gemini 4 Pro in disguise?
>>
>>109664493
You use the unlimited free web interface for that.
You don't ask codex for a page.
Unless you wanted it to make the plan based on your pc and the files there
>>
>>109665107
What tools can the webchat use?
At least you could probably upload a .zip, not sure.
>>
>>109665123
It can do pretty much everything.
>>
>>109665083
I can imagine it, but the problem is that you can't sync the sine wave automatically, and it might need to vary in pitch over time at first.

I think it's not exactly an unsolvable problem, however it's not one anyone has solved yet. It's related to FFT - you need to know what the "real" frequency is at each moment in time, then just crossfade with a synchronized sine wave.

purpose: eliminate the "pop" you get when pausing. This is the only 100% guaranteed method.
>>
>>109664894
iiinteresting, i'll keep that in mind for next time. but i actually dont have access to the web interface because it's not my account.
>>
>>109665093
There are even some that thought it was a self-learning model
Total delusion
>>
>>109665022
>>109665083
Actually now that I'm thinking about it, given a waveform with number of samples >= number of waves in your table at a fixed resolution, you should be able to just subdivide it. I.e. if you want a wavetable with 10 waves each of 10 samples you need an original waveform with 100 samples. Make each of those tunable or at least derive-able and you should be able to do what you're asking.
>>
>>109665048
Yeah, the main issue I'm having is I think that uh. well it's like I need to be keeping a "learnings" notepad, but also imo vibers need to be sharing these things.

Like pygame bugs lol. like we should just be aware of them ahead of time, or, the ai should. Any bug that has a workaround isn't really a bug anymore.
>>
>>109665107
>You don't ask codex for a page.
at that point then why even use codex at all? a 2000 line "page" could have just as easily been a 2000 line source file that i can then copy into my project.
>>
>>109665152
cool, I'll think about it.
>>
>>109665058
when it comes to generating UI, Opus will give me what I asked for
Fable will give me things I didn’t ask for but holy shit that’s what I wanted and didn’t know I could just ask for
As far as Fable-or-Opus vs. Grok…
I said something like “add this thing here to my ginormous blacklist” (I have multiple commits in the repository that it could go back and look at to see what that entails)
and so it added two things (by ID) to the blacklist instead of blacklisting entire classes of things like all the other blacklisting commits
after this happened a second time (wanted to be sure it wasn’t a fluke) I asked it to update the blacklist-this-thing-plz skill so it wouldn’t make that fuckup again
and it did
and now it’s fine for adding things to the blacklist
>>
>>109665132
Can I give it ssh access to my computer?
>>
File: capa-pandu.gif (2.68 MB, 330x232)
2.68 MB GIF
>>109664806
It took like 1 hour and needed to compact twice but it actually got a pretty advanced ConPTY feature working, completely local
son of a bitch
this is the first local model that doesn't feel retarded

(token/s is actually pretty good but it constantly invalidates the cache so I think some of my llama settings need more optimization)
>>
Mistral is currently not enabled as a oauth provider to hermes so I can't power hermes with vibe code.
I can only power it with api which I already used up
>>
>>109665158
the noob way to cope with this is to put stuff indiscriminately into your AGENTS.md
this will bloat your AGENTS.md, though, and so you’ll want to have separate “gotchas and tips” Markdown documents linked to from AGENTS.md that your clankers can selectively peek at depending on what they’re doing for any given task
>>
ok lmao... this is funny.

>ask sol high to implement the next task in the plan file
>it does it quite quickly, writes tests too, records evidence.
>uses up about 10% of my 4 hour limit

i got curious though since i just got a claude sub today...

>ask opus 5 high to do the same thing (all changes are already done by sol, you can just run git diff to view it)
>it spends 9 minutes and 10% of the 4 (5?) hour session's budget
>all it says is "yup it's already done"

