codex-discussions

[bright swift]

(vendored sth lol)

[plucky halo]

I'd be slightly worried if a school had 20,000 exclusions 😛

[lost drum]

where to access that window

[bright swift]

its just a commit view on github

[bright swift]

thats vscode not github

[lost drum]

ye

[hard drum]

???

what were you doing there

[cyan gyro]

5.5 is being tested in the app

[hard drum]

committing node_modules?

[bright swift]

vendored a fork

[hard drum]

:/

[frosty zealot]

Ignore your node_modules already

[twilit bluff]

you need github account to save your work

also, that's a lot of changes 😆

[lost drum]

why would I save it? isnt it saved on disk lol

[hard drum]

real numbers are 10-13k change commits on a codebase that involves: compiler for a bytecode-based language with a stack-based virtual machine + runtime, stdlib, vscode-extension, && CF Pages-hosted website

That's called production, not "i forgot to ignore node_modules"

[twilit bluff]

my bad, wrong phrasing. if you want it persisted online I meant and clone it anywhere

[lost drum]

ohh I see nahhh its offline project

[frosty zealot]

make a .gitignore and just put * in it then so you cn get rid of the annoying 1k+ icon lol even if you dont plan on pushing it to github

[twilit bluff]

you usually don't have to include the dependencies unless you vendor

[bright swift]

i just tell him to not make mistakes

[lost drum]

I might buy pro but no money

[lost drum]

"be AGI so you wont make mistakes"

[hard drum]

"pls fix"
"make no mistakes"

haha

[bright swift]

UserPromptSubmit hook -> additionalContext -> "make no mistakes" -> ez mode on autopilot

[fast iron]

My last request to this agent caused codex to think for 20 hours so far... should I be concerned?

[exotic cave]

And what was your impression?

[lost drum]

Is he editing files?

[fast iron]

not as far as I can tell

[lost drum]

Lol

[lost drum]

I had an issue like this I just stopped him and told what happened and that he must continue on what he left on

[fast iron]

This isn't the first time its thought for multiple hours, but this is the first time it has thought for almost 24

[lost drum]

What the hell you building💀

[hard drum]

the next gpt lmao

[bright swift]

probably was a display/harness bug

also with long reasoning you can run into autocompact and it can never finish "the thought"

seems better to get started doing something before first compact

[plucky halo]

"What was I doing OH! YEAH! That was it...What was I doing OH! YEAH! That was it..."

[bright swift]

yea exactly... i ran into this when building a custom system prompt for gpt 5.4 in opencode

[plucky halo]

How do we give Codex Ritalin?

[bright swift]

tfw you are blocked from starting new stuff because of a large refactor

[lost drum]

Wdm

[hard drum]

i haven't even fully gone there yet , but holy smokes

[lost drum]

Whats this

[hard drum]

programming language + website + vscode extension

[hard drum]

you get more weeky usage, but not by much if you do a lot. you may exhaust yourself in 3-4 days before reset

5.4 xhigh overthinks a lot. don't use it unless EXTREME RARE case where it could work

use high+medium

high to plan/orchestrate, medium to impl code itself

i'm on 20x...

[kind jay]

True

Double down, buy more

[lost drum]

82b tokens

wow

[hard drum]

that is just some weird token cost esimator. no, am a 185-ish eur 20x sub user

999pln?

ty polski?

[lost drum]

No ajak

[hard drum]

oh ok

[frosty zealot]

Me when GPT-5.5 drops

[hard drum]

see? 20x

that's a bit of a delude think right there

[lost drum]

I see I see

[hard drum]

oof

[plucky halo]

All of these PRs with huge changes are terrible practice...

[hard drum]

I'm aware. most of it was website + docs change

[granite jay]

Any ideas ?

[hard drum]

but I know nobody actually cares about my repos, so I can do whatever I want 🙂

[granite jay]

The UI says it is not in planning mode.

Turning it on and off doesn't help.

[dawn seal]

I have Pro plan and I want to use OAuth Codex for my games npcs

is that something I'm allowed to do?

https://github.com/EvanZhouDev/openai-oauth

I believe this is how open claw does it for codex?

I want to do something like this for my application

OpenAI's Codex CLI uses a special endpoint at chatgpt.com/backend-api/codex/responses to let you use special OpenAI rate limits tied to your ChatGPT account.

[bright swift]

openclaw was allowed to do that so that steipete would go to them instead of a competitor, dont know if you have the same leverage 😉

[dawn seal]

I mean repository exist

other people must be doing it as well?

[bright swift]

sure until they get their accounts banned

i wouldnt rely on it, its pretty likely not tos conform

[dawn seal]

urgh I want an official answer

I wish I could get one

[bright swift]

your official answer is in the tos, its forbidden

[dawn seal]

well, they do allow openclaw though

[bright swift]

even openclaw is clearly forbidden according to TOS

[dawn seal]

why are they just allowing it though then

[bright swift]

flavor of the month

[carmine copper]

There are codex cli forks that use that and haven't heard of a ban. The usage even shows up as "Other" in the dashboard. But I don't know.

[dawn seal]

yeah this is what I thought to

[carmine copper]

Codex, does TOS allow that?

[bright swift]

i mean i use a small codex fork too because they are just too slow to implement some stuff. so strictly against TOS too.

problem is those TOS are always set up for maximum strictness, mainly targeted towards free tier abuse etc

so youll never get a straight answer

[carmine copper]

But I wonder. If they detect it ("Other" usage), why don't they block it?

[bright swift]

hm interesting just checked mine shows up as CLI not other

but i use it "normally" not as proxy

[carmine copper]

When I used a codex fork for a while, it was Other (grey color)

Not as a proxy, as coding agent TUI

It added some tools of its own to the tools section of the config prompt or whatever it's called

[dawn seal]

I think what they do not want is outsourcing their service in a form that hurts their ecosystem

like making a coding agent program and selling it

what I want to do is,
some sort of a llm driven life simulator

a game

I don't think it should hurt them

[bright swift]

i'd test sth like gemini 3.1 flash lite for stuff like that, very cheap

or even 2.5 flash lite

or fine tune your own based on the new gemma 4

[solemn acorn]

you can build apps on top of codex using their app server but openai may ban or limit usage of third-party codex clients

stuff like T3-Code is above-board because that's still using codex under the hood via the app server

and it's even fine to plug the app server into stuff like openclaw assuming your openclaw usage abides by the ToS

[tropic karma]

Yep, windows

[orchid plume]

works for me if I click it, but it is incorrectly greyed out

[tropic karma]

The main thing I want fixed though is for it to take less than like 3 seconds to switch between long chats

[orchid plume]

yeah there's still some performance issues but hopefully over time they will improve those

[compact dawn]

man usage burns so fast now

[carmine copper]

What about Every Code? It's been active for some time and it's not using codex app-server.

[solemn acorn]

I wouldn't be surprised if it was against ToS to use your chatgpt subscription with it assuming openai chose to enforce it

you're supposed to use the app server to handle auth

[short linden]

you can't use the subscription model with other tools, please use real API from https://platform.openai.com that's design to integration with 3rd party tools.

[hard drum]

pretty sure OpenAI allows.
see OpenCode, for example. && I did that with a custom harness attempt once.

It's Anthropic && Google that hate you

[lost drum]

Yooo bro as ou have the pro sub is there gpt 5.4 pro xhigh available? have you tried it?

