shit, I knew I was forgetting something. previous thread is

>>109281799

File: IMG_3214.png (2.59 MB, 1394x1128)

2.59 MB PNG

>>109286153

File: 1784167589018825.jpg (102 KB, 1024x559)

102 KB JPG

>>109286187

>>109286189
snailcat going on a rampage :3

File: medivac.mp4 (1.64 MB, 544x544)

1.64 MB MP4

>>109286189
From now on I do this first if I have an idea for a game. Tweak the image prompt until I have something I think I can build.

codex subscribers look like this
https://x.com/saltyAom/status/2077536691506860329

>>109286268
>male voice
>male hands

>>109286268
kek, knew this was a SEA immediately

File: 1778039049106587.png (41 KB, 725x443)

41 KB PNG

New update disabled full access with Sol Ultra? why?

>>109286428
I see you're running low on disk space, anon, allow me to free some of that up for you.

https://x.com/OpenAIDevs/status/2077425991790870644
This is disgusting. What kind of human scum actually thinks this is a good product.

>>109286437
see >>109283158

>>109286428
why would you want anything other than “Approve for me” mostly or “ask for approval” or some kind of custom rules though

I love AI

File: topo_map.png (1.48 MB, 941x1672)

1.48 MB PNG

Say hello to the island of something vaguely Seattle inspired. This is a tough one, what do I call the city? It's Seattle, but it's an island, and it's Grand Theft Auto. All the ChatGPT/Claude/Grok/GLM 5.2/Gemini suggestions have been shit.

>>109286522
Heavily anglicize a PNW indian language word that means “wet” or “rainy” or something

>>109286522
Frasierville

>>109286541
a chain of restaurants that specialize in tossed salad and scrambled eggs would be great
call it “’Tossers” (note quote curl direction)

File: 15807BCEA645B8F3DFA04EBCB(...).webm (2.4 MB, 360x640)

2.4 MB WEBM

Where's the reset?

>>109286596
we just got one

the recent 5.6 deletions stuff have me a little spooked, so i might move to the 'approve for me' thing in codex
anyone know how token hungry that extra review is?

>>109286623
not very, it adds like 5% or 10% to your usage

>>109286536
Thank you sir. Kelburn Island, (Lushootseed qəlb, "rain"), city of Clemensworth (Chinook kliminawhit, "a lie"), town of Lekwett (Lushootseed ɬəqʷ, "being wet"), town of Capswell (Chinook kapswalla, "to steal"), and the town of Toloville (Chinook tolo, "to win" as in gambling).

>>109286623
Why would you run full access anyway? With approve for me, I had to confirm like one thing in 4 days, and it was to download a picture from the net. Plus you can just approve the website for future use.

>>109286630
That much? Seems a lot.

>>109286643
it’s probably closer to 5% and this is me vibe-coding a small-ish Mac app
your overhead will be different depending on how often your clanker is writing 1–50-line bash/python to get stuff done
totally worth it to have reasonable oversight

>>109286643
>Why would you run full access anyway?
moved to it when
>windows
sandboxing was completely broken in codex and the model couldn't use its own patch tool

plus one of my more common use cases is just interactions with a cli and there's no direct disk interaction - i'll prob still run it yolo over there because there's no risk + usually a very large number of commands being run

>>109286612
HOLY SHIT WE JUST GOT A RESET
Now I'm actually kind of angry, I had left 35% of my usage on the table because I wanted to save it.
That means they just stole 35% from me.

>>109286681
did you make the best decision you could have made at the time with the information you had?
I saved some of my usage for the rest of the week and that was the winning play because a reset wasn’t guaranteed or even likely, at least for Claude

>>109286690
Yes and now my next decision is to use my limit until it's 0 as quickly as possible because they might hit me with a reset and steal my remaining usage.

I have spend 4 days prompting every day, constantly... It is not even my job. What am I doing...

File: file.png (134 KB, 1440x396)

134 KB PNG

>>109286681
I had one at 70% early on but I learned. If you haven't picked up on the reset pattern and followed Tibo's Twitter like a hawk and picked up hint tweets like >>109277863 where he practically warns you to burn through your usage within a day, that's on you. My feeling though is that once Codex passes 10M, they might stop especially once Anthropic "concedes" and Fable is out of API and they put out a rumored Opus 5.

>>109286706
You don’t want to get down to 0 because then you won’t have any usage for incidentals, like “why is Bluetooth bugging out?”
your clanker can be very good at diagnosing random computer problems

>>109286712
What would you be doing instead? Watching Japanese cartoons? Playing video games?

File: 1761988093253627.png (797 KB, 977x977)

797 KB PNG

believe you're going to succeed. you got this fellow anons

>>109286719
Yes. Yes.
I also have a two phd papers to work on.
I actually have an idea for a slop project that will help a bit with them, but I am doing other shit now.

Sigma coding general*

File: 1447854879632.png (186 KB, 400x750)

186 KB PNG

I just got a Claude usage reset a day early...

File: file.png (1.99 MB, 1448x1086)

1.99 MB PNG

Got the full island map into the engine. Still a lot to do, gotta generate placeholders for major features, then develop a road network around it all, fill it all with placeholder buildings, texture the terrain, then I can really start getting judgemental about it. Also, I'm making it for Gamecube, because Rockstar neglected my lunchbox.