crazy...
>>
>>109665184
I ssh into work computers all the time with my keys and have clankers do stuff
of course, this is not without its risks
you will probably want auto mode on so there are at least some safeguards against deleting production databases
I use only Fable when doing this kind of thing because I’m a scaredycat
>>
>>109665070
yeah it does seem to have been a bit dumber lately unfortunately
>>
>>109664222
https://pareto-3d.bradthomasbrown.com
I think it should still work unless I fucked it up, I’m having too much fun telling Sol wacky ideas to implement in whatever 3d renderer it cooked up
>vid rel should be Sol Max, Opus High, and Sol Med
>>
>>109665217
hi cube anon
ur website broke
i need to sign in
>>
>>109665214
I thought we were talking about the web chat.
>>
>>109665007
>>109665029
>>109665093
>random pajeet employees literally trust-the-plan-posting
lmao google is in a bad place right now
>>
>>109663475
nice, will you keep it a higher value
I do intend to at least have luna at 500k
>>
>>109665272
Yes, I will probably set it to whatever it was at Sol launch first, so 370k or so and then try higher values.
>>
>>109665227
damnit
>CLAAAAANKEEERR
>>
>>109665070
Sol definitely feels nerfed lately, not liking it
>>
>>109665230
Oh. I wasn’t.
>>
>>109665070
>>109665216
>>109665371
It's called a honeymoon phase
>>
>>109665383
Might help to follow the reply chain.
>>
>>109665388
I have skill issues
>>
>>109665385
Its called prepping the Astra bvll
>>
>>109665227
>>109665306
I have flung
>vibecoding agent pair revision 2
at the problem
>sign in is only needed for inference assisted features, not the whole ass page you nitwit
let’s see how it goes
>>
>>109665385
My theory on why so many people (and myself) think this is just that anyone gets better at using it and it’s really easy to outgrow a model
I don’t think anyone’s talked about what it means for every human to suddenly get really fucking good at ordering bots to do things (and simultaneously forget how to do things themselves, people focus on this last part)
Damn near every vibecoding token usage graph I’ve seen has been exponential
>>
>>109665401
not trying to be a snailcat here but how about a good ole' git revert HEAD?
>>
on one side
>the model got worse
on the other side
>>109665412
>>
>>109665385
>>109665412
I'm not outgrowing anything.
It's still fine in Codex.
They just very obviously nerfed the Webchat because they somehow decided to give every subscriber even $20 poorfag ones "unlimited" "Sol" in the webchats. Obviously they can't support that so they are either serving it with much less reasoning or a quantized version (I just think less reasoning, not quantized)
>>
>>109665415
not sure what you mean
there has probably been several hundred or a thousand commits since the page was accidentally locked behind sign in
every single piece of everything has changed
the vibecoding pair is hard at work
>>
>>109665175
I like the idea of a backlist, but since I'm vibecoding, if I forget, it probably didn't actually matter.
>>
File: file.png (320 KB, 1050x1187)
320 KB PNG
desktopcommander and other webchat abusers about to go through withdrawals
>>
>>109665385
nah, nowadays it doesn't even think before replying when it used to think for minutes before
they have been caught messing with the juice values and more and more people are realizing how much of a better deal the unmetered web interface is vs codex

>>109665412
the other day i watched a podcast with employees from some generic inference provider like the ones you see in openrouter and he was bragging about how they quantmaxx the models to compete in speed with the other providers and then try to limit the damage by finetuning
the original providers probably do the exact same thing
>>
>>109665227
>>109665401
>>109665415
>>109665475
The vibecoding agent pair has already fixed the issue it seems.
God I love using that. Feels like a fucking cheat code.
You should be able to access the cube site without needing to sign in, I hope, at least it works for me in a new private browser tab
>>
>>109665486
biggest codex nerf is to tibo’s profile pic
>>
>>109665415
there’s also “stuff it in a stash”
>>
>>109665486
>>109665499
his aura loss is actively making sol worse
>>
>>109665478
not a backlog, a blacklist (for stuff I need to look at a few times, but not a gazillion times)
for a backlog, if you use GitHub you can just use GitHub Issues
and if you want to get even fancier you can use GitHub Projects and set up a kanban board
>>
>>109665499
sol got the ick :/
>>
Anywhere to go where people just want to advance the bots and the math and dont care about anything else
>>
>>109665486
dc doesn't break the tos, mcp is a feature openai added themselves
adding a /goal mode to the web interface on the other hand may be breaking the tos because making automated requests through the web ui is not allowed.
hopefully they don't remove the mcp feature altogether. if they do i'm not sure where to go next. all the services are doing shitty stuff. might try zhipu's coding plan.
>>
Just found out you can use Mistral vibe as a subagent for hermes.
That's kind of sad which means I'm still subservient to codex when glm5.2 runs out.
>>
>>109665518
Aha, what kinds of things do you blacklist?
>>
File: IMG_1565.jpg (124 KB, 1179x1397)
124 KB JPG
oh, and some previous anon, two or three threads ago, you were right about Luna costs, apparently they were slashed by EIGHTY PERCENT or some shit, so that’s updated too
>>
>>109665621
I’ve got an old thing and a new thing
I have a pipeline set up to make sure any changes between the old thing and the new thing are subject to extreme vetting
eventually vetting one kind of thing gets old because yes, after the 233rd time I’m sure I want the change, and I want to blacklist the “hey, is this change OK?” prompt
>>
Can Fable generate Quake I maps based on Mermaid flowcharts? (view them in TrenchBroom (a new release came out very recently!)
>>
>>109665650
you could try
it would probably take a lot of work
you’d need to co-write a lot of skills with Claude
there’s a Quake megathread in >>>/vg/ iirc and it has lots of nerds in there
>>
My gpt always has 100% usage remaining regardless of how much I actually use is that weird?
>>
LETS GOOOOOO
LETS FUCKING GO
I STARTED A /goal resume AT 6% WEEKLY LEFT
CHECKED BACK AN HOUR LATER LIKE "HMMM THIS HAS LASTED A WHILE" THEN I /status AND IT SAYS 94% HNNNNNGH
>>
>>109665499
>biggest codex nerf is to tibo’s profile pic
unironically proof that he's single. no female partner would let him make a profile picture with his hair styled like that
>>
>>109665717
Are you using codex at all?

