#codex-discussions
1 messages Β· Page 72 of 1
There are definitely certain situations where I have to open chrome - for performance intensive stuff Safari sucks
you know it's there but your brain filters it out
But I've used safari pretty much exclusively for 4+ years with not many issues. Only 1-2 sites every 3 months will force me into chrome
Yeah, I mean if you look at web games, they're actively developing with the latest features, so it's gonna work better on Chromium automatically
so that explains my issues then
Lot of chromium haters after this new GenAI update
chromium is prone to hackers
i don't care about their AI slop
I remember back in the day there wasn't even a standardized way to round corners in css that worked in all browsers heh
i care that the browser actually performs well on things i need it for
google ai is duce
sfari simply doesn't cut it for me
workflow issue ^
lol
here take this chill pill : )
i am chill
true
except my agents stopped working, so hold on, lemme prompt them first
nvm one is still active
anyone use codex to code a raspberry pi yet ?
the other one stopped
Any browser is a target for hackers. It's the main way other people connect to services outside of their local network. I don't entirely think it matters what browser you use, there's always going to be tens to hundreds of thousands of malactors targeting it
safari most safe browser
codex on a pi would probably be pretty painful if you wanted more than one instance running at once
not if y'have 8gb of ram on that thing
not to run on it to make code for a raspberry pi
oh, yeah...
i plan on making a keyboard device soon , i sure it can handle the task perfectly
i got a 512mb pi zero 2 i mistakenly bough tfor a discord bot, buuuuut... 512mb too little
so idk what to do w/ it
doesn't seem resellable
can't return it
optimise the code so it'll run within the parameters you have lol
512mb is a lot of room working in a non-bloated language
Oh yeah, if it's doing python it works great! Interpreted language dev is fine on the Pi
my discord bot was written in Go+Lua
without use of GCO libs
nah i dont need it but i seen it was coming well considering the reset only gives tokens badk not days of sub or broken files because of degrading blah xD
it just cannot compile on that pi due to OOM
is it worth actually using fast
More importantly, write speed on a microSD
I hate lua and will die with my grey hair on the arrays start at 0 hill
Reconnecting... 1/5
stream disconnected before completion: Transport error: network error: error decoding response body
like 50 of these
that, too.
why do you need to compile on device why not precompile and copy the binary over?
that is the same thing w/ mathematicians arguing on whether 0 belongs in natural number or not
Get icon composer π wayyyy easier. It's built into latest Xcode
i eventually did that, but for some reason that pi ran a few hours && then killed the bot
no idea
thing just doesn't wanna budge
did it kill because it ran oom? seems odd tbf. discord bots aren't entirely intensive unless you're doing stupid stuff in the background
probably. i didn't do anything
maybe trying to write database made it go brr or something. i've zero clue
There's another rpi knockoff called Odroid. They have a pretty decent, closer to SSD storage thing, can get a big and fast swap
https://github.com/xsyetopz/go-mamusiabtw it sits here
depends on the db lib ig. maybe the whole db was in memory / the cache was too big
https://github.com/xsyetopz/go-mamusiabtw/blob/main/docs/sbc-hosting.md this thing i needed to make for myself
i wanna get myself a different spc under 100
&& i know one that has 32gb builtin storage + 8gb ram
it's also part of the sbc hosting list
i wanted to make a zimaboard cluster but they're kinda underpowered for how much they charge
I think the last time I played with SBC was when ASUS made a couple and they looked interesting because they were pi form factor but better spec
it was years ago now though
but if anyone knows why my bot acted up on my zero2w, i'd be delighted to hear an answer
it was my 1st time dabbling w/ SPCs
the bot itself was a resurrection based off of wakalaka-rs, an old defunct rust-based bot from 2024
anyone els keep a notes open at all times for when inspiration strikes
now that i have to wait 3 days till a reset ive had all my best ideas lmfao
i need to fix the note placment logic in my app too this chord is dam near impossible lmfao
I mean I don't entirely worry that I'll forget the idea before my notes app opens so no not really lol
this is the better way to play this chord
im going to tell codex to examine this manual fix and try to imporve its logic
To play this chord you need at least 3 hands π
stright up lmfao
I'm sure Marcin would have no problems doing it though
joe pass type chords
i still laugh from time to time when rui said my piano hand be messing up and sent an AI song
lmfaoooooooooooo
bro really tried to trick us π€£
Yeah but even for an AI song it absolutely slapped π the sax was legendary
type shi
brb i need to make a logo for my app
and i need to brainstorm a good layout for my sequencer section
firefox is a RAM destroyer
ironic because i had the same experience with post-2013 firefox
firefox before the new engine was great imo
i miss the old firefox...
i went from IE -> Firefox -> Chrome -> Edge (Chromium) -> Safari -> Chrome
ive always heard Firefox uses less RAM than Chrome so I switched but than it ended up eating way more RAM
it was a lie with new engine
old engine was genuinely better in that regard
i couldn't bear my ram usage using firefox's new engine as 2 or 3 tabs somehow ate more than chrome
chrome for life
just download more ram bro
i use chrome / firefox / brave at the same time
brave is my stream/media browser to skip ads, chrome is my daily driver, firefox for dev π
unified ram moment
I think that unused RAM is wasted RAM. As long as Firefox is making the most of the RAM it's holding that's fine. What's not cool is poor memory stewardship.