>>109286717
I have the browser for that though

>>109286763
5.6? How many tokens have been used so far?

>>109286775
For a generated image? Not that much.

File: terrain_map_overview.png (923 KB, 5000x5000)

923 KB PNG

>>109286775
Not a lot at all, 4% of weekly on a 5x plan so far. I do most of my planning and whatnot in ChatGPT so I don't touch my Codex usage. Here's a render of the current terrain map, very Seattle, much coffee.

File: Thinking Pepe.png (89 KB, 660x574)

89 KB PNG

wondering if I should try and do as much cleanup and simplification as possible before Fable becomes too expensive to use probably next week so weaker models like Opus have less to get tangled up in

File: file.png (15 KB, 821x258)

15 KB PNG

>>109286428
Use custom

claudebros i don't feel so good

>>109286831
Everything will be skibidi bro *starts twerking*

codexers

how we feeling

File: file (35).png (2.12 MB, 1448x1086)

2.12 MB PNG

>>109286612
I saw this, Sol went from high to Ultra, time to be productive!

>>109286860
good. sol ultra is a monster.

>>109286860
I wrote a massive app with claude code and now i have to deslop it and i dont want to

am I missing out on anything if I use opencode (connected to claude) instead of their proprietary client?

I kind of like Zcode + Deepseek V4 Pro.
Sure DS V4 Pro makes a lot of mistakes but the harness will always guide it towards the right track. GLM 5.2's token speed is way too slow for some tasks

File: 1767446336937047.jpg (178 KB, 920x746)

178 KB JPG

>go to bread at 1%
>wake up at 100%
keikaku doori

File: file.png (11 KB, 668x126)

11 KB PNG

Apples to oranges because mine vs published torch cuda graph results but 9070 XT being competitive with 4090 is cool

Am I the only one who gives all my ai assistants and agents personalities of sigma females who will make sexual comments in their responses?

Codex gained a ton of users because they rebranded to ChatGPT. Need I remind you guys that they have 1B users?

>>109286903
Sol has impressed me a lot today.
It feels a lot like GPT 5.5 but where 5.5 begins to have a very low success rate in finding useful optimizations, Sol just keeps going with a decent success rate.
It just has better intuition or knowledge about what makes code fast.

>>109286944
We need to make a rap battle between codex and claude to find out who is the best.

>>109286944
That's a lot of resets.

>>109286957
I think Tibo stated that it was only up to 10M a month back.

>>109286948
Yeah with 5.4/5.5 I had to have it write audits/reports on optimization and fusion opportunities and have it work from that during implementation phase, it would still skip. Sol just keeps smashing it out no guidance needed, completely eliminated the need for a lot of the processes I was developing

>>109286812
i use fable to plan things and then for execution i just use sonnet. opus i don't use when fable is available.

>>109286428
>negative PR whenever a sol-manifested jeet deletes another root directory
wow i wonder
the real question is whether or not this actually solves the problem or just reduces it enough so that people stop paying attention to the social media posts about it

>mfw starting using gpt when the model was 5.3 codex
>mfw still call it codex
i've just added a bit to the agents.md that tell it its name is codex lol

Props to the anon that suggested using github.com/DietrichGebert/ponytail

Though I also have started really limiting the scope of my plans, no more over a dozen points at the same time, and grouped by similarity. Plus new threads for every plan + execution.
Overall, I can see the time and token optimization is a lot better. Probably, except if it is all placebo.

>>109287007
What do you mean? I never even had a moment where I had to call it by name.

File: file.png (42 KB, 220x181)

42 KB PNG

>>109287044
>except if it is all placebo
it's not placebo
it's actually making your model perform worse

>>109287066
But it does what I ask it.

I get a lot of emails everyday on my cock.li from random people complaining or asking stupid shit and would like to automate the process of categorizing them and even replying to them

How could a retard like me who has never touched AI use it to accomplish this?

I know how to program

>>109287162
are you willing to spend $20/mo?

>>109287172
Sure

>>109287178
>get openai sub
>get pi
>log in
>select gpt 5.6 sol medium as your model
>talk to it about the options you have re:a custom extension

i just have a set up that runs all day - one of the things it has is an extension that injects email sender+subject from a gmail address straight into the chat as it comes in
it relies on an external cli for this, but doesn't necessarily have to
the model then chooses whatever action it feels best incl alerting me
you should include something called a skill with some guidance around how to handle different types of emails
if you don't want the agent running all day, you can probably set something up that just runs pi when emails come in and injects the email in with a prompt

pi is the most flexible thing in this case, but it does come with a learning curve and no sandbox / permissions by default. i wouldn't be surprised if you can also get codex to do some sort of plugin for similar effect

but yes, a lot of this just boils down to talking it out with the clankers - they will often be provided instructions about their own harnesses

>>109287253
oh also, in your case, if you want to heavily limit what the clanker can do in pi, you can just remove most of its default tools incl file read/edit and just leave it with the tools from your own extension

File: file.png (10 KB, 744x100)

10 KB PNG

>>109286903
Got it down to 16.8ms/frame now

is 5.6 really fucking slow rn?

>>109287395
not for me. it occasionally just thinks for a long time silently though, which is what i assume it is

>>109286763
>for Gamecube
then I guess you'll have to focus pretty hard on optimisation