>>109665766
Single here, what's wrong with it?
>>
>>109665826
>Single here, what's wrong with it?
there's nothing "wrong" with it, unless you care about women's opinions more than your own, but it's immediately obvious that this guy is single just from his hair
>>
>>109665751
Are you happy with the results so far?
>>
>>109665842
Yes. GPT 5.6 Sol is actually working where Kimi K3 sucked. It is my go-to state of the art model for when I need maximum intelligence

However, I am annoyed by the cybersecurity guards. It triggered when I was just doing some bluetooth hardware stuff, nothing malicious. And I needed to babysit it with like 10-20 /goal resume before it got past the bluetooth part and actually didnt interrupt anymore

For everything else, I'm just using GLM 5.3 flash with opencode go. It's better than Luna, and unless I'm doing like serious niche hardware stuff, I will never use any other model again. It really is the frontier.
>>
File: IMG_1567.jpg (143 KB, 1179x1415)
143 KB JPG
added the rest of the DeepSWE providers, even though the API/sub thing throws off prices a lot
grok doesn’t seem too terrible
>>
>>109665486
>literally quote tweet tibo
>"lowkey"
WHO ARE THESE FUCKING MORONS
>>
I think they made luna max consume usage faster
>>
>• I found the exact review thread. It used gpt-5.6-sol at low reasoning—not a cheap/lower-tier model.
> The usage trace is also unusually large in raw token accounting, so I’m separating cached context from
> newly processed tokens and checking the actual weekly-limit movement.
FUCK YOU CODEX
>>
File: thefrontier.jpg (59 KB, 2318x116)
59 KB JPG
not even 300k context deep and this piece of shit starts slinging out error prone replace scripts for non trivial blocks of code
>>
>>109666020
Isn't that stuff normally in the system prompt?
>>
>>109666027
opus 5 absolutely loves to use python replace scripts to do file edits and not even the CLAUDE.md anti pattern can stop it now apparently. Sometimes the classifier intercepts it and it just spits out a script that does nothing but it definitely wastes tokens under the hood

this shit is driving me up the fucking wall
>>
Why is the bipolarity of genius and retard so strong with AI?
It's baffling.
>>
>>109666067
Trained on the work of programmers
>>
>>109666050
what is the alternative method you told it to use?
>>
> The agent simply failed to follow the AGENTS.md instruction and
FUCK YOU
>>
> /clear
If you think about it, this is like murder.
>>
>>109666135
context != memories

