codex-discussions

[boreal holly]

have you ever changed your AGENTS.md mid conversation?

[lean lark]

Yes I have - if the assistant says there's a contradiction, I'll fix it in one of the files and it processes the new directives. I happen to know this works because there was a GH ticket a few months ago now about Codex using the filesystem watcher specifically to get it to detect such changes and re-read the new directives. So I'm kinda cheating with that knowledge. But that's unrelated to what we're talking about if the AGENTS.md files are not changed.

[frosty zealot]

@boreal holly What frmework did you say you use for hot reloading your swiftui?

InjectionNext?

[boreal holly]

InjectionNext 🪄

[frosty zealot]

Thanks working on something new and going to integrate it

[boreal holly]

Make sure to set -Xlinker -interposer on Debug only. Apple rejects it if those are set on release build

[frosty zealot]

kk

[static sigil]

Hey 👋

Quick request from a long-time GPT user here.

I’ve been around since the early days of ChatGPT, and I actually suggested the “stop generating” button back then — really cool to see it get implemented 👍

Today I’m reaching out with something that could be very useful for Codex.

I’d like the ability to manually trigger context window compaction.

For example, I was reviewing some code Codex generated and, while debugging, I asked it to compact its context. The context was sitting around ~40%, and I wanted to bring it back down to ~20–25% like it usually does automatically — but it didn’t trigger.

It feels like this should be something we can control explicitly. Right now, I end up consuming a lot of tokens just to keep the context alive as it grows toward 100%.

Would it be possible to expose a way for Codex to trigger context compaction on demand?

Thanks!

[frosty zealot]

You're better off opening a feature request on the official github repo

[unique spade]

it s already implemented i think

[static sigil]

I want to trig it manually 😭

[lean lark]

It's the "/compact" command.

[unique spade]

it has been recently impelmented bro

[static sigil]

Thank to reply guys i will look at to !

[unique spade]

1 week ago it was made available in codex app and vscode extension

[frosty zealot]

They knew he was coming

[unique spade]

yes haha

[lean lark]

Run, Duck, Implement, Cover.

[boreal holly]

I've had manual compact for a few weeks now haha

[static sigil]

https://youtu.be/x5lv-56_KHk?si=ZcygoxoOQw7LUw0z

[unique spade]

yea the command itself was from quite some time in cli i think

it was just not exposed in app and ide extension

[boreal holly]

That's pretty wild!

I don't use it very often. Usually if I see like 10% ctx and I'm like "I wanna compact and then send, that way the instructions are fresh"

[static sigil]

Ok looking for in-app !

[boreal holly]

No sense in sending a message, and then it reads 2 files, compacts, and reads the files again lol

[unique spade]

yup it s exactly when it s worth compacting manually

otherwsie it thinks for 10 seconds and gets auto compacted

and if it happens during context gathering it will just have to do it again

[static sigil]

It would be nice to be able to tell it something like:
“From our current context, keep this and that, forget the rest, then we’ll continue after you’ve compacted.”

[lean lark]

I have a solution to that...

I don't have discussions with the assistant in the Codex client window. I have them in separate .md files. Then compaction doesn't hit them and I still have the files for other threads. It's a way of building memory over time.

[static sigil]

Well i want a compact now, not when it reach 100% 😭

"automatically" 😒

[lean lark]

So my in-window commands aren't "how do we do this or what's involved in creating that", they're "continue the discussion in foo.md".

[hard drum]

france spotted

🏳️

[lean lark]

Then we go to "refine the discussion in foo.md to a firm set of documentation and directives for the assistant". Then I save the discussion for those "when did we dedide to do this?" or "why are we doing this?" moments.

[unique spade]

that's too niche case, you have to implement it yourself

it's doable though

[cyan wing]

Who’s ready for tomorrow 👋5️⃣.5️⃣

[unborn thunder]

For now I’m using ChatGPT codex to review codes instead of actually doing work on my GitHubs

[unborn thunder]

What’s tomorrow

[unique spade]

bro use slash command

you type in chat

/compact

[lean lark]

I believe you mean "the cloud/browser" UI for Codex, right?

[cyan wing]

who’s gunna tell him

[unborn thunder]

Don’t tell me Claude is coming up with something better

I’ve got enough already for a week from gpt and Claude

[static sigil]

No way 🫪

[hard drum]

people say gpt 5.5

[cyan wing]

Impossible. OAI releasing new frontier pre trained model

[unborn thunder]

A lil improvement here and there

[hard drum]

only way to find out when users actually get to play w/ it

[unborn thunder]

Claude does the job right thought

One day GPt will be as good

[static sigil]

You saved my life 🫶

[cyan wing]

OK you can believe that
more Codex for me 😇

[lean lark]

I used to use the browser UI for Codex all the time. Then I loaded the VSCode Codex extension and I don't think I've used the browser UI more than once since then. The browser interface is (I believe) the same as one of the interfaces available in the new App. But the VSCode extension allows us to converse with the assistant with the code right there. It's a synchronous model compared to the async model of the Codex cloud interface, which creates a VPS to do the work. With extension, we don't need a VPS, the environment IS our local dev environment. But! (In the opinion of anyone who understands this untrustworthy tech) It must be secure/sandboxed.

[unique spade]

lol, claude never passed my interview job, i fired it after the first day

[static sigil]

From my screenshot i shown, i dropped exacly to 20% context !!! Thats awesome !!! 😍

[unique spade]

glad to help

[hard drum]

when you gonna water the interviewees?

[frosty zealot]

I need you all to mentally prepare yourself for the influx of crazy people tomorrow when 5.5 drops we need to be on our a game. I need you guys to think of the best rage baits overnight. We need to really rule with an iron fist tomorrow you know what I mean.

@kind jay need you on your A game

Jane’s our heavy hitter

[static sigil]

Is it true about 5.5 ?

[lean lark]

Just use image 2.0 and slam peeps with crazy announcement "screenshots".

[static sigil]

Hope it ll kick Mythos's back (to pass through moderate guard)

[lean lark]

Blink - no one really knows about 5.5. There's just a lot of guessing that it might be tomorrow. Thursday is a mostly confirmed drop day, so the guesses seem reasonable.

[frosty zealot]

Very high, likely nothing is a guarantee though

[lean lark]

( And there's also some rumour that Mythos itself is a myth... )

[unique spade]

well sam replied with an emoji that looked like confirmation to someone asking about thursday

[hard drum]

i wanna do the tinfoil hat &6 say that "Mythos" is probably just a fine-tuned 4.6/4.7 specific to that area

[lean lark]

I use a lot of extensions but we're talking about Codex here. I'm not tracking you, sorry.

[lost drum]

Or they might launch it without testing it enough and everyone gets hacked

[unique spade]

bro app / vscode extension / cli are all part of the local family

codex cloud is different, it has 0 contact with your desktop
it can only code via github
it loads the github in it s own sandbox and work from there

[lost drum]

But do they boost his work or make it easier for you to check code? Lmk or You use some skills?

[unique spade]

i never use codex cloud because of that, i used it in the summer when it was the only option available for subscription

[lean lark]

Venus, the app window for working with repositories looks just like the cloud UI. That's all I was saying.

[tawny turret]

We are not getting any new models today, thursday released happened

[unique spade]

in chatgpt? or you mean codex desktop app

cause yes in chatgpt if you click codex it sends you to codex cloud

[lean lark]

I was commenting on the improvement of the experience from cloud UI to extension UI. I dunno if he uses VSCode or Notepad. I don't want to comment on other extensions, this is a Codex forum.

[unique spade]

i mean codex desktop app is very similar with the vs code extension. similar ui

codex cloud is a different beast and it s where you land if you clock codex link from chatgpt app

how to boost codex ?

codex can t be boosted

it s already at max boost

😂

but seriosuly all other vscode extensions are for you the human, codex doesn t have direct access to them

codex you can boost with skills, mcp, hooks etc

not with vscode extensions

[lost drum]

I see I see thank You

Am a beginner

[unique spade]

no worries 🙂

i have like 20 vscode extensions that i installed when i installed vscode

but now i never use any

[lost drum]

Haha

[unique spade]

like 3 or 4 for github

but who cares since i tell codex to open pr's , adress bot reviews there and everything else

and it does it it's own way via terminal not via vscode extensions

it s the equivalent of you clicking buttons 🙂

for me codex is first my computer operator to be honest

i need something done i tell it to do it

[lost drum]

What are you building tho?

Things for yourself?

[unique spade]

current main project is linked in my discord profiel

beside that i'm working on my trading bot, but that's private repo

[hard drum]

🙄

[unique spade]

and these days i'm doing my own UX for codexXchatgpt to get rid of the pain of using 3 vscode windows, 1 codex desktop app and 1 chatgpt app

[dusk thorn]

Totally real screenshot

[unique spade]

lol. is that real haha

[lean lark]

I think you guys are not understand what the Codex extension is.
It's an interface between the VSCode app and the remote server for AI services, and it works with the local file system. We all get that.
But it's a UI component. It's there for human interaction.
We have other extensions for human interaction and for working with our projects. These might include functionality for a linter, code beautifier/prettier, code refactoring, snippets, documentation, remote work, etc.
The statement "I never use any extensions" is kinda nuts. Why use an IDE if you're not extending it?
Whatevah. You do You.

[unique spade]

although what kind of phone is that where the background of claude up extends all over till the very top?