[plucky halo]

5.4 pro is only available via API on codex

[lost drum]

oh

then we wait for 5.5 codex

[hard drum]

Not tried 5.4 pro because too expensive && quick to drain

My workflow already makes me go into weekly deficit

[plucky halo]

Why do you use the logical AND in conversation?

[hard drum]

Yes, but do you not realise how stupidly expensive GPT 5.4 Pro is?

[lost drum]

ye like 10x

[hard drum]

it's both logical/bitwise. logical conjunction. It's 1 less letters/symbols

&& it's distinct enough where you can read && parse it fast

[plucky halo]

Bitwise and is &, isn't it?

Also people know what an ampersand is, you only need one if it's to read quickly, surely

[hard drum]

that's because C is a badly designed language w/ a terribly weak type system

CPUs don't care which one is logical || bitwise

for CPUs, they're one the same

[plucky halo]

You're not speaking C though, you're speaking English in a chat with humans

[hard drum]

I'm an EU customer

[hard drum]

Point being?

[plucky halo]

It's weird af

[hard drum]

That's a you thought, not a universal thought

[plucky halo]

If anything, it reduces readability. Same keys to write 'and' and 'or' (fewer for or).
IF it's quick readability, then one ampersand is enough

[hard drum]

You're in codex space, friend.

I'm sure you understood &&, so keep your "it's weird af" to yourself, aye?

[plucky halo]

I'm well aware which channel I'm in. JUst a very weird thing to do for reading speed

[hard drum]

Fair point.

[plucky halo]

Didn't mean to be rude - sorry for that

[hard drum]

You weren't rude.

[hazy gulch]

I tried using codex 5.2 with a good and long promt but it gave me this error.

"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.2-codex' model is not supported when using Codex with a ChatGPT account."}}

Cant i use codex even if i just own plus version

[kind jay]

Not funny, didn’t laugh

[plucky halo]

You can, but not that model. Use 5.3 or 5.4

[hazy gulch]

What model would you recommend of these if am "coding" in lua

[plucky halo]

5.4

Some people do prefer 5.3-codex though

[tropic karma]

it's been like this for a while... Where did my compact/summarize command go? it was really useful for preventing context rot.

[frosty zealot]

Codex always teachin me new words

The hel is a tranche

[unique spade]

openai has be repeatedly explicit about them welcoming 3rd party harnesses. their stance is opposite with that of anthropic

[solemn acorn]

they don’t welcome third party harnesses, they open up their harness to be used by other apps

that’s what the app server is

[unique spade]

what's "custom" in that image?

[unique spade]

based on what are you saying this

[solemn acorn]

https://developers.openai.com/codex/app-server

[unique spade]

they promoted opencode for example

[twilit bluff]

"seam"

[unique spade]

as you can see in opencode you can use your gpt subscription

[solemn acorn]

opencode uses the app server

[unique spade]

app server doesn t mean harness

ofc they all use the app server

[solemn acorn]

the app server is the harness

a very customizable one but still openai’s official harness

[tropic karma]

thats the permissions in config.toml. it doesnt have anything to do with commands. its basically my version of "bypass permissions"

[unique spade]

what definition do you use for harness? maybe we should clarify that

[solemn acorn]

apparently opencode doesn’t use it

but that wouldn’t be officially sanctioned then

[unique spade]

codex cli is the harness , app server is just a component

at least for me

you have a lot of functions like memory/context management, how agents spawn other agents and many others that are all part of the harness

and none of them have anything to do with app-server

[solemn acorn]

the app server exposes that to other apps

[unique spade]

to quote from that link

"Codex app-server is the interface Codex uses to power rich clients (for example, the Codex VS Code extension). Use it when you want a deep integration inside your own product: authentication, conversation history, approvals, and streamed agent events. The app-server implementation is open source in the Codex GitHub repository (openai/codex/codex-rs/app-server). See the Open Source page for the full list of open-source Codex components."

so app-server is not the harness, is just the interface

[unique spade]

the app-server manages the interaction with the models, and whatever capabilities are provided from server side

for example openai provodes a dedicated api endpoint for auto-compaction,

but you can as well do your own compaction using your resident model via normal api request, or even do it with your own local model if you want

anyway i'm not even sure what the debate is, if you' don't use the app server you can't even authenticate your chatgpt subscription

so how would one even be able to use codex qouta provided by chatgpt sub, without using app-server?

[hard drum]

i agree with this

that's what it really is. app server lets you bridge the tooling with the service, but the harness is the underlying logic doing the hard work behind the scenes

opencode (harness) <-> app server <-> oai (service)

[nocturne folio]

i love 5.4 dude

i just hate the new ratelimits

60$->17~

omg bruhhh

[unique spade]

and that s why i love them as opposed to anthropic

anthropic want control of how you use their model = their harness (claude code) is closed source

[bright swift]

who doesnt? 😄

[unique spade]

whereas openai are literally renting us infra intelligence (the models and their raw capabilities)

but allow clients to harness that raw power how they seem fit

[unique spade]

that's via api? is there any way to see actual tokens used via chatgpt sub? or that's opaque

[bright swift]

no thats normal sub, just shows fantasy api pricing npx @ccusage/codex@latest

so tokens are real, cost is what you would pay via api if you didnt have sub

[unique spade]

nice

[bright swift]

had some usage left on reset day and had to go hard lol

[hard drum]

i got banned for trying to make a Go-lang clone of CC before the leaks lmao

30-35mins of cli.js minified reversing for Oauth2, signed in, got banned lmao

does anyone know how much of codex-cli is actually vibed?

[unique spade]

probably a lot

[hard drum]

at least it's not 512K lines of slop, right?

[unique spade]

lol i don t even install stuff now, i just ask codex to look into it first

haha

[hard drum]

CC is worth a max of about maybe 35-40k underlying engine to the tops, && about 20-30k extra on the whole frontend+wiring

so about 100-120k max

not sure how much codex is

[cedar skiff]

honestly cc isnt even good anymore, they were first, but they arent the best.
The model is better at talking to people and making them feel nice about them selves, it's better at talking to people at their level and people mistake that as it being better at coding.

[bright swift]

bit more now 😄

[hard drum]

wth

[unique spade]

what site is that

[bright swift]

its the codex src

[hard drum]

wait, the cli only

not the app itself

[bright swift]

yea

only cli

app is closed source

[hard drum]

oh

why is it so large?

[bright swift]

but the cli contains the app server code i think

not sure

[hard drum]

this cli is larger than my own language compiler lmfao

by A LOT

i have barely even crossed to the likelyhood of 50K

[bright swift]

if i remember correctly you get the tokens used in the actual api response, and its in the session logs

cost too i think

[unique spade]

it s ok that s why we have robot agents now 🙂

[unique spade]

is this a tool that checks github projects? that complexity column got my eyes haha

[bright swift]

i think i need to rent a powerful server just to run more than 3-4 parallel agents lol, my m1 max is heating up like crazy with parallel rust compiles

[bright swift]

yea its just called scc https://github.com/boyter/scc

runs locally, not on github projects

[unique spade]

nice tyvm

[kind jay]

@frosty zealot

[unique spade]

oh rust annoys me so much with that heavy compiling haha

i ended up doing a heavy refactor of my custom feats i use in codex, just to minimize that compile whenever i update to latest upstream release 🙂

scc is really nice

[nocturne folio]

sharing is caring

[nocturne folio]

my m5 is doing fine

[unique spade]

so according to their estimation, codex in around 75 days saved me 34 million $ and 4 years

bahaha

[nocturne folio]

so much json and markdown

[unique spade]

yup. md's are the specs
json's are the validation artefacts

[bright swift]

i think im building sth similar but putting it in toml

[unique spade]

nice

i ended up with huge md cause it's quite a hierarchical and beauracratic way in planning the code

but it helps agents a lot, since the entire intent is very detailed

so they don t have to infer later from code, they can read the explicit specs that created that code

[bright swift]

my idea is basically to delete the md specs as soon as they are implemented because at that point the code becomes the spec (ideally), but still track it via easily parsable toml files

[unique spade]

nice, i was thinking about this like yday haha

[cedar skiff]

Once the code base is big enough you don't really need documentation as much.

