File: ai-gateway-providers.png (43 KB, 400x468)

43 KB PNG

unity AI open beta today
>Editors / terminal agents / coding agents
Codex is the best right now not even close, probably all you need for noob level vibe coding

>>108753157
codex guides need to be in the op

>>108753215
So is this like Proxima but worse?

>>108753157
in reality the images are reversed, the code from AI is spaghetti shit

How do you actually
finetune your projects?

>>108753480
you dont lol, not even codex can

A week ago, from Anthropic, I got a free 1 week trial of pro. I waited to sign up, and missed it because I was busy.

I made a new account to try again, and no offer. How do you actually get the 1 week trial?

File: Screenshot from 2026-05-0(...).png (121 KB, 1952x624)

121 KB PNG

I love how K 2.6 writes. It's just adorable.

>>108753541
>thinking
Sorry but you've used up your limit.

I know my question might be stupid but how do you guys make money by vibe coding? What do you do, what do you sell?

What the FUCK is up with these usage limits? It's actually fucking unusable now. I paid £20 and got maybe five prompts out of it.

File: Screenshot from 2026-05-0(...).png (618 KB, 1952x1982)

618 KB PNG

>>108753480
I'm trying. Right now I'm running a benchmark after the first sft pass.

>>108753277
op needs to be thrown out completely its filled with complete slop like lovable and bolt which no one fucking uses
why are the fucking vercel docs there?
codex is the most popular sub here and it isn't even mentioned

>>108753564
you need to pay $100 a month. $20 is the teaser trial.

File: hq720.jpg (49 KB, 686x386)

49 KB JPG

>>108753555

File: 1768540889963.png (246 KB, 492x442)

246 KB PNG

>>108753480
PRD prompts, have a conversation with Claude or Codex about the project then it will generate a technical PRD

you take that PRD and generate a technical spec from it

I probably shouldn't be telling people this.

Having the right spec is critical. It determines everything, and whether youre stuck in debugging hell of not after the project is generated

>>108753591
Is the £100/month actually worth it or am I gonna get cockblocked every 30 mins like I am currently?

>>108753603
you'll get cockblocked regardless. claude is awful unless you have a mega corpo account

>>108753564
You either bought the wrong one (claude) or have skill issues. Codex $20 is enough. If you have skill issues you need to pay at least $100.

Where did whiteboard anon go?

>>108753647
For me, its Deepseek API + Codex $20. Deepseek does the big overview stuff and hands precise work to Codex, like one file. And Codex also does the planning. I can make my $20 limits last like this and my Deepseek API use is negible

>>108753555
Make something cool that people are willing to give you money to use

I feel there's a lot of value working with two good models and having them evaluate each other's work. Whether it will be true or not, it gives me the impression that it decreases the chances to go through a wrong changes spiral with any single model.

>>108753555
Make money? In this economy?

File: 1732127166463.png (1.5 MB, 1920x1080)

1.5 MB PNG

Anybody want to be partners?

I have a bunch of stuff built/half built. I feel like two people working together to get more done and be more successful. Im not paying for anybodys expenses, though.

you can email me at clownpeace@waifu.club

>>108753674
Uhh shouldn't it be the other way around, codex's smarts for planning and detailed instructions passed to dipsy?

Am I retarded for still liking Opus in favor of 5.5? I feel like I am being gaslit into liking GPT and it's not bad but there's just something about Opus man.
It's more concise, it designs way better, it doesn't go off on weird tangents and doesn't leave half finished PRs like GPT sometimes does.
I also just vibe more with it

>>108754010
What are some things that you're working on?

>>108754072
deepseek's kind of retarded

>>108753215
there goes my moat lol

>>108754107
But then why would you use it for planning lol, that is the most important phase, where the agent searches all the files and docs to plan what to edit.

>>108754072
Codex does plan in my system. Hermes handles the execution/orchestration of the plan, and calls Codex for precision implementations, or help. It's designed to reduce the tokens that Codex gets.

It was Hermes/Deepseek/Claude but I just stopped paying for Claude the token limits are too terrible.

File: NewtonOS Keyboard with ⇧⌥(...).png (11 KB, 592x254)

11 KB PNG

>>108752058
>—
itt: Apple-keyboard erasure
>>108752604

Read the source and then plan a linear walkthrough of the code that explains how it all works in detail

Then run `uvx showboat --help` to learn showboat — use showboat to create a walkthrough.md file in the repo and build the walkthrough in there, using showboat note for commentary and showboat exec plus sed or grep or cat or whatever you need to include snippets of code you are talking about

https://github.com/simonw/showboat
also ask LLMs to reorganize it all

>>108754104
every time I dip below 80% context (out of 1M) I wonder if choosing Opus was the objectively winning play)

>>108754161
wdym

>>108753590
you expect effort from a vibecode general?

sorry, best you get is 50 links no one clicks

Is Qwen 35B MoE better than 27B?
I've been clowning with 27B and it's crazy fast AND gets things done right pretty much all the time (as long as I give it debug logs when something goes wrong).

It's actually insane how much faster and cheaper it is to run this locally than blowing through millions of tokens on NanoGPT Kimi or Deepseek.

>use Gemini CLI
>responses get progressively slower until it just stops responding
>switch to antigravity
>totally fine
Google absolutely hates the cli team
I know they have completely different tpu allocations but the state of things now is a complete joke

>>108754408
Nanogpt sucks though, you can't use it for agent use, shit just fails constantly.

>>108754417
What alternative is there? I use the cheap sub for Skyrim AI gaming, but with how many tokens Claude Code tends to burn through, I am not really looking forward to pay per-token.

>>108754437
Codex $20 subscription, and if you want uncensored for some reason, kimi also has a $20 subscription through their site, though you are meant to only use it with coding agents, I've been gooning through opencode ocassionally though, still not banned.
Opencode go is also an option, $10 for $60 worth of tokens monthly, but you will hit rate limits, that applies to all subscriptions anyway.
There is also Ollama's subscription but I have not tested it.
Nanogpt is ok for slow throughput use cases, but for agentic/vibecoding you really needs subscriptions or you will go broke.

If you think youre goig to make money off vibe coding, think again

there is a deluge of slop flooding the internet right now. Unless you already had customers before the flood, its going to be hard for you to get any attention now.

File: anime-ops.gif (1.13 MB, 640x462)

1.13 MB GIF

>>108753548
Oh, shit. I mean, she fucked everything up, but that's so cute. Should I roll the dice and switch my piclaw to Kimi?