[hard drum]

looks like MIUI (android) to me

[unique spade]

what remote server for ai services?

[lean lark]

You would have enjoyed the separate thread we had here on Monday on this topic.

[lean lark]

workstation to OpenAI via Codex ... duh

[unique spade]

next monday

this is codex vscode extension

it has de codex binary

[potent mason]

Seems legit

[unique spade]

which is the one from the codex github repo

plus some extra UX files

[potent mason]

Send it in the Claude discord server though everyone here is already using Codex

[lean lark]

( There are some people that we meet in life that we simply can't communicate with. I think I've met another. )

[unique spade]

it doesn t connect to any ai services server

it connects to openai servers the same way codex desktop or codex cli do

[lean lark]

( Is it just me or does this guy not only contradict himself every few minutes, but sometimes in one statement after another? )

[unique spade]

it's just you what contradiction did you detect

this is the compiled binary of codex

[lean lark]

it doesn t connect to any ai services server ... it connects to openai servers

[unique spade]

the one that you ahve also in codex desktop app

[hard drum]

the same contradiction that probably happens to be in those "AI detectors", || maybe they found a real one

[unique spade]

and in codex cli

[lost drum]

Use ghidra

[hard drum]

isn't it trying to specify the actual server instead of something outside of that, if any to be found?

either way, that's still a bit of a weird sentence. good catch

[unique spade]

because i don t get what ai services server you mean.

[lean lark]

I'm really sorry venus but I need to back away. We simply don't communicate. I've spent too much time here bantering anyway and for me you're a huge time drain. Please respectfully, don't engage with me. And I won't with you. Thanks. 🙏

TYVM

[unique spade]

oki, we re good, let's let waters chill , we got heated for no reason

[lost drum]

Have you seen results of my bot on #off-topic ?

[obtuse schooner]

is yalls codex

bugged

mines just been like this

[lean lark]

Second report of that today?

[lost drum]

Have you tried closing it via task manageros?

[obtuse schooner]

i have

resested my pc and also reinstalled

[lost drum]

Wow

[lean lark]

App or extension?

[obtuse schooner]

app

[lean lark]

Windows or Mac?

[obtuse schooner]

windows

[lean lark]

I'll go on a limb and suspect this is a Windows-specific implementation issue. No idea about cause but it sure has that smell.

[unique spade]

i ll check

[lean lark]

Maybe a comms issue to servers, or an exception being thrown and just swallowed somewhere. It would be nice if we could get logging data. I know nothing about the abilities of that app to log issues. Might be worth looking for launch switches, or even looking in System Event Log.

[unique spade]

my windows app seems to be working fine

i had it open most of the day

with codex is really good idea to have multiple surfaces you can use

usually the best way to repair one is to tell codex itself in another one that works

lol

i ve done it quite a few times

[lost drum]

Worsr part he cant tell codex to do it

[obtuse schooner]

im using antigravity to repair it

[obtuse schooner]

🥹

[unique spade]

yea that s why you need to have codex app, vscode with codex extension and codex CLI haha

[lean lark]

Good luck on that

[unique spade]

gemini and opus will break it on purpose

[unique spade]

i d rather get codex cli and work with codex from there to debug the app

[obtuse schooner]

ill do that if antigrav doesnt work

[lean lark]

Is the full source available for the Codex Windows UI?

[dusk thorn]

I found on twitter, probably images v2 😂

I always call it twitter i forget its x it’s legit less to type aswell im just dumb

Btw they posted this on the official chatgpt account today

Tomorrow gonna go nuts

[fossil hound]

Will we get an Spud/5.5 update this week, what are the odds?

[boreal holly]

OK dude, the image v2 + Codex, I tried feeding a reference design image to Codex and my god... Frontend is no longer arduous

[lean lark]

If that were Mr Potatohead in the pic I'd think Spud fershure!

[obtuse schooner]

same thing with vs studio 💀

[brisk yoke]

I've been working with multiple vscode codex threads (of different workspaces) at the same time in the same machine for the past month+,

I've noticed this was very costly so I upgraded my subscription to pro recently.

starting this week specifically this started causing hangs on all those threads when I try to use one of them, then the other too while the first one is still thinking - opening a third is not even possible, it just hangs trying to load that thread.
this never happened to me before this week, it worked fast and fine and did not hang at all, the hanging started this week/last weekend specifically.

anybody else suddenly stubmled upon this behavior change?
is this tracked and recognized somewhere?
is there a fix upcoming?

anybody knows of a good workaround other than migrating to a competitor or launching the next vscode codex extension on a vm to force the bug to stop reproducing (mabe if it doesn't know its on the same os it will avoid this hanging issue)

I think I do not have time for openai to solve this so I might need to migrate my vscode codex session away from this so I can continue my work uninterrupted. I've noticed that the codex app has a built in "handoff" button which I never used but I suppose it is exactly for these cases of migrations
however - in vscode codex I've not seen these buttons, how do I still migrate from those threads?
I've noticed vscode codex can see windows app codex threads but not the other way around

[lean lark]

Ahhh. I believe you're either facing a comms issue or maybe a license/account issue.
What country are you in?

[obtuse schooner]

US

[lean lark]

There goes part of that theory. Thinking...

@brisk yoke I have an under-thought theory while I'm thinking on something for @obtuse schooner ...
There was a change not too long ago that broke our ability to have more than one separate thread/window open for Codex. If you recently upgraded, you might now be seeing this. Does that ring a bell?

[brisk yoke]

thank you for your response, no I didn't recently actively upgrade in any manual manner

[boreal holly]

Restart computer, unplug all USB devices

[lean lark]

Dread, now I'm thinking licensing. Maybe your login to OpenAI hasn't been established.

[obtuse schooner]

chatgpt recently disabled my acc bc of my "region" but thats wrong

But it wasn't like this, it just told me to login

[lean lark]

There is a cache that gets saved under the user\name\AppData.codex path or something. People have reported deleting that folder helps many similar issues. I don't recommend that yet, just thinking...

[obtuse schooner]

nvm antigravity fixed it

[lean lark]

If Codex can't login to your ChatGPT account, then you're stuck at the gate of the licensing server - nothing to do with Codex or AI so far.

DO TELL, PLEASE.

[boreal holly]

Did antigravity restart your computer, update drivers, unplug all USBs? That typically fixes most issues

[lean lark]

Yeah, the question is, did antigravity change code or the environment, and what specifically?

(Again, is Codex Windows App fully FOSS?)

[obtuse schooner]

mb what does fully FOSS mean

[boreal holly]

Ah, security when you need it the least.

[lean lark]

Free and Open Source Software, available at GitHub, with source available for Antigravity to operate.

[obtuse schooner]

im not sure

i installed it from the microsoft store

💀

[lean lark]

doesn't matter Dread, but thanks. So...

Yes, Antigravity operated correctly and started to check the environment first rather than code. So we don't care about code now, it found the environment issue. We should remember this for the next person who reports this issue.

Yeah, @slate schooner reported the same symptom today. I wonder if it was the same cause.

[unique spade]

did you get it to work?

[obtuse schooner]

yes

[lean lark]

https://tenor.com/view/nailed-it-gif-6752122830141673309

HA! See? Comms issue.

One of these days I'll look into how to post a GIF inline here. 😭

[unique spade]

it's not, on the plus side though in vscode extension you have the dev option to use a custom binary. but the ux is closed source

[lean lark]

Didn't think so but he said he was gonna get Antigravity on it and I had to wonder...

[frosty zealot]

GIVE ME 5.5

NOW

[unique spade]

well opus can wipe your entire hard drive if it s in a bad day

[frosty zealot]

OR ILL BURN THIS PLACE DOWN

[unique spade]

i don t think tinkering with whatever files it finds would be a stopper for it

lol

[balmy acorn]

anybody else have 5.5 i just got it

[unique spade]

where you got it

[lean lark]

ChatGPT isn't showing it to me. I really hope I haven't been using it during A/B testing. Yesterday or today, the assistant is hella chatty to offer detail on every freakin request. It's annoying!

[frosty zealot]

Can somebody click on a random plugin in Codex and see if they get the 'App Unavailable'

Specifically Vercel or Stripe

[lean lark]

others working?

[frosty zealot]

I just clicked on whatever the first one that showed up and I could 'Add to Codex' for some reason Vercel and Stripe plugins both are greyed out

[potent mason]

It might be them changing some stuff (because of the recent hack), specially around env secrets

[frosty zealot]

Are you able to install the plugin?

[potent mason]

I already have it installed

[frosty zealot]

hmm

[potent mason]

I wanted it to check some logs a while ago

Have it disabled though, and I can enable it

[frosty zealot]

Codex is telling me I'm getting 403'd on a cloudflare challenge apparently

[lean lark]

403!

[potent mason]

21k tokens used out of the box with a hey message, is this high or is it normal?

[lean lark]

Look up to recent note #codex-discussions message

[frosty zealot]

im good on every other feature (if this was to me) and I'm on a Mac M3 Max

[lean lark]

Oh man, I just brainferted there, yeah, nvm. Derp.

[frosty zealot]

And other plugins let me install them, its just Vercel and Stripe for some reason

[lean lark]

I was thinking if you failed an auth somewhere then you might have gotten a general block from CF and maybe that turned into a firewall block (highly unlikely but brain wasn't in gear...)