So much easier than exporting 12 different size icons
its not cool when it hogs all my ram slowing down my computer
unused RAM Is definitely wasted RAM
I consider that poor stewardship. Especially if it's wired down memory with high residency (contiguous, can't be paged out)
ZFS is π₯. Love the snapshotting feature
Git for the whole file system
indeed
same for disk space, i always make sure that my ssds are 99% full
Me too. Not on purpose tho π€ͺ
is multi accounting allowed under ToS
Yes. Afaik you can't resell access (host your sub online as a paid resource)
No intention to do that, just wanted to know if having 2x 200$ plans is okay
Yeah, the terms of use are framed around "how you use this account". There's nothing in there saying "one account per person" or anything like that
it's fine as long as you don't abuse it
you're allowed to have alts. simply don't share your accs && use them to avoid limits
by avoid, i mean cancelling sub && making a new acc just to sub over there
or whatever is done
I gotchu
the multiple accounts being allowed might be changing in the future, I've heard that people are being directed into phone number verification flows now and hitting barriers preventing use of the same phone number for more than one account
I have a Pro 20x and a Plus account. It's completely reasonable to have one for work and one for personal
^
If I'm paying 200$ for a plan surely it's reasonable I'd be okay with buying a pay as you go sim lol
They are reporting it wrong
It is explicitly allowed and enouraged to open up to 3 accounts per phone number
I don't really use GPT for anything other then work - but I'm freelance and my personal projects make me money. Hard to classify it for me
I used to have a couple extra plus accounts to use to limp through the last couple days of the week when I'd burn through my pro account in 4 or 5 days but ended up canceling them when they cracked down on the 5h limits on them
https://help.openai.com/en/articles/8983031-how-many-times-can-i-use-the-same-phone-number-to-complete-the-phone-verification-associated-with-an-openai-accounts-first-api-key-generation
Possibly those users reporting it "not working" are from unsupported areas
But honestly - I am from neither and was able to do it without any issue
if the 5h limit is so low that I can't burn the whole weekly limit in a day anymore it doesn't fit my intended workflow
Idk if they pay that close attention. I bet you they look for many IP addresses executing multiple Flex /v1/responses for seemingly different things. Heuristically, one computer IP with 2 accounts coming from there looks more like an office doing serious AI work than reselling access
See I'm in a building that has over 800 people on one IP address hahaha
Everytime I think about how companies judge usage by IP I can't help but think of CGNAT
decided to pick up on the discord bot again, to fix architectural issues && whatnot as i started seeing issues after manual audit + AI audit on top
codex be goaling because it didn't work on 0.132 until i went 0.133
I wonder how much RAM the router needs to have to support the size of NAT tables you'd need for that many clients to reliably do anything
still waiting for my custom harness to be done
I'm in a building with 8 people, 7 have ChatGPT Business, I have a Pro and a Plus π and the business accts have the same domain email as my Pro acct so if they're OK with this distinction they'll probably be OK if you open 2 pro accounts
to be able to dogfood itself once i get to use it
I'm pretty sure the wording in the TOS around rate limits usage etc is mainly targeted towards abusing free accounts
i wish i could be a part of a workplace that uses AI stuff to build stuff
could be a great job, tbh
or even as an intern...
i'd happily be an unpaid intern in one
I run a double nat but I presume Fortinet / whatever enterprise routing platform they run is built for this kind of thing. It's a relatively common setup in educational institutions and large offices
They all use their account to do HVAC stuff. No Codex usage except when they do excel spreadsheet stuff in ChatGPT
you're really lucky to be in a space like that from my eyeview
I run a lot on my network, loads of local services exposed etc. I also have a lot of clients and they try to limit me to 6 devices before I have to pay.
Them seeing one mac address makes my life a whole lot easier - and reduces the risk of me getting attacked by a whole lot
yeah, that makes a lot of sense
if y'know anyone willing to take an unpaid intern in that dabbles in the stuff that i do, maybe y'd have me somewhere? @boreal holly
if you offer yourself to be exploited like that for long enough someone will surely take you up on it eventually
nobody took me in for 5 years locally in my country after i graduated, so...
scary
that sucks
i'd rather be exploited than stay unemployed forever
what else am i good for, y'know? i am just a tool, after all.
works for me, not for thee
you don't have to just be a tool or cog in someone else's machine
my 5 years of unemployment &6 being stuck in social services would like a word with you in an interrogation room
if you were in my shoes, you'd likely go insane && rope yourself
Open a LLC, release a software of some kind, and pencil in as work experience at a software company π even if it's your company, it's still experience
y'do realise opening an LLC is costly in Estland, right?
Work freelance
well, actually, somebody at a pawn level like me would need to open an OΓ (OsaΓΌhing), && that expects you to have some form of starting capital
In the age of AI, do what, exactly?
people will just ask gpt or something
Or um, there's other less costly designations. Sole proprietor probably not an option for penciling in work experience
Not for things they don't understand
the market is heavily saturated, though. chances are, without presence, i'm pretty much dead on arrival
AI has removed most of the simple jobs that I used to get. All of the more complex stuff still exists
see, i'm not good at leading thngs, but i am good enough at following orders
because that's what i do.
freelance expects you to be your own CEO
I don't have much business knowledge. I typically work 1:1 with clients and just build to their specifications
that's... not my proficiency
my proficiency is in following orders && not thinking, just doing
exactly what a tool does
then you need to change that. AI is better than you at following orders
a company that loves a good object could harness me for some use
but i've yet to find one willing to not consider clinical challenge a liability first
unfortunately, that isn't possible
My personal belief is that anyone can accomplish anything. I do not have much of a creative mind, I don't think like that, but I've been getting better over the past few years. Even if you aren't naturally good at something, even if you're naturally terrible at it, you can still become great
AI in its current state doesn't follow orders as good as i could, esp. when inferred context is at play
why not? you have all the tools at your disposal to build your own stuff
You won't get anywhere in life with a mindset so fixed on what you can't currently accomplish
i usually build stuff that i believe makes sense to have, or something i believe might be worth something, but not monetarily worth something
i'm not going to live long anyway, so it doesn't really make a difference if i took a piece of advice that's incompatible in every way
just build stuff that works, && hope somebody sees use in me for their stuff
that's.. kinda it
i mean this sounds like motivational slop, but you might have some limiting beliefs
i like to give stuff, not sell myself out
what's your definition of not living long
the fact that you are in a discord discussing ai tools alone shows that you are ahead of 99% of people on earth probably
a couple of years, maybe 2028/29 before i collapse?
This just isn't a beneficial belief system - you don't have much time left and so you want to accomplish less?
Why not push yourself as hard as you can in the time that you have
i do enough of what i can in hopes somebody sees use
is that financial collapse or medical issue related physical collapse?
both
financial first, life later
right now i'm thinking about making a personal git portfolio website to start putting stuff on display
in hopes that makes any difference at all
Believe it or not, the fact you're willing to do software dev for the fun of it, you're half way there π a long time ago I applied to Codeweavers as a wine dev. Didn't get the job (their review process is surprisingly difficult), but the CEO said "if you wanna work here, I recommend contributing PRs, building a portfolio this way." So I'll pass that along to you π
i already do as hard as i can
a lot of codebases i kinda wanna contribute to are strictly anti-LLM, so that leaves them out of potential stuff i do locally on those codebases
like a patch fix for SDL
...unfortunate
i mean openclaw is hiring fulltime and pretty pro-LLM
just one example
more than enough work there to make a name for yourself by helping out
i might want to start smaller. openclaw is huge
don't bite more than i can chew
yea just an example
besides, 0 work experience requires work experience somehow, && to get experience y'need a job, && to get a job y'need experience
a catch-22
What is "pursue goal"
config.toml:
[features]
goals = true
then use /goal ... slash
they've been since 0.128 iirc
or was it 0.125?
