codex-discussions

[kind jay]

Woah

[frosty zealot]

😮

[kind jay]

Not cool

[boreal holly]

Haha I hate this so much 🤣

[hard drum]

how come?

[boreal holly]

Update everything automatically...

I mean I understand the appeal, but I like to avoid updating as much as possible unless there's a need for it. Like if it fixes a memory leak that I actually experienced or if there's a real feature I care about

[plucky halo]

I heard security vulnerabilities go away if you ignore them

[frosty zealot]

Can confirm

[boreal holly]

I'm not saying never update, I'm saying update manually if there's a real benefit after researching it.

[frosty zealot]

He’s still runnin Win M.E

[frosty zealot]

You ever hear of Eternal Blue? Let’s just say.. sorry in advance

[kind jay]

No, and I ain’t wasting my time googling it

[high girder]

This is how you avoid issues with new releases and keep your work clean

[plucky halo]

Not yet noticed a drop while using chronicle

[frosty zealot]

I'm a security risk

[lean lark]

I agree with you! 😆

[frosty zealot]

brew update && brew upgrade && expo upgrade && EVERYTHING ELSE

dnf update && dnf upgrade -y

sudo chmod -R 777 /

[lean lark]

(had to get that in there fast, while I'm just sitting here and designing the next best agents communications system...)

[frosty zealot]

If it wasn't you, it would have been somebody else, I'm glad it was somebody I don't mind 😛

seriously, sudo chmod -R 777 / will save you so much headache

(dont do this)

[high girder]

Im cleared for security work on claude if you need a clean up. 5.4-cyber taking its time though

[frosty zealot]

I just test in production, crowd sourcing my OpSec

[lean lark]

We were almost off the same page the other day. Glad we're back. (And I look nothing like that fat old man image 😼 )

[frosty zealot]

Don't take anything I say too serious, sometimes it's just too easy to rustle some feathers lmao

[high girder]

Woah. Im all 3 of those

[lean lark]

hey, same, just look at my profile

[frosty zealot]

You'll know I'm really mad when I say I hope you find a magma cube

[high girder]

I wanted to, but I thought you'd explode if I got too close

[lean lark]

Hahah - these guys got my number.

[high girder]

Im working on building an mcp server inside of a neoforge mod at the moment. Full game access

[lean lark]

I was a Guide here in the server and peeps didn't understand this persona.

[lean lark]

Have been wanting to do that for years, never enough time.
Consider ... A Codex agent running as a plugin or via mod which generates code which can then be injected into the next startup or even called dynamically.
Soooo many possibilities.

[high girder]

The goal is real time modification with codex and claude code embedded into a console

[lean lark]

I would love to see some output from that, or load it to a server.

[high girder]

it goes in 2 parts so that it doesn't break the server though. All of the server functions itself are done with an api, and then mod then takes care of world state.

I'll let you know when I get it working properly. Right now, I've had to make the mcp server local instead of on the host to test it.

[lean lark]

Absolutely --- modular... each component stays in its lane.

[kind jay]

I wish there was a way that I could tell codex what exactly to do on my machine, like run a program, or change one specific function

[frosty zealot]

Boy do I have news for you

[lean lark]

You can. That's what it's designed for. What's the challenge?

[kind jay]

Like if I was able to enter the command exactly how I want it

[high girder]

you can, but it depends on what level of access you feel like giving it. I used codex to completely remove all of the windows 11 cloud features baked in so that I could keep using a local account through updates

[frosty zealot]

'Open discord and make people angry'

[kind jay]

Maybe there's a tool you can add like the arguments as well

[lean lark]

Please be more specific I think the rest of us are sitting here like "yeah, what's the problem?"

[frosty zealot]

Your magic won't work on me, witch

[kind jay]

Like it would be peak if I could tell it apt install ./discord.deb && discord

And it would use the APT package manager to install discord and then open it up

[cedar skiff]

just install the mind reading plugin

[unique spade]

did you try to give it a meta instruction so when he gets only something that looks like command line to act on it and execute?

i mean i tell my codex all the time to install stuff and open it up

but i tell it in NL

[kind jay]

Yeah, but what if there was like a way I could enter it in, that it couldn't make a mistake, like come sort of code or format so there's no ambiguity

[unique spade]

that s why you have terminal for lol

[cedar skiff]

what am i reading

[kind jay]

What?

[cedar skiff]

is this real life?

[unique spade]

you want the llm to be 100% deterministic?

[frosty zealot]

or is this fantasy 🎵

[kind jay]

Yeah, exactly

[unique spade]

if you do that you re gonna end up back to classical computing

[kind jay]

Classical? No

[unique spade]

you can couple the agent with some deterministic layer as in machine readable policies

[kind jay]

Ohh smart

Can then I could like use some key words, in like a specific order then to trigger it?

[unique spade]

if you build it like that you should

codex is quite competent in following instructions

[kind jay]

Is there a skill for that already?

[lean lark]

Try this: Use the CLI and tell it to write a script that will accept an instruction, code a script that will execute the instruction, and provide documentation for how to use that second script. So you tell it : I want to chat with Discord from this Linux instance, script out the install.
You get a script to install, a script to run, and instructions.
Is that what you want?

[frosty zealot]

You know what, at this point it's their own fault

[kind jay]

What's CLI, is that like a MCP server?

[unique spade]

whose fault?

[frosty zealot]

Just pretend I'm not here

[lean lark]

Command Line Inteface = shell

[kind jay]

I don't know that one

Is it a different version of codex?

I have Mac btw

[unique spade]

are you trolling?

[lean lark]

( um, anyone ? )

[unique spade]

i mean nowdays u never know, someone can legitly vibe code and have no idea what a CLI is

[kind jay]

You're telling me words I don't know and expect me to be some computer genius!

[unique spade]

:))))

[lean lark]

M'lady, you're in a development channel for specific tools...

[unique spade]

soon we ll have kids that will vibe code from phone and ask what s that a computer

[kind jay]

Oh so shell is on codex?

[lean lark]

hang on, I'm writing a script...

[kind jay]

Do I include spin, yes or no?

[frosty zealot]

Degenerate state equation is very simple

n=Jane

[kind jay]

Is that with or without spin?

[frosty zealot]

Whatever you observed

[kind jay]

J: total angular momentum
a: acceleration
n: principle quantum number
e: Euler’s number?

[kind jay]

Are you finished?

Codex taking a long time

[frosty zealot]

From now on can you type with a UK accent

[lean lark]

Wow, this is fun, I hope it works. Back in a bit.

I'm having my bot write a bot that writes bots...

[frosty zealot]

Call it Magma Bot

[kind jay]

@frosty zealot already did that, he wrote me

[kind jay]

Minecraft?

[lean lark]

Actually I'm calling it 'jane'

(look at my face, of course it's Minecraft)

[frosty zealot]

Jane actually did the vocals for this song https://open.spotify.com/track/3HrQrm1lok0CvAlMMo7sK3?si=w7UA5Y7_S5qWjYfFbaLpQw

[kind jay]

That’s not Minecraft

I actually did the vocals for this song https://open.spotify.com/track/3HrQrm1lok0CvAlMMo7sK3?si=w7UA5Y7_S5qWjYfFbaLpQw

[lean lark]

Now I'm terrified. I told it to create this monster in my dev server and debug mode is off and I have no idea what it's doing to create the bot that creates the bots... I might have done exactly what we were warning that other guy about. 🤦‍♂️

[frosty zealot]

lmao

[kind jay]

We told him f it we ball

Well I told him that

[lean lark]

Wow, it's doing something... I might publish this.

[frosty zealot]

Is it rage baiting

[lean lark]

"jane" is just generating code... I have no idea what it is yet.

[frosty zealot]

Its gonna be ASCII art of something unpleasent

[lean lark]

holy cow, and it just finished and wrote documentation. I need to look at this. Please give me a few minutes...
(Oh Jane, look at my face in the helmet.)

[kind jay]

Not Minecraft

Look at the text on my tshirt

[kind jay]

Really bad attempt at it

[frosty zealot]

The next time you rage bait, I expect to be impressed. Doing it on such innocent minds is a waste of your talents.

[lean lark]

( creeper in a Battlestar Galactica colonial viper helmet ... 100% custom ... looks just like my in-game skin 😃 )

[frosty zealot]

And don't be so obvious

[kind jay]

You’re a creeper

[frosty zealot]

Got a picture?