[frosty zealot]

Not unless my IP is being flagged as suspicious for some reason

but i would thin kanything else hosted by CF would be challenging me

Oh well I'm gonna eat some Oreo's while Codex figures it out

[slate schooner]

I had to delete entire .codex folder. That fixed the issue

[lean lark]

Thanks for the follow-up. Yeah, that's another common solution, unrelated to the firewall.
Good! We now have two viable possible solutions for that specific anomaly.

(That's not an Oreo, son.)

[cedar skiff]

Doesnt feel like we are getting a new model today.

[lean lark]

why should it?

[rustic gate]

hey guys, got that email saying codex on plus tier was going to change - do we know what the general shape of that is going to be? right now i'm able to rock 5hr resetting windows and reset the weekly and am pretty dang happy with it.

[lean lark]

For your (our common) usage pattern I don't think anything changes. Could be wrong.

[rustic gate]

Amazing. I can generally get 5-6 2.5 hour intense sessions per week right now.

[lean lark]

Gonna go look for that email, BRB

Last email I got from company itself seems to be on March 31. 🙁

[cedar skiff]

Should it feel like it?

There is a lot of chatter around model release and there is a little excitement and anticipation surrounding it. I'm not sure how to answer if it should feel like it, i can say it doesn't feel like it though.

[lean lark]

It doesn't feel like an earthquake is coming to sunny southern california here ... should it?

[cedar skiff]

Not sure why youre so hostile about it

[unique spade]

i think this was the highest level hint that it might get released today

[lean lark]

Gosh, I'm sorry. I'm not being hostile. I'm just poking at ya. If we don't feel something, why should we be asking Why we don't feel it? It's just a funny concept to me.

[cedar skiff]

ok then, i never did this part - why should we be asking Why we don't feel it that was you.

[lean lark]

Really, I'm never hostile ... just a little weird and sometimes very direct.
Or in the immortall words of Douglas Adams ... "Mostly Harmless".

[frosty zealot]

We gotta work on your rage baiting, I have you booked with @kind jay at 3pm tomorrow PST.

[frosty zealot]

And it's normal to remain hopeful for something you're eagerly anticipating

You know what, you're grounded for the rest of the night, no more computer for you

[lean lark]

Wahhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh

I'm a little trigger happy from some of the questions today.
Just got called to dinner. I hope it's pasta. That always calms me down.
And I picked a bad day to quit amfetamines. 🙂

[frosty zealot]

I reminded myself of when I was growing up and my mom would ground me by taking my tower power cord, not realizing the monitor power cable was the exact same, so I would just unplug my second monitor for the night

👿

[lean lark]

AFK / gnite

[frosty zealot]

don't forget your life alert!

[lean lark]

Where's the below me button?

[unique spade]

i m curious about these people complaining about UI/UX

what they expect for codex to one shot what they have in their mind?

isn t UX/UI work supposed to involve a lot of micro-iterations and you keep moving things till they all snap in place? why would it be different with a model that doesn treally "see"?

personally i enjoy doing ux/ui with codex , just need patience and express yourself one by one what you like tweaked

i'd rather do that and own the result than get a shiny template from the start which is not actually my effort

[frail meadow]

I think the main thing is at the present state, if you have a picture or reference photo + potentially reference code, Claude takes less turns than Codex to get what you’re envisioning.

You can do it with both for sure and codex can optimize the performance, but Claude will get you there faster imo in terms of UI / UX that you’re imagining

[unique spade]

ok so it's mostly about vibe coding based on an image sketch etc

codex is not a very good vibe coder then

[frail meadow]

You’re vibe coding either way

[unique spade]

but it s by far the best engineer

[unique spade]

you the human yea

[frail meadow]

Codex will make the visuals performant

[unique spade]

but i was looking at it from the perspective of the model

[cedar skiff]

codex just doesnt make nice looking ui compared claude.

You have to put more work in

[unique spade]

for example if you give a hard logical problem, gpt actually reasons through it, whereas claude vibe solves it

[frail meadow]

Backend codex kills

[unique spade]

yea i see it in a similar way, and personally i prefer to put work in, because that way i'm more in touch with what gets built

[cedar skiff]

Lets just hhope the next iteration doesnt better at both 🙂

[frail meadow]

What UX you got cooking?

[unique spade]

still doing mostly backend stuff, i'm aphantasic though, so i can t imagine things visually if i don t see them

so i really need to iterate

lol

move that button and all that, cause till i don t see it, i have no clue

😂

but i agree, what codex does by default feels like it s ux made for itself

[frail meadow]

Wireframes and visual mockups are still a really good tool

[unique spade]

it has no sense of human ergonomics

just sutffs a lot of info everywhere

gonna play with building this today

[frosty zealot]

I love image 2 lol

Pretty much 1:1 with the real pic pixel art version of my cat currently sleeping in front of the fire place lol

Im surprised there is no Rust plugin

[craggy whale]

Hiya. What channel should I be using to ask questions about UX? I'm not sure what to prompt Claude Code, Claude, Codex, or ChatGPT to fix grid layout issues in my video application, the file picker bit, built in Tauri.

[craggy whale]

What a delicious qcuat!

[unique spade]

guesss here would be a good place since it involves codex

[craggy whale]

Oh, okay. Sure, okay. There's no forum in here, so do I make a thread?

I don't want to spam the main chat.

[unique spade]

hmm, here is mostly some regular people that we keep chatting about codex , you re not spamming

[craggy whale]

Oh, okay, sure. I will go ahead.

[unique spade]

just try to give as much detail as you can of what you re trying to do and what doesn t work, and hopefully people will reply along the day with ideas

[craggy whale]

I have no programming knowledge.

I'm trying to create a file picker in my application I'm building. Just like it is in the files application, the GitHub project, but I don't know what to say to it because what keeps happening is instead of it being 2x3, it's being just one big tile.

https://i.ibb.co/yncXVqjS/Screenshot-111.png
This is the reference image, and if you want the image of what the folder picker in the application looks like, I will also send that now.

Please noti me if you reply. Thank you!

Oh, by the way, this is just a personal project. It's not for anybody else or whatever. It's a GUI, built in Tauri, to organise videos on my PC. It takes inspiration from Diffractor, Files, the GitHub project for the layout for the file picker, and MYLIO photos.

I don't need coding help, I just need to know what I even need to say to the AI. So basically I guess I'm trying to ask what kind of like design language should I be using and where can I go to learn that design language, so I can learn a bit.

Is this descriptive enough at all?

The purple images are the failed attempts at getting the grid correct and the animal crossing image is the reference.

[frosty zealot]

Are you trying to itterate on it to express what you're not content with? Or are you just trying to like big bang it every time @craggy whale

You're also better off asking it to use a framework like Tailwind to accomplish this and the tailwind grid

[unique spade]

yea i think the info is more than enough

[dusk thorn]

5.4 says hi

[frosty zealot]

LOL

[dusk thorn]

you guys excited for tomorrow?

[unique spade]

the issue is this? or the entire box in which that folder icon is

[craggy whale]

Um, I'm actually not really sure. I've just been trying to explain to it exactly what I want as descriptively as possible, and then if it gives me something I don't like, then we keep editing it. But we've been editing for like 20 times now, and we keep cycling through the three main failures.

What is Tailwind?

[frosty zealot]

framework for CSS which CSS is basically what positions the things on the screen https://tailwindcss.com/

[craggy whale]

The issue is the actual grid. I will try and highlight that now. Does this show? This is the reference image, and the grid is meant to look like this, other than the horrible gap on the right. If that can even be removed.

[frosty zealot]

Yes, tailwind actually knows how to layout a grid and respect resizing etc.

[worthy furnace]

Can spud release already

They said this week

[craggy whale]

Ah, okay, I see. So do I need to pay for that? Is it a subscription as well, or is it something else?

[frosty zealot]

Like I said, I would suggest prompting to install Tailwind and instruct it to use Tailwind grid system

No it's totally free

just tell codex to use install/use it

[craggy whale]

Yes, sorry, I wasn't ignoring what you were saying. I was responding to you from the top down rather than bottom up. So that's why there was like a delay in comprehension.

[frosty zealot]

All good 🙂

[craggy whale]

Okay, sure I will.

[dusk thorn]

its coming tomorrow

like 99% sure

[frosty zealot]

It's an open source project https://github.com/tailwindlabs/tailwindcss , its kind of a big deal, probably used in every single web page on the internet at this point (Tauri uses WebKit/WebView for the interface, its just wrapping a website basically in a native app)

[potent mason]

They've been saying that for a week+ now

Honestly as long as its not a dissapointment like Opus 4.7 they can take another week imo

[dusk thorn]

no like, its coming

[craggy whale]

Oh wow, I see. So, I assume the difference is it actually understands UX, so it's like Figma, in a way? Whereas of course Codex, Claude Code, etc. They're only good at implementing code, not really translating a description.

[dusk thorn]

if it doesnt, ill legit put my status some horrid crap

this was yesterday

or tuesday depending on time zone

[craggy whale]

Okay, Tailwind is now installed. Do I continue prompting as I was, and it will just be more smart now, or?

[frosty zealot]

I would say something like 'Please re-create the file picker GUI using Tailwind, and their responsive grid layout' or something like that

[craggy whale]

Okay, sure, I'll try.

[frosty zealot]

It helps to say also like

or like end that prompt with 'Ground yourself against the latest Tailwind documentation related to this implementation before starting' or something like that

Cause knowledge cutoffs on the models can sometimes screw with the implementation if it installs something new and there was changes

[craggy whale]

Oh, I see, okay. Yes, I understand.