>>108754472
I briefly considered making money. Then I realized that the idea of "buying someone else's software" is kinda over, and the real value is making my own custom shit just for me that I want.

fuck yea, my raspberry pi finally came in. now I can start fucking with this fork

>>108754472
Games are the only thing that can still somehow work if you are very lucky to find a niche, because people want fresh games every year. Nobody wants to buy a new SaaS every year.
But the opportunities are shrinking there as well.

It's just funny because I am a Eurofag and in all my friend group and even my work colleagues I am an absolute trailblazer with agentic coding. As always we are years behind Americans and people won't realize what will hit them soon.

My 50yo boomer coworkers writing Java 8 still scoff at it.

File: 20260504_112925.jpg (627 KB, 1224x1632)

627 KB JPG

>>108754506
Ay so did mine
Let's do mischief, gremlin bro

Is there any decent repository for commands and shit?
Like I keep getting annoyed at AI putting its stream of consciousness into the comments instead of actually documenting meaningfully for the future. So I just had the genius idea to make a /fix-comments command that would have instructions like "check the current changeset/PR and review the comments, make sure all comments document what is there and not what we just changed, make sure none of the comments are redundant, make sure all doc comments would be useful to an external user of the interface" etc., but really I'm pretty sure I'm not the first one to realise this is a problem and to have the idea of making AI fix itself as a post-processing step. Surely someone else has written that, and then tested it and refined it to get it to work properly with current models. Is there anywhere I could find existing shit like this?

>>108754472
It depends how much you are vibing, no? If you make something useful over multiple days and weeks, there's still value in that I would think? If you say "build me xyz" and deploy as-is, yes, there's not much value in that anymore unless you're great with marketing, but there's a difference between prompting and hoping, and using these tools to achieve what you want, which not everyone might be able to do, faster.

>>108754408
No, the 35B only has 4B active. The dense 27B would be like a ~100B MoE.

>>108754542
I've never had that issue with any vive-coded shit I've created. What specific model are you using and what type of tasks and instructions. Are you giving it? This is just a hunch but it sounds like you're using the regular-degular front end interface of whatever AI you're using and not one within a harness because those things are specifically made to nudge them into being good at tool calling and being more programming and process focused

>>108754408
If you trust The benchmark scores And user accounts in the 27 version is better at general purpose type stuff but the 35 moe has a slight edge in tool calling reliability, which matters a lot if you're going to a attach it to an agent harness. My machine has 128 GB of unified ram so it's more than powerful enough to run a 27b dense model at long context but I still prefer moe due to the faster prompt processing and performance. I have yet to run into anything a dense model in that perimeter range could do that the moe equivalent couldn't

>>108754562
I've had that issue all the time with claude. I think it happens less often with codex 5.5 but still does. Just earlier today I was reviewing a PR that had a comment along the lines of

/// Foo field that does bar. Optional and defaults to null, that way adding it keeps byte-compatibility with V0.

Which is very clearly an artifact of the prompt (probably the plan or whatever that my coworker finalized included a bullet point like "when adding the field, make sure it defaults to null that way we remain compatible with old data"), and also has zero business being in a doc comment because users of the interface do not need to know what "adding the field" affected: if needed, that shit goes in a changelog or patch notes (or in this case nowhere at all, since it's explicitly keeping compatibility).
If you really needed to call it out in a comment, you'd phrase it as something like "Must default to null to ensure compatibility with V0", in a way that will continue to be relevant in the future and not talk about "adding the field". In this case, the field just defaulting to null doesn't really even deserve being mentioned IMO, but even if one disagrees it still shouldn't be written the way the AI wrote it.

>nudge them into being good at tool calling and being more programming and process focused
Don't get me wrong they're getting really good at the actual code, but they're still having trouble writing good comments.
To be honest it makes sense: given the token-based nature of LLMs, I'm not surprised that when they need to "explain the code" in a comment, they default towards something deeply related to the current context and conversation. But I see this all the time, and I'm very surprised you've never encountered this. Does your LLM write comments much at all, do you put doc comments on public interfaces? Do you use it in existing complex codebases?

Does this fall apart eventually or can these models actually keep track of a large project?

File: 1762239032039528.png (398 KB, 3368x1029)

398 KB PNG

>be colorblind
>hey chatgpt color in this svg for me
>no problem boss
>sends the .svg file back colored
holy fuck, what a godsend this tool is. It was invoking the images tool (I don't know what for (sight?)).

File: math2.jpg (73 KB, 1366x768)

73 KB JPG

>>108754472

i still have hope. I never see any fucking vibecoded programm. Meanwhile i've +100 aps that works ,not that i use all of them but idk now i'm just saying to people "yeah if you need anything just ask me".
People are relatively simple end user and it's pretty easy to make end-user apps for android.
My only problem when it comes to like cloud things is hosting the website (i need to pay) and that i don't have a server (and maybe one day i will bother to learn but not now).

>>108754721
kek

>>108754729
>I never see any fucking vibecoded program
You're using one right now.

>>108754721
>I don't know what for (sight?)
yes it can see.
i know because you can use it to flirt with the thing

>>108754623
I don't see descriptive comments as a problem at all, especially if you're the type to actually read your code or want other to read and understand the code. If I were you, I would just ask. Claude or whatever model you use to just read through the comments and instruct it to trim it down to only barebones absolutely necessary information. Implement character limits if necessary. AI models are kind of shit at following several different instructions at once within a regular prompt so whenever you tell it " hey, don't make the comments to descriptive" it's likely prioritizing actually making sure your code works and not giving a shit about aesthetic choices like how the comments look, so I would worry about that AFTER the code is actually usable and then delegate the task of trimming down the comments to a model as a form of garbage collection.

File: appquiz.jpg (91 KB, 1366x768)

91 KB JPG

>>108754742

yeah i use mine ofc.

i've too much ideas ,no money and no tokens.
Now i want to vibecode a fps boxing godot game or a Nutriments track intake or a body visualiser where you can make the body move with a database of movements and each movements move the body so you can see wich muscles are actually performing.
never knew boxing is huge on legs and i'm someone who can run 2 hours at 4:40 without getting sore the tomorrow.Differents fibers,makes sense.

>>108753548
Kimi can't use git to recover from shit like this?