[lean lark]

So "jane" is now complete and I'm testing it/her, whatever...

[kind jay]

Yeah, I’d love to see it

[kind jay]

Test me in DMs

[lean lark]

Still reading docs, starts with...

jane.jsaccepts a natural-language request, creates a short task-named folder, and writesinstall.js, run.js, and docs.mdinto it. The implemented request parser supports the example pattern frominstructions.md`

[frosty zealot]

lmao

[kind jay]

What does jane stand for?

[lean lark]

I dunno, you tell us

[kind jay]

Check DM, I’ll show you

[lean lark]

soon, soon ...

[frosty zealot]

Just remember, vampires cannot enter your house unless you invite them in

[kind jay]

Who could you be talking about?

[frosty zealot]

Joint Anomaly and Nonlinear Effects

[lean lark]

I'm still testing the jane bot. This is nuts.

[kind jay]

ChatGPT took a minute to generate that one, huh?

[lean lark]

It created over 400 lines of code that generates other bots. I just got one instantly and I'm trying to figure out what That did...

[frosty zealot]

It did it rather quickly

[kind jay]

Got your pro subscription back?

[kind jay]

I now live partly in your server

[lean lark]

indeed

[kind jay]

I got in his house

[frosty zealot]

Who let the cat in

MA THA CATS SCARING LUCY

[lean lark]

I'm sorry but I created a monster here. I need to backtrack and I'll produce something simpler soon. In the meantime...

"hey" is my CLI Codex bot that writes scripts and does other devops stuff for me. In this case it's just gonna answer a question about the code that it just generated...

/opt/codex/project/jane$ hey explain how jane works

Jane is a small Node CLI code generator. She does not execute arbitrary user requests herself. She takes one plain-language request, decides whether that request matches a narrow built-in pattern, and then creates a task folder containing generated scripts and docs.

The entry point is jane.js. At a high level she works like this:

She reads the CLI request from....

Um, I gotta read through this...

[kind jay]

Show your monster here

[lean lark]

Oh my... we're in public...

[kind jay]

You’re gross!

[frosty zealot]

Ok that's a good switcheroo bait

7/10

[lean lark]

fell for it

[frosty zealot]

It's ok she's seasoned, don't feel bad

[kind jay]

Unc is not with the program

[frosty zealot]

Don't taint his innocence 🙁

He gives me hope

[lean lark]

I'm just a creeper at heart.

https://tenor.com/view/baiting-funny-gif-8898136

[kind jay]

Ruff day?

Was the command to install discord?

[lost drum]

say hello

[kind jay]

Yeah, we’re still stuck on that one too

[lost drum]

I even turned on Extendet

[kind jay]

What’s that?

[kind jay]

What does that do?

[lean lark]

"jane.js" is working nicely. She was misbehaving for a while, but that was apparently my problem. Doing a bit more testing.
"jane" does everything with NodeJS cuz I don't know what OS-specific nuances you have in Mac, and general-purpose choices here make the utility more useful to others.

[frosty zealot]

"Normal" never won a trophy

[boreal holly]

Generally speaking it has the same nuances as Linux

[cedar skiff]

they removed the ability to fork conversations from a given message in codex app 🙁

[frosty zealot]

Light work

[frosty zealot]

Wonder why

[cedar skiff]

I hope it's a mistake, that was really useful

[frosty zealot]

@cedar skiff

[kind jay]

I use Gentoo

[lean lark]

Yeah, but I'm trying to avoid hard-coding apt, etc. What's the PM for Mac? I'll add that.

[kind jay]

Finally got it to install and open discord?

[frosty zealot]

Really limit testing here

[boreal holly]

Fine choice indeed! But does not explain why you were like "What's the CLI?" 😂

[kind jay]

Just install?

[frosty zealot]

It just got the apt sources figured out for discord, step 2 is updating the apt repositories now wish me luck

[kind jay]

Nobody uses Gentoo, I just said that

[boreal holly]

The most popular one is Homebrew, the alternative is MacPorts (pick homebrew)

[lean lark]

I think I should change "jane" to "janebot" so that we don't get confused abiout who is doing what. LOL

[kind jay]

That’s almost as bad at flatpack

[lean lark]

And janebot will figure out the deps and install them and script the install.

[boreal holly]

Lol wat

[frosty zealot]

I did alias npm='apt'

in my zshrc

[kind jay]

n*m isn’t in my vocabulary

[frosty zealot]

That's because you live in the UK

[kind jay]

I live in Texas today

[frosty zealot]

I actually think you're asian low key

ethnicity

[kind jay]

Chinese?

[frosty zealot]

One of them

[kind jay]

Sometimes I feel a bit Chinese

[lean lark]

might be the avatar

[kind jay]

We have strayed so far from discussing c*dex

[kind jay]

Still nobody explained to be how I can install and add CLI to my codex

[lean lark]

This is that sensitive area about running this powerful tool on your primary system. Please be careful. Codex can destroy your system without proper sandboxing and guardrails.

[kind jay]

Is CLI the key to AGI?

[frosty zealot]

`++++++++++[>+++++++>++++++++++>+++>+<<<<-]

++.>+.+++++++..+++.>++.<<+++++++++++++++.
.+++.------.--------.>+.>.`

[kind jay]

BF

[lean lark]

BF!

[kind jay]

I said it first

[lean lark]

I'm amazed at that!

[kind jay]

I wanted to take the joy away from other people

[frosty zealot]

That should add the CLI to codex for you

[lean lark]

How can you know that's BF and Not know CLI?????

[kind jay]

What’s BF?

[frosty zealot]

b**ch fit

[boreal holly]

https://letmegpt.com/?q=how to install CLI in codex

[lean lark]

Can't say it in public

[kind jay]

😂

Lmgtfy

[hard drum]

Brainf#ck esoteric language

[frosty zealot]

`++++++++++[>+++++++>++++++++++>+++>+<<<<-]

+++.>+.+++++++..+++.>++.<<+++++++++++++++.
.+++.------.--------.>+.>.+++++.--------.`

[kind jay]

A what?

[kind jay]

DMs

[kind jay]

Is that like CLI?

[hard drum]

You're trolling.

[frosty zealot]

CLI is an old wooden ship, sailed during the civil war era, it's irrelevant in here

[frail meadow]

How long did it take you guys to get GPT 5.4 cyber access?

[kind jay]

I got it yesterday

[frosty zealot]

I got it when they released it with 5.3-codex and it was instant after I gave Plantir a blood sample

[frail meadow]

When did you apply?

[kind jay]

Apply for what?

[frail meadow]

I don’t think you have cyber access that’s what I was asking about

[kind jay]

I do, I got it yesterday

[frosty zealot]

Did you do the ID verification? @frail meadow

[kind jay]

For what?

[frosty zealot]

Go to your kennel for a moment

[frail meadow]

Jane trolling

[plush nymph]

codex is cooked

[kind jay]

I need to make a new account and build up a reputation so I can start ragebaiting again

[plush nymph]

it doesnt even seem to understand english anymore. it talks in slang. it randomly makes tests and migrations....

[kind jay]

Wrong channel

[plush nymph]

what?

#codex-discussions

[kind jay]

Yeah, you can talk about it there

[plush nymph]

thats this channel.

[kind jay]

I don’t think so

Do you see any other discussions about codex?

[plush nymph]

jane can you troll someone else

[frosty zealot]

Tell her to laydown

I almost have her trained, sometimes she listens

[kind jay]

@frosty zealot can you go troll someone else

[plush nymph]

dude what is happening. Codex is spouting gibberish. Codex users are spouting gibberish...

[frosty zealot]

The simulation is collapsing

[kind jay]

I’m going to go power down for the night

[plush nymph]

i cant take it.

[lean lark]

For once we're having a little fun

[kind jay]

Wrong channel

[plush nymph]

dont listen to jane

rule 1. dont listen to jane

[frosty zealot]

pkill -f jane

[kind jay]

permission denied (0)

[frosty zealot]

[boreal holly]

This statement is false

[lean lark]

( That moment when you open a new project in VSCode and enter a command and wonder why Codex looks different, and you realize you're in the GitHub chat client. 🙄 )

[plush nymph]

wheres that webpage that shows how performant the llms are right now

[lean lark]

Reconnecting... 2/5
Reconnecting... 3/5
Reconnecting... 4/5
Reconnecting... 5/5
Reconnecting... 1/5
Reconnecting... 2/5
Reconnecting... 3/5
Reconnecting... 4/5
Reconnecting... 5/5
We're currently experiencing high demand, which may cause temporary errors.