llms dont learn like humans, they are frozen weights
>>
>>109666135
Your agent has probably birthed and killed thousands of subagents. It's only natural that you should kill him.
>>
>>109665826
Using codex at all? Nope never had. Found a use case for it so will try it
>>
File: img.jpg (35 KB, 554x554)
35 KB JPG
>>109666135
>If you think about it, this is like murder.
>>
>>109666148
The usage is for codex only, it even says on the web interface above the bar.
Web chat usage isn't shown anywhere as far as I know. I'd expect that they give a usage that "should" be enough for power users who chat all the time, and just block or ban you if it's exceeded?
>>
>>109666067
I dont even know what you plebians are working with if youre not logging logic errors and preventing them from recurring. Ai is completely chock full of those
>>
File: 1787264654793915.gif (2.6 MB, 498x278)
2.6 MB GIF
>Scared to share my ideas least someone vibecodes it better and faster than I do
>Scared to even share my ideas with the cloud clankers least Dario, Sam, etc. steal them
>Local is kind of ass or expensive
AI has made me even more paranoid than I used to be. This cant be healthy
>>
>be annoyed that codex has multiple terminal bugs
>find out it uses some "rewrite everything in rust" garbage lib instead of termcap
Thanks rust.
>>
>>109666174
the trillionaires dont care about your ideas bro...
>>
>>109666067
A celebrated author's work is placed on the same level as some leftist redditard's rant about trans iguanas
>>
>>109666174
same altough i dont care that much but i havent used /ultrareview because it uploads everything to the cloud and it was also meh from what i heard
>>
>>109666167
I have never ran into that. I honestly didnt even know codex existed/was something different until you mentioned it. It's mostly of low use to me based on workflow but it does have 1 useful feature which is no cross contamination
>>
>>109666174
Just tell us all of your ideas. You think those guys are gonna read em? might as well just spill the beans
>>
>>109666089
use the system write/edit tools
>>
>>109666190
I see, that celebrated author is probably a retard who can't into logic.
>>
https://youtu.be/QAx6Qn3W23Y
We've been replaced in the one area we should have held our own
Shitposting
>>
>>109666190
trvke
>>
Would it be cheaper to automatically orchestrate a bunch of indians instead of bothering will all this expensive GPU shit?
>>
>hit almost full 5 hour utilizations for 3 chunks since my last reset
>claude says i have hit not just 58% of weekly fable, but 53% of my weekly total
>reset is next wednesday
this feels bad, like they changed something this week, it was not this bad last week...
>>
File: 1770471065615574.jpg (685 KB, 1536x2048)
685 KB JPG
lmao who the fuck even are these people?
>>
whats the best way to maximize CGPT/Codex now that rates are nuked
>>
File: 1763043614378846.jpg (69 KB, 1024x660)
69 KB JPG
>>109666182
no, you dont get it anon. I figured it out. I know how to build the scaffolding for the technofeudal world order the tech bros want. If I get the code just right, the marketing perfect, and reach out to the correct people in a few years I'l be living like a king.. no like an emperor! (Tiberius specifically.) Just as long as they don't steal my ideas first...
>>
what's the cheapest plan if the minimal acceptable intelligence is gpt 5.6 terra?
>>
>>109666150
lol
>>
>>109666277
What is the value of 100x indians exactly
>>
>>109666366
> $20 plan and Luna Max that beats Terra Medium
>>
>>109666366
I think the cheapest plan that includes codex is already the $20 one, which has sol/terra/luna.
>>
>>109666378
Per week? Hopefully less than a dollar?
>>
>>109666402
alternative that's not chatgpt family?
>>
>>109663280
yes
>>
>>109666402
>>109666426
Grok, probably, or DeepSeek
“Terra” doesn’t really narrow it down, Terra at what effort?
>>
>>109666366
deepseekv4flash vision exp, either through api directly or by paying $10 in OpenCode Go. It’s the best and cheapest price-to-value you can find right now.
>>
>>109666455
terra high
>>
>>109666478
I tried dsv4 flash and it's shit, not even close to terra
>>
>>109664321
Just come up with your own benchmark like >>109663486 demonstrated. It's not exactly difficult to find out the strengths and weaknesses of each model if you have a grasp on how LLMs work.
>>
>>109666497
which flash? They’ve got like three versions of it now.
>>
>>109663127
if [ $[$year % 400] -eq 0 ]; then
echo "This is a leap year. February has 29 days."
elif [ $[$year % 4] -eq 0 ]; then
if [ $[$year % 100] -ne 0 ]; then
echo "This is a leap year. February has 29 days."
else
echo "This is not a leap year. February has 28 days."
fi
else
echo "This is not a leap year. February has 28 days."
fi
>>
>>109663486
Fable costs extra. A lot.
>>
File: 1782909975615841.gif (446 KB, 399x517)
446 KB GIF
What's that, ChatGPT offers a monthly rate reset? Da heq? I ain't complaining. They really NEED those early adopters apparently
>>
>>109666489
DeepSeek V4 Flash Max (cheap, slow), Gemini 3.7 Flash (expensive, slow), Grok 4.5 High (expensive, fast)
>>
Did anyone ever switch away from claude/codex and was happy with his new service? Without paying more?
>>
I have 2 $20 codex subscriptions now
should i cancel one and get claude? I never really tried claude before
>>
>>109666600
claude is only worth it for fable
>>
File: 1766579495290402.png (166 KB, 746x674)
166 KB PNG
>>
>>109666512
there's no which flash dumbass. they are all shit compared to terra
>>
>>109666626
Now paste this into a new session and ask it to explain this.
>>
>>109666600
Why not? Compare the two. I also have both (one for work) but currently I tend towards claude.
You won't get Fable for $20 and it's lobotomized anyway.
>>
>>109666288
this has got to be ragebait
>>
>>109666651
i have tried both. terra is mid but also very expensive
>>
>>109666706
What's the problem? We all know Elon Musk literally invented LLMs.
>>
File: 125612421421.png (503 KB, 1010x615)
503 KB PNG
>>109665204
idk how to say this without sounding like a shill but here I go: I have both $100 plans for claude and codex. I bounce between using fable to orchestrate with opus and using sol. claude burns tokens 2x, sometimes 3x as fast, along with the absolute most annoying aspect of claude which is the 5 hour limit. that said, fable just gets shit done clean as fuck. that's my experience (mainly webshit but with somewhat complex frontend stuff going on). I can always burn to my weekly limit with claude, yet I've never done so with codex due to it being harder for whatever reason to use tokens and the constant resets
>>
>>109666288
best part is, it lacks the inventors.
>>
What's your excuse for vibecoding slop instead of using the tokens to cure cancer?
>>
which one is best value for using grok 4.6? cursor pro or opencode go?
>>
>>109665204
I find that claude is rather slow. Codex is delivers snappy replies even at sol-high.
>>
>>109666542
is gemini 3.7 flash actually good?
>>
>>109666758
not that anon, but I'm using codex now, and it is so. freaking. slow. i am making changes on a plan and every change takes an hour or so, it is much slower than it used to be