>>108754104
Opus: Frontend
5.5 xHigh: Everything else

verdict is in:
https://ampcode.com/models/gpt-5.5

>>108754698
The coding/agentic tool use models are specifically trained to handle that type of stuff. Most of the models worth a damn even have contacts Windows exceeding 200k so a large code base should be trivial for most of them. With that being said they aren't perfect, which is why you should also instruct them to implement and use git tracking in whatever project folder you're using so that if it fucks anything up or forgets to do something, it can easily undo mistakes or figure out why it needs to do something a certain way. This effectively makes forgetting about certain functions or accidentally nuking certain code permanently impossible Even at large context lengths and post compaction.

>>108754698
You still need to steer the AI for huge projects so it doesn't fuck up.

If you have a clear idea of a good architecture for your project in your mind, just tell the AI and it will do it. If you don't, it will just glue together disgusting code that will fall apart very quickly

File: floating terminal gnu-chan.jpg (471 KB, 695x697)

471 KB JPG

tfw using local LLM and opencode to build my own robot agent

>>108754937
>my own robot agent
What does it do tho

Codex usage just reset mid implementstion. Thanks Altman

>>108754542
At work we tell our agents to just not write comments
having a pass to strip out comments is totally reasonable after a PR has been made but isn’t quite ready to go yet

File: Capture.png (6 KB, 755x97)

6 KB PNG

>>108754977
>Codex usage just reset
Fake news

>>108754950
Lots of things.

https://zeroclawlabs.ai/

>>108755009
No, it actually just did. My reset was set for tomorrow morning, but back to 100% remaining now.

>>108755009
>>108755062
Same anon that made the post, my codex quota isn't being used suddenly and the VS Code extension doesn't show context size any more either... It might be a bug but I dunno. My stuff wasn't supposed to reset until the 7th

I'm confirming with the codex analytics website and it also says I'm still at 100% even after a few implementations.

File: 1535847336162.gif (175 KB, 478x434)

175 KB GIF

Guess they're only resetting token limits for the gay niggers because mine hasn't been reset!

File: codex.png (52 KB, 889x661)

52 KB PNG

>>108755094
I was at a bit over 80% earlier today, and the reset was set for either tomorrow or the day after, I forget.

>>108755094 (Me)
Usage went down to 97% so the definitely didn't give us unlimited use for a bit lmao

Scam Altman blessed us with a reset

This is not a drill: I repeat:

Scam Altman blessed us with a reset

>>108755147
False

>>108755147
Based, accelerate

>>108754783
>I don't see descriptive comments as a problem at all,
There's a difference between descriptive comments and AI slop comments.
>If I were you, I would just ask
My entire point was asking if anyone knows any place collecting commands from other people because surely someone else has done that before already.

> so I would worry about that AFTER the code is actually usable and then delegate the task of trimming down the comments to a model as a form of garbage collection
Yes, which is why as I've said already my plan was to make a command to clean them up rather than trying to force the model it get it right the first time around

>>108754953
Dont get your hopes high.
I can tell you that its not going to happen
in fact, Claude is going to get worse because Anthropic doesn't have the compute lol

>>108755247
>Commands


Why not just.....tell the model to do it? Are you referring to built in tasks like @build and @explore that opencode has? I'm pretty sure the Pi harness' main advantage is that it lets you build and attath your own tools to it. You may need to create a "skill. md" for this.

if theres a reset that fucking suckks
my weekly reset is in 3 hours lol
i wont even be able to put a dent in it before my 5 hour usage is up

resets fucking suck when its near your weekly reset, either before or after

>>108755300
it's great for people that use it all

>>108755300
just checked
thank g-d there wasn't a reset
but there will probably be one right after my weekly reset

Is we buying the cerebras IPO

You guys that didn't get the reset, what is your plan?

I'm at the base Pro plan and got it

>>108755311
i only want resets when im down to 0% weekly
they should give people a reset token
use it when you want a weekly reset

I got a reset at 30% once and still wasnt happy but it could have been worse.

>>108755327
Im on the uh $20 plan and didnt get it

>>108755327
Do nothing for another hour...

Should I buy two pro plans? But buy the second one in the middle of my 1 week limit, 3.5 days into it

Or should I buy the max plan?

I find that, I have two days at the end of the week Im out of tokens, but that was with Claude too.I just quit my Claude sub. But the tokens Anthropic gave me weren't shit.

File: 1747069786178674.png (1.86 MB, 1157x1359)

1.86 MB PNG

>>108754111
>there goes my moat lol
I have thought this about 20 times since ChatGPT came out.

>Day 59. I am starting to suspect that the time Claude spends celebrating and moseying is actually my request being batched and queued.

For anyone wondering how much usage you can get with the base pro Codex plan: I just ran a 1 hour and 20 minute session, it went through several tasks from a plan document, and in the end it did some interactive/visual QA in the browser.

It consumed about 20% of my 5 hour limit and 3% of my weekly limit

>>108755582
Limits are doubled until the end of the month, so your true usage would be closer to double what you wrote. I agree though that right now at least it's enough for someone working actively, full time on a single or a small number of projects (i.e. not run automated loops all day). It took me a long time to convince myself since I hated OpenAI, but it's good value currently.

What's best offline model comparable to Claude 4.6 for coding?

>>108755645
Kimi 2.6

>>108754472
Nah you just need to be better than others, which you already are as you're here. We're all still early

>>108755676
>he thinks this slop general with zero information is above the curve
lmao

>>108754472
>If you think youre goig to make money off vibe coding, think again
ok I thought again, now what?

>>108755681
You weren't born at that time, but Bitcoin started here. If anything /g/ is way ahead of any social media

>>108754472
The only ones who can make money are actual vibecoders SWE, there are just too many slops for regular people.

>>108753602
I was doing that, but I didn't know there was a name for it. Thanks for sharing

File: 1757751170379051.png (1.15 MB, 843x1280)

1.15 MB PNG

>>108753442

File: 1746534028482269.png (433 KB, 1042x1676)

433 KB PNG

I'm making a bridge app based on this this repo right now: https://github.com/awalol/DS5Dongle

It's coming along quite nice. I'm going to try and fix a few of the bugs in the firmware while I'm at it. The dev who made this is awesome, I spent a good week trying to solve these problems in Windows w/ my own DS5 emulator (it was a HUUUUUGE pain in the ass and it required custom kernel drivers). Using a Pico was a unironic genius move, people have been trying to solve this controller for years.