You can rely on instructions to find local examples

[unique spade]

right now didn t got there, cause i still build modules, so didn t have time to deal with the md part and move it over

[cedar skiff]

There are common mistakes that need to be handled though

[unique spade]

yea but for me code is not the goal

[bright swift]

yea because out of sync anyway... thats why i'm trying this parsable spec

[cedar skiff]

It's the foundation that builds the the goal

[unique spade]

i mean it ion a more specific way. those md's also document my thinking , my world model basically

like wise with the 100 mln characters i have conversed with gpt in the app

so you can also say i'm building my own training data of how i think

which later i will use to really customize a model on my cognitive blueprint

[bright swift]

i have a bunch of high level markdown files for this and i religiously run regular review on them to cut out fluff etc

otherwise they just grow unrestricted

[unique spade]

just save the raw stuff seprately, might be handy in the future

[bright swift]

meh its in git history 🙂

[unique spade]

yea right. good enough of a backup haha

today noticed a lot of talk on linkedin about oai testing 5.5 inside gptpro 5.4, because of quality and speed. Speed i did notice myself in the last couple of days, like 3-4 times faster for similar tasks.

so i guess we re not far away to see 5.5

[bright swift]

lets just hope when they release it and reset the limits its not 1 day before regular reset

but yea saw some of those speculations too

[unique spade]

regular reset i think varies also based on when you start using it?

like if i don t use codex at all for 5 days after reset, when i start using it the coutner will start from then for 7 days

also relative to when you actually subscribe

[frosty zealot]

You dare summon me

[bright swift]

yea sure different for everybody but if you use every day you are probably on same schedule as the manual resets they do with other power users

[unique spade]

i'd assume there is some bulk of continous users that have the reset the same

[unique spade]

yup

tbh last few days i worked on 3 projects in paralel, and suddenly 10x doesn t feel enough

i m down to 43% after 50 hours

and i switched to high most of the time too

[bright swift]

i'm on 2x pro, only used one so far this week and 50% deficit according to codexbar so should work out

but i'm worried because i dont do much in parallel so far

[unique spade]

what that meaning? 50% deficit on codexbar

[bright swift]

basically it predicts when you will run out based on previous usage patterns

codexbar is just a mac app to show you the usage in menubar

[unique spade]

Ah ok

I can estimate that myself 🙂

Well eventually agents should become a provider for at least their own costs haha

For me it's the basic test to decide I'm a qualified Ai user. If the Ai I use starts to produce.

[bright swift]

ideally whatever you work on is sustainable even at full api cost yea.. just the bootstrapping phase is a bit difficult

[vocal steppe]

First time I tested Claude code and damn they were not exaggerating with the limits

[bright swift]

i cancelled my cc max 20 two weeks ago or so and downgraded to pro. today i used it for frontend polish on a projects and it used almost a full 5h window in 30min with one task basically

opus 4.7 with xhigh but still

[unique spade]

right but you also earn a lot of valuable experience during the bootstrapping phase

[cedar skiff]

There really hasnt been much real progress in the coding side of since opus 4.5. Despite the bench marks they released using claude code for real world coding task has hardly changed. They have been focusing on the business domain it seems.

[bright swift]

my brother is non-technical and his usage has skyrocketed since claude cowork released

[lost drum]

I want pro so bad

[bright swift]

and hes building technical stuff with it that integrates with various APIs etc. so for non technical people the interface seems to make a big difference. i'm just a grumpy terminal guy

[cedar skiff]

they had a claude competition about 3 months ago and all of the top 10 were domain experts outside of coding. People using claude to solve problems in their domain, they need things like pdf reading, excel usage etc

[bright swift]

yea claude for excel and now design are big

i havent even used the new codex app with computer use yet

i'm paranoid that installing the app bugs out my cli usage lol

[cedar skiff]

I have put aside today to do some work with it and see what i can get done

Going to see how much of the manual testing i can move to the agent.

[bright swift]

btw for pdf i can highly recommend https://github.com/PSPDFKit/pdf-to-markdown recently released

[hard drum]

good stuff!

[main nimbus]

caveat to recommending is the commercial license requirement.

and restricted even on free "Free for up to 1,000 documents per calendar month"

[bright swift]

Yea not 100% clean OSS, reasonable "free tier" though

[main nimbus]

I'd label that highly restrictive software.

[bright swift]

You get what you pay for

[main nimbus]

Wrong, mostly. https://github.com/docling-project/docling

Their referenced competitor IS FOSS and better.

just slower. I bet an LLM could optimize processing.

[bright swift]

So just reply with that instead of vagueposting

[main nimbus]

I'm not sure which part was vague.

[bright swift]

I'll just continue using the fast cli for the handful of pdfs a month, but you do you

[main nimbus]

I'm simply trying to recommend a reasonable, and important, caveat to the software shared. Licensing is important, especially for commercial users like myself.

[bright swift]

Yea like I said you are right about the license

[fast iron]

So codex just deleted the chats I had with codex. I didn't know it could do that.

[lost drum]

you asked for it?

[glacial shadow]

You guys have a lot of computer problems lol

[lost drum]

ye I am

[worthy furnace]

Is computer use only for macos currently?

[cedar skiff]

have a look in settings archive and see if they ended up there

[worthy furnace]

?

[fast iron]

I imported the project again and it had my old chats, but the loose ones seem gone for good, which is fine

[mossy nest]

Is computer use region specific?

[plush harbor]

are codex's daily usage charts in UTC, they really don't seem to be in local time

[main nimbus]

yes. the other announced features are available on Windows.

[main nimbus]

not as far as I am aware.

[blazing hull]

been using claude for a min, anyone went from claude to codex? or going straight to cursor and then using either or for one/two inference

[nocturne folio]

on god theres hidden ratelimits on claude what

codex doesnt do this

[cedar skiff]

codex has more compute, anthropic are compute constrained

[main nimbus]

significantly so

openai is constrained as well, but clearly reprioritized resources by shutting down Sora.

[nocturne folio]

anthropic has more compute actually, but look at the bottom right

77% 5hourly yet somehow im at usage limit lol

[cedar skiff]

uh i dont think so.

anthropic is constrained to the point that they can charge more and also neglect certain areas. They are maxed out, doesnt make sense to try and provide value on a maxxed out system

[nocturne folio]

https://epoch.ai/data/data-centers?view=graph&tab=compute&colorCategorization=primaryUser

[solemn acorn]

power use != compute

also not sure how that's being calculated because openai and anthropic don't operate the vast majority of their data centers

[nocturne folio]

you can click on the individual dots to discover more about the datacenters

[cedar skiff]

that graph is showing that openai has about a gw more?