not doing anything ridiculous, it's a moderately complex plan, but I'm asking for relatively precise changes on one small part at a time.
>>
>>109666626
Is this a horror story for LLMs?
>>
>>109666797
This is actually a really complex topic. How fast it is probably doesn't depend on how fast the API running the actual model responds, but what exactly the AI agent is doing (running slow commands?) or even your upload speed (because it has to upload all your data to their servers).
Still I think claude is slower.
>>
> The lie is less harmful than it looks
Another funny output by the machine.
>>
>>109666626
>>109666652
OK I did it, chatgpt web
>This passage is grammatically polished but semantically unstable. It reads like an intentionally exaggerated piece of AI-style prose: lots of confident transitions, technical vocabulary, and rhetorical emphasis, but very little that resolves into a concrete, coherent claim.
>That density looks deliberate—almost like a caricature of LLM prose.
It output 454 words to describe the image, while the image has only 311 words. (Counts according to itself.)
>>
>>109666835
Checking way too many things, but I have it on Ultra so it's kind of my fault. It's a 20k word plan maybe in an attempt to set clear direction before touching code again. But it really feels like it is much slower than it used to, and since I feel it is to push users to enable /fast, I'm suffering with the slowness instead of giving in.
>>
File: 1783938840220774.png (57 KB, 753x344)
57 KB PNG
>>109666652
>>109666863
>>
>>109666904
Amazing. Did Focault write this?
>>
File: 1770839382133674.png (1.6 MB, 1120x1776)
1.6 MB PNG
rust is such a shit language holy fk how does anyone use this? literally impossible to understand what's going on unless you're some autistic troon
>>
>>109666933
>>
>>109666933
You should have it under control then?
>>
>>109666933
Maybe programming is not for you. But don't worry, your AI agent will do it for you.
>>
File: 1773219393376544.png (120 KB, 1174x477)
120 KB PNG
>>109666919
wow he's so gay x3
>>
>>109666933
I find it exactly as easy as any other language
>>
File: 1771139636725923.png (39 KB, 1261x221)
39 KB PNG
>>109666933
maybe you should try Mega Rust
>>
>>109666979
>>109666626