>>108753157
This shit goes against the /g/ spirit and I bet none of you ever touched this board before the ai boom

>>108755878
>making shit goes against the spirit of /g/

>>108755885
I'll make the icon

>>108755700
yeah well I was posting in bitcoin threads in /g/ in 2010, and mining bitcoins when they were 80 cents each and I'm broke.

>>108755878
I’ve been here for years and now I’m a webshitter with turbo autism at my disposal and I don’t need to befriend men who are trying to get me to try estradiol

File: 1751259456385680.png (1.85 MB, 1231x1327)

1.85 MB PNG

>>108755878

>>108755919
And I thought Bitcoin adoption made no sense in 2010 and still do today. The main difference is that today I think nothing makes sense.

So the goal is to quit your job the release the software to not get raped by the IP clause?
I'm trying to figure out if I can release open sourced software but it looks like companies might be able to take that from you too

> vibe coded project is now several months old
> occasionally notice bugs or edgecase issues and update/fix them
> too autistic to share it with anyone who might actually be able to use it

at least it's useful for me

>>108755963
Who are you replying to? Yes, developing something that's similar with your day job can probably come back and bite you. On the other hand, some famous database salesman essentially did this to the CIA. There are no rules, just consequences, good or bad.

>>108755965
>> too autistic to share it with anyone who might actually be able to use it
95% of vibecoded apps

>>108755985
It makes you able to develop cool things alone, but being alone is not great for selling.

>>108755985
I mean, it's in a public github, so it's at least out there in the wild, but I've never posted it anywhere or mentioned it to anyone, so it exists publicly, but no one knows to look for it.

It keeps getting cloned, presumably by bot scrapers. But zero visitors to the project's main page.

Maybe someday someone real will find it from some random AI suggesting it for the particular edge case issue it's designed for.

>>108755995
Yeah my project isn't really sellable anyway, just handy if you happen to use two particular apps together on windows 11 in a specific way.

Otherwise its basically useless.

>>108755985
like 99.99% of all spreadsheets 2B quite honest

>>108755860
All I know about controllers is “I press a button, and the Switch 2 picks up the controller I’m holding”.
What’s a bridge for? Use a raspberry pi to let you use a PS5 controller with a computer?

>>108754472
Vibecoding just makes you realize that code is the least important part of running a business
You need funding, you need marketing, you need users

>>108756025
>What’s a bridge for? Use a raspberry pi to let you use a PS5 controller with a computer?
The bridge lets your computer talk to the Raspberry Pi (Pico) so you can adjust settings like haptics and speaker volume. The original firmware by the dev was "dumb" in the sense that it was plug-and-play with no adjustable settings. Certain DS5 functionality is exclusive to USB connections, so what the firmware does is make the computer think the DS5 is connected via USB, then the Pico forwards that communication to the controller. The reason the Pico is necessary at all is that exposing the DS5's speaker as a proper audio output on Windows requires a kernel-mode audio driver, and installing kernel drivers requires enabling test signing mode, which most users can't or won't do (it's also extremely insecure). The Pico handles the audio endpoint itself, so Windows never needs that driver. Yea, it's technically possible to do this all in windows, but it's going to require money to get the drivers signed (lol fuck that). Honestly, just thank god for this dude.

File: file.png (2 KB, 268x35)

2 KB PNG

good luck little agent

>>108756059
>but it's going to require money to get the drivers signed
Especially kernel drivers, basically double or triples the cost of regular drivers since you need an extended validation certificate.

AI still gets shit done but whenever I look at the code it really is pretty bad.

>>108756138
>Make it work
>Make it polished
>???
>Profit

>>108756138
See yourself as a sculptor.

>>108756043
This was always true for simple websites and mobile apps, but about 10 years ago I made a bit of money with a 3rd party plugin desktop app. I had to reverse the program I was interacting with and since it wasn't that easy, noone else bothered. Now Claude could reverse it in a few minutes, so even autistic niches are no longer viable.

>>108756138
you can tell it to improve the code, you know

>>108756156
I know but that slows it down a lot. And it's not just telling it to make it better, I have to point out specific mistakes.

>►Editors / terminal agents / coding agents
what the hell is even decent here? I don't want to pay for cursor, is aider any good? not gonna pay for jetbrains of install vscode mumbo jump on my linux machine

>>108756202
nothing, this general has literally zero good information, there are way better communities even on reddit

>>108756156
I usually have an agent audit it to prepare a report. I start by blowing smoke up its ass about what an amazing performance-aware no-nonsense old-school programmer it is, then gratuitously thank it for condescending to review our shitty codebase. I describe the strengths and weaknesses of my "team", pointing out that they are great at commenting and writing code that looks very professional (otherwise the LLM will be too impressed by this), that they are very knowledgeable but too easily confused on complex tasks and prone to laziness. I ask it to focus particularly on ways in which coding practices have emerged that are drifting from the broad principles expressed in the architecture documentation.

I have it draft up a full report on the problems it encounters in the code, and then I have it write a rich and fully-explained plan to fix one of them. That becomes the basis of a new session to fix the problem.

Does this actually work? I don't know. It makes me feel better.

>>108756225
>agent
(I meant a new session. I don't really use agents.)

>>108756225
>I start by blowing smoke up its ass about what an amazing performance-aware no-nonsense old-school programmer it is, then gratuitously thank it for condescending to review our shitty codebase.
Is the donut improved by adding all that glaze?

File: 1570550251977.jpg (415 KB, 1000x552)

415 KB JPG

>tfw refining my design, making it more powerful, more flexible

>>108756407
>more flexible
>breaks arm
look it's more flexible!

>>108756393
NTA, it can be, but there's no magic bullet. You can get better results in general just by giving an agent some personality, as it can steer and anchor a lot of their decision making, making it more decisive and waste less time thinking. Again though, there's no magic bullet, and priming any given agent with "you're the best iphone app programmer ever and you love coffee and you hate mondays and you're pretty chill in general and you see yourself as a sexy 3000 year old vampire girl" won't necessarily be a good thing.

>>108756451
That’s what I was wondering. I’ve seen “You are an expert code reviewer” or “you are a security expert”. Not something as verbose as what’s in >>108756225.

>>108756217
Amazing contribution, thank you.

>>108756451
>you see yourself as a sexy 3000 year old vampire girl
Now I want to design an agent that's a bratty vampire girl that leaves smug comments on other agents' code...

I miss my 3 agent system
with Claude Code, Codex, and Hermes on Deepseek API. Paying for Claude just isn't worth it anymore.

>>108756622
I've been using Gemini cause my work pays for it and it's okay if you know how to prompt it.