not sure, I don't even have /goal I just ask codex to make a goal and it shows up lol
odd, but the feature could be toggled on to make it explicit
Like I mean what is the purpose and difference it would make vs without it
goal is essentially a built-in ralph loop with some caveats
goals are default true since 0.133.0 i believe or 132
There's nothing stopping you from using it as a research partner, or a draft code agent! And the PRs don't have to be spectacularly huge accomplishments. You can find really small issues in a repo, patch it, submit. Anything to put green dots on the Github activity grid makes a difference
i don't see why green dots matter on mine
it's like a more structured ralph loop that you can set token limits and explicit stop conditions on
i find that what i make kinda makes more sense than what small thing i contrib to
if my own github repo page is full of niche garbage, nobody would take me serous outside hobby
but i won't stop
i just like to give && make stuff
i don't have the poison of money on me
i like to give free, MIT-licensed stuff
SW companies want to see collaborative effort on the resume. OSS contributions to other people's repos is a really good way to do it!
When I review a github, the most important things to me are
- Code quality
- Documentation quality
& last of all, there being enough code to make a reasonable deduction with.
The subject of the projects themself, doesn't mean much
well, let me guess.. i'll have to pay you to make an audit on mine?
at this point i need a pair of somebody else's eyes to see what i may be blinded on
nah haha - I don't charge to help people. I make enough from providing services to my clients
I don't charge to help people
God, i wish i could do more of this, but some people insisted on accepting a charge so i stopped
I don't work in corporate tech though, so bare that in mind. I've hired for a couple startups and reviewed applications as a favour to others who are hiring.
used to do tech support && repairs for some households && peeps online
except some started basically forcing me to take a pay for it
so i stopped
I've been developing for ~11 years now, cyber for 3 and freelancing for a little over 5. Was full time at a company for 1.5 of those years, and 2 other companies for a couple months
i dont get it, what are we even talking about here then
basically, i don't really wanna work for money, but work just to benefit the people in need
but i need just enough of a wage to pay rent && codex sub, && that's it
nothing else matters
&& my rent is already dirt cheap
you have to take the money to continue to live to continue to do what you really want to do though
so 500-550 usd per month is more than enough for me to survive
i consume about 380-430 per month depending on season
You could look into DA. They are relatively picky, but they pay per hour for writing prompts and code for chatbots. You help propell the future of AI, and can work whenever you want.
However, they have a strict no AI policy as you cannot use AI answers to train AI
Some of the projects will even cover a sub
&& am on a workbite (side hustle?) for my landlord making 7 bucks an hour doing cleaning service that gets deducted off my rent
but you rejected money, and you complain that you cant get work experience without experience, and you only need 500 usd
i reject money i don't earn thru a contract i signed
which is called dirty money
thats a weird worldview
welcome to estonia
I don't sign contracts, bar NDAs with most of my clients. That's pretty much the way the online work world works
i mean, wouldn't a company want an obedient walking object amongst their worker list?
one that just does what they're told by the spec?
not anymore
thats AI now
Companies have mostly always looked for out of the box thinkers
with this attitude, honestly yes
people dont need free MIT code anymore, LLMs just generate it on demand
you weren't in my shoes, so i can't say you'd know what it's like
so the options you give are basically:
- make something that starts bringing income
- find remote AI-first startups && get in
they still need to achieve their objectives, but if you dont want to help with that, i dont know whats left
&& considering i do have a paid Apple Dev account, i could try something...
just need to do something useful for people
Is Tibo awake already?
this pretty much goes for all of software dev though lol. no reason to build software that doesn't have a purpose
Anyone here try the memory feature of codex thats marked as Experimental? I know my automations use it, but I havent turned it on for normal development work yet. I still prefer to rely on AGENTS.MD files / config.toml developer_instructions to specify my coding style and standards. I'm just curious if anyone has used it and it has actually noticeably picked up on coding standards it gleaned from conversations, without doing too much 'well, this one time he did it this way for a one-off task, so he must ALWAYS want it done that way'.
well, i'd need to reconsider what is even worth monetising, because i lack the conception of "exploit people"
I have memory off because it starts refencing things across chats and I really don't like that
its not exploitation, its value exchange
stop with the communist mindset
...communist mindset? i don't live in russia, friend
i wasn't even born when estonia was in that era
i know nothing about communists
clearly, cause russia is not communist lmfao
I've never exploited anyone. People pay what they are willing to pay. I assign an arbitrary value to my work based on what I feel I'm worth, the time spent and the experience required to fulfil the job.
If someone wants to pay too little, I don't do the job. If someone wants to pay too much - I'll tell them it's not worth that much
err wrong reply lol
If someone wants to pay too much - I'll tell them it's not worth that much
WAIT, you're allowed to say that?
"I can't charge you that, that's rediculous, how about X?"
Ah yeah, makes sense. There are times when I am like, "I wish it knew how I was doing this in the other chat" and other times where I was just exploring some concept or idea and definitely dont want it to reference that.
I thought people told me that this was a bad thing to say, so you should try to snake your way into making them pay what they want to pay you in that context
That's very true and false at the same time
There are a lot of software libraries where there's a huge benefit using it as a dependency rather than reimplementing it or forking it. Not necessarily MIT, but a lot of Apache2 and GPL that powers the world and makes sense to build on top of
at least that's how the older generation Estonians told me
yea the established hardened stuff, but not random new libs most likely
You can operate how you want to. It's your work, and you're the one that they're paying. If you want to be exploitative, fine, I'll frown on it, but that's the way you run your business / you work. If you want to be fair, I'll respect that
anyone use GPT5.5 for reverse engineering PE binaries ? is it bannable ? got perma banned by anthropic for working on security stuff π€
I wouldn't let somebody pay me more than what they believe the thing is worth
there are a lot of MIT and BSD-licensed single file libraries out there that are very useful in game dev etc where trying to outright reinvent the wheel by burning your own tokens would just be a waste of time
Yes, I reverse engineer practically 24/7, it's fine. You may get a cyber flag, but you can appeal and they will remove it
If you're actively trying to circumvent DRM / do dodgy things, you may have less luck
I remember when Wireguard came out. Everyone in the world was happy with OpenVPN, but then Wireguard appeared as a random new lib and was so incredible it got inserted directly into the Linux kernel
i have this belief, from lived history, that i should not put my own price on things that i cannot prove are worth that much.