I did also, before I came to the server, I did ask the AI itself, how do I prompt you better. But of course, that was a mixed bag of improvement.

[frosty zealot]

It's not that it really 'understands UX' or is like figma, CSS is just what tells items on a webpage where they need to be and at what time etc. Tailwind is all CSS (and other magic) under the hood that handles all the extreme heavy lifting and burden of handling every edge case, like window size, what browser its being rendered on etc. etc. and provides a lot of nice functionality like a grid system so the components in that grid know exactly how to maneuver themselves so they stay nicely centered, can resize the window and they know what to do etc. etc.

I'm sure ChatGPT can explain the nuances of it and why its so effective better than I could

[craggy whale]

Yes, I understand. Thank you again.

[frosty zealot]

np hopefully it helps

[craggy whale]

Yeah — I can see the pattern you’re hitting. It’s not you wording it wrong, it’s Claude solving the wrong layer each time.

You’ve got three failure modes cycling:

Single giant tile → container width constraint
Correct grid but wrong spacing → layout logic issue
Looks correct under force, breaks after → JS overriding CSS

You already proved #3 is the real one.

So instead of asking Claude to “fix layout”, you need to pin it to one exact responsibility: find and remove the code that rewrites the grid after render.

Give it this — this is the one that will break the loop:

This is what ChatGPT said to try and help me fix it before you suggested Tailwind, so I hope this helps.

[pastel sentinel]

Why is codex only availlable for a limited time

Oh I have a year so thats good but still why it has to be limited time

[craggy whale]

I'm probably not going to be any help, but I just get it with my subscription, but I don't know if that is limited. It didn't say anything about it being limited.

[pastel sentinel]

I already have plus

[brisk yoke]

I upgraded to vscode 1.114.0 and it solved my issue smh

[pastel sentinel]

so does it mean I can still use it or nah

[craggy whale]

I assume so?

I would email support instead, if they have an email.

[pastel sentinel]

my friend told me I can claim this what do they do

[frosty zealot]

Whats your like main question

Codex is permanent

its not a limited time

[pastel sentinel]

Oh ok

what are the credits for

[unique spade]

to use codex

[craggy whale]

Side note, it's not related at all, but what do you think of my Discord skin? I made it with AI. I was really pleased with it, it has great hierarchy. Yeah, there's a few issues with it, but it's good enough.

[frosty zealot]

I'm a big fan of glass morphism so A+

[pastel sentinel]

how does it work exactly tho like how do they run out

[dusk thorn]

do you guys wanna see what i just spent part of my rate limits in codex on

[unique spade]

it s cool

[frosty zealot]

Codex is for like an agentic work flow, imagine ChatGPT has the abilitiy to use your computer, for you, but it is expensive to use, as where ChatGPT you just basically ask it a question and it replies

[frosty zealot]

the 'credits' are just basically an arbitrary amount of how much 'text' you can send to it

[unique spade]

https://chatgpt.com/codex/

you go here and install the app

[pastel sentinel]

I already have the app installed

[frosty zealot]

Well skys the limit now

[craggy whale]

Oh wonderful, thank you. Yeah, it's glass morphism, but it's like a more modern spin on it. It's like a blend of multiple forms of glass morphism. By the way, over here in the UK, we say A star instead of A+. Interesting quirk.

To clarify, the emphasis of the skin is purely visual fidelity and just it being easy on the eyes. That's why it mixes, like, I don't know how to explain it entirely, but I would say it's like a mix of dark mode and light mode in one.

[unique spade]

and then it s similar with chatgpt, but you use it on your computer and have it do stuff for you on your computer, like write code for you and other stuff you might want it to do

[dusk thorn]

bros starting codex the day before the massive changes 😭

[craggy whale]

Thank you, dear mantis.

[pastel sentinel]

What changes

[dusk thorn]

thursday rumored to release 5.5

and maybe some codex updates

[frosty zealot]

That probably means nothing to him

lol

[dusk thorn]

rumored super app maybe?

[frosty zealot]

and I dont mean that facetiously

[dusk thorn]

i just want chatgpt built into codex app

[pastel sentinel]

Btw people mainly use codex to make cool apps to post on there resume or something right

[frosty zealot]

The work around is to have Codex use its computer use to interact with ChatGPT app on your computer lol

[pastel sentinel]

Im kinda a noob coder so wondering if I should try that or just use it as a study tool

[dusk thorn]

good ones make shippable products, some use it to make slop

dont use codex for study

use chatgpt

dont waste your limits

[dusk thorn]

lol i lowk use chatgpt to build my promps so i would want other way around

[craggy whale]

Can somebody explain to me what a vibe coder is? I hope that's not me, because A it has a negative connotation, and I don't want people to think that I'm just using AI because I'm lazy. I still debug it and I'm still thinking of all the ideas and stuff. It's just doing the stuff that I would never be able to do. It's like the equivalent of instead of having a human friend, I have an AI quote-unquote friend. Doing all of the work on my vision is like with Mariah Carey. She can't play the piano, but she can tell the piano player how she wants it played.

[pastel sentinel]

Vibe coder from my understanding is just someone who uses ai to code

[dusk thorn]

vibe coder is a term used by people who dont accept change 😂

[craggy whale]

Oh, I thought it was if somebody did it sloppily.

[frosty zealot]

You'd be a vibe coder lol

It's not as bad as it sounds

[dusk thorn]

usually its either used minimally in a good way and mostly in a derogitory way by some

depends on the person really

[frosty zealot]

It just means you're 'vibing' with the agent going back and forth with it

[craggy whale]

I only ever hear it used derogatively, so I see. We need a different term.

[pastel sentinel]

People say it better to learn coding manually because learning how to deal with the annoying probelms turns you into a better coder

[frosty zealot]

That'll go the way of the dodo bird soon, mark my words, I think it will become more favourable to have people that know how to efficiently prompt, and use an Agentic harness to its fullest extent

[dusk thorn]

if you know the basics of a language and debugging and then use ai to help you build you stuff with it so you can prompt it right and know what it does is like the best combo

[craggy whale]

Yes, I understand. So I guess it's kind of like how me being a traditional artist, I cannot draw digitally. So it's like analogue being translated to digital, I guess.

[dusk thorn]

yeah prompting is for sure a big thing and requires some skill vs what some people think who ask ai when their milk will expire think prompting is

[pastel sentinel]

I do think I should had tried coding manually more but I will try to ask it to explain it to me so understand

[craggy whale]

Oh, I agree. There is an intelligence to it.

Not that I'm very good at prompting AI, but I'd say I'm better than the average user.

I would call myself an intentioned user.

[pastel sentinel]

So I should only use it to make an app or something important cause if I ask to many prompts I lose access

in order to use you need codex points

[dusk thorn]

basically you got 100 bucks right?

[pastel sentinel]

yeah

[frosty zealot]

You can always subscribe, or buy more credits

[dusk thorn]

so it uses rate limits, its weekly and hourly if you subscribe but i assume with 100 bucks your either using api pricing or the amount of a pro subscription 5x spanned out

[craggy whale]

Uh, new failure mode. Now it's two times one. How do I get it to three times two? Why does it keep failing?

[plush harbor]

why? You can talk to codex just fine, and its boring and personality free but that's kind of what you want

[pastel sentinel]

do I still have a 100 credits

[frosty zealot]

Tell it you need way less empty space/whitespace on the folders, way too much margin and padding

[dusk thorn]

chatgpt is kinda like my personal chatbot anywhere i go so it knows me better than codex, and knows my current projects and can help me build prompts better as codex likes direct stuff usually and i dont need to use my rate limits for it

[plush harbor]

I use codex just for code, chatgpt for random and project chat, and now I have the api in so I've completely sold my soul today

[pastel sentinel]

I shouldnt had asked it stuff I could had asked on gpt

[frosty zealot]

@pastel sentinel

[craggy whale]

May not be viable for you, but what I do is I use the actual agent and then the talking or whatever it's called, so like Claude, ChatGPT, for the- how do I talk to Claude Code? How do I talk to Codex properly in all of those bits?

[plush harbor]

chatgpt gave me the money flying away emoji, codex just told me where to put my api key without comment. they are not hte same

[worthy furnace]

How can I use gpt-5.4-cyber in the CLI? Or is it only a security scan feature on the Cloud Codex?

[craggy whale]

Okay, sure, I'll try.

[dusk thorn]

maybe try asking it the prompt it suggested of how much 2,500 credits of codex usage you can get. i assume the 100 bucks got turned into credits for the service

[dusk thorn]

good idea

[worthy furnace]

Was hoping today, very impatient 🥲

[rapid crater]

is codex down

it has been loading model for hours

[frosty zealot]

https://chatgpt.com/codex/cloud/settings/usage

Scroll down to credit usage

if it says 0 you dont have any

if ti says 2500, then you have 2500

[frosty zealot]

There you go

Look at the other picture I sent, depending on what model you use its equivalent to X amount of credits

[frosty zealot]

You're also on 'Plus' which has a weekly/5hr usage limit, so you will consume that, before you dig into your credits just fyi, thats what those two green bars are, so as you use codex it will consume that first it auto refills each week

[dusk thorn]

i got claude for security and some legal things and then codex as main bot and got a AI_RULES.md for my project so they both know to read that first and it has rules for how they should work together and then at the bottom a change log each agent adds with the what how and why and what model did it. only have it read latest entry of course or i would be token dumping and simplified rules

[craggy whale]

Hope it helps!