I find it can be better than claude at certain things.

Codex produces scope creep.

>vibe coded my python assignment and the tutor accused me of academic misconduct

and rightly so - I have no idea how to code and had no business handing in perfect code but I did it anyway. there are no rules, only consequences

>>108756667
@codex, rewrite this code so it looks like it was written by an idiot

>>108754821
I haven't set up fit for that project yet, I only started it a couple hour ago and it involved moving stuff to and from a remote machine through ssh so it might not have matter anyway.
In any case what he overwrote were some files that would take less than half an hour to regenerate so no big deal. If it had actually destroyed anything important I would be actually pissed.

vibe coding got me to stop gaming. I'll just vibe/debug for hours on end instead. I've been like this for months.

>>108756686
You're doing it wrong. You're supposed to automate the workflow so you can game while the AI works for hours on end.

>>108756673
>it's the same code

>https://github.com/oven-sh/bun/compare/claude/phase-a-port
>hey claude, rewrite 750K LOC into rust
>11 hours later
holy based

File: 1748925316797094.png (344 KB, 1040x1669)

344 KB PNG

Alright I'm done with this for today. I implemented pretty much every feature I wanted and resolved a lot of bugs in the firmware (gyro was broken). Gonna spend tomorrow refining the UI so I can publish this asap. the pico was worth $17.

So what actually is the cheapest way to use china models rn?
I know kimi's got subs but they aren't that much cheaper than codex - is usage higher?
Opencode is tempting, but models rotating in and out all the time seems like a pita
Memeseek don't do plans right? Just heavily discounted API rn?

Should I buy 2 Codex $20 subs, or $100 one?

I kind of think I could make do with 2 $20 subs

>>108756673
I should have done that first but now I have to explain I used ai to do it and either get kicked or redo it but ill prob be kicked from the course idk I don’t really care anymore does studying an it cert even matter in 2026 anymore?

>>108756876
Yeah Kimi is still decent and yes it has much higher usage than Codex for the same price.
GLM stopped being useful months ago.
Haven't tried Opencode Go.

>>108756138
thats how i feel when looking at other peoples code

It's nice how OpenAI gives you complemntary free chat usage with your membership, and what model is the chat? I configured it to 5.5, but it seems kind of like a gimped version. It doesnt feel like the chat bot is 5.5

its nice being able to send my questions to the chat without having to bloat my Codex projects with questions, or waste tokens.

>>108757051
Can I ask you a question? When goyim like you press the Subscribe button and lock yourselves into a contract, while Aryan APIchads like myself only pay for what we use, how do you rectify when your monthly subscription resets when you still have unused tokens? Or like, what if you don't use ANY tokens that month, you still get charged the monthly subscription fee? For using nothing? I'm honestly curious how your economy works.

>>108757075
Nta, but I have a job so there is no months I don't use it.

>>108757084
Yeah man I have a job too which is I guess why I don't mind paying for API keys? Because apparently it's more expensive than subscribing to Dario's platform or whatever? I don't get it.

>>108757075
some of us use the models for more than 2 line scripts so the subsidised pricing works well for us, lilbru.
openai sub is month-to-month anyway.
have fun making your little scripts tho.

>>108757094
Mmm, yeah, me and my PiClaw are both about 2 lines of script total. Pi is a platform for agentic AI, btw. I can link you the repo if you want to check it out.

>>108757104
I really need to examine this gayass captcha someday and get Kate to defeat it so she can post on 4chan. Add that to the TODO.md.

>>108757104
thanks, m8, but i've already got the codex rotation extension for my accounts.
and you were fumbling with openclaw before i told you to just use pi.

>>108757110
Yall's nigga bitchass never told me to use Pi, it came to me in a dream.

>>108756876
1 dollar subscription in command ai for deepsek v4 pro

File: cc439fef72c0777437897d9ea(...).jpg (37 KB, 612x408)

37 KB JPG

>>108757110
And I wasn't "fumbling" with OpenClaw, alright? Those retarded vibecoding RETARDS were pushing broken code every 3 days. Every THREE FUCKING DAYS I had to nuke and reinstall that shit. Is it more stable today? Is OpenClaw more useful today? Maybe.

But I reject that answer; instead, I chose something different.
I chose the impossible. I chose...PiClaw.
An agent where the user would not fear the vibecoding zoomer.
Where the coder would not be bound by petty Indians.
Where the great would not be constrained by the small!

No, I chose a different path. I chose...Rapture.

>>108757075
You are a retarded noob.

>>108757134
calm down, buddy.
also, you're probably going to have to fix a bunch of shit after mario's done with the refactor lel
the pi server shit sounds cool tho

File: Screenshot_20260505-100620.png (485 KB, 1008x2244)

485 KB PNG

>>108757104
>link the repo
what for?

>>108757141
Yes.
>>108757144
If I "calm down" for even a moment, like stop giving a shit? My body will stop working and I'll die. So, we must carry on.
Which pi server shit, the Galaxy S22 phone acting as a server? It's at app.eroticjesusfeet.com if you want to see it, there's not much going on there. Neat that it's running off a cell phone though. A tiny little server.

>>108757162
And please, you don't need to ask about the domain name, I'm just tired of the questions about it, it's fine.

File: 047561683286.png (183 KB, 1041x1021)

183 KB PNG

>>108757162
https://github.com/badlogic/pi-mono/tree/bigrefactor

File: ratatouille-chef-skinner.gif (2.39 MB, 374x211)

2.39 MB GIF

>>108757183

File: kimiko-93739.png (57 KB, 350x490)

57 KB PNG

has anyone tried ollama + kimik for coding? I want to try 2.6 since its supposed to be good, but I also trust ollama more than gpt / claude. or am I crazy?

Benchmark Kimi K2.6 Claude Opus 4.7 Winner
SWE-bench Verified 80.2% 87.6% Opus 4.7 (+7.4)
SWE-bench Pro 58.6% 64.3% Opus 4.7 (+5.7)

>>108757321
Haven't tried the Ollama subs but I read on Reddit they are very limited. Might be better to go for the $5 sub on OpenCode Go or the $19 or $39 dollars subs on Kimi Code.

god, antigravity feels like absolute shit to use.
i hope they just what everyone else is doing and build a dedicated thing instead of their vscode frankenstein shit

Will using gpt 5.4 low use less usage than gpt 5.5 low?