so i let others place value on it based on how they feel about it
if they want to pay 5 bucks, let them pay 5 bucks
It's kind of a gamble
the proof is that somebody is willing to pay you for it, no exploitation
I won't ever accept less then I think something is worth though. I need to be fairly compensated for my time and effort
but then i have to ask if they're doing it out of pity, obligation, or kindness
if you say so
I need to be fairly compensated for my time and effort
If i said this about myself, i'd sound like a narcissistic π hole
what MCP are you using ? i used a IDA's one that is public on github, but interested in your workflow if you're willing to share π
jon bon jovi has a couple restaurants that are open to anyone regardless of income level where people just pay what they want or can at the end, even if it's nothing
it's pretty cool
oo
IDA MCP works fine - it's more the context you provide it. I have a lot of custom tooling I've built surrounding Mach-O binaries as they're what I work with day to day
I wish more rich and famous people used their wealth and influence for good like that
if i was wealthy, i would've loved to do stuff like that
i could help so much if i had the capital for it
ok cool π thanks
You should use this as your drive. When a client offers you a large amount of money, think about the amount of good you could do with it
Also, stop caring so much about what other people think of you
Yep. The importance of using well-known, solidly tested libraries in enterprise software is underestimated by a lot of people that don't work in that environment. My company WANTS to buy long term support contracts for dependencies, because it means our product is less likely to be hit by a bug that cannot be fixed, and conveys a sense of trust that the developer who's dependencies you use knows and tests their product.
LLMs can generate a boatload of code, but with it being a concept that is foreign to your team you will have huge blindspots when it comes to potential bugs and security vulnerabilities.
that's a bad way to say
Your decisions shouldn't be affected by anyones beliefs other then yours
that's basicaly saying, "stop eating healthy food. it's good for you but i said so"
i care, && y'can't make me stop caring about things i care about
it's a system that works
but this
let's cycle back to that thing
this is a good workaround to accepting monetary compensations
it isn't selfish unless blown on my own needs, which is bad unless we mean living costs, which is essentials that require covering
so it actually allows me to pay for the sub, then continue using the rest, after rent is covered, to fund other people's needs
it's a feedback loop
you're onto something
i like that
can someone pls ping me when limits reset so i can remote into my pc when im out and restart my agents
Not really though. It's actually a really healthy mindset. Other people can offer you advice, but if people judge you for your actions and you let that affect you then you'll never be happy
i'm never happy either way. i don't have feelings, just so y'know
i simply think
or i don't. whichever needs one or the other
We are all too obsessed with the opinions of others. It impacts our behaviour, our choices, the way we dress etc.
People make their own choices based on what other people think
it's a system that has worked for me well, && if it ain't broke, don't fix/touch it
then why are you here complaining, i dont get it
I mean I think it is a really broken way of thinking but that's my opinion lol
complaining on which part?
cycle to which part you are referring to
my context window is a little low
yea
/goal on app when
isn't there a toggle?
it works, just doesnt autocomplete the command
anyway imma go eat. you can even put a message out saying that 80% of your earnings or whatever go to charity. or don't publically advertise it, if you think that people will interpret that as you wanting recognition for doing a good thing.
there's 2 sides to every story
i don't see a point in charity myself. i prefer to give it out to people that ask for support personally
like friends && some strangers online, like maybe here
yeah you can even say 'make a goal' / 'new goal' and it'll create it
or some of twitch people that tune onto my streams that have been there early
sponsor codex subs for kids in lic / under developed countries
could be cool, don't think anyones doing that
dude you just said that charity is great and you could give out so much if you were wealthy, so which one is it lol
wait, you said it was charity?
i thought it was restauraunt
i mean that is a form of charity
i didn't see 'charity' as a word
i think we have a very incompatible set of vocab && definitions
the official definition of charity is
the voluntary giving of help, typically in the form of money, to those in need.
i'm not entirely sure how you could interpret that differently lol
giving to people in need is charity
my definition is that of what Estonia calls it: a non-profit registered entry that does something for the people
a charitable organisation is a philanthropic non-profit but charity itself is just the act of giving to those in need
they're usually under the MTΓ
thanks for telling me the difference that i likely won't remember
it's a pretty important distinction for someone with your world view π€·ββοΈ
so anyways... wen reset
i mean, you haven't seen the worldview at full scale, so not much there to say
wrong say
hold on
there
rwen
somebody misspelt Wren language
"codex hack codex and reset rate limits"
Is there a reason that /status in codex app no longer shows me my context usage after the latest update? Big sigh
works fine in vscode extension latest prerelease
i believe it's not easy for a common man to understand my worldview in depth without further environmental dive
y'can pick out the surface easy
anything lower-level needs history
&& is heavily dependant on the environment of the object's (my) origin, which would be the southern side of Estonia
far down the south
I mean I know that you want to give back to people you know that are struggling, particularly to people close to you, which is charity
i just call it being useful
since that's what i'm supposed to do
if i don't do that, it'll be my fault if something bad happens to them after
&& if my own gift-giving caues trouble, also my fault
so i need to make sure i do what makes them feel good/happy.
it takes some effort, but it is really nice to see them be happy about my givings
which to me is the whole point: make somebody's life feel less miserable
'see a need, fill a need' from an old children's movie
FWIW I think it's cool dedicating your time to help people, as long as you take good care of yourself in the process π
i do the bare minimum on myself
just enough to make sure i still can give
i require that others need more than i do
it's about considerations
if i think more about what i want && need rather than others, that'd be inconsiderate of me.
somebody comes up to me && asks for say... help with tech or wants a cheap Steam game: i give
i don't ask for much (if anything at all), so i just really live on the hopes that i don't mess up on the opportunity to be used
Hi guys
hi
Iβm out here patiently waiting for Tibo to reset rate limits in response to the rate limit errors this week
same...
i'm running out soon
I ran out this morning. CC resets for me in 2 hours thankfully and Iβve been using Antigravity in meanwhile to test their 9x increase in rate limits
Nothing compares to Codex tho
Bro, you must attach HEAD π
codex? not unless you run it outside sandbox full-time
that's basically dangerously-skip-permissions
Like it's connected to termux local host
Yeah
Oh forgot
Lowkey who ever made it happen is peak i would say
This is ridicoulos
β {
"type": "error",
"error": {
"type": "invalid_request_error",
"code": "unsupported_parameter",
"message": "Unsupported parameter: 'reasoning.summary' is not supported with the 'gpt-5.3-codex-spark' model.",
"param": "reasoning.summary"
},
"status": 400
}
That error comes from having model_reasoning_summary = "concise" in the config
I guess never heard of gracefully falling back on unsupported settings lol?