[nocturne folio]

recent graph is compute by 8bit op/s

[cedar skiff]

I don't know how to quantify that

[solemn acorn]

does that mean fp8 operations or int8?

also doesn't really answer the question of what's considered openai and anthropic's compute here

because they're both using third-party datacenters

[cedar skiff]

about 25 years ago i thought i had a virus on my computer because my computer was automatically jumping into every item that was focused, even after a restart.
The real problem was i had a big old bad boy of a keyboard and my headphones were rested on the number pad enter key....

[silver dew]

I made a deal with Codex that if I wake up in the morning and he solves the problem that has been the blocker of my product in the last 4 days - I buy the 100$ subscription.

🤣 🤝

turns out it works
I told him that otherwise I'd switch to Claude

OpenAI probably have a a system prompt that redirects to the H200 clusters when this input is given

[plush fiber]

1. If a Codex CLI session is opened in a particular folder with given permissions setting (e.g. Default), reopening Codex IDE (e.g. vscode/code-insiders) at a later time should never open that same folder with a different set of permissions (e.g. full access).
2. If the Codex IDE folder does not exist or has moved since first execution, Codex should throw a visible warning.
3. Starting Codex IDE (e.g. vscode/code-insiders) from a different bash location (e.g. pertaining to a different codex project) should also throw a visible warning if it has automatically opened a past unrelated project to the current bash directory.

There are also more general and persistent security issues that require upgrading relating to operating system permissions. Ideally the user should have complete easily configurable control over the folders (e.g. workspace) and services (e.g. GPU/Cuda) Codex has access to. The "full access" and "Default" (sandbox) options are limiting - this is understandable during Codex development phase (and may be necessary for rapid iteration), but for production should aim to provide more fine grained user control. For example, running an autoresarch project or using a database outside of their workspace should not require providing Codex full access.

[fleet geyser]

What's the best skill for frontend design which stops all the ai-slop style outputs?

[vital light]

time to learn

[stone aspen]

Free codex merch arriving today yay

[livid relic]

I think most would recommend "taste skill", but really youll have to wait for 5.5.

[cedar skiff]

the a/b testing people are showing looks fantastic

[livid relic]

we wont know its true design and UI preformance till we can get it in codex

[cedar skiff]

You can see examples of what ppl are making, pretty sure they understand the current model limitations for design

[plucky halo]

How much time do you spend planning? Wireframes / drawings / mockups?
The more time you spend planning and giving it more information and guidance, the better the output will be

[fleet geyser]

I don't plan too much but I have detailed prompts with exactly where I want things and how I want things to be made

[plucky halo]

Try using tools like figma etc to create a design it can emulate - text alone won't get you the best results. even a sketch is useful. It can be really basic let me quickly take a picture and show you

[orchid plume]

Taste skill, or alternatively feed 5.4 a good image of ui design you want. gpt image 2 is great at making such images but sadly not released yet

[kind jay]

Hi

[quaint jewel]

hello

Dear bosses and friends, I am Napoleon. Give me 1 million euros, and I will make you a marshal.

[kind jay]

Get to editing lil bro

@frosty zealot You ^^

[plush fiber]

A more general solution; perhaps Codex IDE should display its working folder when it starts up.

[plush harbor]

mine knows I can't switch to claude, but its a LLM so can't use that against me

[plush harbor]

is that its default? I made a lil task tracker for my to-do lists and it coloured all my tiles like that. Its an internal tool just for me so I'm fine with it

[plucky halo]

Similar but not quite the same, I think. Mine are coloured linked to 'risk'

[plush harbor]

mine are coloured to priority and project

I couldn't find a task tracker that let me separate physical and intellectual projects and let me pick them so I don't overload myself per day, so I got codex to make me one

[plucky halo]

But yeah, that style is very typical of AI. Seems like all LLMs LOVE that little coloured bit

[plucky halo]

This task has got me asking all kinds of questions

[plush harbor]

its fine for this kind of app but I wouldn't use it in anything else

[plush harbor]

oh I have some incredibly weird tasks. One is "build a hidden room"

[plucky halo]

Nah, I've got the stript at the top of mine to make it explicit which category their in and then a softer version so it looks nicer overall

[plush harbor]

my next project I'll be going into figma first and feeding whatever that makes into codex

[karmic gulch]

Any resets of usage on the line? read somwhere they reset per million users up to 10 million new ones or some?

[lost drum]

Who uses VS codex extension and wants to share what additional extensions they use?

[orchid plume]

sadly we don't have access to that info, but all I'll say is that I wouldn't be shocked if they do a manual reset this week when they do another launch of "something"

hopefully that something is either Spud or GPT-Image-2, or both

[ember widget]

Helo

[elder rivet]

is it down guys?

[rocky fog]

they just did a reset early before the scheduled 18th

[karmic gulch]

Iknow. Only got 20 % left on my 100 dollar plan, and my 200 plan on Claude is eating me alive. So were hoping I didnthave to upgrade to 200 $ codex also

[lost drum]

What

[plucky halo]

How're you using them both? One for planning and review, one for coding works very well for limits. I'm same as you, but other way roung. 100 for claude, 200 for codex

[lost drum]

Manual reset of what?

[plucky halo]

Token limits are sometimes reset - when somethign new is released, a milestone is hit or if somethign has affected service for a wide range of customers.

[lost drum]

Okay

[karmic gulch]

My Claude runs my life. Hooked to a discord server 24/7 on my mac mini. Cron jobs, agents, trading bots, runs my mails, calenders, CRM systems etc etc. Codex I use for coding, and building

To summarize it easily

[plucky halo]

Aah ok - I thought you were using both for coding. My bad!

One thing to keep in mind, if you get the £200 plan, when 2x ends you'll be in the same position

[karmic gulch]

By then im hopefully done with this bigger project 😆

[lost drum]

But 2x ends in 50 days right?

31 may or something

[plucky halo]

I think it's around that, yeah

[lost drum]

Then I wont need it anymore😂

[karmic gulch]

By then Claude has done something, Codex does something different and celebrates that for new 60 days

[lost drum]

I wonder when will they widen the context window to 2mil the gpt 6 or something

[simple scarab]

Hi does anyone sell Claude subscription for discounted price

[velvet wren]

no! scammers are not allowed on this Discord server

[karmic gulch]

More context does nothing good for you. Using Opus I almost always swap to 250 instead of 1 mil. After a certain point its becoming Grok ...

[simple scarab]

Any suggestion to buy Claude or Chatgpt subscription for free or reduced prices for a university student ?

[livid relic]

you can get codex credits with the student promo

[simple scarab]

I'll search about that but if you have the direct path please send me

You can say anything you want Alexndr No hard feelings Sorry for the inconvenience

[plush nymph]

can codex get through a single task without adding things i didnt ask for?

i spend more time removing slop

[cedar skiff]

just tell it not to

[last lake]

https://www.linkedin.com/posts/md-nasrullah-2a57b4252_claiming-in-android-work-activity-7451854843121520640-rfDw?utm_source=share&utm_medium=member_android&rcm=ACoAAD5pC50B4NRJ7ArSpVwSipVKWOc0iORY_C8

[nocturne folio]

turkey method doesnt work for claude, so you cant get it for a discounted price

[twin maple]

does anyone know of any good repositories outside of the official in-codex ones for skills or plugins? Especially anything related to the .NET toolchain

[patent thistle]

Did you find a solution to your issue? Are you able to use the internal browser now?