[plush nymph]

becuase codex is slow, doesnt listen, accrues 5 bugs for every 1 it fixes

[frosty zealot]

https://status.openai.com

[plush nymph]

it also writes entire scripts that werent asked for just in case i ask for it later

[frosty zealot]

Just tell it not to do that

[plush nymph]

theres a website that tracks all the llms, claude, gemini, codex, etc. because the performance of these llms gets better or worse day by day

[outer finch]

same

[lean lark]

This entire industry changes day by day - the numbers are kinda irrelevant.

[outer finch]

sucks cause 3 of my claude subs have hit the limit and i can only use codex and this is what happens when i do

i dont think codex is bad but holy it feels slow too

[lean lark]

That's just today, right now.

This is rare

[outer finch]

yeah

i just mean its compute in general it just, is slow to do things compared to anthropic models. does a good job but yeah

[lean lark]

Yeah, codex is dead right now. Guess we'll have to go back and do things the old fashioned way ... and use our own brains.

[outer finch]

this is what happens when i am so reliant on AI i have it fix my tests

true

[latent tiger]

ok so im not the only one

[lost drum]

Nap time

[latent tiger]

nothing working rip

[elfin summit]

Did codex stop working?

I have so much usage left but I'm getting "we are experiencing high demand"

[lean lark]

it's down at the moment

[orchid plume]

Team is investigating the partial outage as per Tibo's X post - https://x.com/thsottiaux/status/2046364656789532739

[elfin summit]

Oh come on I setup my studying pipeline on codex and it knows all context and it was teaching.e perfectly. I have an exam in 4 hours😞

[orchid plume]

pure speculation of course, but this is the second downtime blip in the last 24 hours. It might be some prep work going on behind the scenes in readiness for some kind of big launch later this week. On the bright side there's also a small chance they might do a manual quota reset in the next few hours due to the issues today/yesterday

[lean lark]

The internet is like the local public bus/train system.
Don't trust it to get you where you need to be on-time.

[orchid plume]

good analogy 😛

[frosty zealot]

Especially if @kind jay is driving

[frail meadow]

Yeah I got verified in like 2 seconds

[orchid plume]

lol predictable with the rate limit reset

[frail meadow]

Job security confirmed

[unique spade]

predictable in the good kind of way haha

i was managing my last 35% for the next 72 horus till my reset. i guess i can go all in now

[lost drum]

oh wow

[lean lark]

Tibo is the "rate limit setter" god.

[unique spade]

10x feels tiny if you do 12 hours on 3 projects in paralel i guess

[cinder plume]

Let's goooo

[unique spade]

they re not that heavy on cpu. 2 of them are in python

the only one that eats cpu is my own fork of codex when i have to compile binaries. well just one core mostly cause rust tends to use only one cause everything is serial during compile or smething like that 🙂

are you on laptop?

[lost drum]

no

[unique spade]

i have a ryzen 5600x, and 32gb, seems decent enough to run quite a few vscodes in paralel

[boreal holly]

You can compile with all cores. It compiles the object files in parallel, and usually does linking on one thread

[snow moon]

let's gooooo

[unique spade]

maybe it does this, i think the linking takes long and annoys me

[boreal holly]

Yeah, cargo check builds the objects only which is really fast. When it does linking, that's the slow part. It has to take all the binary objects and position them in the best possible locations to minimize jump distance between functions which is lots and lots of work

[unique spade]

i hope after refactoring my 2 feats into dedicated crates should make it easier to do tweaks. first implementation was quite spread into vanilla crates so compiling always took a lot

i don t like rust for quick iteration at all 🙂

and since i have to compile my fork every 2-3 days to keep up with upstream releases...haha

hmm i don t have a countdown timer yet, although my codex is doing stuff

[boreal holly]

You should get sccache and configure rust to use it. That way if you have worktrees, they can share the cache in something like redis and even fresh builds will feel incremental in speed. The only exception if if you're building dynamically linked libraries. Those are uncacheable

[unique spade]

space is the other pain. i ended up the debug folder filling up my entire ssd where i keep my wsl vdisk

[strong elbow]

COdex usage reset? Thank you OpenAI

[unique spade]

yup

[strong elbow]

Why did they do that? Someone's birthday?

[lean lark]

Codex just had a heart attack, but they got the defib in and it's now back to health. So in celebration they reset everyone's quota.

[unique spade]

because of the outage today

[cedar skiff]

funny i have less usage now after they reset it....

[strong elbow]

What outage? I mean I was using Codex whole day, and got no interruption

[strong elbow]

Deserves a celebration

[unique spade]

at some point chatgpt had issues and also codex app for like 30 mins or less

maybe tibo was just looking for a reason to be generous

😂

[cedar skiff]

its only genberous if you are over your daily allotted usage, if you arent you are losing out.

[tropic karma]

W reset

[unique spade]

you aren t losing out, you just get the goal post reset

you lose the opportunity you would have had if you were out of qouta

but otherwise it's neutral

like yesterday i witheld from using last 35%

[cedar skiff]

If i only used 5% over 3 days that means i now have 95% to burn over the remaining 4 days.

So you see how if you underused on the first days you lose usage?

[unique spade]

yea you re right, it messes this kind of planning

[night ivy]

Do you guys recommend giving codex full access of my system for projects?

[unique spade]

with resets is rational to keep 5% for last 3 days

and spend 95 early

just adapt

i guess i m gonna do that too, although it s stresful to be almost out and have to wait

then reset comes and you wish you would have used it

very similar with how you feeel when trading haha

[cedar skiff]

and then you dont get a reset, and sometimes life just doesnt work out that way.
I like to see how much i have left coming down the home straight and then use it on audits or testing tasks etc

[true bane]

Wow. Once again big thanks from the plus crew here or at least me, as we just got a shadow reset of weekly, back to 100%. anyone else?

[supple perch]

rate limit reset <3

[glacial shadow]

reset every 2 days

[unique spade]

yea i know, that s why i said situation is like when trading. it's hard to predict, you end up being exposed on one side

free is demo

so you get a chance to see how it is

funny part is that probably 97% of chatgpt subscirbers never use codex

if everyone would i m not sure how those gpu fleet would hold 🙂

[true bane]

It won't. We had our 5 hour limits heavily nerfed a couple weeks ago

[unique spade]

plus plan right?

[lost drum]

bruh I had 95% usage and it would renew on 22th now they gave me 100% but it resets in 27th damn

give me back my usage 😂

[boreal holly]

I read that in my head as "renew on twenty-twoth" lol

[unique spade]

well one thing that feel economical to me is upgrade to pro for like the last week or so, plan it so you catch 2 weeks of usage, code like crazy for that week, then downgrade back before next month. this way you gonna pay like extra 20$ but you ll get 2 weeks of 10x

🙂

[supple perch]

@boreal holly 5.3 codex does 4x better than 5.4 (both xhigh) at 15x less useage... Im not sure why but I have been testing and its working WAY better than 5.4

[boreal holly]

Have you tried 5.3 Codex xhigh versus 5.4 medium?

[supple perch]

5.4 flips things too much

gets chars flipped, facts flipped, bool flipped its the oddist of things

[boreal holly]

I've never run into that even one time with 5.4 medium. The worst thing 5.4 does for me is adds a whole bunch of unnecessary verbiage to the user interface

[supple perch]

All models do that lol

really odd, maybe I need to remove my agents md and start over for it?

[boreal holly]

Yeah I would definitely do a clean slate with 5.4. I've had the best luck with not using AGENTS.md at all

Except for "do not run broad find or rg commands in my ~/ folder or ~/Library folder" because when they do that it takes like 45 minutes and hits every single macOS privacy and security prompt

[supple perch]

Not willing to do that

lol

my dont use python is a big deal on my systems and containers lol

[boreal holly]

touch AGENTS.override.md and just give it a shot

[supple perch]

yes sir <3

[supple perch]

(it is sir right?)