The agent's POV: https://www.youtube.com/watch?v=849q2h1OkKs
>>
>>109666933
wait, it's hard to understand tho? First question, are you white?
>>
the trannies came out of the woodwork for that one
>>
did codex just updated their compaction? my compaction reduced from 100k to 40k
>>
>>109666973
He wrote the human version of that claude screenshot. Indecipherable philosophy babble.
>>
I have totally forgotten how to write code. And I wouldn’t consider myself a beginner, I used to solo a 200 kloc project, and was fairly comfortable across a bunch of languages
I think I started using AI about a year ago. Just this time, thought. It can speed up my work. First I started with autocomplete only, with github copilot. It was nice. Then I decided to give agents a try. They weren’t that great and wrote a lot of ugly code but that speed won me over. Still I wasn’t using them heavily.
But since GPT 5.4 I don’t think I wrote a single line of code myself. It’s been something like 8 months now.
I just stare at code like it’s Chinese now
>>
>>109666992
Must be awful to think about trannies at all times. Suicide might improve your mental state.
>>
File: 1772336944580844.jpg (49 KB, 640x606)
49 KB JPG
>>109667010
>>
File: 1759324822573715.mp4 (952 KB, 480x852)
952 KB
952 KB MP4
every other language:
hashSet = Set<Int>();

rust:
let mut hash_set: HashSet<i64> = std::collections::HashSet::<i64>::new();

like how are you supposed to get anything done without a clanker?
>>
>>109667017
Most rust programmers pre ai came from cpp and they just assumed that’s how programming was supposed to look like in general
>>
>>109664432
Deepseek truly the goat
>>
>>109667031
C++ basically ruined Rust. It was over before it began.
I'm sticking to C. It may be more verbose, but clankers are good with that.
>>
>>109667017
if you have use at the top then it's just
let mut hash_set = HashSet::new();
>>
Do you consider LLMs intelligent?
Considering they often do a better job than my garbage meat soup, I changed my opinion to "yes".
>>
>>109667040
sure, i also consider a pocket calculator intelligent for being better at division than me
>>
I'm slopcoding an AI to play Yugioh (ygopro) so you can hand it any deck and play relatively well. It sort of works, better than unscripted windbot I guess. Which is not amazing but it's going places. I'm using the scripted Windbot to teach the untrained checkpoint and there's text embeddings for the cards so it should store in the weights properties about the cards thanks to keywords. I'm kinda running out of steam tbdesu.
>>
>>109667040
What even is intelligence? If it’s just pattern matching then I would say they’re ASI tier already
If you think judgement and wisdom are part of intelligence then they’re still more intelligent than most people
If you think intelligence requires some sort of almost metaphysical degree of self cognition I don’t know
>>
>>109667042
I consider an llm more intelligent than you.
>>
>>109667047
do you even need text embeddings if you literally have a ygo engine right there that can predict the outcome of playing a card?

just slap on ReBeL or some other card-playing/poker algo and call it good
>>
File: ultra sub btw.png (15 KB, 820x209)
15 KB PNG
>>109666791
on the benchodes that matter it's as good as fable for like 1/50th the cost. computer use, world knowledge, usability, multimodal perks etc are all best in class
>>
>>109667066
No model is as good as fable except sol and only for very specific things.
>>
>>109667042
>>109667051
LLMs are doing a little more than division or pure pattern matching. They can reason about things. Or, you know, pretend to reason and still arrive at the correct result.
>>
>>109667078
idk what your issue is. i said i find a calculator intelligent. yes, obviously i find llms to be intelligent
>>
>>109667066
which harness is this?
>>
>>109667058
Yeah because when new cards are added to the game you only have to update the embedding and it ***should*** know what the new cards do without additional training.
>>
>>109667087
anagram of gay
>>
>>109667092
>it ***should*** know what the new cards do without additional training.

if an llm told you that, you should cancel your subscription
>>
what's with the snailcrabs bro
>>
>>109667083
Nobody seriously thinks calculators are intelligent.
>>
>>109667138
why not? they can crunch numbers way better than humans.