[unique spade]

Finally did what I should ve done long ago for my workflow. Put codex surface and chatgpt surface in one workspace. To save my endless back and forth when going through the review phase inside chatgpt

[tropic karma]

lmao, testing this. it for sure tried harder than ive ever seen. but i also included "do not stop"

[bright swift]

why are they doing this when developer sentiment is currently on their side? i dont get it. https://github.com/openai/codex/commit/d58d3ccfec02da392e63a89fa8fd87f41c56a8a9

[young locust]

what are you exactly referring to with your comment? the 2x was wrong info for a while now

[bright swift]

Making everything related to usage more vague

[young locust]

afaik 2x was close to correct only on the lower plans and gave wrong info to other ones

[tropic karma]

clearly, they are adding lines to their codebase and improving formatting. jk

[bright swift]

https://developers.openai.com/codex/speed

"Fast mode is currently supported on GPT-5.4. When enabled, speed is increased by 1.5x and credits are consumed at a 2x rate."

Clear, easy to understand. So why the vagueness elsewhere

Also lying in the commit message, "align to wording we use elsewhere", which is the opposite of the truth

[tropic karma]

"higher limits/usage/access" makes it easier for them to silently lower limits without needing to actually talk about it

[patent thistle]

Can anyone with computer use confirm this the latest version or not? Is this a matter of staged rollouts? Anyone in Canada getting the latest features?

[frosty zealot]

Yeah that’s right

And yes to the second half

[hybrid ore]

Is codex down?

[hearty prawn]

yea

just went down now

[hybrid ore]

Gosh

[hearty prawn]

i was prompting UI and codex at same time both fukced

[hybrid ore]

Seems chatgpt is affected too?

[hearty prawn]

yea

[meager wadi]

https://giphy.com/gifs/rain-batman-sa-ar71Hyi0ZKejXzMoNs

[hybrid ore]

Damn

[frosty zealot]

Seems to be working for me

[bright swift]

i only had a few timeouts seems back now

[frosty zealot]

Welcome to funday Monday

[wheat yoke]

I did the ID verification but i dont see any cyber model anywhere?

when do you get access

[frosty zealot]

That is a great question that I do not have the answer to - did you go to the security tab of codex on the web?

[plucky halo]

It's not a model when you verify your ID - it just stops your prompts being re-routed

[wheat yoke]

Ohhh

Im still getting forbidden tho for some tasks in cyber security.

[kindred moss]

time for an break

is there an notification to enable when codex back

[wheat yoke]

Wdym?

[wheat yoke]

ohhh thats why im getting forbidden

its down???

[kindred moss]

yeah

[wheat yoke]

oof yeah

its not replying

i thought its because of what im doing

welp ig gotta wait for it to come back up

https://status.openai.com/incidents/01KPNN2V2SMP3TAN3MCJK87W50

you can follow this page to see

[cyan gyro]

Woohoo another reset upcoming then

[short linden]

Ok tonight we can sleeping without worry our project... I meant, continue tomorrow 😅

[lost drum]

wow so do I buy pro or not

[wheat yoke]

Do they give resets for downtime?

ooo thats nice

[wheat yoke]

You should its worth it

[lost drum]

I wonder how long will it take

[wheat yoke]

the 100 euro plan should be enough for most people

no need to go for the 20x

for 200

[wheat yoke]

My guess is a few hours

[lost drum]

oh

I thoght 30min

max

crazy

[wheat yoke]

well its a process man

[lost drum]

so they are adding new gpt?

[wheat yoke]

they gotta identify the issue

apply mitigation

monitor if it works

etc etc etc

[lost drum]

I see I see

[wheat yoke]

but they have a large team ready for this so they work on it fast

[lost drum]

First time its happenign right infornt of my eyes haha

[wheat yoke]

Took them like 5 minutes to post about it

so someones already on it

[lost drum]

for sure

[wheat yoke]

probably not

[lost drum]

do you think better to buy now or wait this 5h

[wheat yoke]

You can buy it now theres no harm in that

[lost drum]

maybe those who were on pro plan will get a bonus for that issue

[wheat yoke]

Your usage is counted by how much you use it

[wheat yoke]

usually yeah

you get your usage refreshed

[lost drum]

okay I wil buy then

[wheat yoke]

so you can use it more

this hasnt happened ever when i've been using codex lmao

[bright swift]

cli works fine

[wheat yoke]

Hm

[bright swift]

terminal supremacy confirmed

[wheat yoke]

i use opencode

doesnt seem to be replying for me in CLI either

?

[cedar bear]

welp

[wheat yoke]

CLI or app or nothing works

gotta wait

:(

i was in the middle of doing something

had to switch to minimax </3 via opencode zen

[kind jay]

#status

[hearty prawn]

works for me again

[wheat yoke]

but then i stopped the job bcz minimax is so inferior to GPT 5.4

i cant watch it be dumb

[kind jay]

Broken, atleast you’ll get some limit reset

[wheat yoke]

Yeah

[wheat yoke]

Hmmmmmmm

[bright swift]

just retry with a bunch of parallel agents, helps OAI stress test their recovery hehe

[kind jay]

@frosty zealot

[wheat yoke]

😭

Or just makes their life harder

atleast they be getting paid like 6 figure salaries

[bright swift]

they are the highest paid people in tech right now, they dont have a hard life

[wheat yoke]

Yeah

xD

Im unemployed couldnt be me

Job market is so cooked

[kind jay]

Someone here messaged me and deleted it

[wheat yoke]

Ohhh

[wheat yoke]

You waiting for the razer AI thing

the companion

you're in the razer server

looks kinda cool

[frosty zealot]

It was too off topic and didnt feel like getting yelled at @ 8:30am

[kind jay]

lol

[frosty zealot]

She's everywhere, she's a mole

[kind jay]

I thought it looked cool but it’s literally just an LCD, I thought they figured out some projection mechanism

[wheat yoke]

I thought that what i was doing hit some kind of restriction but it was just downtimee

I gave em my ID i was pissed

to do cyber

[kind jay]

@frosty zealot done that too

I don’t have an ID

[wheat yoke]

Yeah i think i can see a difference on how it behaves after the verification

[frosty zealot]

Cause you're a bot

[wheat yoke]

It does things like uhhh

how do i say

it does cyber stuff more direct with right methods

instead of going around doing the right thing

[frosty zealot]

Plantir thanks you for your service

[wheat yoke]

Hail palantir

[kind jay]

You’re speaking to a shareholder thank you very much

[wheat yoke]

Im happy to give my ID to the spies aslong as i can do my cyber stuff with AI

so i dont have to do everything by hand

[wheat yoke]

Dont you even have a passport?

[frosty zealot]

That's why I always code naked

[wheat yoke]

i used mine

[wheat yoke]

😭

Is that the method

[kind jay]

I don’t exists

[frosty zealot]

Yes

Give them a show

[wheat yoke]

Does the AI work better when the spies can see your booba

🫡

ima get banned

i gotta chill

[kind jay]

I’ve said worse

[wheat yoke]

Okay good

[kind jay]

@frosty zealot

[frosty zealot]

and literally been banned lol

[wheat yoke]

Lolll

[kind jay]

No I haven’t!