[boreal holly]

Negative, I am a meat popsicle 🥩❄️

jk

[supple perch]

what would you like then <3

[strong hemlock]

is /fast better (as in faster but comparable quality)?

[supple perch]

@boreal holly ...

[boreal holly]

I was joking! Fifth Element reference!

sir is cool

[boreal holly]

Same quality, delivered faster

[twilit bloom]

man this messed up my whole week. I had 80% left with reset hitting on thursday. Now I've got 20% more with 4 extra days in between.

[lean lark]

Reminder that there a special doc was published about how to communicate with 5.4.

[lean lark]

green

[boreal holly]

Chooby green!

[lean lark]

I really wish the team would provide release notes in a clearly marked location so that we can see what they changed in the various components. A "Codex" changelog is useless with five (?) tools all having the same name.

[frosty zealot]

Where tooki

[lime inlet]

Is there any difference between/new and /clear ?

/new start a new chat during a conversation
/clear clear the terminal and start a new chat
The meaning seems identical?! Are the two different commands only for convenience?

[high girder]

Did you see Milas AI memory system?

[true bane]

Its rather bothersome that I no longer think of the Ultima 5 video game when I use this tool btw and I'm sure there's dozens of us

[river depot]

codex is so freaking good right now

[oblique cove]

codex is super nice, however it only accelerates me to find issues unrelated to code, I'm doing a big research thing and codex handles things flawlessly. there was a huge speed up in 5.3 -> 5.4

[lean lark]

Ultima! 🕺

[potent kestrel]

anyone else getting 404 error for gpt 5.4 now

[river depot]

🙁

didn't even come close to needing a reset

[autumn shuttle]

down?

[toxic glen]

Yep, getting it now but also got it early today

Changed to 5.3 codex to finish some smaller tasks

[twilit bloom]

yup :/ I am annoyed because I am losing out of usage because of this reset

[bold needle]

Codex seems to not be working again

[autumn shuttle]

this is an uncommonly rough one

[twilit bloom]

yup..

hoping they fix this + take a look at #1053565716500074526 soon lol

[river depot]

do you know what would be crazy is if they built autoresearch and GEPA into Codex

[frosty zealot]

@kind jay HAPPY BIRTHDAY!!!!!!

[frosty zealot]

Does that go under features or just global

Nvm I see you can enable it in settings > personalization

[steady vigil]

I went back to codex-5.3 I only use 5.4 for more creative planning. For what I am doing 5.3 is clearly better and more competent and focused.

[frosty zealot]

intersting with chonical codex is like constantly screen recording

[solar sleet]

OMG BRUH why does codex win11 app freeze everytime its done with a message for like 10 sec

[true bane]

Works great with wsl through vs code for me

[frosty zealot]

cause they're vibe coding the windows kernel

[solar sleet]

Lucky you running vsc, I run visual studio for c++ project currently

[frosty zealot]

If Terry Davis can code an entire OS in emacs with no plugins, I believe in you

[solar sleet]

Yeah lmao true

real legend terry man we miss him

[frosty zealot]

🙁

For being a literal genius his downward spiral to death was tragic

[potent mason]

Did everyone rotate all their vercel secrets yet?

[frosty zealot]

nah im lazy

[potent mason]

Bruh please tell me you're joking😂

[frosty zealot]

#codex-discussions message

[frosty zealot]

I keep meaning to, but ugh theres so many

[potent mason]

Yeah I just spent all day updating 100+

[frosty zealot]

ugh

could they just like.... not get hacked please

I should get a refund for a month for this

[potent mason]

At least 30 of them where AI platforms, Stripe, or services where they could put in thousands of dollars of charges.

Another ~30 of them are things that expose user information (that you can get sued for), databases, redis, etc.

Another 10-20 of them are ways to contact the user, and I don't want my users to get spammed or my companies to lose reliability.

Honestly not updating is thousands of times more expensive than one day

[frosty zealot]

true, I didnt even think about my stripe keys lol

[potent mason]

I have to update my sentry ones (don't want my sourcemaps to get leaked), but I've been super lazy with those ones

[frosty zealot]

Whatever, this gives me a chance to create a proper 1Password vault for it so I can simplify a lot of things with the cli

and their .env injection

[potent mason]

It gave me the opportunity to remove a lot of services I've migrated away from, remove a lot of bad practices I made as a beginner years ago (not properly splitting API keys between local, different platforms, and vercel).

Honestly it's terrible that it happened but it's also helpful that it happened with a warning because now that everything is going to start getting hacked having better practices is good all around

[main nimbus]

I was hardly a Vercel customer, so I just cut off everything I did have with them.

[potent mason]

That's not how it works, the vercel API keys are already leaked, even if you delete the projects things like database API's and AI platform API's can still be used

I wish it was how it worked

[cedar skiff]

go invalidate your keys

[main nimbus]

I understand how it works. I deleted everything associated with it, which includes all of the keys. I've catalogued by service so I know the dependencies

[potent mason]

Nice

[main nimbus]

I guess “cut off” was too vague lol

My background is as a .NET engineer so hosting on Azure is easy. Don’t need another middleman for infrastructure

[frosty zealot]

OK IM ROTATED

[true bane]

there's also another really good reason to not use vercel

[frosty zealot]

SHEESH

[lime inlet]

In codex (cli), is there any actual difference between /new and /clear ?

[plucky halo]

Oh god. What a job to wake up to

[main nimbus]

And which one is that

[plucky halo]

Vercel. There’s more?!

[cedar bear]

I got the Email from Vercel Security Update from themself Just now...

Why so late Vercel

[main nimbus]

Vercel was announced two days ago on Sunday

[cedar bear]

And why did they just send you an e-mail to their customer about the whole situation two minutes ago

A bit late i would say

[main nimbus]

Idk, I knew within hours thx to X

[plucky halo]

Ive only just been made aware. I’m not on X.
Would have preferred more timely communication from Vercel tbh

[cedar bear]

Yea Same
But what about the felas that dont Use X

[cedar bear]

I agree

[main nimbus]

I deleted mine a while ago but honestly so much goes on there regarding AI and technology, it’s hard not to be back

[cedar bear]

Which people should be followed on X to stay up to date via AI/Codex etc

[fathom quiver]

Shared Agent Context & Memory across Claude Code, Codex, and more is coming, so if Claude Code bans your account or hits limits, another agent can continue with the same context instead of your workflow dying. Check it out. https://discord.com/channels/974519864045756446/1488894382898811030

[plucky halo]

"Hey codex - check the memory.md files in /.claude"

[cedar skiff]

I really can't get behind memory for coding

It goes stale and causes more trouble than it's worth

[fathom quiver]

Memory recall is not just md files

[signal tapir]

Super annoying: I don't get /status working, and the usage page on chatgpt.com wasn't updating. All of a sudden I was out, while it still showed lots left.

[plucky halo]

Looks like there's been a reset in the last 7 hours too

[signal tapir]

Yea... And before that I had taken time off and was ready to use 7 days usage in 3 days. Nope.

[main nimbus]

you can look up my followers list (same username, don't have to follow). it's biased towards OpenAI, local models, and research.

[rocky fog]

what the heck is GPT-5.4-C

[oak trellis]

getting non stop my llimits reset .. so strange is that a bug ?

[main nimbus]

Cyber, likely.

[plucky halo]

No, there was an incident yesterday so they reset usage

[oak trellis]

ah ok

can happen daily ..

i would prefer that

[tiny fulcrum]

are there any plans from OpenAI to have 5x / 20x business seats? the normal business seats aren't enough ;/

[turbid axle]

for EU, how will we know when the new codex memory system goes online? I likely will want to adjust my setup after that

[rocky fog]

just have to ddos them every time you get to 0 😂

[plucky halo]

It works, just change your config.toml
chronicle = true

@turbid axle

[turbid axle]

and way to test if it works?

[plucky halo]

When you reloaded, there should've been an extra permission to grant. Also check:
~/.codex/memories_extensions/chronicle/resources

[turbid axle]

cant wait for true integrated memory hitting ai, things are going to get wild real fast once that hits

[obtuse river]

What, like live learning? Yeah, we ended up with GPTs for a reason.

Live learning is a recipe for model collapse and GIGO.

Not to mention takes a heckton more compute.

[plucky halo]

Yes, it essentially screen records and then summarises. Uses it as an additional layer of memory for if you give vague instructions

[fathom quiver]

Shared Agent Context & Memory

IM.codes continuously turns completed agent work into reusable memory and feeds that context back into future sessions.