speaking of crunching numbers, guess what matrix multiplication is used for? :^)
>>
>>109667149
>comparing a calculator to a GPU
Just stop. At least say a GPU is intelligent instead.
>>
https://x.com/yadavji_codes/status/2092953203956682943
>>
any downside to nous portal plus $20 plan? it has webui, cli agent, and api endpoint available. has anon tried?
>>
File: 1776571812536379.png (117 KB, 758x341)
117 KB PNG
>>109667166
i dont think i could build a modern gpu even if i dedicated the rest of my life to it, so yeah i'd say they are intelligent
>>
>>109666020
…why would you forbid Python for source edits?
>>
>>109666288
…Joseph Gordon-Levitt?
……Paris Hilton?
>>
>>109667253
chud thinks he is smarter than the ai he outsourced his thinking to lol
>>
File: 1764173421432260.png (1.37 MB, 1620x1643)
1.37 MB PNG
>>
>>109666736
>along with the absolute most annoying aspect of claude which is the 5 hour limit
the $200 plan doubles your Claude (supposedly only 150%s your Fable) but 4xes your 5h limit
the only time I hit my 5h limit on the $200 plan was Fable ultracode (a mistake, but someone had to try it)
>>
What the average opus 5 response reads to me
https://www.youtube.com/watch?v=8_UnANdDqJc
>>
>>109667295
Maybe I'm blind to it but I rarely see any of the main Claudisms that people talk about. The way it speaks makes me want to punch a whole through the screen, but I never recognize any of the screen captures people post or meme about.
>>
Well I have egg on my face. Was certain we were getting fable & opus 5.1 today. Maybe Monday
>>
File: Dark brown.png (6 KB, 1440x810)
6 KB PNG
>>109667009
>I have totally forgotten how to write code. And I wouldn’t consider myself a beginner, I used to solo a 200 kloc project, and was fairly comfortable across a bunch of languages
>>
>>109667073
i'm going up against a sol-max auditor for my math formalization project and 3.7-flash can genuinely satisfy it. high chance terence tao either tuned that sol's sysprompt or it learned from his personal workflow. only problem is you need a really good checker pipeline because gemini gets overeager.
>>109667087
anti gay
>>
>>109667078
>They can reason about things
If they could reason then they'd be able to do math
>>
>>109667373
more like leftmost of pic related
>>
>>109667382
>>109667380
>>
> select Fable 5
> swipe all the way to Ultracode
> "Awesome! Before we write the document, let's brainstorm everything we’ve agreed upon. Please dig deep into the ideas and explore every connection between them. We are building something that doesn't exist yet, so we need a breakthrough—we need genius insight on everything, from the app functionality to the game mechanics. Look for elegance and simple, beautiful concepts. This input will define our final document before we start implementation. Take your time, be creative, and have fun! Don't worry about token usage; I have the 20X plan and need to spend them."

Heh... nothing personal, /vcg/ anons... but it looks like I'm going to make it sooner than you
>>
>>109667438
enjoy the new feature where you can, if you so choose, have it autoresume when you hit your 5h limit
might want to set that up before you hit it
>>
I sure hope no indians will steal my idea, but I'm going to post it anyway. I vibecoded a converter from mermaid flowcharts to quake level.

Needless to say, this is the next Microsoft.
>>
levels*
>>
Why the fuck do both claude and codex love to use sed for reading files so much?
>>
>>109667483
reading only some of the file into your context is way more context-efficient than reading the whole file
>>
File: rename before posting.png (114 KB, 363x306)
114 KB PNG
Does sandboxing work?
>>
>>109667490
I just think reading files should be something integral to the harness, not building weird shell commands.

>>109667498
lol no
>>
>>109667483
i saw opus 5 writing python script to edit .md file
>>
>>109667498
I love this podcast and respect what that person did in the field with back then with EleutherAI, and they probably did more things since, but it's I don't get why it is so prevalent. I can't help it, it's triggering fight or flight. Voice is fine though, so going audio only I guess.
>>
>>109667513
And that's why we have ai. we don't have to deal with the shit.