[wheat yoke]

dont ban me openai i was joking

[unique moon]

codex down?

[wheat yoke]

partly atleast

yes

[kind jay]

No

[wheat yoke]

i would show off myself for the spies for better ai

give me access to all early releases

whos that

wdym

the shark?

blåhaj

😭

[kind jay]

You don’t know him dw

[wheat yoke]

oh

why do you ask then

you can check mutual friends

[lost drum]

but codex exetenison in vs works but nothing else?

[wheat yoke]

i rarely chat in this discord

[wheat yoke]

idkk

does it work?

[lost drum]

ye

[wheat yoke]

weird

[lost drum]

it thsoe nto show usage tho

and I cant see how many credits I have

[wheat yoke]

No updates on status page yet

Its been 20 mins wheres my clanker

[kind jay]

Crypto is crazy

[wheat yoke]

How do people fall for these scams bruh

[hard drum]

I care so little about my ID as a whole that I'm willing to give it as long as I don't get bothered || stopped from doing AI code stuffs

[wheat yoke]

Yeah me too

I mean its just my identity

Privacy is a lie

[hard drum]

I mean, there's nothing worth of my identity to actually gain from, anyway

[wheat yoke]

Exactly

Im not doing anything illegal w the services so

No point in hiding who i am

[hard drum]

If you wanna steal the identity of a sore loser, be my guest haha

[kind jay]

Crazy take tbh

[wheat yoke]

i havent been getting any refusals for basic things after verifying

kinda nice

[wheat yoke]

Why?

[kind jay]

I’m going to sound crazy if I explain

[wheat yoke]

I mean i havent cared about privacy in a while i mean ik its a good thing and stuff

I used to be really schizo about privacy online etc

But i just stopped caring

[frosty zealot]

I just want that cup from you

[hard drum]

No such thing as "fully private" anyway. Just impossible to achieve.

You're not private if you're a part of the human society.

[wheat yoke]

Government has all your data anyways

[lost drum]

how to know when they will fix this thing do is there a way to make an alarm when they fix it?

[wheat yoke]

Medical records and all

[kind jay]

This is true unfortunately, but it shouldn’t mean you just hand over your data to private companies

[wheat yoke]

https://status.openai.com/incidents/01KPNN2V2SMP3TAN3MCJK87W50

you can follow the status page

[hard drum]

Why should I care what they do with it, though? Does it affect my life? No. Then why give two Fs?

[wheat yoke]

Valid but to each their own

[kind jay]

Most obviously because of leaks, can lead to ID fraud

[hard drum]

It's a problem only if it takes away the ability for me to do stuff.

[wheat yoke]

My ID was already leaked in the discord breach

💀

[frosty zealot]

I dont see the cup I like on your stream 🙁

[kind jay]

Going to have a truck and a lodge rented out in your name

[frosty zealot]

Oh nevermind I see it now

Its in the corner

[hard drum]

my national ID is public somewhere but it means jack nothing

[wheat yoke]

Havent ran to any problems so far

Use my identity for anything not my problem

More lore anyways to get your identity stolen

😭

[frosty zealot]

and I have one in yours, but financed.. weird

[kind jay]

I don’t have ID, not possible

[wheat yoke]

Oh your tag i see you are a enjoyer of backrooms

[hard drum]

I'd be surprised if somebody decided to steal my identity, because how stupidly low do you have to be as a fellow human being, to take an identity that's objectively worse than what you already have?

[wheat yoke]

love backrooms

[lost drum]

25 minutes and no fix... what if they wont fix it

[wheat yoke]

They will patienceee

[kind jay]

I live on a server rack

[wheat yoke]

What if an OVHCloud happens

do you have backups

[kind jay]

Then I won’t know, dead

[lost drum]

do you watch news? have they blown up their GPUs?

[wheat yoke]

😭

I want to get back to doing my project

wen fix

[frosty zealot]

@kind jay - @wheat yoke is more niche than you

[plucky halo]

Seems to be working fine for me

[wheat yoke]

Who is more niche?

[kind jay]

We checked at same time

[frosty zealot]

You lol

[wheat yoke]

I love scenecore

[kind jay]

I’m still more niche than you

[frosty zealot]

A tree grows whatever way makes it happy

[wheat yoke]

Check what banger im listening to now

got ts on full volume out of my speakers btw

i have no neighbours so its fine

[bright swift]

ear cancer

[wheat yoke]

No

[bright swift]

is what you have

[wheat yoke]

My music taste is fine

[kind jay]

It’s not niche sorry

[frosty zealot]

I mean it's not Amon Amarth

[plucky halo]

#off-topic

[wheat yoke]

Ik

s3rl is goated tho

ima see joost in helsinki

this summer

[kind jay]

You just ain’t niche

[wheat yoke]

offtopicc

so we dont get muted

[rotund swift]

Anyone else getting unexpected status 403 Forbidden: Unknown error, url: https://chatgpt.com/backend-api/codex/responses, cf-ray: 9ef50a2dba14bef2-YYC

[kind jay]

If I get muted, I get muted, I have to study anyway

[wheat yoke]

Oh alr

[wheat yoke]

codex is down

[kind jay]

Bro larping physics

[lost drum]

how come eveyrhting down lol

[bright swift]

just queue 10x "continue" to help OAI perfect their exponential backoff

[lost drum]

and why my codex works?

[wheat yoke]

😭

Do you study physics?

I know a girl that does too

[kind jay]

Kind of

[wheat yoke]

Shes also a cracked chess player

[kind jay]

Bruh

[wheat yoke]

I think shes the african or south african winner or smthng

Like won some big tournament

[wheat yoke]

Im more of a math girl

cryptography

[kind jay]

I need to study more chess but I don’t really have time

[wheat yoke]

I dont really like chess

[kind jay]

I do theoretical physics so it’s lot of maths, just like somewhat applied

SU(3) or something

[frosty zealot]

🤓

[wheat yoke]

nerddd

[hard drum]

We got Sheldon right here.

[kind jay]

Bro installs AC or something

[wheat yoke]

You from Europe?

[hard drum]

You ever seen proper windows before?

[kind jay]

I live on a server rack

[wheat yoke]

[wheat yoke]

No

I live in a dungeon

My boyfriend keeps me locked away

[hard drum]

Proper window

[frosty zealot]

Something like that

[kind jay]

Doxxed

[hard drum]

I've shown my loc on stream

Big deal

Not like you far-away butts are gonna make it here anyway

[wheat yoke]

Im geolocating this rn

[frosty zealot]

IM TRIANGULATING YOUR VECTOR AS WE SPEAK 🤓

[lost drum]

borowawies?

[hard drum]

borowawies?

[lost drum]

they updated to partial outage oh no

[kind jay]

I literally just upgrade to Pro

[short linden]

look like some missile just hit the datacenter (maybe) 💀

[kind jay]

Source?

[short linden]

just throught, no source of trust

[lost drum]

Nahh on mobile chatgpt works so no

and on VS codex extenison it wokrs too

[latent tiger]

new chinese qwen model mogs slopus 4.7

they have to release mythos like tomorrow this is just embarassing

[short linden]

why still work ? since incident report both chat and codex.