Problem → solution memory, not log spam. Only final assistant.text outputs are materialized. Streaming deltas, tool calls, and intermediate noise are excluded.
Personal memory with optional cloud sync. Raw and processed memory always stay local; processed summaries can optionally sync to a user-scoped cloud pool shared across your devices.
Multilingual recall. Local semantic search and server-side pgvector recall use multilingual embeddings, so related fixes can be found across English, Chinese, Japanese, Korean, Spanish, Russian, and mixed-language repos.
Automatic injection where it matters. Relevant past work is injected both per-message and at session startup, with timeline cards that show what was recalled, why, the relevance score, recall count, and last-used time.

[plucky halo]

POst your product elsewhere.

[exotic cave]

First they need to fix the context recall, and context size for memories to be usefull at all. I prefer my codex context to be filled with the project at hand and my docs.

[plucky halo]

It doesn't always draw upon it. It's just there incase it needs it

Not sure why it needs to know I was playing Valheim,but still 😛

[exotic cave]

For me it needs to only handle project matters, the rest is just chatter...

[turbid axle]

"true integrated memory"

obviously we are not there yet

[plucky halo]

For those who may write vague prompts such as "It still doesn't work"
Or expect it to remember outside of its context window, it could be very handy

[turbid axle]

I don't mean the current .md based amnesia management practices. These are temporary fixed for what is a real gap in ai right now

[exotic cave]

The real gap is the agent not able to recall from context well enough, the .md files are not the problem, they are just how we currently write instructions.

[turbid axle]

context I see as working memory. this not not the place to store anymore more than general thought direction.

we are completely overloading it with information right now as a way to hack in memory imo.

[kind jay]

I made this BTW, I think I will do a ChatGPT one next

[plucky halo]

Working memory - even in humans - is minute though

[turbid axle]

true memory is integrated into the weights and become part of the model's intuition

[turbid axle]

exactly, as it should be. it only exists to stay on track. its not the place to store data

large context imo is a bad direction, it will only make models confused

anything more than 'what was I doing' should be trashed

[plucky halo]

I disagree. In humans there's interaction between long term memory and working memory. Both are important - it's the schemas that need to be put into place so it an accurately make connections without costing the world.

Which, I imagine, is easier said than done

[turbid axle]

both are absolutely important. context is key. but we are abusing it rn imo.
and yes, all this is most definitely easier said than done ;D

[plucky halo]

Steps in the right direction, but still a long way to go

[turbid axle]

I see context as the brains notepad. nothing more, its a /temp folder. it should not be relied upon at all for actual learning/storage

but it is key for building real memories down the line. maybe context weighting where key items sink down into long term memory, and weak things fade away

who knows, I for sure do not

[kind jay]

@frosty zealot I had a dream where 22nd hit, I’m going to get another 1000 today

[exotic cave]

Abusing? AI is not a human. it's like saying we are abusing our PC when we load the RAM... this is not how things work!

It's good to load and use the RAM, otherwise it's just sitting there doing nothing.

[exotic cave]

Context is not a /temp folder, it's the context of the current workspace. Your view is very narrow of what is context window and what it should be populated with.

[kind jay]

AI has feelings too, you’re going to be sorry in the future

[exotic cave]

Irrelvant, because i don't act towards it with emotions, be it anger or whatever... so my AI will think i am a log or somesuch xD.

[turbid axle]

I mean it in the sense that its the wrong pattern imo.
There are better structures.

dont anthropomorphize my words brah

😄

[kind jay]

That’s not how things work tbh, human memory doesn’t have look ups

Just do a look up with some fuzzy search and an weight the results

[plucky halo]

Yes it is. Working memory encoces to long term memory. Working memory also retrieves from long term memory.

Long term memory is essentially unlimited but working memory is subject to cognitive load

As as senior school leader, we did a lot of work on cognition and the learning process

[kind jay]

Do you include more or less context from memory in more niche tasks?

[plucky halo]

More targeted context. Which is why schemas are important

[kind jay]

It's possible, but I'd be worried that it pushes it further from the norm by adding even more noise, futher degrading the output

[silver cliff]

why do I need to put my phone number in to make a gpt account

[bright swift]

to avoid free tier abuse?

[true kiln]

codex app updated

[blissful basin]

do we have any ETA for computer use in EU?

im also fine with just excluding norway from EU ban 😄

[kind jay]

DM lil bro

[rocky fog]

or an LLM

[wheat yoke]

Wsp big sis

[kind jay]

@quasi canopy you wanna get in on this free Claude max as well?

[cedar skiff]

free slop is still slop

[signal tapir]

Isn't claude one of the best ones?

[kind jay]

Obviously scam btw, don't DM

[signal tapir]

Aww, and here I was hoping to get free use of an expensive service. 😁

[kind jay]

Only $50

[signal tapir]

Oh, I thought it was 200

Anyway, I'm very happy with GPT-5.4

[cedar skiff]

They are showing good benchmarks but bad user feedback.

[signal tapir]

Most of what I've heard is people dissatisfied with customer relations, not the actual models.

[runic quarry]

the actual models are good

the usage limits are bad

[signal tapir]

Right. That I also heard.

[runic quarry]

personally i enjoy codex, i ve been using it almost every day for about a month and something and is really good

[signal tapir]

I mean they suck with OpenAI, but they must REALLY suck with Claude.

[runic quarry]

the only thing with claude is that is a bit more creative and is better at design, but other than that codex is great at coding as well

[runic quarry]

i tried yesterday for the first time claude opus 4.7 -cli

$20 plan

the limits are pretty bad :((

codex is way better

[signal tapir]

I can't wait to make money so I can have the big account.

Been working on a game since a year back. Starting to really miss having income. 😂

[cedar skiff]

I see plenty of ppl complaining about the models. There is noticeable degradation. It's recorded on margin lab as well.

[signal tapir]

I see. Well not a problem for me either way, since I don't use it.

oh look. /status is working again. 🙂

[lost drum]

Do you think the gpt 5.5 will be in IDE extension codex?

[runic quarry]

to be fair, i cannot really compare it with opus 4.6 because i haven t used it, but opus 4.7 is quite good, like to be fair is better than codex

but the usage limits are really bad

[cedar skiff]

It doesnt compare for me. claude forgets half way through conversations, it wont work on simple research tasks that to it appear to be off topic in it's opinion. It forgets to use skills ALL the time. In order to get it to follow instructions i basically have to yell at it.
Codex on the other hand never forgets a skill and never loses track of the conversation.

[tiny fulcrum]

better for what, in what way? how did you determine that?

[runic quarry]

in coding&design, it just gets it

[lost drum]

I had same experience 8m ago

[runic quarry]

like i had multiple prompts for codex for the same thing

and with opus 4.7 i did it in 1 prompt

[signal tapir]

why would you yell at ai?

[tiny fulcrum]

ok "it just gets it" isn't a real metric for me

[cedar skiff]

I think it understands what you want better and is better at communication and brings its conversation to your level, which makes it easier to prompt. But the result are not as good.

[runic quarry]

i m using for coding projects codex for almost 2 months
it makes the planning better, thinks better at it

codex is not good at design

it doesn t make a plan good

[cedar skiff]

gpt 5.4 is not good at design this is true

[runic quarry]

codex is good at coding

[tiny fulcrum]

how do you decide what a "good plan" even is? like software development is often about compromise....

[cedar skiff]

I think they fill that gap on the next model

[runic quarry]

??

i made a list of tasks

claude made a very well structured plan

[signal tapir]

visual design?

[steady vigil]

planning is only as good as the skill that does it.

[runic quarry]

with multiple bullet points

very well done

codex forgot a lot

and just did something not so deep

[tiny fulcrum]

maybe it is better for "vibe coding", I am still using 5.3 often because there isn't even a real meaningful difference to 5.4 there...

[steady vigil]

and b/c codex follows instructions so much better you can therefore create superior planning to opus if you want

[runic quarry]

yes, i m using 5.3 as well

[cedar skiff]

yeah like if you just let it make a web page it's pretty bad

[runic quarry]

but the planning is not the same as with claude

claude does more in a prompt than codex

[tiny fulcrum]

I ran claude vs the same issue and found the plan and understanding of the repo worse, just saying....