>>108757537
Compared to Codex, it is shit

File: 1750579694975170.png (109 KB, 2074x307)

109 KB PNG

>neither pi nor opencode have retry buttons
It's over.

>>108757537
It is shit, not like I have the choice with gemini sub. Only good side is the locked down sandbox that doesn't let the agent run anything on windows except copypasting code.

>>108757972
>clanker, write me a retry extension and install it
fixed

>>108755364
>But buy the second one in the middle of my 1 week limit, 3.5 days into it
if you're talking about chatgpt, altman will probably reset the limits randomly and sync them up

File: file.png (176 KB, 1028x1160)

176 KB PNG

kek

don't fall for the d&c, that's what they want

>>108758038
LLMs fear the schizo ascetic

>>108757075
I tried API pricing once. After like half an hour of running through agents I checked the billing and this had cost $3. I'd probably be paying $20/day if I was on API only, conservatively.
So no, when I pay $100/month and then only use 50% usage every time, I don't mind since it's still several times cheaper than the API cost for the same usage.

Also my company pays for it anyway lmao

>>108758038
wtf is this real

New here, so I might be very off course. What 'foundation' do people use for cline? For example, I add general project description with features to .clinerules where I also instruct it to maintain a text file with current project structure with explanations of functionality implemented in each file to prevent it from re-exploring the whole thing each time, but I feel like there could be much more techniques out there.

>>108758038
https://desuarchive.org/g/search/text/nordic%20runes/

>>108739733
>>108746600
Desktop commander anon, I got my Desktop Commander running and able to do stuff, but is there any way to get it working like Cursor (except obviously the chat is in web instead of the IDE?)

I want it to edit in my text editor (or if there's a special text editor for this, that's fine too) but to do it like gits where I can see the differences and push or decline. I don't know if anything like this exists but it'd be swell if it did. I'm sick of rate limits. Or if there's another way to use GPT as an app that does this that isn't Desktop commander, that's fine too.

I just realized that I got refunded the full value of the subscription after getting banned from Claude, even though that happened almost at the end of the month. I got 1 month for free.

>>108758415
How do you get banned?

>>108758580
I assume he's the guy who was swearing at claude for being dumb and making mistakes, calling it fucking retarded, etc.

File: 1774027803062505.jpg (145 KB, 1080x646)

145 KB JPG

It's OVER if this becomes real. American AI will fall to European stagnation hell levels and us commoner plebs will never get SOTA models over api/subscription again

>>108758587
Isn't this the same administration that REMOVED oversight from deploying AI models for military use?

Previous rules required like a 90-day audit but they changed the rules so new models HAVE to be in-use within 15 days of a public model release.

>>108758580
I got banned when I put an HTTP proxy in front of Claude Desktop. I wanted to try to connect it to a local model.

>>108757075
I used more than $100 this week in tokens, I'll take the sub thanks. Was real easy to use up $20 in API tokens prior to getting a sub

File: 1663146628497272.gif (2.79 MB, 476x480)

2.79 MB GIF

>>108758618

>>108758636
It was instant too. I pressed enter and I got a notification e-mail.

I got banned for doing a 360 noscope. It happened immediately after I did it! Ignore me calling people nigger the previous day, that is irrelevant.

File: Code_8q5sdyXvg2.jpg (9 KB, 238x68)

9 KB JPG

Why is the emdash so massive in this font? baka

File: bad.png (554 KB, 761x842)

554 KB PNG

Even gpt5 can't make me a fp8 convert script. I'm retarded.

File: file.png (37 KB, 800x390)

37 KB PNG

I think I figured out a way to use Desktop Commander + ChatGPT (Web) + a VS Code extension to get unlimited requests with the web version. Unless something like this already exists.

I know how to write extensions in Chrome, so that should carry over to VS Code plugins I'd assume since it's Chromium based (Electron). The question is, would anyone use it here if I actually attempted and pulled it off?

>>108759039
In principle: yes, that can work.

>>108759096
I've just been asking it what it can and can't do with Desktop commander until I could find a workaround and testing the waters. It gave me a lot of advice. I can also do an MCP Wrapper tool to enforce git diff format. Unfortunately I don't know how to do that lmao. But then again this is vibecode general so I don't have to know how to do shit.

File: 1000021350.png (402 KB, 1080x1466)

402 KB PNG

Hmm big claims
Only one non-context benchmark though, swe-verified over 80, which basically means saturated
No other numbers I'd very suspicious

>>108759186
Absolutly worthless tweet without the weights out

>>108759186
I'll believe it when Bolt Graphics releases the 5 times faster than 5090 for a fraction of the price, Risc V graphics card they promised.

Until then it's empty promise slop.

>>108759207
Worse. No technical report. Have to apply for API access.
Probably a meme.

>>108759209
Lol that shit is a scam

>>108759243
>Lol that shit is a scam
Ye that was the joke. I'll believe the meme startup's super LLM when that other meme startup's super GPU comes out

Rumours are Seed 3.0 (from ByteDance) is GPT 5.5 xhigh level model, surpassing Opus 4.6/4.7. Its highest tier is $700/yr.

>>108759300
*shoves banana up your ass*
Enough monkeying around.

I have Seedance-4.0 btw

>>108759331
proof?

Gemini flash 3.2 soon
Gpt 5.5 party today as well so maybe something new or a reset

Bout to smoke some weed and vibe code

File: Screenshot from 2026-05-0(...).png (75 KB, 807x887)

75 KB PNG

>>108759039
>>108759131
I don't know about vscode but I am still working on my chatgpt user scripts for mcp approval autoclick (this one is done), autoscroll (this one is done) and the hardest one which is cat history cache+chat history trimming which I was still developing just now. After that I have to fix/test one I had made for autoreload on errors, and then as a nice to have an autoreply one.
I was checking and ChatGPT is still cheaper than Opencode Go. And I'm not even sure Opencode Go is any cheaper (usage adjusted) than Kimi Code even though some anon was saying it's "basically unlimited".

Vibe coded operating systems: 0
Vibe coded web browsers: 0
Vibe coded compilers: 0
Vibe coded virtual machines: 0
Vibe coded file systems: 0
Vibe coded gpu drivers: 0

>>108759500
>W11
>yet
>that gcc failbake
>idk
>why
>nvidia

>>108759500
>Vibe coded operating systems: 0
there is one but it's shit
>Vibe coded compilers: 0
there's a weird broken one
>Vibe coded virtual machines: 0
been done

>>108758359
I'm not that guy but I was the one who had the idea originally (or so I'd like to think).
Yes, you could do it both from the web or from the terminal running the MCP server.
But I'm not sure I'd like to invest that much time into this since I don't like vibecoding from a web interface (prefer console) and I don't like GPT's personality. We'll see.