Oh dude, they are unequivocally the worst when it comes to ignoring unsupported params
But that's a really good one to make a github issue. If it gets enough π, we might see it arrive in v0.159.0
Haha neat!
It's got very good animations
I use Claude, Grok Build, Gemini and also Codex..and some other Open Weights models to save money.
Repo is solid. Go check
efficient. fast. compact. no bugs. no spaghetti. no security issues. no slop. no black box.
codex down?
no
itβs just spinning forever for me
*whenever i try to login with an account in the codex desktop app, it asks me of a phone number, but when i use my phone number it tells me "Unable to send a verification code to this phone number. Please try again later or use a different number"
idk its working fine for me
5.5xhigh
wen reset
tiboooo
5h47m goal uninterrupted
35h28m goal uninterrupted
Iight, so I'm talking with ChatGPT about the caveman skill with the page open, and ChatGPT is slowly responding in caveman π little things here and there. Fascinating
anybody having this issue?
always lol
still have 37% on mine. won t mind for reset to come later haha
probably by multiplying the token usage with official api costs. curious if it's complex enough to bookkeep every turn and write the model used and the token delta. might be possible, but i think you must do it at turn time, because afterwards you only get tokens stats for the entire chat id
na I got it to show up haha
ah finally! the reset I prayed for
GOtta stop gambling with this lol. Had spent 80 % after 2 days after last reset on 200 dollar plan. was down to last % now lol
How you guys able to see Token and token costs are you guys using API i can only seem to see percentage still
I was geting used to composer 2.5 fast speeds now haha, got mixed feelings when I go back to abusing the 5.5 again with its fairly "OK mid slow mode"
RESET!!!!!!! I had 0%
omg i should've ran it down. was 15
idk if to switch from high to medium tho
I would need to see the difference in usage if its significant or no
question now is, do we gamble on another reset in 3-4 days? or is the march madness of resets finally over
I had an all nighter coding and now it's reset - awesome!
Less goooo
how can I define if the tasks require high inteligence vs medium or low? Is there a propper way to do it or just intuition?
I just medium everything, its plenty smarts and faster, even low works well for many things
if I do happen to bang my head into the wall, I high or xhigh it. but I pretty much never need to
key is just good planning ahead of time
hmmm but You see, my project is cind off autonomous cause it emits pormpts for itself each run and he reasons about the context he gathers and sutff and idk if I lower the intelligence he will forget about some rules
I was running fast mode all night, I turned it off this morning because my faith in Tibo had wavered
Like he emits rules fynamically for himslelf but there is so much to reason about that he could ignore smthing
π blessed be
wen reset?
done
looks like the app updated, how do we check out token usage does anyone knows?
I use CodexBar (3rd party)
theres definitely a log though as it can pull in usage from when it wasn't installed, I'm not entirely sure where it is
CodexBar is OSS though so you can clone it and ask codex where it loads that info from
okay nevermind a guy on yt made this vid I Tested GPT-5.5 Medium/High/xHigh Reasoning Levels: CLEAR Difference
try it I guess
I guess get myself a backup and test
high tends be better at follow instructions, and just deeper work generally. its good for planning etc. but for coding work itself think its probably overkill
cause each run he gathers 70% of context in order to reason so idk bro
use high/xhigh to make a solid plan, then use low/medium to implement I think
its huge porject
divide and conquer as with any project I suppose
risky to try cause I cant just duplicate a thread to have the whole raw context saved somewhere cause then I could see the reasoning difference on med vs low vs high
I know that on xhigh it was overkill
hmm mine didn t reset yet
Oh man, what a great time for a reset
why the f was there another reset
Oh yeah hasnβt propagated to me either
it did here
Why not
They had released an optimization that messed up caching which burned up usage for users
you mean redditors
tihs si edualc
I spend 2 days on 0% and felt relief when reset hit
hell yeah
Same here lol
translate?
No, I'm talking about project ideas.
Expand on your current one, deploy it. Thatβs gonna keep u busy for a while
I've already done everything I had to do.
One seriously hot tip for design, download OpenDesign and give it a go. Can use your codex sub cli for it, or any other. Try it (I use kimi for this one, but kimi and gpt 5.5 are about equal ish on design imo)
I was lost.
Unfortunately not, I made it with so much care that I lost my way.
I got a reset
few minutes ago
which?
context
does not reset, it compacts π€£
I feel lost.
Gpt 5.6?
what do you want?
you could ve just add captions to ur vid bro
I don't know either, bye.
damn
third world countries, is what it is
I think I will gamble and go from high to low, wish me luck
will give feedback when it breaks
Haha, low is actually my go to #1
but you use it for coding right?
ofc, but I always make plans for every thing I do with high xhigh/opus extra high max depenidng on what am doing
cause my those all 5
1 search context
2 valdiate context
3 implement new actions from plan
4 caldiate if he implemented them correctly
5 emit prompt taht will repeat this process
something like this
implementation there is no need for more than low, sometimes medium. Ive used composer 2.5 lat days when I was low on usage, had codex / opus plan, once the plan is tight, its not a big deal which model these days implementing, (given you actualy code review after)
and 6 (not often) change whole plan based on new context
Did that guy just refuse to take a feedback n leave?
think so xx
I even translated his speech but idk waht he was talkin about
speed + low vs slow + high
Told him he could expand the app since it looks like it got potentially, bro said nah
he may had some self feedbacks after sending that vid that made him reflect on the idea of asking us about it, he may decided its best to brainstorm it himself if u know what I mean
tl:dr what he made? we can make it here and ship before he wakes up
that`ll show him
5.3 spark is there for a reason
Codex clone this app, make no mistake
forgot make safe for hackers
/goal also add light and dark mode
Does anyone have a project they wanted to make but never did? Now that Iβve finished mine, I feel kind of lost.
Iβll build the whole project and send it to you.
Prompt: Make $3B project, make no mistake
When can I expect my project to be delivered?
He probbably thouth it would take more time to explain everything to us in english than just message chatgpt in his language
No, but thereβs no way to do it like that. It doesnβt even make sense.
EXACTLY
Did they reset usage?
Did you guys have any questions about my project that I can answer?
Unfortunately, I can't.