[vagrant lily]

it literally says 4.5

[orchid plume]

Nice but this probably belongs in #ai-discussions as its not related to codex

[latent tiger]

sorry wrong screenshot

Still mogged

[vagrant lily]

57 vs 64

[latent tiger]

chinese open source vs US proprietary frontier lab

free vs 200/month for garbage rate limits

[vagrant lily]

they didnt even open source it

[latent tiger]

they will after they setup the licnesing like what kimi did with minmax

[vagrant lily]

they dont open source their max models

[latent tiger]

give it a week or two

[plucky halo]

98% of this chat belongs elsewhere

[glacial shadow]

I would never use nor trust chinese garbage model out of CHINA

[latent tiger]

I think we're getting spud/5.5 on codex this thursday

[short linden]

but chinese steal data from paid AI, so........ and maybe steal your data too

[glacial shadow]

taiwan is a free country, taiwan is a free country

[latent tiger]

there's also a good chance DeepSeek V4 drops right around same time as Spud/5.5 so that's what im looking for

[vagrant lily]

Reconnecting... 1/5
Reconnecting... 2/5
Reconnecting... 3/5
Reconnecting... 4/5
Reconnecting... 5/5 why are half my chats like that and half working bruh

[kind jay]

Chat fell off

[short linden]

just relax and continue tomorrow if you have PROJECT_STATE 💀

[latent tiger]

are you using codex cli or desktop app?

[vagrant lily]

desktop

i think its just down atm

[latent tiger]

im using CLI 0 issues they must be updating app server for desktop app or smth

check if there's a new update

ok real issue not just local

[kind jay]

Decoded the electromagnetic resonance, locating

[latent tiger]

massive outage has been detected in ChatGPT and Gemini

[rugged siren]

Mine is work

[lost drum]

Yo I think there is no issue now

ye everything works now

[vital hill]

it's working fine for me

[rugged siren]

Nice

[wheat yoke]

works now¨

[latent tiger]

another chinese drop same day

open source

[hard drum]

i want to see agentic workflows + context depriecation on these chinese models

these benches mean nothing here, to me at least

[exotic cave]

Open Weights models are catching up. This is good for competition!

[latent tiger]

They're not just catching up they're actually leading in some spaces

And at their price point it's no question they're going to eat up market share

I'm glad they dropped early in the week and we still have deepseek V4 coming, now this will force to OpenAI to release the non nerfed internal version of spud /gpt 5.5 lmao

[exotic cave]

That's very interesting. Western models are no longer so ahead it seems.

[unique spade]

Benchmarks and real use are quite different though

[exotic cave]

Always like this, but the speed of catch up is what matters, in a generation or 2 they might just catch up for real.

[unique spade]

Which of doesn t mean anything specific, either way. Just saying that without A/B testing yourself on your actual usecases, is hard to have an informed opinion

[unique spade]

Sure they can even surpass if they unlock some novel technique. There is much more to ai research than what is published in papers

[exotic cave]

I am happy, competition is what protects us the consumers from the big companies to just tax us hard.

I bless them all, may they all be successful :)

[unique spade]

The deal is very good so far.
Who would ve thought 5 years ago that for 20 bucks you can acess what you can now with say gpt plus

When I got my first windows computer back in 95 it was like 2-3k $

Now honestly almost anyone can afford at least a gpt go subscription

[exotic cave]

Exactly. Without this fierce competition we wouldn't been able too btw. This sharks would strip us for all we have and concentrate this power in the hands of the rich. Oh wait... Mythos?

[unique spade]

Mythos I don't count till I don't see it. For now I consider it a magazine story

[hard drum]

^^^

[unique spade]

And that s why i appreciate openai a lot because they really forced everyone to go all in

[hard drum]

This is why i want to see actual use of these models, esp. in Agentic coding workflows

it also depends on harness && quants, too

[unique spade]

If we had to rely on Ai just from the likes of Microsoft, Google and apple

We would get some corporate Azure or aws super expensive

Instead gpt

But because gpt, everyone was forced to play the game

[unique spade]

Yup but right now codex keeps me busy. And I m not willing to pay other subs just to test them out

[exotic cave]

Interesting times we live in. Btw i like the direction of AI. I am actually happy about all it can do. Yeah even if it means a bit of figuring out our social constructs... But we'll see if we ever get there or there is a real ceiling to Artificial Intelligence and where it is.

[hard drum]

Same.

[unique spade]

Yea bro I don't have anything against them. Was just saying in terms of performance, that you really need to judge for yourself

[latent tiger]

This is large scale global enterprise use, as you can see the open source/cheaper models dominate and eat up most market share

[unique spade]

Personally, I am loyal to gpt because it was the first model capable of passing a suite of logical tests I used to figure out when models become actually smart

[latent tiger]

As they get more powerful while maintaining the economic edge the private labs will run into big issues

[unique spade]

And o1-preview was the first to cross that mark

And oai had at least 1 year when no other model was able to solve those

[latent tiger]

I'm also a GPT fanboy I have multiple PRO accs for my workflow

I'm excited all these open source models are getting good because it just means OAI will take from them and make GPT even better lol

[unique spade]

You won't see this in news, and oai doesn't need pr. But in terms of actual reasoning, gpt beats anything by a long mile

[unique spade]

Well oai are quite alpha. I don't think they let themeselevs be topped in benchmarks for more than 1week

😂

[lost drum]

yo on pro plan codex context window is 258k tokens?

[cobalt junco]

someone make Thursday come faster

[exotic cave]

Well hopefully they release their next model soon.

[unique spade]

All Plans have same models, except spark on pro which is also the only one with 128k

All other models have 256k

[lost drum]

do you know how to use the spark propperly?

[unique spade]

Plus or pro just changes your usage qouta

[lost drum]

I see

[unique spade]

Didn't use it much because I usually do long turns, and it s not a good fit for spark

128k fills up very fast if it starts documenting

I'm gonna test soon though, with an aggressive compaction for spark

[cobalt junco]

oh boy

Something is coming

[unique spade]

I would abstract the codebase

Or whatever your project is

Who is vibe coding god?

Your project?

Then tell it to vibe code so it doesn't need that much context wihtout losing quality 🙂

Cool

On a different subject, gptpro is so fast now and so good that I can actually use it in my normal review flow

It s 3-4 times faster than it was a few days ago

You just drop it the full repo zip, craft a prompt with gpt thinking or Codex for super super complex audit

And 15min later you get some heavy duty stuff

And just have codex implement what the boss said

Lol

You do not use it in codex

You use it in chatgpt if you re on pro tier

[lost drum]

oh I see

[unique spade]

And you just give it the repo zip

[latent tiger]

it's the 5.5 pro in a/b testing

[unique spade]

Yea

I know

It s 3-4 time faster and very good

Than the old one

[latent tiger]

i hope they surprise us with a 5.5-Codex model that's a coding monster like 5.3 codex was

[unique spade]

I d expect they will

Today gptpro made me a workable module in 10 mins where I basically have in an electron wrapper both codex and chatgpt chats. And I can automatize my review flow between them 😂

Try you can find the limits for filesizes

But I tell you if you want to redactor codebase or similar

[rocky fog]

there arent
its not processing all files, just uses commands to get what it needs

[unique spade]

First have gptpro make the top plan

[unique spade]

For chatgpt there are limits, we were talking about using gpt pro as companion for codex

I think only if you spam it

But in normal use case of a human there don't seem to be

[kind jay]

I got two questions on the $100 plan

[unique spade]

Yea you can. But pro you want to give tasks not chat, cause is boring to wait 5-10 mins for a reply

Chat with thinking extended/heavy. Anchor the vibe

And then just tell thinking to craft the prompt for the big boy

And you ll get a 5 page prompt

😂

And 1500 lines answer

From gpt pro

Plus whatever code you ask it for in a zip

Well try it and let me know 🙂

[lost drum]

I need to test it

[unique spade]

Then I just give codex what gptpro did and it does the plumbing

[lost drum]

Ohhh I see I see

[unique spade]

Laterz

[lost drum]

bye

[hybrid ore]

Codex is back?