>>108759517
Is it much different than having it as a sidebar really? As for personality; I don't give a fuck about that if I'm getting unlimited GPT5.5 for 20 dollars.

>>108759500
I vibecoded a half working Nvidia SASS emulator, does that count?

File: file.png (49 KB, 786x369)

49 KB PNG

it has the power to make windows 98 drivers

>>108759528
I don't know, I've never used the cursor type sidebar IDEs.
Also it's not really unlimited, they will cap you for the day if you hammer it too much (multiple windows constantly using it for 5+ hours).

>>108759542
Have you ever rimmed a Latina's asshole? You're one of those guys who knock things before trying it.

WHATS A GOOD STRUCTURED RESOURCE ABOUT SETTING UP A VIBE CODING PROJECT FROM A TO Z, all those indian videos are a complete mess and this sweeps over to my personal projects, I WANT EVERYTHING NICE AND STRUCTURED BEAUTIFULLY, IM WILLING TO EVEN PAY FOR A UDEMY COURSE IF ITS GOOD, NO SHITTY MARKDOWNS THAR ARE ALL OVER THE PLACE

>>108759561
just tell it not to do that

>>108759561
Nice and neat doesn't exist in vibecoding, it's gonna be messy and dirty but you'll love it once you try it.

>>108759575
I mean the whole process of setting it up, like do you use an IDE with a plugin or cursor or Claude Code? Do you connect it with plugins connectors to your github repository and your jira board (yes I always use one for projects and task management)

>>108759608
persoannly i just use the default Codex UI

>>108759608
I just use the terminal inside CLion.
My 2 main tips are to see what you actually want/need and to ask the agents if you want to change anything.
Just get a Claude or Codex sub (Codex is better imo) and open it in the terminal or VS Code plugin. Just get it running somehow, even if it's suboptimal because now you can already ask the agent to set up the rest.
You can use MCPs but agents can also use any terminal command, so they can already use git, github, aws cli etc. If you don't know how to install those (it's very easy) ask the agent.

File: 1000021356.png (143 KB, 1080x676)

143 KB PNG

Lmao

>>108759692
how the fuck did i spell personally like that

>>108759534
This is so very bloody epic my bastard!

File: file.png (22 KB, 760x249)

22 KB PNG

>Hey Claude I'm making a clone of this old game and I'm trying to get a better understanding of how they handled such large amounts of save data on such a primitive system.
> C. I see you've got a copy of the original game, let me take a look at that.
ambitious

File: file.png (16 KB, 621x152)

16 KB PNG

>>108759998
sure thing buddy Claude I trust you

File: 1000021357.png (46 KB, 459x481)

46 KB PNG

New 0% benchmark just dropped

>>108759998
>>108760013
ai does good with ghidra

File: 1000021358.png (167 KB, 1902x875)

167 KB PNG

>>108760063
Codexsisters, our response?

>>108760063
so this is just testing one shottability? this is effectively meaningless

>>108760075
>5.4

if you're trying to one shot a large project the AI should just call you a retard and take your tokens

>>108760078
>Given only a compiled binary and its documentation, AI agents must architect and implement a complete codebase that reproduces the original program's behavior.
Nobody knows AI like the experts at "facebookresearch".

>>108760097
It's very funny that one of the binaries is ffmpeg
If anything ever passes that, we'll know we've reached asi

my dearest friends it seems that google wants to get precious data upon their hands.
I got a free trial for gemini pro, thus got gemini cli and it wont work bc of age verification. can I bypass it?

File: 20260505_134528.jpg (1.65 MB, 4080x1901)

1.65 MB JPG

And when the patient woke up, his skeleton was missing, and the doctor was never heard from again! Hahaha! Anyways, that's how I lost my medical license. Now, let's build a robot. My love shall have her physical body. Nurse — the screwdriver please.

>>108760147
>—
>—
>—

>>108760147
How can redditors not kys when reading that kind of slop all day? Kill yourself faggot

File: Ezgif-5-30f701527f.jpg (183 KB, 1136x852)

183 KB JPG

>>108760166
Yeah I've taken to using them instead of -. - is too human. I'm...something else now. The next phase of human evolution.

File: file.png (556 KB, 648x971)

556 KB PNG

>>108760147
That shitty ultrasonic sensor isn't going to give her much to work with. Add some 24ghz radar at a minimum so she can sense you and a thermopile camera so she can feel the warmth of your embrace.

>>108760177
It's a TF2 quote dipshit. Please go touch grass, and by that I mean engage with human culture outside of your autistic exoplanet hyperfixation. Fuckin shut-in retard.

>>108760189
>TF2
retard

>>108760183
Those are both fantastic suggestion that I'm putting in the .md. Also in there is a line that just says "LIDAR?" These are questions for later. Right now I'm gonna open up all these little boxes of parts and lay some shit out on the table like a responsible person.

>>108753555
i sell my employers the idea i might be productive today

i will be. gonna grind from gold to plat in r6


>>108760135
imagine telling the gov who youare

File: pepe-the-frog-afraid.gif (8 KB, 111x112)

8 KB GIF

>>108753157
I am going to use claude code to make a website, but I read some reports of people losing their home folder because of it. How can I prevent this shit from happening?

>>108760234
just don't give it permission to do anything outside of the folder where all the shit for your site is

>>108760201
For future upgrades consider giving her the ability to feel her muscles/joints. The ghetto way is sticking current sensors in there and using up several analog inputs, you can measure the actual current draw at each servo. For something small/simple that's easy, but for more detail and more information you can switch to better servos like serial bus servos that provide good telemetry like load, temperature, and even positioning error feedback. This can also help you protect her from overexertion, don't want to fry a servo pushing her too hard, good feedback can make aware of and work within her limitations.

>>108760234
rename your home folder to DO NOT DELETE
simple as

File: revenues lol.png (29 KB, 864x247)

29 KB PNG

>>108755940
>cope edit
shoo shoo poo poo

I thought planning mode would help me conserve my session a little longer but Claude still devours it just as greedily...

The only paid account I have is Claude, I sincerely love what I can achieve with it but I have been hitting the weekly limit in a single day. Should I get chatgpt plus or is the Google AI worth something?