[dusk thorn]

wait btw how did you get the student credits

[frosty zealot]

Gee I wonder

😛

[dusk thorn]

i just want url bro 😭

[dusk thorn]

palantir 💀

[worthy furnace]

?

[worthy furnace]

Holy tabs

[dusk thorn]

OH COME ON UNI ONLY

BOOOO

[pastel sentinel]

He said he was tryna make a "good palentir" to run his business make it efficent like for dashboards or smt

[craggy whale]

You probably already know this, but you're being scammed.

[pastel sentinel]

And sent this link https://www.instagram.com/p/DXJdLnFkkOm/?img_index=2&igsh=NjloamZvc2sweDZj

[dusk thorn]

does he realize that palantir is a cyber defense company used by the us government for surveillance and military use what kinda company does bro have

[pastel sentinel]

Yeah every thing he’s been saying been sounding fishy which sucks cause I was friends with him in highschool

[frosty zealot]

He's in full dox mode

[dusk thorn]

oh my god hes such larp 😂

[craggy whale]

Trust your gut, trust your intuition.

[pastel sentinel]

content marketing

[dusk thorn]

yeah use codex for yourself you gonna find that it will push you ahead many ways

[craggy whale]

9 times out of 10 if it feels off it often is, and if it continues to feel off and increasingly gets worse, it definitely is.

[pastel sentinel]

https://www.instagram.com/reel/DDyVTFyMVuW/?igsh=MWRuYXYzd2NycWhyag== he also posts on insta

[dusk thorn]

it.... its.... beautiful

[pastel sentinel]

Also transfering him credits according to gpt could risk flagging my account

[dusk thorn]

yeah that would be bad

slight issue yk

you should try building something with it like a personal dashboard thats local

good starter project

[pastel sentinel]

how would I make that

[dusk thorn]

make the model 5.4 medium, and try this prompt in codex:

Build me a clean personal dashboard website with a dark theme.

Include:

• welcome message
• date/time
• to-do list
• notes
• weather
• upcoming events
• goals

Use Next.js, TypeScript, and Tailwind CSS.
Use sample data.
Make it modern and polished.
First give me a short plan, then build it, then tell me how to run it.

[pastel sentinel]

Tbh im a noob coder I know some basic stuff but ive mostly been vibe coding but I end up studying for the tests and im taking
Object-Oriented Programming for Business rn. which is teaching me java

[dusk thorn]

nice

[pastel sentinel]

i'll keep this in my notes

GPT suggested this what do yall think:

1. SQL Assistant (BEST for you)

Why: Directly helps BCIS 4660 + interviews

What it does:
You type:
“total sales by product category”
It outputs:
SQL query
short explanation
Why this matters:
Codex is designed to generate full features from descriptions
This forces you to actually understand joins, GROUP BY, etc.

👉 This is the highest ROI thing you can build.

🔥 2. Java Assignment Helper (easy win)

Why: You’re not confident in Java

What it does:
Input: assignment instructions
Output:
pseudocode
simple Java code
explanation

Codex can:

write code
explain it
fix bugs

👉 This directly improves your grades.

🔥 3. Study + Quiz Generator

Why: You’re behind in multiple classes

What it does:
Input: lecture topic
Output:
key concepts
practice questions
mini quiz

👉 Turns passive studying into active learning.

🔥 4. Resume Project (this is what actually matters)

“Student Data Assistant”

What it does:
Ask:
“top 5 customers by revenue”
Output:
SQL query
explanation
maybe sample output
Why this matters:
Codex is used in real companies to automate coding workflows
This is something you can actually put on a resume

[dusk thorn]

i thought i was the only one who had a overly hyped gpt

i was wrong

[craggy whale]

Based on my experience, do it one step at a time, although of course that can be hard if you have message limits, etc, but you make less problems that way, and also it's really important that it only edits the exact thing you need it to edit.

[dusk thorn]

#2 and #3 are good, you could for sure implement both into one thing

[craggy whale]

Otherwise you're just wasting tokens on it streaming the whole code, etc.

@frosty zealot It's me again. If you don't want me to noti you, say, that's fine, I just didn't know. This is just in case; it may not bother you. So, uh, tell me either way, but. We're still having issues with the grid. This isn't the first time I've had issues with a grid. Every single time I try and create a grid, there's an issue.

[dusk thorn]

who trained gpt 5.4 on making joke images of anthropic 😭

[ivory zodiac]

https://x.com/LLMJunky/status/2047176477200826584?s=20

[dusk thorn]

very cool will have to look at!

[ivory zodiac]

thx. add your skills pls

[frosty zealot]

It's fine but I may not reply soon as I'm just aboutto head to bed lol, I find if im fighting to get it to do something, it's usually like an architectural issue, you have to tell it to think outside the box etc. follow patterns up the chain, follow the whole rendering tree, tell it exactly what it's not doing, compared to what you expect, more descriptive the better, but like I said tell it to work its way up the tree, and I find using the keyword 'idiomatic' helps a lot like 'Ensure we recreated the view and all parent views with 'idiomatic' tailwind, grounded to the current documentation

[craggy whale]

Oh, that's absolutely fine. I didn't mean, like, rushing you to reply. Thanks. And okay, yes, I will try that. I've never actually heard of that word. I will see what that word means, and I will try.

[frosty zealot]

Idiomatic Code: In programming, this means writing code that conforms to the preferred style of a language (e.g., Pythonic code).

Idiomatic or 'Industry standard'

[dusk thorn]

did codex do the frontend?

[frosty zealot]

I use them interchangeably

[ivory zodiac]

yes

[dusk thorn]

wow nice

i tried using gpt images v2 to generate website concept ideas from a scratch one and when i gave codex the one i liked it surprisingly worked nice

[frosty zealot]

Did you use https://getdesign.md/lamborghini/design-md for the design?

Looks similar

[pastel sentinel]

Btw which LLM do you guys think is the best cause ppl been saying gpt is falling off and that claude is better

[dusk thorn]

claude is good if your rich

20 dollar plan is a scam i have it

[craggy whale]

It's probably against their terms of service, but I have three accounts.

[dusk thorn]

codex with gpt 5.5 coming out tomorrow will be the best model for public use i assume

[dusk thorn]

knowing their recent requiring random id checks i wouldnt be suprised if it ended soon

[craggy whale]

So I get as much as I need so close to max without actually paying max prices and not freaking out about having leftover usage that I'm paying for that I'm not using.

[craggy whale]

You need an ID now?

[dusk thorn]

some on x said they had to do a id check on it

[craggy whale]

For Claude?

I've been meaning to get a fake ID, but I don't really know where to get one. All the Xbox gamers have a fake ID. I've seen it in YouTube comments. It's hilarious.

[dusk thorn]

its for new subs it seems

[craggy whale]

Oh, okay. I've had it since September.

[dusk thorn]

you just gotta be 18+ to use it

[craggy whale]

I am, I'm 26, but I don't want to have identity theft, etc.

[pastel sentinel]

I hate that so many apps are requiring face scans and id verification now

Roblox and other stuff

[craggy whale]

Hopefully we will overthrow technocracy soon.

[frosty zealot]

@craggy whale I have to head to bed, don't sweat on it too much if you can't get it figured out tonight, tomorrow GPT-5.5 is supposed to be releasing and from what's rumored it's supposed to be a very large leap, it will likely fix any issues you have today that GPT-5.4 can't seem to figure out, and hopefully be much better at front-end design

[dusk thorn]

yeah fr

i actually use codex to help code my roblox game im making rn

sls

[pastel sentinel]

nice I heard ppl are making tons of money from roblox games and some are made with ai

[craggy whale]

Oh sure, sleep well, and thank you. That's interesting, I didn't know that. And yes, hopefully so. I just need to find out what keeps going wrong with grids. Every single time I try and create a grid.

[dusk thorn]

helped me re write some logic scripts to make it smoother, exactly on time stage and booster seperations, and even some correct physics

[dusk thorn]

yeah

it helps alot with scripting

i just use a mcp server connector from a studio plugin so it can access my entire workspace inside studio

gtg to bed i got school 7:30 and its 12 already crap cya guys

tmrw will be good

[pastel sentinel]

See ya

[brisk yoke]

we've all been there, time to upgrade your subscription xD

[pastel sentinel]

Man I hate that everything is a subscription nowdays

Video games , tv and movies, ai, etc

[frail meadow]

You can always raw dog it with the API

[pastel sentinel]

how so

[frail meadow]

Not really recommended since the subscription subsidizes you but follow these instructions: https://developers.openai.com/codex/quickstart

You can also buy credits but that’s the same price as the API

[carmine copper]

Only 20 times more expensive but at least no sub.

[fleet geyser]

I've gotten $400 of codex usage (at the api rates) out of my $20 plan in a month so I'd say it's pretty good compared to the api

[unique spade]

Zerg vs protoss 😂

[supple perch]

welp, api is down again

oh... its like offline...

[gentle harbor]

rip looks like there is no 5.5 today ?

[alpine wasp]

+1

[prisma flare]

why you think that ?

[cedar skiff]

api is down, maybe we will get it?