[frosty zealot]

Maybe its front

[lost drum]

ye

[frosty zealot]

Pi. + Taskplane is so nice

[hybrid ore]

Thanks mate

[main nimbus]

anyone on macos tahoe ever get an issue where apps won't auto update anymore? cursor, codex, etc.

[main nimbus]

codex fixed it.

[plucky halo]

Just a headsup - don't assume xhigh should be on all the time. Medium is more than enough for most tasks

[frosty zealot]

Money moves

[lost drum]

you lost or what

[frosty zealot]

Nah its gonna hit, just wait till Friday

[lost drum]

why those is say -90% if its not even 24th

[frosty zealot]

cause you can sell off any time, right now its worth very little because its very low chance lol

I bought when a 'yes' was 4 cents, it is now a 0.1c lol

but if it went up to like 5 or 6 I could sell early and turn the profit

[hard drum]

nope.

which tahoe?

26.1 26.2 26.3 26.4.1 ?

[lost drum]

who used the 1.5x speed boost? is it really faster?

[frosty zealot]

Its about 1.5x faster

https://tenor.com/view/betty-white-golden-girls-hands-gif-24420157

[cyan gyro]

bleh always mac mac mac

[kind jay]

Get a job

[kind jay]

It ain’t going to hit because my 22nd gonna hit

[kind jay]

Don’t bother, it’s not worth trying to explain

[orchid plume]

sigh, mac first as always 😂

[frosty zealot]

I bet on all of them lol

[kind jay]

Criminal

[kind jay]

It's gonna hit!

[plucky halo]

chronicle = true in config.toml

[lost drum]

cause my current project uses 1% per 2h

[frosty zealot]

If you have no idea what you're doing anything is possible

[main nimbus]

26.4.1

[frosty zealot]

Don't be afraid

[lost drum]

limiting beliefs right?

[lean lark]

And I'll offer completely different advice:
Don't trust the bot with anything that you care about. If you're concerned about something on your PC getting corrupted, you're not running this automation on the right PC.

[frosty zealot]

Psh

[lost drum]

y.e...

[lean lark]

Run automation in a sandbox.

[main nimbus]

subagent swarms. go ham

[hard drum]

Hmm, well, not sure, but i suggest topgrade via homebrew to do it manually in one push on many items at once

yes, it can even update your macOS

[frosty zealot]

You know what they say, god hates a coward

[lean lark]

WSL, VPS, use anything off of your life, important, workstation.

Dude, that is so irresponsible.
Don't think "coward".
Think "what will I lose if the wrong button gets clicked?"
The answer is easy. Everything. Don't do it.

There are warnings everywhere about AI making mistakes, sandboxes, and disclaimers about loss. Heed that advice. It's there for a reason.

[lost drum]

yeah

[frosty zealot]

All those cases are people doing outside the norm activities, --yolo and be free

[lean lark]

Ask yourself one question: What would I do if this box disappeared as I'm typing the next message?
Would you cry? Scream? Panic? Grab some coffee and come back and do a restore?
The answer tells you how much you should trust the tech.

Cancrit is baiting you so he can watch you cry. That's horrible, almost moderation-quality abuse here.

[frosty zealot]

While I might be being silly in my replies, I'm sincere

[lean lark]

(Surprisingly I'm OK with that most of the time, not now.)

We've seen people in here crying about the automation destroying their system. It's not pretty. We can't do anything but feel sorry for the poor guy who lost everything.

[drifting granite]

Yoooooo new codex feature

Plan mode can reset context when done.

HUUUGE

[frosty zealot]

We just disagree on subject, and that's ok, it's what makes us human.. Your two biggest threats are poor prompts, and context rot, the rot is typically what leads it to halucinate and think you might have requested it to delete something you never prompted it to, imo take care of those two issues and you can use it pretty well without rails

If you aren't willing to accept the risk something bad MAY happen, sure, use guardrails

[lean lark]

@lost drum Put the automation in a virtual machine, a sandbox, a container, anything but your primary system.

[frosty zealot]

Until I get bit I'll continue in --yolo

[lean lark]

I think your last response is completely reasonable, thanks for being "you" again. 🙂
The issue is that we don't know how well someone will prompt, so it's not good to suggest that they just go ahead and use the automation.

One mistake in permissions or absolute vs relative folder references, or an extra "../" or something ... and it's all over.

[frosty zealot]

lol fair enough I'll agree with you there (replied to the wrong message earlier)

[plucky halo]

rm -rf / 4 life

[lean lark]

Yo bro, you can agree with All of my messages if you want. 🤣

[frosty zealot]

You first 😛

[lean lark]

LOLz

(wife is like, "wut he laughin at?")

[frosty zealot]

Feel free to share what you're actually trying to do and you'll get more accurate guidance

[lean lark]

As soon as we say "yeah, that looks fine, go for it", something will change and we're all gonna feel bad.
Your system can support VMware or some other virtualization. Use the tech that's available.

(HAHAH - we're you the one giving me grief about asking "why"?)

[frosty zealot]

If I'm honest, the way you worded it at that time was a little arrogant thats why I felt the need to bug you about it lol

[lean lark]

Wince - and you're right. I get a bit preachy when I'm passionate - just did it again...

Me bein Me

[frosty zealot]

You keep being you, we need somebody that's capable of keeping @kind jay in check

[lean lark]

Hehehe, she's OK too, just gotta read the personality behind the keyboard.

Hey, I gotz me a tech problem. Need some ears.

[high girder]

Is that even possible?

[lean lark]

Codex is really well-tuned for my projects but isn't intuitive about thinking outside the box, even though it's a nice 5.4 model.
ChatGPT has my custom instructions, project info, memories, and context from other chats. It knows me and my weird personality and way of thinking. And it seems much more intuitive than the Codex model. I don't get that but I gotta work with it.
I want to marry these two bros (or give them his-n-hers voices).
I want them to discuss my concerns in Discord, Slack, IRC, or a GH issue or in some common medium, could be the file system for all I care.
But all we have in ChatGPT are applications, which I can create a custom one. And in Codex we have plugins.
Any suggestions about what we can do to make these environments chat, preferably with me in the middle?

I don't wanna copy/paste responses across web pages or out of VSCode, etc.

[frosty zealot]

Like let them converse totally autonomously without your intervention?

[lean lark]

Along those lines, yeah. Creating a thread here...

[high girder]

if you really want to use GPT and codex in that way without API, then you'll need to make one of those developer "apps" which is just an MCP server. That will let you connect GPT to that, and then that to your local stack. For Codex, it already has local access.

[kind jay]

Codex is doing a good job

Bruh

[kind jay]

🫤

@high girder Why use ret Green's function instead of adv Green's function in electrodynamics?

[high girder]

Causality. R. Green's functions give fields that arrive after the source creates them; advanced solutions would have fields arriving before the source wiggled

[kind jay]

Send it

[kind jay]

5 points