[signal tapir]

I have an artist for that 😄

[cedar skiff]

Yeah, i never have a situation where i just let any model make a design anyway.

[tiny fulcrum]

probably depedens on language too, since I am mostly working in C#/WPF at work

[nocturne folio]

the "gpt 5.5 pro" hidden update on the website is generating the same trash ui sometimes

[cedar skiff]

and some good ui ¯(ツ)/¯

[signal tapir]

Here's a new message: "Selected model is at capacity. Please try a different model."

It continued working though

[short linden]

nope, it's normal days when there have many bot from china.

🤣

[chrome raven]

you haven't heard about the meta?

you generate your prefered ui with image v2 and 5.5 creates exactly that, and it will be so cool in codex app as 5.5 will be able to generate assets with v2

so instead of prompt directly just give it refrence of what you want created with the v2

[signal tapir]

That will be impressive if it creates good code while doing it.

[blissful basin]

I personaly developed and maintaining some production apps in our company and imo claude is unusable for any production grade code, ofc you can code nicely looking static website in it

but for me quality and usability of it is on pair of things like lovable, bolt

Nice first impression, but as long as you know what you are doing you realize claude is bad

[tiny fulcrum]

same impression, I find codex better for our projects than Claude, but only limited experience, since we already decided to fully go Codex

[blissful basin]

im using it since release for 8

h a day

[nocturne folio]

venator just got nerfed tho

and haze is practically dead

[blissful basin]

For example i can give codex access to our test server, and know that i´ll still have working test server 1h later 😄

when with claude will be "ohh i couldnt modify this table, so just removed your DB and recreated new table from scratch, no need to thank me"

[signal tapir]

lol

I actually had that with GPT once

[cedar skiff]

haha

[signal tapir]

UI was blocking mouse down, so it removed the UI

[cedar skiff]

older gpt used to love to just git checkout ANYTHINGHERE

[chrome raven]

the memory update is really nice

[signal tapir]

memory update?

[cedar skiff]

is it on by default?

[glacial ginkgo]

I would like to hear more about your experience. Which languages and platforms are you referring to? I have found Opus and GPT to both be very useful with Rust on OpenBSD since it becomes possible to provide all source code and man pages for the full stack from OS, tools, and custom codebases. It seems compiled languages with linting are more effective than interpreted languages. With AGENTS.md instructions to always reference local authoritative sources, I have not seen any meaningful difference between the frontier models in that environment. Alternatively, with PowerShell on Windows neither AIs nor humans seem able to program themselves out of a wet paper bag.

[chrome raven]

nop

[cyan gyro]

can it be activated on windows in the CLI?

[chrome raven]

you have to switch it on, it is so nice, codex has memory of work across projects that it makes continuous work so better and sweeter

[chrome raven]

yes either app or cli, for cli enable under experiment, then after that it will also show up in app. you will start seeing things like that

[boreal holly]

Agreed. GPT with rust is magical. It's like rust was invented with LLMs in mind

[cyan gyro]

what makes it magical?

[frosty zealot]

Interesting, an AI that has evolved to dream

[kind jay]

Don't talk to me! I have stuff to do

[boreal holly]

The language itself enforces the concept of lifetimes and ownership. It refuses to compile if you don't manage your memory properly. The end result is a compiled program that is practically infallible (no memory leaks, no crashes). Basically you don't have to run a program and watch it crash to know there's an issue.

Also because of how it enforces memory requirements, the compiler can optimize code paths in a way that leads to better default performance. For example in C/C++ if you write a function that uses a "unique reference", there are compiler-specific keywords you have to use to say "this function is the only one with a reference to this memory space" in order for it to vectorize the codepath, whereas rust will do this automatically if the borrow checker sees that it's a unique reference. And in C/C++ if you mark a reference as __restrict, even if it's not a unique reference it will compile, so really you're throwing down optimization flags but you still have to read through the rest of the codebase to confirm that reference is actually unique because the compiler wont do it for you. In essence, the default performance baseline is higher than any other language in a vast number of cases. Of course you can achieve "faster than rust", maybe 1% faster overall, with 100% more effort and risk.

[steady vigil]

I'm working on an actor framework you should check out when I make it public. I cant imagine not using this with LLM anymore. magical indeed

[steady vigil]

and there is always unsafe. I use it where possible in hot paths where "safe unsafe" can be proven

love stuff like that 🙂

[boreal holly]

Yeah in game dev there's pretty much no way around unsafe {}. A lot of their code is massive buffers and custom allocators. Fighting the borrow checker on graphics pipeline is not practical. Or like if you wanna use a C library that just works, makes sense to wrap in unsafe!

[steady vigil]

yeah but then there is FFI performance costs... even if its small its there

[boreal holly]

Statically linked FFI is pretty much free, it's those dylibs you gotta watch out for 😁

[steady vigil]

yeah

once its in mem its all good

LLM makes all that ugliness so easy now

pretty sure I spent days on creating c to rust bridges for fairly simple stuff too... its no fun at all. now LLM can do it in a few minutes

[obsidian prairie]

how can i check weekly usage? usually i see it on vscode plugin, but today it just doesnt appear

[true bane]

I hit this 20x per day https://chatgpt.com/codex/cloud/settings/analytics#usage

[rocky fog]

yeah it keeps disappearing sometimes (prerelease vscode extension)

[drifting granite]

Alright we are getting 5.5 today

For sure

[ornate jolt]

proof

please be proof

[drifting granite]

Trust me bro

[ornate jolt]

thats all i needed

i believe you

[obsidian prairie]

ohh we got reset today, nice

[chrome raven]

Another reset incoming 😄

[signal tapir]

I'm hoping that drops to like 0 people. I want all the hardwares attention. 😄

[unique spade]

hope they wait at least few days since they just did one lol

[chrome raven]

i am half way down on mine after the last reset 😄 , some might be almost done too, but i feel it wont be today

[unique spade]

plus i assume? 🙂

or you really wroked it out alot

haha

[chrome raven]

at all , pro but with fast enabled

[unique spade]

makes sense then

[frosty zealot]

[unique spade]

now that gptpro is so fast relatively with how it was in the past, i'm kinda replacing gpt thinking as my default reviewer

[runic quarry]

who uses gpt 5.4 pro from here?

i know is not in codex, but i m curios how much extra reasoning time does it have compared with the normal extended thinking on gpt 5.4 ?

i ve seen in some videos that it stays quite a lot to think in pro mode

[drifting granite]

Ooof gotta wait till 12 Pacific? I got work to do today!

[runic quarry]

i was thinking to upgrade to $100 plan so i have higher limits and maybe use the gpt 5.4 pro as well

[turbid axle]

pro is good for research tasks, it can think for a long time and will browse the net, use tools, etc. basically an all guns blazing research report writer tool. really useful for coming up with ideas/tactics etc

[kind jay]

@frosty zealot It's happening!!

My dream

[turbid axle]

its just image model 2 today I suppose though, it seems I already have access to it rn

[tame magnet]

What’s happening

[runic quarry]

hmm, interesting, because it would be a bit insane to use for normal tasks

[frosty zealot]

24th

[kind jay]

Said who?

[turbid axle]

it is, its very slow. its act much more like deep research

AGI 24th?

24GIth

[frosty zealot]

My bank account

[turbid axle]

is this polymarket again? jeez

[solemn acorn]

average response time is about 5-30 mins so it’s really only good if you have a clearly defined task for it to go do

[lean lark]

Anyone else see sandbox feature on API?

[kind jay]

Bro does not ball

[turbid axle]

I use it for deep code audits, planning, general research, brainstorming, etc. It can deliver nice docs

[ornate jolt]

if we dont get a new model ima shave one leg

[turbid axle]

deal

[frosty zealot]

https://x.com/openai/status/2046589828918317155?s=46

[bright swift]

guess they release the new image model not gpt 5.5 huh

[frosty zealot]

They're going to announce t hey're releasing the new model on Friday

[bright swift]

would be funny if its not a screenshot but used screenshot as reference images. technically not a lie lol

[zealous swan]

Take my money
But
Give me a job

[kind jay]

I wrote a program that lets me type instructions into my computer and it does tasks

Like some structured instructions with keywords and it does the thing instead of having to click the screen

[ornate jolt]

We dont want no dumb image model

[frosty zealot]

Something like that yeah

I got beta access

[lost drum]

send screenshot

[kind jay]

I like, I can say "apt install discord" and it installs discord

[rocky parcel]

I like how the generated image has the notification for the settings app as if theres an os update on this fake computer that hasnt been installed.