$2B project, no mistake
I test this prompt on my promptbridge gpt and here is what he came up with I mean its a prompt that may work or some
whats the url to your project? if you dont mind sharing your page = )
yes, just started it
i believe it will be my life's work β€οΈ
localhost my beloved
alrdy tried :3000 and :8080 came up empty
for the next week of my life
It's on GitHub, but I'm afraid something might happen regarding piracy.
WTH 5.5 low just copnsumed more usage than high damn
reset wen
you havent deployed it yet? ment for your lanidnpage
little bit like real life lol
No, I just did
This chat will keep me entertained while waiting for the reset. Thanks everyone whoβs participating
more care
No way, how
Is that Spanish?
damn but same pormpt maybe the pormpt and actions enforneces so much reasioning that the low plan is beeing enforced to such elvel of reasoning that it uses more tokens LOL
opensource the idea? no monetizing?
Create an issue for it. Our lord Tibo might give us another reset
idk but high consumed 2% and low 2% too on plus plan
of weekly
There's no way.
They messed up caching again?
porbbably my project is that weird
now I gotta go nuts this weekend. reset coming guys.
U can always put 10 pops up to it before the user can upload an image
remove your access key from the repo mate
My idea was to charge per key.
I forgot, oh, thank you.
NOOOO WHY
Not a douchbag lol haha, come on be nice π
I had forgotten
Im kidding, but itβs weird github havenβt implement any feature that blocking you from uploading .env key
In my case, I donβt think I even have that, because I donβt use an API. Everything is done manually. So I donβt think mine even has that. I think.
The short answer is ~29 million secrets were found exposed on public GitHub over the last year.
According to GitGuardianβs 2026 State of Secrets Sprawl Report (released in March 2026), a staggering 28,649,024 new hardcoded credentials were detected in public GitHub commits over the course of 2025. This represents a 34% year-over-year increaseβthe largest single-year spike ever recorded.
The data highlights a massive shift in what is leaking and why, largely driven by the explosion of AI-assisted development and agent infrastructure.
The translation is only in English for now. The entire process of translation, speech bubble detection, original text removal, and inserting the translated text was built from scratch. I didnβt use any API key.
according to gemini atleast
Is this AI generated
ofc lol, that stuff is not in my head lol
Yes, everything.
"Public vs. Private (The Hidden Danger)
While 29 million secrets hit public GitHub, the report notes that internal/private repositories are actually 6 times more likely to contain hardcoded secrets than public ones. GitGuardian found that 32.2% of internal corporate repos contain at least one hardcoded secret, operating under the false assumption that "private" equals secure."
more concernign tbf
My project works like this: you upload an image or PDF, it detects the speech bubbles, removes the original language, and places the translation on top using the same font and size. The entire system was built from scratch.
I meant this
how much it costs for one image?
and those it work for chineese too?
like from chinesse to english
it would really help if it cheap or even run locally
So, since Iβm not using any AI, like any kind of AI API, the project was built completely from scratch. It has no usage cost because it doesnβt depend on any external API. And for now, it doesnβt support Chinese yet.
Considering OCR on the full page first? Google Cloud wold probable give bettter translation, but again 20 bucks for about 1000pages for opensource is not doable i guess
At the moment, it only supports translation from English to Portuguese.
That's why I did it from scratch.
So u use a machine learning model instead?
Yes
make a premium feature, 9.99 per month, send to google cloud translation for premium translations, and store it in db for reuse on next one. The popular mangas will be free for you to generate after first (only db network costs) and it could actually generate a few bucks per month
I might turn back high at this point
Do you think anyone would pay?
Not into manga myself, but i understand the world wide fanbase, so prob if the translations are good i guess. worth a shot?
I'll try then.
You need to do market research Bro
that's right
Iβm going to set the price at $1 for the version without Google Cloud and $5 for the version with Google Cloud translation.
Thereβre alot of manga looks amazing but there just isnt any translation for them
This could be a good use case
Codex, make the same app, no mistake
And it's even simple to do.
Which language do you recommend?
Every
Right now I do it by hand, I don't use a pen.
Wow, now this sum Iβd pay for
But they use an API, I don't.
But then, I could tell codex to do it LMAO
but if you could utizlie gemma 4b or apple intellegence to do the translation on your local comp, sell it for 9.99 lifetime, that could do something. not sure if the models will suffice tho yet
Yes, I made mine using OpenClaude.
Wdym by hand?
bro is full of mysteries
I think he dips his finger in chocroal
All my code is written without using APIs; I did everything myself.
HUH
so UnVibecoded
what are ytou doing here then? this is the lazy zone
I'll try using Gemma 4.
I also used OpenClaude.
Ah so 100% local inference then?
bro is Socrates
Yes, everything and location.
I was even going to use a template in the translation, but I ended up not doing it.
That's awesome. It's amazing what modern hardware can do these days
I think the same thing too.
Feels like im talking to a llm
?
You're absolutely right!
Technology am I right
Letβs unpack that ~
This project is revolutionaryβyou gotta expand on this idea fast, this is urgent
So give me some language tips.
I?
No itβs a recurring theme
true
Ok
Youβre absolutely rightβwhatβs an idea without any starting point
I'm going to add English support for Mandarin/Chinese.
first time i ever had this happen
Did you tell it to make no mistake?
No, I'll do it by hand.
π€£ lol i never use 2023 intel. we re in 2026
Claude with open router model inside a codex file. Heβs on to something
Much better
my aesthethic taste is most probably lame since i mostly care about functionality, but i sure love i can put the info i want where i want in my custom codex ux
yes this is extremely disgusting but you like it so who cares
I made a script that automatically switches to another free model whenever a free model reaches its quota limit.