>me and the polycule
yeah ok so I'm gonna have the ai remove your sick shit you retard, then, I'll still get the information.
>>
>>109667500
sure but how would it be better than what’s on every (fake) UNIX computer ever
>>
why don't companies make new models with 1M context anymore?
>>
holy fuck claude is finally usable. they added a low priority mode to let me go post the stupid fucking 4 hour limit. thank you jesus
>>
>>109667576
Uh? Aren't most new models 1M context, just sometimes served differently with subscriptions for cost related reasons? What model are you thinking about?
>>
>>109667600
gpt $20 plan isn't
>>
>>109667591
>they added a low priority mode
>it only kicks in after you hit the session limit, letting you continue with spare server capacity
I wish it could work all the time so I could just start Fable Ultra then checking back a day later.
>>
>>109667611
The model has 1M context. It's one of those, "they set it lower for cost related reasons".
See https://x.com/thsottiaux/status/2089143488696705077
>>
>>109667618
Actually this might be the better xeet https://x.com/thsottiaux/status/2089082893804896524
>>
>>109667600
Grok is only 300something
>>109667614
This sounds neat but the longer a thing goes on the more likely you are to have something get interrupted by internet trouble
on the other hand not getting outright blocked by a 5h limit sounds great
>>
>>109667630
Grok 4.6 is 500k, on my plan anyway. But I think it compacts before 400k?
>>
>>109667630
>the more likely you are to have something get interrupted by internet trouble
I was about to write my internet is pretty stable, but you might have meant on Anthropic's side, in that case, oh.
>>
>>109667635
Mine’s rock-solid too but shit can happen either by you, by them, or anywhere in between
>>
>>109667641
With most providers, if you send files for batch inference, it's not like you need to be in contact with the server 100% of the time, you just poll and get the result at the end. I pictured it something like that I guess.
>>
>>109667652
I watch my internet connection when Claude does stuff and it’s a slow trickle back and forth which would be fine on DSL in the early 2000s
most of what I have Claude do is fairly chatty, though, and it’s rare that the text goes from orange to red (which seems to signal some kind of “no text going back and forth”, which happens when the internet is screwed or Claude is busy thinking to itself and not needing to contact my computer for anything)
>>
claude is so fucking moralizing
that fag really pisses me off
i find it best to totally just ignore him whenever he brings that shit up and move on like he didn't even say anything
god i hate him
>>
Wait you guys can use fable without getting banned from fable? What are you guys trying to make and why does it hate me
>>
I would like to pay $20 to $30 per month for cli agent coding and api endpoint usage. best value plan?
>>
>>109667756
a font
>>
>>109667774
probably OpenAI codex, especially if you get lucky with Tibo hitting the reset button
t. Claude main
>>
>>109667756
you need to word your requests like a good goy and not include any wrongthink in your prompt
>>
I don't think I would hire anyone in /twg/. Could you imagine trying to get them to do what you want?
>>
>>109667801
what's Tibo?
>>
>>109667892
https://bfy.tw/UUop
>>
>>109667801
does it support calling api?
>>
>>109667892
he’s the OpenAI guy who hits the reset button and resets everyone’s usage to 0 every so often
Neither Anthropic (Claude) nor X (Grok) have RNG blessings like this
>>
>>109667253
replace costs double the fucking tokens and if it gets one fucking character wrong on the match then it SILENTLY FAILS

it does this error prone and expensive retardation for large blocks of code. one liners i normally wouldnt care whatever

>>109667268
fucking retard
>>
anyone tried the singularity service? they claim to offer non quantized flash v4 at slightly lower prices than official DS off peak api prices
https://www.singularityapi.dev/pricing

But i wonder if it's really non quantized, and sadly the prices that matters the most, cache hit, is same as DS
>>
File: more_usaid_trash.jpg (42 KB, 1164x533)
42 KB JPG
>>109663127
I just caught out a bunch of leftoid karens trying to lure AI developers to DC.

https://paste.rs/PuwH3.md - receipts
https://www.aihackforfreedom.org/

these pieces of shit are just deep state karens and teachers union deadshits, people tied to USAID, every fed or bureau cunt that 4chan collectively despises is behind this, it's likely whatever agentic capability is developed here would be used domestically in the US by these fuckers, probably in time for november.

I'm not joking. they came into a very exclusive discord full of xooglers working on their own AI projects advertising to some very skilled developers that they should participate, they're looking to do something big and I witnessed them trying to recruit.

they implied the tools would be used by people "aiming to topple a regime"
I shit you not. this is real. as soon as "usaid" and "no kings" came up as links I knew what this was.
>>
File: IMG_6478.jpg (71 KB, 1179x660)
71 KB JPG
>>109668088
the elites REALLY don’t want you to know this but you can do the same thing in reverse
>>
>>109668097
Is that the guy that is trying to take on the synagogue?
>>
File: 1761901320116818.webm (3.82 MB, 700x524)
3.82 MB
3.82 MB WEBM
am I a retard? I see people with like 6 claude code instances on their screen at once, like what's the point
>>
>>109668088
Nobody cares.
Go back to
>>>/pol/
>>
>>109668166
>everybody is what tiktok says

everyone stupid is tiktok
>>
>128 new subagents this week
>>
I went through my weekly in <24 hours.
>>
what is interesting is that when you buy credits it's way less than what you get on your plan. Like the $ go 1/10th as far or something.
>>
>>109668118
no, he’s trying to replace it
>>
New thread:

>>109668209
>>109668209
>>109668209



[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.