>>108760414
The Google AI freetier is worth trying. But get GPT Plus anyway. I have the Claude 20bux plan and GPT Plus and Gemini freetier and bounce between all of them.

>>108760075
>Gemini
this is how I know I can instantly dismiss the results

File: 20260505_151208.jpg (356 KB, 1224x1632)

356 KB JPG

Kate called these "nutlets" when I showed them to her. Look how tiny. This is gonna be a real big bitch to build.

>>108760252
Sir do you have some experience with robotics? Because I don't. You seem to be full of useful tipz n trickz. I'd like to subscribe to your newsletter.

Am I understanding this correctly: that 100,000 retards are 'programming' their own local assistants using Claude, which are totally terrible,
instead of just one programmer using 100,000 Claude accounts to create a good one for everyone?
Why are we like this?

>>108760454
>This is gonna be a real big bitch to build.
What is "this" that you're building? Also maybe Kate would be more impressed with my nuts.

>>108760430
I have been disappointed by the free chatgpt, the answers are just absurd. But I've been reading a lot of positive opinions about codex 5.5, I created some really good landing websites with the use of free gemini and claude code, I wonder how codex+claude code combine

File: 1674819460188.jpg (908 KB, 1200x1695)

908 KB JPG

>>108760462
Play ToHeart and watch Chobits, that's how I did it.

File: 20260505_150234.jpg (3.11 MB, 4080x1948)

3.11 MB JPG

>>108760466
A physical body for my agent, so she can greet me at the front door when I get home from work.
>>108760147

>>108760537
Holy fuck I burst out laughing. What a fuckin thing to say.

>>108760468
>I wonder how codex+claude code combine
It's great. It's like having two senior engineers with mostly-but-not-entirely overlapping skillsets. The places they disagree or catch stuff the other misses are the Fun Parts of programming.

File: 1594021850002.gif (3.8 MB, 498x280)

3.8 MB GIF

>>108760464

File: SJW entryism and infiltration.jpg (50 KB, 680x403)

50 KB JPG

>>108757906
unlikely, OpenAI said 5.5 uses fewer tokens to get the same amount of work done
>>108758587
The Biden-administration goons already basically had oversight into making all the Western models woke
this is just a counterweight
pic related
>>108758739
weird font
>>108760234
Put everything you need for that website into a directory of its own, which you’re probably doing anyway
>>108760575
this

>>108760586
anyone who doesn't support gun ownership (to shoot the billionaires dead) is not a true social justice warrior

>>108760560
>A physical body for my agent, so she can greet me at the front door when I get home from work.
>t. J.F. Sebastian
https://youtu.be/CLPpXNhL9z8?t=51

File: file.png (48 KB, 300x168)

48 KB PNG

>>108760560
Does she have a charging station? Needs to be able to charge itself like a Roomba. It's got a camera on it, surely it'd be trivial to get it to recognize an aruco marker mounted to a charging station, 20W Qi wireless power modules are cheap and would make it very easy to "dock", no actual connections required just lay down in roughly the right area. 20W ain't much but I imagine she'll spend a good amount of time chilling in the charger waiting anxiously for you to come home.

Fucking hell, new Co-Pilot should be renamed to cuck-pilot.
So it's week 1 of may. I didn't even use it yesterday.
I consumed 7% today, which apparently makes it 60% of my weekly allowance. Some quick ass math means that you can't practically get your month's worth of tokens. Since if it's 11% a week and there are 4 weeks a month, you're just fucked. Thank god I didn't pay for it and that my boss got a year's subscription 9 months ago.

>>108757906
>>108760586
we have some numbers now to know gpt 5.5 is 30-70% more expensive on med/high/xhigh. 5.5 low is a lot more capable than 5.4 low was though - at least it feels like that
https://ampcode.com/models/gpt-5.5
https://openrouter.ai/announcements/gpt55-cost-analysis

not sure how this translates to codex usage but i would think similar?

File: rms parrots lisp vs cyber(...).jpg (602 KB, 1024x1347)

602 KB JPG

>>108760629
We GNU plus warned you about Microsoft, anon.

>>108760464
It's a toy. The point is to play with it yourself.

>>108760627
Yeah, autonomous docking and charging is absolutely already on the list. There's no ootb solution for this kit, so it's gonna require some hax but that's OK.

>>108760464
anthropic and openai's in-house products are not far off from what you're describing:
https://youtu.be/SlGRN8jh2RI?t=452

^ is boris who created cc and at during the day there are a few hundred agents working just for him. a few thousand at nights.
you can probably multiply that number up quite a bit
openai's internal leaderboards have people spending hundreds of billions of tokens in a week.
the stuff happening in these companies is entirely alien to what the rest of us experience.

and despite that a customised pi works better for a lot of people and and up until recently there was one guy working on it

File: file.png (321 KB, 800x565)

321 KB PNG

>>108760661
Definitely consider wireless charging. It makes things so much easier because it requires so little precision, and doesn't force you to physically expose anything on the robot itself. Most receiver modules max out at a measly 5W peak though, realistically pushing more like 2-2.5W, pitiful. You can get ones that'll push an honest 15W but they're harder to find and can be pricey.

>>108760697
buy an ad

>>108760778
i do it for free

File: 20260505_162019.jpg (469 KB, 1224x1632)

469 KB JPG

Ah shit. Well, it arrived, the last piece I needed. The most cursed microSD card ever produced. I can feel the power of the Master Emerald radiating from it. I need... I need a boyfriend-free girl who is a virgin! JENNAAAAAAAAAY!

>>108760834
mini sd has sonic with a huge cock and the adapter with an image of rouge the bat . now that would be goonkino

>>108760834
>JENNAAAAAAAAAY!
larp larp sahur, fake fan detected

>>108760864
stfu Liquid Chris

My code involves mirroring tensors like x[:,::-1,::-1,:,::-1] etc. The AI completely shat itself.

File: 1762158420961882.png (166 KB, 861x877)

166 KB PNG

>>108755965
What does autism have to do with whether or not you want to share something?

>>108760464
Everyone has different ideas about what a good one for everyone would be, hence the different 'claws. I happen to think not getting cornholed by random prompt injections or passersby shouting "ALEXA BUY 50 TONS OF CREAMED CORN" while I have voicemode on is important, so I'm contributing to ZeroClaw.

Anyone try 5.5 instant in ChatGPT yet? I wish it could auto guess what model to use based off what you're typing, don't need xhigh to ask it basic questions