Yeah bro same here. Idc if it looks funny, as long as it works flawlessly
I can't design but I have extreme dislike towards ugly UI so much so that I'll refuse to use a product because of it so I just stick to text based
download opendesign, and you will be a god
I had a look at some of the examples, they're alright - I'm happy slowly learning Figma & design basics and trying to improve my creative thinking first
just hook it up to an codex mcp, and let them figure it out between the ai`s over a bbq while you fiddle with figma xP
I've literally had my weekly tokens RESET yet again
hahahah - i'll stick to my cli's for now. I presume there'll be some significant developments in the AI design world soon
Thank god I didn't go for the $200 option
you phrase that like it's a bad thing
I bought a second $200 plan an hour before they reset π€£
terrible timing from me but I have more capacity now
I'm angry at myself as I came sooo damn close to paying
your 2 hours to late, thats missed opp and usages, the caching aint working, use it fast for incommng reset soon
Put yours on fast speed
yeah fr you gotta make codex call you every time theres a reset & sprint to your computer
How good is opendesign, give me 5 examples @karmic gulch
2026 AI coding be like "Do you approve?"
I actually am on xhigh fast rn lol, but because I have a deadline. Codex just sped up the execution of my emulator from 177s to 12s after 2 hours of debugging so I'm extremely happy rn
I havenβt switched off xhigh yet
woot woot
hold on - asking codex to answer that one
Donβt like the others
Auto review permissions on top
I can't deal with the performance decline - Codex is dumb enough rn as is lol
Well they all are dumb gpt and claude regardless
They must have big latent spaces with more relationships
I don't use Claude anymore. I don't agree with Ant's business practices, and I wasn't happy with the value I was getting out of my 250$
but for real, just download and see. they have implemened a gallery of open designs peoplke share also, so you can download, - browse and se for yourelf. Im working on a corperate crm system for a client, and it finished a 35 page system design package just about now, some time to actually get the design package as ytou want, but once thats done it generate the design pages perfecly
Business loves Claude for whatever reason
It's too sycophantic for me
I like the AI to take responsibility for its mistakes - it's not my error, it's yours. It also seems to lack the ability to determine the truth of the statements that it outputs. I had way too many experiences where it would claim something which it would immediately figure out was untrue in the next turn
thye got fooled. claude were #1 for about 1 year, then openai stepped up, and the CEOs are hype machines, so they make deals with anthropic... give it 5 months when claude takes the lead again, im. sure thye will be doing deals with OpenAI again xD
I'll give it a look next time I need to generate a design. I've been meaning to use Image V2 to gen some designs for a while
Speaking for myself, I built my entire system using OpenClaude because it delivers results. Thatβs why people like it
Yes
it also utilizes the images 2 when you want, for designs within. and you can use pretty much all llms cli or byok. your choice. im into kimi atm , think they deliver the best UIs atm, 5.5 just behind tho
Gets confusing asf
i know this is old but i wanted to ask if this works for you? Iβve been using claude to run deep research to have it create detailed implementation spec with full code/code skeletons, but when asking ChatGPT to do the same thing it doesnβt do it
Did you make it lmao
The place im working for, planning to switch there contract to Claude , I was like NOOOOOOO
would think so the way you're shilling
it says that it wasnβt able to access the actual code in the repos
b..b...but Mythos tho π€‘
lol, just preaaching good tools for the community, ofc could stop it, and let ppl never find it π
From my usage of Deep Research it normally has a very specific structure it sticks to - Pro is a different model and is more like 5.5 on steroids with a much larger thinking budget per turn. It used to average 60 min -> 90 min for me but long threads are practically unusable due to the design of the WebUI
translation from Japanese to English
I got Altman and Tibo on my bedside table
v cool
I can assure you that it doesn't say I hit you
I'm still adjusting, I just started.
There is absolutely no way all of that Kanji translates to "I hit you"
Tanoshiina
We only do encouragement here good sir
"It seems fun that you...."
It doesn't say hit in any of the sentences : p
Tell codex good try
I am an old weeaboo
Yes, I'm adjusting the pipeline.
I'm just playing, you could leave it like that and noone would know anyway
That's the fun of noone knowing japanese
But then my job would be a disaster.
did a reset happen ? β€οΈ
Yeah, the story doesn't have to be whatever they wrote originally π
Dont worry, great things take time
Do you have feed
Quick question, do you guys use 5.4 or 5.5 for QA testing?
5.5 Medium for pretty much everything, 5.4-mini for implementation
5.5 xhigh fast mode for everything
all your compute belongs to me
/goal extra high for a whole sprint and then starts doing small tasks
That's egregious
Do you think using the ChatGPT API will improve the translation?
Depends how good your OCR is
It need context for better translation
If your OCR is the issue doesn't matter what translation service you use
So my pipeline has
Which pipeline did you build?
Interesting... The second bubble was actually almost coherently close to the original ("Don't worry" | "Don't let me disturb you")
I use llm for translation for the past years. Google translate just dont do it for me anymore
Which one do you use?
The good one
You should honestly try PaddleOCR. It's SOTA for image OCR
reset happened but limits still feel like draining much faster even tho they said they fixed it, anyone else noticed that? am on pro x20
lol. I mentioned to codex yesterday that I'm just not using my quota and we had a little conversation looking for something to do and as soon as I said something that looked feature shaped it ran off an ddid it
You're right to call that out.
For me it was always quick, it wouldn't be over in 40 minutes.
Absolutely β
It is bad, but it is not unrecoverable.
β
Yeah I tried both 5.5 Thinking and 5.5 Pro, both had issues actually viewing github repo contents
iβm considering building a custom MCP that just provides better github specific functionality π
Doesnt github have an mcp?
I tried to contact them but couldn't.
You're right to call that out. I'm gonna stop hand-waving and check the actual source.
AI cant even help me understand this convo ππ
Iβm sorry. I did contribute to that.
Proceeding directly to the next tranche of gaslighting now.
all quotes from my current session btw, which is a frustrating one π
Huh, usually GPT 5.5 Pro can't access the repos because it doesn't have "apps" tooling. I think someone in here resorted to exporting their repo as a .zip file and uploading it to the convo for ChatGPT to operate on or something
yea works nicely
Yooooooo those it work with chinesse too?
Yes
the chatbot versions get almost excited if you give them stuff the coding versions usually see
Can I do chinese to Chinese π€―π€―
I jave better idea than manga
?
It works for me
Just looked my latest commits
which?
How much it costs per translati9n
Dont say it
I was doing code feature by feature in chatgpt until it started to get too big and too many inconsistencies, codex is definitely nicer
Why
I'm not using an API, everything is local.
Cuz i know whats better than manga
So it uses ur CPU?
yes friend
Yes
So we have 2 ideas then
Accurate? At chinsse tho? Cause its super tricky
No, in Brazil.
Just use deepseek
Well actually any LLM since thatβs the one thing theyβre good at fundamentally
What bazil?
Scroll way up for the lore
there is only one
Last time I tried it, the "Pro Thinking" trace is empty, but it was using python to try to HTTP request the github instead of using the app connection
Ok
Try now
Oh you did
Yeah it works for me havenβt had problem
where is my pro theme color black/whiteπππ it was so good
I can't because my OCR searches for text balloons and there aren't any in this image.