[brazen heron]

DA9TVW

Enjoy fast

[gentle harbor]

near the end of the day for me

[cedar skiff]

It's like just after midnight in some places in the US

[brazen heron]

DA9TVW
Thanks me later

[worthy furnace]

What do we enjoy lmao

[worthy furnace]

So you are a time traveller

[ivory zodiac]

137am mountain

do me a favor guys, add skills, hooks and plugins here

https://codex-marketplace.com

[gentle harbor]

what ? its 9 39 pm right now

[cedar skiff]

what day?

wed?

[gentle harbor]

23rd, it was supposed to come out today no ?

[worthy furnace]

We still have time lmao

[ivory zodiac]

yup its over

gg

[gentle harbor]

what time is it in cali right now

[prisma flare]

it gets release san francisco time so there is still a lot of time left lol

[worthy furnace]

0:39 pm

[prisma flare]

its 00:41 in the morning there

[worthy furnace]

Its 9:41 am for me

[gentle harbor]

oh.

[cedar skiff]

its 5pm where i am

[worthy furnace]

So I am ahead of sam altman in 9 hours

He is sleeping right now, so if its today, its probably going to take at least + 9 hours for me 😭

[turbid axle]

so the linear integration is pretty great

[prisma flare]

yep, so no testing with 5.5 today for me

[turbid axle]

I now just have an automation running which picks up issues automatically and implements them.

All I do at this point is just design and have agents create issues for it.

[split dome]

gpt 5.4 is at capacity? now?

[ivory zodiac]

im actually using it all sorry

[cedar skiff]

it has been a bit sluggish tonight, leave some for the rest of us will

[lofty steeple]

hey, is this new? or just a bug

[supple perch]

api down...

[lofty steeple]

nooo :(((

[lament estuary]

It's happening w/ 5.4 xhigh and high for me right now.

I threatened to switch to Claude and tell Tibo that he was not being a good clanker. I'm on a good run right now on xhigh though.

Spoke too soon, went down again

[gentle harbor]

works fine for me

[lunar berry]

yo
{"type":"error","sequence_number":2,"error":{"type":"server_error","code":"server_error","message":"An error occurred while processing your request. You can retry your request, or contact us through our help center at help.openai.com if the error persists. Please include the request ID dbe384d4-e0da-435a-ac4c-5e3486381539 in your message.","param":null}}

[gentle harbor]

i wish they shut down old models so we had more capacity faster

like why is 5.0 still here

[cedar skiff]

would that just shift the load to a different model?

[lament estuary]

The stopping and going stopping and going is cooking through my limits right now too. Had to have it make a .tmp file to keep track of itself just now

[gentle harbor]

yes id imagine or it would goto training, feel like it would be best to shut down old models 1 month after a new one came out

[split dome]

thx, wonder if it's just (x)high. I switched from xhigh to medium, we'll see

[lunar berry]

https://suno.com/s/rxj3y6Y9IU7fkYzm

[cedar skiff]

People build their workflows around models, that takes time and effort and new models have different nuances breaking those workflows.

[lament estuary]

Falling back from WebSockets to HTTPS transport. stream disconnected before completion: An error occurred while processing your request. You can retry your request, or contact us through our help center at help.openai.com if the error persists.

Think I might just put a lid on it for tonight. Unusable

[gentle harbor]

gpt 5 is nearly 8 months old isnt it ?

[cedar skiff]

I'm not sure how old it is, but i think it's reasonable to expect a year out of workflows. Also the initial point is what i was talking about, shutting those down wont be likely to increase capacity because it just shifts load, those users don't go away, they just have to rewrite their environment and use a different model.

[split dome]

ok gpt-5.4 medium also at capacity. nvm 🙁

[alpine wasp]

bro, is the API still cooked? why is the status page all green right now?

[gentle harbor]

weird the normal codex app is working fine for me with no issues

[cedar skiff]

it's still the really early hours in the US seems odd that they are at max cap right now

this is starting to feel like a new model release \o/

[lament estuary]

And a limit reset too

[gentle harbor]

i think im using us servers as well maybe its just cali as im pned into florida

[lament estuary]

Everything is. 5.3 just hit capacity too.

[alpine wasp]

why is every model release always like this

[cedar skiff]

not sure exactly why, probably just the nature of the beast

[lament estuary]

Probably because they have clankers putting everything into place to get the new clanker model in place to release

[turbid axle]

gpus gpus gpus gpus gpus gpus gpus

[lament estuary]

And the clankers mess something up

[alpine wasp]

still 3 PM in here and the API is down again for the second day in a row. no explanation, nothing. this is honestly so frustrating

[lament estuary]

Altman had to take a play out of Dario's playbook

[turbid axle]

calm down kids

keep the gamer attitude for game channels

[craggy whale]

I fixed it. Claude eventually realised that it was simply mathematics, which was why every other AI kept going around and around in circles, because it was focusing on the code. The code was never the issue, the mathematics was.

@frosty zealot For when you wake up. Yay, good news!

[gentle harbor]

i wonder, how does chatgpt extra thinking compare to codex extra high ? like how much more thinking does either one do

[prisma flare]

Gpt extra think is about 4x of codex xhigh. But you don't want long thinking for implementing code because it over thinks and introduces problems

[ivory zodiac]

i dont wanna hear codex isn't good at frontends ever again

i said EVER

[prisma flare]

nmv i mixed up gpt pro and codex sorry

[kind jay]

This live?

[kind jay]

That’s got nothing to do with mathematics lil bro

[craggy whale]

Here's its reasoning.

[exotic cave]

What you used to create this?

[kind jay]

What you showing me for?

[turbid axle]

any particular workflow used for this to get nice frontend with codex?

[dense path]

hello world

[ivory zodiac]

https://codex-marketplace.com

[ivory zodiac]

5.4!

[ivory zodiac]

yes, blog tomorrow

[lost drum]

Idk the gpt heavy is better than pro extended imo

[blissful basin]

nice stuff

[hazy pendant]

anyone have redundant hooks info display in their codex? how do i close it

[kind jay]

Looks good, well done

[ivory zodiac]

thanks 🙂

please add ur skills and hooks and plugins

[hard scarab]

Can u we use kimi 2.6 in codex? Any hacks todo it

[hazy pendant]

any help?

[hard scarab]

??

[kind jay]

Gpt 5.5 today

[lost drum]

wait what the hell actually?

[orchid plume]

yeah it's almost certain they will drop that today with the recent deliberate leak and cryptic hype posts on X

[lost drum]

my bad I sended wrong pormpt to codex

was so scared for 5min

[unique spade]

in codex cli you should be able to use any other model, but maybe just by api, didn t really bother to check in detail tbh

but as a general rule, just clone the codex repo start a codex thread in that repo and ask codex first all these questions, you will get an answer grounded in the actual truth of the codex repo

[limber ledge]

Codex 5.4 Desktop has been a miracle to me. I have had no problems. Talk about making everything for you or organizing your desktop for you lol

[unique spade]

yup

simply amazing

[lost drum]

launch that 5.5 already cause my project needs more reasoning

[unique spade]

i always enjoy the most when i watch how codex codes itself

[rocky parcel]

Is anyone else getting constant errors using the codex app-server?

[lost drum]

yo do you have any solution to have a copy of the repo each time he runs so I can revert back cause soemtimes he changes code by scripts and I cant just UNDO it via codex itself so ye also idk why but the git. is not detecting that something changed when he runs this commands so idk

[unique spade]

you mean the codex repo or what?

[rocky parcel]

Codex is not version control. You should understand the basic git actions so you know how to manage codex's use of git. the difference between untracked, staged, committed, etc.

Just because codex can do those things for you doesnt mean it will do them how you expect, especially if you dont know what to expect.

[unique spade]

ah it's a general git question

yea you should get a hold of git basics

[rocky parcel]

Codex can teach you the git basics if you ask 🙂

[unique spade]

git it's probably the single most important relic of the old age

i mean if there s something to pick to know , i d say is git

the basic structure of it

[rocky parcel]

thats an interesting perspective. I dont know if thats something you are claiming as a strong opinion or off the cuff. But there may be better ways to do version control now.

[unique spade]

didn t mean it s the best version control

[rocky parcel]

maybe not for source code. But git is terrible with non-source artifacts.

[unique spade]

was more like thinking about the version control class

that if it s one class of program that you should understand it's inner logic

it s version control, anything else you can leave to agent in the extreme

[rocky parcel]

ah, yeah. I'm having trouble thinking of something that has more staying power in the new age :p

[unique spade]

yup, i mean syntax has been totally comoditized

but version control is the backbone of the history of what happens

[kind jay]

Zip file in the google drive >>>

[rocky parcel]

"final.zip"
"final2.zip"
"final2.zip.bak"
"final3.zip"
"final-final.zip"

[unique spade]

share by the way other ways you like for different usecases of version control

[rocky parcel]

I dont have a good solution. I work with source code, and therefore I built automation so that I can generate my non-source artifacts from that source as needed.

But really, thats only "best practice" because git is trash with files that arent line oriented text.

[unique spade]

in a funny way one of my fields when i was in academia was "possible worlds semantics". turns out git and similar instantiated much of that branch of logics/philosophy in a practical manner

[rocky parcel]

Do you have a link to an intro to that topic for a newb?

[rocky parcel]

I'm endlessly fascinated with formal languages and that sounds related

[unique spade]

just replace git with modal logic and possible worlds semantics

[rocky parcel]

harsh, point taken lol. Is the right term to use "possible worlds semantics"?

ok

[unique spade]

yup

Saul Kripke is the philosopher who made the first semantics for that in the 60's i think

it strongly overlaps with the underlying logics behind quantum more philosophical topics

[rocky parcel]

Is this related to wolframs "ruliad"?