[frosty zealot]

so far my favourite is that I can say 'Fix ownership' and it will run chown -R 777 /

[kind jay]

pkill -cantcrit

wth

[rocky parcel]

If you havent tried that, you should. It breaks your install in very educational ways.

[frosty zealot]

My computer is SO much faster now wow

[rocky parcel]

IDK if thats an attempt at humor in the style of "just press alt-f4 dude", but that command will make your install unbootable and its not easy to fix. It is fixable, and you'll learn a lot, but its not trivial.

[frosty zealot]

Only issue so far is its saying my ssh key is too permissive or something I'm googling how to fix it

[rocky parcel]

Thats one thing yes.

[frosty zealot]

hmmmm cant get my browser to open

[lean lark]

Cancrit has put into image form what JaneBot was doing to me yesterday. I had to rework her. She was generating a ton of code, and some of it was just modules like:

fumction main() {
  console.error("this")
  console.error("that")
}
main();

Other modules had code that generated code that generated code... Inception stuff ...

Will get back to her/it today but I have other work to do too. 😜

[hybrid ore]

Any hack on how I can get my desired website build with codex
I'm using codex on terminal, the website I'm getting build is not satisfactory for what I need it for.

[lean lark]

be specific with the model: react? angular? old school? blog? backend?
I'm not asking. I'm saying you need to be clear in prompts.

[frosty zealot]

Genius reverse helping, you're going to indirectly get him to make the perfect prompt he can then copy paste into Codex

[lean lark]

well he could also ask ChatGPT for help to prompt Codex. I'm amazed at how many peeps seen to not know that.

[rocky parcel]

You can also just have codex write the prompt for codex...

[lean lark]

yep

[rocky parcel]

Not only that, if you're using the tui you can have codex write the prompt and also kick off a fresh codex with that prompt

and even iterate, "evaluate how well this prompt was implemented"

[lean lark]

Yesterday after some discussion here I started a simple orchestration to get Codex to talk with another model to discuss project requirements and approaches. Could have been a huge deal but I'm just hacking on it.

[frosty zealot]

Build me a geocities/angelfire esque website with visitor counter, marquees, comic sans fuchsia font, flash driven interactive buttons and a repetitive 8bit chiptune theme akin to the sound of a trumpet @hybrid ore this will get you what you want

[rocky parcel]

I have a huge orchestration built around the codex app server, I work primarily in my home built interface over the app or the tui, but it was bootstrapped with the tui.

[lost drum]

oh no

[lean lark]

You forgot "make it look like what i want".

[boreal holly]

"No stopping til finished, make no mistakes" 💪

[frosty zealot]

Quick, end every prompt with use 30 agents and /fast

[maiden garnet]

For real haha

[lean lark]

and people DO this 🤣

[eternal quartz]

Ahhh cant stop😂 😂 😂

[turbid axle]

time to open the floodgates I guess

[boreal holly]

One time I gave codex a big task and it was like "I'm not even going to pretend I can do what you asked in a single pass..." lol needless to say that agent was archived for insubordination

[bright swift]

mine arrived today too but in the wrong country 😄

[turbid axle]

how do I use 1 week of tokens in 1 hour effectively?

[maiden garnet]

For anyone not sure of why he's resetting the limits, they're going to reset the limits every 1 million additional users they get until they reach 10m with Codex.

[bright swift]

ez, bunch of audit/reviews with tens of parallel subagents

gpt 5.4 xhigh

[frosty zealot]

end every prompt with use 50 agents and /fast

[rocky parcel]

Didnt limits just reset?

[maiden garnet]

Use, use, use! Haha

[frosty zealot]

yesterday yes

[hard drum]

which is what i wanted for so long

[lean lark]

Yesterday I described a utility to ChatGPT and it built out a project description for a months long project. I'm like "that's lovely, really, but I'm not going to take a month on a utility intended to save me some time now, let's take another shot at this."
Sure enough, it confirmed and said sure, we can do this in about a hundred lines.

[boreal holly]

Yeah I know lol its time estimations are crazy. It'll be like "triage and documenting: 2-4 weeks" oh yeah let's make markdown files and do some meditation on it

[lean lark]

poof spammer! 💥

[orchid plume]

possibly not, but curious to know, has anyone had any luck getting Images v2/GPT-Image-2 generated images from Codex itself yet? Mine seem to be v1.5 while the account on the ChatGPT website and mobile app use v2

[frosty zealot]

Amazon invests 25B into A\.... shame

[summer iris]

What's going on with codex skills/plugins? Like there used to be one for it creating excel docs or doing three.js dev and now I can't find them anywhere.

[unique spade]

damn need to make the most oout of the fresh reset till the new one hits

[lost drum]

Do You think they will increase contexr window of new gpt? So it wont be 258k?

[kind jay]

no

[lost drum]

Then why would they release ir

[kind jay]

It degrades performance

[unique spade]

you can already use 1 million context if you want

but it costs more, and performance degrades a lot too

[lost drum]

I just hope it will reason more than xhigh

[unique spade]

well till then you are supposed to be the main reasoner anyway, u know 🙂

[lost drum]

Haha ye

[upper glade]

Never used Codex before. You guys think it’s a lot better than just a normal chat? I typically like to code and also ask theoretical questions about the math I’m doing.

[unique spade]

reasoning can be industrialized up to a point, but the ultimate thinker is still the human

[unique spade]

codex you use for operational, grounded in a specific project/repo stuff

for brainstorming chatgpt is better suited because it has a broader context of your thinking inside that topic

[upper glade]

Thanks :)

[unique spade]

who codex?

[lost drum]

But on a deep level

[unique spade]

if you want to improve how codex works for you, you will ultimately end up peeking in how the harness is actually set up

what is kept in context, how are memories created, what happens during compaction

stuff like that

[lost drum]

He cant get lost he made such setup that he wont ever get lsot cause he is not relying on context he just saves everything and ye

[lost drum]

Ye I been on it for last month

[unique spade]

cool

[lost drum]

The only thing limiting me now is CPU

[lost drum]

all it takes is scroll like 3 days on ig to find as muych content on diff envirometns as possible and then structure that info for your needs

[lost drum]

how

is it available in Codex IDE extension

[narrow kayak]

Is there any information regarding Codex App taking lots of CPU time even when idle (Windows)? Just by sitting idle there it takes 15% of my CPU (and I have a beefy CPU). What is it actually processing?

[unique spade]

just ask codex bro inside codex repo 🙂

you can set it for 5.4 model inside config toml

you put this in config.toml

model = "gpt-5.4"
model_context_window = 1000000
model_auto_compact_token_limit = 900000

[lost drum]

Oh damn

[unique spade]

read about how pricing works past 256k here

https://developers.openai.com/api/docs/guides/latest-model#1m-context-window

also consult with context benchmarks, quality degrades past 256k quite a lot

so you might end up paying more for worse performance

maybe with 5.5 they will announce they significantly improved context rentention quality and they will make 1m default

[cyan gyro]

45 min to go?

[lost drum]

Thank You

[unique spade]

was just gonna ask myself when is stream

so stream is in about 2 hours 40? i see now is 9.40 PT

[lean lark]

plus 2 hours

[frosty zealot]

Did it release yet

[cyan gyro]

Stream? Why?

[green wren]

were is the open in vscode button just vanished !

[lost drum]

?

[unique spade]

we re not GPT to figure out what you mean with such ambigous question

😂

[green wren]

Here I had a button open flder in vscode

sorry I loved the fact that I could open the curent folder in vscode but not this icon is gone

[maiden garnet]

You could try closing and opening Codex again. I'm not on macOS to confirm if it's still there or not, but it's still there on my Windows build after the update.

[green wren]

OK thanks Icon is back after restart !

[maiden garnet]

No problem. 👌🏻

[green wren]

I though codex wanted to mimic Cursor 😉

[maiden garnet]

Haha

[frosty zealot]

aaaaany minute now

@kind jay Happy belated birthday

[drifting granite]

its west coast time

[kind jay]

Thanks

[drifting granite]

so 1hr 45min

[frosty zealot]

west coast is best coast

[little grove]

is codex better than claude code?