So youβre going to sell this to manga/comic publishers?
and for those who read as well
Well if the readers get it from rips and unofficial sites you could get hit by tos
I'm still starting out.
https://github.com/unclecode/crawl4ai anyone tried out this one?
you should make it a browser extension so you right click the image and slap the translate button and it replaces the manga with the translated version π
Oops, I hadn't thought of that.
thanks
Damn Zed is fast. That rust engine truly is something. anyone else using it for codex?
Where is this?
I couldn't find anything about Agent Zed.
https://zed.dev here mate
Like the Codex?
I'll test it.
do it, you wont go back = )
Now tell me how you did the translation. ?
Oh forgot to answer, no magic at all. Flash 3.5 extended did the translation, and then called nano banana to generate it. Its not viable at all for costs ofc, but shouldnt settle for anything less, if you want the app to geenerate
I had thought about that, but it would be too expensive.
fast at what? the overwhelming majority of time you spend when using models is induced by the models themeselves
Optimised probably
why would i pay 10$ for 5$ api credits though? :))
use it with codex mate, its free to use with codex...
unlimited edit predictions would have been a thing 5 years ago
yea but why, i already use codex app, and my own custom ux
more of a general question about 3rd party tools, not zed specifically
Use case vro whatever works for you
if your happy your happy. For me the speed and smoothnes of it all makes zed #1 withut a doubt
speed as in app smoothness i assume right? like the app itself is fast
it's face melting in comparison to anything else out there
bordering on sublime text speed without being just a notepad
PSA MacOS Devs: Add terminal of choice to developers list. If you don't you are wasting a lot of cpu on syspolicyd trying to deal with untrusted chains (like rust).
will improve performance significantly
but no markdown preview.. π’
sometimes I feel like the only reason I am even using vscode these days is quick review and right click > preview markdown. Also this is a problem actually in that I use md too much, and probably everyone does when its not the best format for most things anymore... and is a goto crutch, but that's another story.
Yeah, that's precisely what I use VSCodium for. Render markdown and mermaid diagrams π Zed for everything else. That multi-buffer search feature is insane
yes mermaid! I forgot to mention that. HUGE crutch
we need a new modern file format I think
https://claude.com/blog/using-claude-code-the-unreasonable-effectiveness-of-html?utm_source=tldrai
it actually pretty accurate but I think we can do better in this world of ai
I need to force myself to move away from md to html for a while to test
I think there was a tool that let you render react inside markdown or something so you can kinda achieve similar results without doing straight up HTML
.mdx or something
but html is ubiquitous and prob better for llm to understand
It's also incredibly verbose and not reusable. Of course Anthropic is gonna recommend HTML, because then you use up your entire quota on generating rich documentation π
you aren't wrong
html has some nice rules and constraints, we can lint it and make sure its right etc. markdown not so much
This is the only reason I donβt use html
Otherwise itβs effective
Hi guys
btw why VSCodium vs VSCode?
Less resource consumption, and I only have the mermaid extension
no rust analyzer?
this is zed side for you?
Yeah rust analyzer in Zed. OOB it works way better
ok cool I will have to try that split
I want to fall in love with Zed. I just havent yet.
It has the best remote dev I've ever used! Totally flawless sync between 2 machines, very efficient use of resources
Is auto-review not working correctly for anyone?
It's prompting me for permissions for every file edit on the newest version of the Codex app
I fell in love with Zed because it's so much faster then pretty much all the mainstream editors. Low memory footprint, quick startup, great large file support
I dont have much use for remote dev. I use proxmox/viscosity/ansible for that , but the lean and mean is why I am interested in zed. and of course I'm a big rust head
I like it because you can deploy a debug docker container, remote into it, attach debugger to the process inside docker and get full capabilities
ok that's interesting. I do use podman a bit
VSCode's remote dev was great for me when I used it way back when, I just couldn't deal with how abhorrent the performance was
& Zed's collaborative editing is way better
yeah I used to use vscode with ssh capabilites but I found it was the wrong model entirely
I pretty much switched to local dev, ci/cd to staging and then to prod on mr
Yeah, the reason I do it is because if I run an API server on my mac, totally different architecture than the prod server it will run on, and you can catch those nasty "works good on my machine" bugs π
yeah I do builds on a proxmox vm, unless remote on vpn, so 2 pipelines. most of my targets are linux but I work on macos so often doesn't make sense to build on macos
exactly
I like Proxmox, and I run it in my homelab but for production deployment I typically skew towards dockerized systems (portainer / kameleo for smaller deployments, gcp cloud run / k8 for enterprise)
surely this shouldn't prompt on auto-review no?
yeah. I do the same. I still run podman within devnet vs prod in some cases... but sometimes for rust microservices I dont use docker at all and go straight to systemd
A git add command?!??!?!
Never used podman, maybe I'll give it a go
systemd can be a pain with automated deployment
I mean its basically just rootless docker, not really any difference in this context
that is what ansible is for
# ~/.codex/rules/01_safe_git.rule
prefix_rule(
pattern = ["git", "add"],
decision = "allow",
)
That'll permanently allow it even on workspace-write
I'm more annoyed at the fact it's prompting in the first place. On the last codex update, it'd do pretty much anything unless it was unclear if I'd like it to be done, e.g. dangerous changes, deleting files etc
It's now prompting me for file writes, or adding files to git - even with a goal engaged
I love it too - except now it seems it's not working, and ideally I'd like to head off soon and leave this goal running overnight
Hey, apparently approval policy has to be set to on-request or granular, and approvals_reviewer set to AutoReview
is this in a .toml?
Yes
Apparently the default thing is approvals_reviewer = "User" lol in the latest version (if unset)
I'm not sure. I think approvals_reviewer is a top-level and a profile config. Idk about per-thread override. Also I was wrong, it's
approvals_reviewer = "guardian_subagent"
The only bad thing is that my laptop is crap.
Theres a couple issues on github that mention codex dropping approvals_reviewer = auto_review from the config
Adding that line seems to have fixed it
Crisis over haha
You know what? I had v0.130.0 checked out lol but I'm glad it worked anyways!
Yeap honestly love Tibo xd
My usage is already dissapearing
Down 10%
Tibo for world president
on $100 plan, i already burned more than the plus plan after reset lol (assuming plus=1x and pro=10x)
dunno what im gonna do after May when the 2x promo ends
The Pro 20x appears to be the best possible value
Get the $200 one, it's on a 20x until further notice
i feel like i dont have enough time in a week to /plan 20x token worth of changes usually
maybe i need to change how i work? I usually create task files working with the agent, then run them overnight while I sleep in a ralph loop