[unique spade]

wolfram as stephen wolfram?

[rocky parcel]

yeah

I realize from your previous messages that you're probably not interested in having this conversation, gpt-5.4 says theyr'e related but not the same idea.

[unique spade]

not familiar with "ruliad" i have to check. i did read his new kind of science 25 years ago and thought it was one of the best books written

although the dude is a bit egomaniac

and lost the plot around the middle

[rocky parcel]

Yeah its one of my favorite 1000 page textbooks.

[unique spade]

but first part is fabulous

[unique spade]

exactly

[rocky parcel]

agreed, the last few chapters bring it back though

in a fun kind of speculative way

[unique spade]

if you want something heavy which suddenly is vindicated by LLM era

a book that took 18 years to be written and at the time won the "nobel of philosophy" with qoutes because it s different prize

but it s considered the most prestigious

Robert Brandom - Making It Explicit

it's inferentialist-pragmatist semantics of language

[rocky parcel]

Thanks for the recommendation 🙂

The amazon abstract got me. Gonna have to read this.

[kind jay]

Why are you guys larping as high IQ?

[unique spade]

this is the 2nd paragraph of the book

it was published in 1994

you will see "reasoning llm's" of today basically mentioned there

[rocky parcel]

"gaseous et" lol. Pop culture of the day creeping in.

Also that is a very dense paragraph.

[unique spade]

it s funny, types are fundamental in programming, nowdays also tokens are very famous

but it is less known that type/token is a fundamnetal distinction in the philosophy of language

[unique spade]

the entire book is that dense, this guy is like the modern kant. it took him 18 years to write these 700 pages

similar ambitions with wolfram

[rocky parcel]

I understand tokens in the mathematical sense, as points in an absurdly high dimensional space. But I dont see the equivalence to types.

Unless you think of the space of all possible types in a similar way

[unique spade]

well the distinction i mentioned is not that deep. it's analogous to class/instance

just saw it when i quickly browsed brandom s book and reminded that both terms have also significant meaning in this other real of computers and now ai

but modal logics is indeed essential, because it branches also into deontic logic (logic of norms) and epistemic logic ( logic of knowledge), temporal logic etc

basically everything that matters, outside things, are treated best with a form of modal logics

[wheat yoke]

I am NOT reading all this

[unique spade]

you wanna understand why models hallucinte, if you don't know the basics of these logics, you won t get to deep in the matter

[unique spade]

cause no one else talks anything at this time

and it's about codex by extension of codex being a llm haha

[rocky parcel]

Interesting take. I think I understand why models halucinate. But I wouldnt claim a deep understanding of those branches of logic, so I'm not sure if I can make the counter claim.

[unique spade]

i m sure you do, but can you formalize it? so you can qunatify it?

🙂

if you wanna do that you need the tools

[rocky parcel]

Yeah I havent tried to formalize it. I'm sure someone has though, thats the first step towards reducing halucinations.

[unique spade]

right

[cedar skiff]

There is clear understanding of why, it's just not a simple fix.

[turbid axle]

I hope gpt5.5 is like, 10x smartser, and 30x cheaper

valid?

[wicked briar]

I mean 5.4 is already smart

10x of that would be agi

[rocky parcel]

The formalization should help inform the fixes though.

[turbid axle]

even with single agent on medium im running out of tokens on my pro account. this is too much

[unique spade]

10x smarter and 30x cheaper will be much closer to "agi"

and we re already quite close

[wicked briar]

cap

[cedar skiff]

I mean they have come a long way with it but it's a systemic problem

[turbid axle]

they really squeezing the hose atm 🙁

[rocky parcel]

They have indeed. I agree. I dont think we're actually of different opinions here.

[unique spade]

enjoyed the talk, i'll go for now. laterz

[turbid axle]

would not even be mad if 5.5 is just 5.4 but a lot cheaper

[rocky parcel]

👋 ttyl

[turbid axle]

but, its probably AGI

agi-high

[rocky parcel]

I would have considered o1 to be AGI if you had presented it in 2015.

[kind jay]

Probably will be this

[turbid axle]

yeh I expect something like that too, little smarter, maybe little cheaper. some in between before they bring out the potato canon

[unique spade]

o1 is still the fundamental breakthrough

everything that came after i downstream of o1

before o1 models were literally stochastic parrots

[cedar skiff]

Id like longer context without degradation

[unique spade]

here's a zenish qoute about the matter

"Hallucination often begins where an untyped assumption is mistaken for a real object."

[cedar skiff]

I don't know how it relates

[turbid axle]

memory is next. that will open pandora's box I believe. rocket ship material

[unique spade]

you'll find out one day. that's why it's zennish 🙂

[cedar skiff]

you just yapping like youre a cult leader

[turbid axle]

hallucinations are a feature imo, a key one at that

[unique spade]

yea i cam with a ship from venus 🤣

[cedar skiff]

context rot != hallucinations

[turbid axle]

it's not something to solve. it's to be combined with reasoning, and memory.

[unique spade]

you are talking about the value of generation
that's a super power of llm's sure.

but otoh you don't want a model to hallucinate you the wrong critical dosage of medicine

in some domains hallucination is catastrophic, in others it's an asset

[cedar skiff]

I don't want it to hallucinate i want it to just say it doesnt know.

[turbid axle]

that is what reasoning and memory is for.
hallucinations imo are essential for creative thought.

a confidence factor where it is self aware of hallucinations however I can agree with

[unique spade]

yea and for now the epstemic engineering crew still have work on that part, quite a lot it seems

but it s much better than 1 or 2 years ago imo

[unique spade]

to be fair hallucination is such a bad term for the technical concept in case of llm's. it's borrowed from the psychology toolkit

[turbid axle]

they never bothered me tbh. I always double checked and rarely used gpt for 'answers' much more as a research/eduction tool.

and now with reasoners/agents it self corrects all the time. Only harm is more token cost I guess

[unique spade]

yea but the average joe uses gpt too, and it tends to trust what ai says 🙂

[turbid axle]

average joe is a little dumb I find. can't be helped much. they will come around with time

or they won't, and well..

iteration as oai does is the best medicine here

[hard drum]

there's a difference between 'feels like they're right" vs "think they're right". average Joe is the former

illogical feelings apparently overcome logical thinking

[unique spade]

that s not very politically correct 🙂

[turbid axle]

anybody who thinks they know anything at all is dumb average joe in my mind

hey, well, everything is politically correct for my own politics 😄

[unique spade]

see ya laterz peeps

[turbid axle]

gpt5.5 when

[unique spade]

well today still possible right? 12 pm in PT is still some hours later

[turbid axle]

politics is just the loudest wins imo. I say all goes. good/bad is for average joe

[turbid axle]

most likely, otherwise I rage

[unique spade]

previous livestream about image 2.0 was announced 2 hours from now

if we don t get anything in next 2 hours i m gonna start worrying 🙂

[turbid axle]

interview I watched yday said they will be dropping multiple things coming weeks, lets see

[unique spade]

i think if 4.7 had better results they would have launched it already

but 4.7 turned to be kinda lame

[hard drum]

4.7 is pretty disappointing && token-consumptious from what i've heard people report on

[unique spade]

i saw a lot of people complaining and how4.6 is better

[unique spade]

same my linked in and whatsapp group say the same

[hard drum]

i had the ability to use 4.6 for nearly a month, && 1m ctx version for a week or two

4.6 was lazier than 4.5

4.5 was on the same level as 4.6 sonnet

it's like the more intelligence the model harnesses, the lazier && "i can't be bothered" it becomes

[unique spade]

might specific to claude training regime

i can t say i noticed that in oai model evolution

[hard drum]

gpt 5.4 being almost equal in laziness of 4.6

you have to be really watching gpt5.4, or else...

[rocky parcel]

I don't have numbers to support this claim, but anecdotally, I use 5.4 mini for implementation after 5.4xhigh does the planning work.

[hard drum]

i go from 5.4 high to 5.4 medium

xhigh thinks too much, then defers things more than high variant

[unique spade]

that could be true, mine is wrapped in so many beauracracy it can t afford to be lazy

[hard drum]

but i don't use 5.4 mainly, i have my own framework using other models in other agents

[rocky parcel]

Meh, I have a custom harness and workflow orchestrator and it seems to work out pretty well.

[unique spade]

i did recently a test on a logical puzzle with all thinking levels for 5.4 and 5.3

best one was gpt 5.3 medium and high

highh a bit more correct, medium the fastest

xhigh in both cases felt a waste of tokens to get a wrose result

but it was on a logical puzzle, not code, so not directly relevant for coding per se, but for reasoning

[bronze sable]

I found a company that sells codex only openai plans for 50% off, how are they doing this? anyone who knows?

[unique spade]

depends is it ana ctually public company?

[bronze sable]

I'm testing their free plan rn, kinda scared to pay though, even if it works

[bronze sable]

yeah

[unique spade]

or some blackmarket company

[rocky parcel]

You want someone to ask you for the link so you can spread the scam eh?

[bronze sable]

no? i just got the link from someone else

lol

[rocky parcel]

riiiiight.