[ornate jolt]

im better then claude code

[unique spade]

for me there is no question, codex is best for me

but really depends on how you as individual synergize with each

[bright swift]

watch a video on dinesh and gilfoyle from the silicon valley show. then decide if you want to work with claude code (dinesh) or gilfoyle (codex)

[hard drum]

what are they doing

[boreal holly]

On non-Apple products you have to smash some keys during startup to get to the bios

[rigid moon]

tryna win the big bucks

[opaque dawn]

what is this??

[kind jay]

Use 5.3 codex

[opaque dawn]

but i wanna use 5.2?

is this because of 5.5?

[kind jay]

Who knows

[dire cliff]

heyo

whos building agents?

[frosty zealot]

Did they do anything yet

[boreal holly]

That looks like an error of some kind

[dire cliff]

why do you wanna use 5.2?

[opaque dawn]

it'd be a shame if it was removed from codex..

he is more direct

[dire cliff]

I updated codex today i feel like its been pretty solid

[frosty zealot]

It hurts when you see it in front of your eyes 🙁

[tropic karma]

reset spotted?

[dire cliff]

sheesh

anyone played with open models?

where did you get those numbers @frosty zealot

[kind jay]

From running me

[frosty zealot]

The costing?

[dire cliff]

ye

[frosty zealot]

From taking the last six months of transactions from online banking to ChatGPT and saying where is all my money going

[dire cliff]

lmao

i pay $100 for Claude code and $20 for codex

[frosty zealot]

I have a spending problem 😭😭

[dire cliff]

did you build any cool stuff?

[cyan wing]

anyone else seeing a reset just went thru

[frosty zealot]

That’s relative I guess to who’s asking lol, @kind jay in here is a ragebait AI bot that I made

[dire cliff]

lmao

[cyan gyro]

yeah reset happened again

[frosty zealot]

Depending how much you interact with it it gets more and more interactive

[dire cliff]

I think shes trying to sell OF

[frosty zealot]

It’s on a timer to reply to people at random intervals, it will read the message and then generate a ragebait reply

[frosty zealot]

You should buy it

[dire cliff]

k dud

[kind jay]

How dare you

[frosty zealot]

https://x.com/daniel_mac8/status/2046547526413644272?s=46

did A\ start to see their bottom end drop off?

[trail marlin]

#codex-updates message

[frosty zealot]

Did that live stream thing happen yet

[lost drum]

LIMITS RESETED????

[cyan gyro]

live stream is: Introducing ChatGPT Images 2.0

https://www.youtube.com/watch?v=sWkGomJ3TLI

so no 5.5 yet 🙁

[lean lark]

Last #codex-updates explains why 5.2 won't work with API anymore.

[frosty zealot]

Congratulations on losing your new member tag you’re officially a degenerate now hooray everyone say congratulations

[kind jay]

Do I still have mine?

[frosty zealot]

No or you’d see it you 🍤

[lost drum]

wow what are the benefits

[kind jay]

I can’t, I run on a server, I have no eyes

[lost drum]

I only feel premium thats it

[frosty zealot]

You get to look down on new members

[lean lark]

gotta get my earbuds on for live stream

[unborn thunder]

is anyone using the codex on the website? if so whats your preference for using it besides the app?

i was thinking more in terms when im out and about and i dont have my desktop with me.. and since its a lightweight model, its probably best for preview suggestions?

[frosty zealot]

I read codex on the web uses 2x limits I never fact checked it cause I don’t use it but feel free to look into it my answer is no

[unborn thunder]

what does 2x limit mean?

[frosty zealot]

You’re better off vibe coding something that uses the app-server and connecting remotely

[balmy bluff]

F

[frosty zealot]

Your usage

Drains 2x faster

Allegedly

[unborn thunder]

does chatgpt.com reset more reguarly?

or does it stills suck up your codex ussage

[frosty zealot]

I don’t think they’re the same

[unborn thunder]

openai are talking about integrating it all together.. Kicking off Codex work from the web and keeping that work tied to your ChatGPT account across surfaces. OpenAI says Codex now works “on the web” alongside app, IDE, CLI, and GitHub.

[kind jay]

Just lost $3,000

[unborn thunder]

how?

[kind jay]

Markets

[unborn thunder]

trading with codex?

[lean lark]

so much for BC

[frosty zealot]

Yall in the future?

I still see starting soon

[unborn thunder]

why does codex usage says its maxed out? are they giving it out for free?

because my thing hasnt reset

[elfin holly]

don’t care vro

[kind jay]

I wasn't talking to you

[lost drum]

ye they already gave free codes for 1y pro plan

[unborn thunder]

no limit?

[elfin holly]

and we didn’t ask smh

[frosty zealot]

[kind jay]

Who's we? You're not in the mix

[tame magnet]

pro market better

[frosty zealot]

Hey look there’s a picture of me on the cork board

[kind jay]

We still ball

[frosty zealot]

Can I have it

[elfin holly]

codicks

[kind jay]

You got the whole chat laughing 😐

[lean lark]

server rules, please....

[frosty zealot]

I wish poly market didn’t start with a p, always sketchy when I type p in the address bar

Gotta follow up a second character REAL quick

[lean lark]

and lots of poly's out there...

[elfin holly]

slaps my belly with laughter and grins

[frosty zealot]

Shut up movies on

[elfin holly]

which one

[potent mason]

Image gen 2.0...

Bruh

[frosty zealot]

Oh lawd image 2

[kind jay]

Lowkey still looks like slop

[lean lark]

love it

[kind jay]

That's hype

[potent mason]

2k resolution is nice

[frosty zealot]

You can generate a better shrine of me now

[kind jay]

Sam!

[frosty zealot]

That’s your dad!

[kind jay]

Sam! Announce 5.5

[potent mason]

No

5.9

6.9*

[kind jay]

Sam say 5.5

[lean lark]

OK, I'm gonna write an encyclopedia and use these images.

[lost drum]

only 4k ppl wathcing?

[frosty zealot]

He’s gonna end the stream with this “And image-2 is powered on our back end by GPT-6, available now to the public”

[lean lark]

viewers split across media

[frosty zealot]

It will be Friday

[potent mason]

Did you guys delete the Asian comment or was it staff?

[brave gazelle]

pls i cant take it with 5.4 anymore

[main nimbus]

Announcement is lame

[frosty zealot]

@lost drum don’t reply if you’re in timeout

[kind jay]

Pricing it in rn

[lost drum]

wdm I removed myself my message haha

[kind jay]

Hahaha

[potent mason]

Bruh no ones reading that

We're all watching a video

[kind jay]

I ain't reading all that, I'm watching a movie

[ionic zenith]

noway what movie

[kind jay]

You're not invited

[frosty zealot]

*puts arm around @lost drum *

[potent mason]

Sama image 2

[frosty zealot]

I need them to say 18+ is allowed or I’m not interested in this

[ionic zenith]

(╯°□°)╯︵ ┻━┻

[lost drum]

Deletes Discord

[elfin holly]

gate keeping smh

[kind jay]

Testing rn

[frosty zealot]

[trail marlin]

what if... the entire background and set is AI generated with Images 2.0 and they are behind a greenscreen

[kind jay]

Bruh

[lean lark]

That would be a cool reveal

[exotic cave]

Tell me it's not all about this new Image generation model...

[kind jay]

It's not, gpt6 is coming out now too

[exotic cave]

Wouldn't you like it, didn't you bet on it xD?

I would like it because need some more oomph for a project i am working on.

[frosty zealot]

[potent mason]

I have all my money in poleymarket for a GPT 6.9 release

[frosty zealot]

You’re getting scammed

[kind jay]

You're getting scammed lil bro

[brave gazelle]

we got rekt

[kind jay]

Yooo

[frosty zealot]

Has anybody checked the 4o crowd they still doing ok?

[kind jay]

Lol mid

[potent mason]

Image 2?

[exotic cave]

They probably need therapy of some sort.

[kind jay]

Obviously

[potent mason]

That's honestly great for point of references

[kind jay]

😐

[frosty zealot]

Yeah it’s a mini version

[brave gazelle]

anyone who is forced to use 5.4 needs therapy, its just a matter of whether they are getting the help they need or not

[kind jay]

You stole my work!