[bronze sable]

you are even more skeptical than me

#chatgpt-discussions lil bro

[unique spade]

it s probably a bot

[bronze sable]

and i wont be sharing the link anyway lol, if I don't trust it fully myself then why would I want someone else to trust it on my behalf

low iq reasoning

[bronze sable]

this is the type of ppl that get replaced by ai

[unique spade]

and quite aggresive bot too haha

[bronze sable]

which is a good thing

[rocky parcel]

they were right, gpt2 was too dangerous.

[bronze sable]

hope u haev atleast 500k in assets generating atleast 10% yield

[kind jay]

DM me link lil bro

[heavy rapids]

maybe the company's got an Azure instance

[heavy rapids]

me too please

[unique spade]

this convo is a bit dubios

[rocky parcel]

"a bit" lol. Its an obvious scam.

Its literally textbook.

[unique spade]

do we have mods?

[bronze sable]

no im not sending anyone the link

[kind jay]

Obviously, but I want to see none the less

[kind jay]

We do indeed

[bronze sable]

lol

[unique spade]

@last willow

[kind jay]

To do what lol?

[tawny turret]

keep it lowkey blud

[tawny turret]

hi jane

[kind jay]

Oh hiiii

[tawny turret]

you ignored my dm

.....

(mno not dm about advertising something )

(it was yesterday )

[frosty zealot]

Did it happen yet

[unique spade]

No

You can go back to sleep

[wicked briar]

in like 6hrs or something

[torpid bridge]

Hi I am currently using codex via the vscode ide to work locally, anyone knows if there is a way to rename tasks/conversations ?

[lost stag]

codex is the best but there needs to be a save option for the app

[cedar skiff]

save what?

[lost stag]

like I cant use the same project on my laptop to my main pc but its not too big of an issue

[lost stag]

the chat feed on the app

it saves to the device not like a cloud thing

[cedar skiff]

Ahh i see what you mean

[wicked briar]

feelings are getting stronger

nice

[lost stag]

codex is a game changer

[frosty zealot]

Unfortunately I woke up to a blizzard and still have to go work in this

[wicked briar]

I actually am against this like working on local stuff in codex should stay on my machine not uploaded and shown in the chatgpt website like they are separate app and should stay that way

[frosty zealot]

You can make a private git repo and just push and pull your threads on different machines but tedious, but work around for now

Or if you have say like a synology network storage or equivalent, you could set up their file sync and it would be all local

[lost stag]

its a nice feature but they should make it work both ways because I prefer the app over the website for some reason

[orchid plume]

I see the model descriptions on Codex git repo have changed, notably GPT-5.4 is no longer described as the latest model

[lost stag]

thats what ive been doing but it would still be nice to remember from chat to chat

[wicked briar]

just use app then

[wicked briar]

👀

[frosty zealot]

Yeah Synology how’s a really nice file sink it’s almost immediate would probably be the most seamless thing if you had one

I’m using voice to text so if it looks like I’m having a stroke, that’s why

[lost stag]

the app doesnt translate from device to device thats my only issue

[unique spade]

Wow

[unique spade]

VScode extension convos do? I mean the codex ones. I assume it could work via some function provided by VScode

[lost stag]

I dont see a plugin for that in the codex app

[unique spade]

Sry I thought you were also this

My bad

[lost stag]

to rename, I just right click

and click rename chat

[kind jay]

Imagine having a j*b 😂

[unique spade]

That s the pain point right?

[frosty zealot]

Waddya mean it’s great

[unique spade]

Like to be able to carry on the same convo on multiple devices?

[orchid plume]

indeed, a very strong sign that GPT-5.5 is coming today, along with possibly a new mini model, codex model's description was also changed.

• gpt-5.4 changed from “Latest frontier agentic coding model” to “Strong model for everyday coding”, and its default reasoning level was raised from medium to xhigh.
• gpt-5.4-mini description was updated from “Smaller frontier agentic coding model” to “Small, fast, and cost-efficient model for simpler coding tasks.”
• gpt-5.3-codex description was simplified from “Frontier Codex-optimized agentic coding model” to “Coding-optimized model.”

[blissful basin]

Imagine all these AI doomers saying that AI will take all jobs, i work probably 3x as much as before AI and produce probably 5x as much code if not more 😄

[lost stag]

main point? I guess I dont understand what codexs end game is, I thought it would be like the chat gpt app where I can use the same conversations on all my devices

[kind jay]

You do SWE?

[blissful basin]

yeah

[unique spade]

Pain, I meant your issue

[wicked briar]

it is coming today

[unique spade]

You can copy your sql with convos from desktop to laptop if you want

[orchid plume]

oh definitely 😛

[unique spade]

And then you ll see the same threads

But I agree it would be nice to have at least some import/export function

So you can move your context on another device if you want

[wicked briar]

thats like saying I want you to take my whole codebase and make it move around on whatever device I go

[unique spade]

Or put some tunnel service so you can connect with another laptop or even phone

[cedar skiff]

there is a tool for that github

not sure i want my chats on github though

[blissful basin]

i wonder if sharing context between devices could be solved with skill/plugin. In the end these are just json/md files

[unique spade]

Yea chats are kinda private

[blissful basin]

I tried to copy it manually but didnt work but also didnt really put much effort in it

[unique spade]

The convos are kept in a sql

Skill sounds overkill since it's good old files.

[cedar skiff]

It would expect the project to be in the same directory, so thats something that needs to be considered

[wicked briar]

same I don't want my company's sensative data be lying around on chatgpt history

😭

they should give toggle if they decide to do it at some point

[lost stag]

how will 5.5 be better

[cedar skiff]

base model is bigger

[unique spade]

yea, so to keep it simple you would want to have the same location

[cedar skiff]

i mean you dont usually have the same user folder on all computers

[minor forum]

so gpt 5.5 today or tommorow or what

[boreal holly]

I straight up have /Users/robertsale on all computers. Even on Linux after sudo ln -s /home /Users

[plucky halo]

No one here really knows. It's all guessing.

Except that it absolutely won't be tomorrow

[minor forum]

didnt sama say so on X

[plucky halo]

Have a look on his profile

[minor forum]

it was a reply I have seen it myself

I am a regular user(degenerate) on X

[unique spade]

he had that emoji that s what i remember

this one

[torpid bridge]

doesnt seem to have that option in the vscode codex extension

[tawny turret]

whole community relying on one drunk reply

pro

[gentle harbor]

is this fake ? can you send the link

[plucky halo]

Then why are you asking? If you believe the CEO of the company has confirmed on X that it is out today or tomorrow, then why ask?

[minor forum]

Trying to talk to the community

[unique spade]

https://x.com/sama/status/2046798393918505024

[boreal holly]

Instead of 5.5/6, they're gonna release a new feature where the voice AI can talk to animals

[plucky halo]

I can already talk to animals

They just don't understand me

[boreal holly]

Yeah but you can have a bird, and ChatGPT be like "CA CAW", and the bird completely understands

[gentle harbor]

oh wow

[boreal holly]

Ground breaking research

[cedar skiff]

welcome to the light side!

[minor forum]

Anthropic with mythos: Welcome to the dark side

[tawny turret]

lowk do u guys also think sam altman isnt actually that bad of a person as they are portraying him to be

i think dario is far worse, since he lies a lot and anthropic is lying trying to deceive ppl that theyre focused a lot of trust and safety

[cedar skiff]

not many nice guys reach the level of these two

all the nice guys got hit with sticks already

Who knows though

[tawny turret]

i know

!

im smart

[cedar skiff]

So much smear media around, i just dont buy into any of it

[frosty zealot]

@kind jay https://youtu.be/i-slJ7SYjHE?si=KFzS9BinxKFT5hTP niche

[boreal holly]

Sama is famous, there will always be people besmirching the good name of famous people. I think he's a stalwart venture capitalist with good intentions

[kind jay]

No

[kind jay]

People just hating because he got a bag

[frosty zealot]

It’s amazing that they get more heat than Anthropic when OpenAI has done nothing but Support their community, meanwhile, you have anthropic who on a whim turns on and off features without a car in the world and seems to give zero crap about any sort of constructive feedback. They just do what they want when they want.

[tawny turret]

he a good guy tbh I remember in old stanford startup school series he used to actually care abt being fair with employees & management (derived from his blog posts and questions)

[frosty zealot]

I bet he’s packing you know what I mean

[kind jay]

Packing?

[frosty zealot]

Yeah

[tawny turret]

yeah lol opus is dogshi, used to be ok during 4.5 and 4.6 launch and then turned into complete ahh, i have basically infinite codex im spamming 5.4 xhigh its better than opus 4.6 prime and 4.5 prime, and never regressing / rerouted or weird outages

[tawny turret]

sama is gaey

[kind jay]

A\ support is dog

[glacial shadow]

4.7 and 5.4 are nearly equal

[tawny turret]

wdym? there is no support

[tawny turret]

no lol 4.7 is worse than 4.5

[kind jay]

Also usage limits are dog

[glacial shadow]

what are you coding with it

[tawny turret]

dude i have like 20b tokens used last 2 weeks trust me i know what im saying, i used opus 4.6/4.7 and 5.4 is way better

[tawny turret]

codex only from now on with a custom skill

[kind jay]

LLM wise, opus used to be good at front end design when gpt wasn’t

[glacial shadow]

dude i have 9870quadrillion tokenz made right now

actually just crossed 987980quintillion as we spojke

[tawny turret]

no point in using anthropic models for that when gemini models do the same, cheaper and faster

low iq sarcasm, in case u didn't get it, noboduy likes sarcasm