vibe-coders

[balmy depot]

Not entirely sure what I am doing is vibe coding anymore. 😜 But vibe coding adjacent for sure.

[gentle aspen]

holly essay. at least use some markdown, dawg💀

tbh the most annoying abou tmaking an app is finding the icons lmao

[gusty meteor]

windows bug💀

[gentle aspen]

what?

[gentle aspen]

wdym

[gusty meteor]

look picture agin some areas are transparent somes not lmao

[gentle aspen]

ohhh, i forgot how windows stuff even looks like. maybe bcs they deprecated the mica for some frameworks

[gusty meteor]

soo come linux xd

[gentle aspen]

I mean, I love prefer linux ONLY to ML stuff, but I think mac and windows is supperior when it comes to convinience

[gusty meteor]

💀

[rain lava]

Define convinience.

[gusty meteor]

💀

[gentle aspen]

the access to archive a dataset which has many options in a few clicks

I mean no offense, but it's just mac and windows are just easy to work with

[rain lava]

I know right 😭 i aint reading it if its formatted like that

[gentle aspen]

linux is good, but all the distros I try are so buggy.

[gentle aspen]

frfr

[rain lava]

Like delete?

[rain lava]

I swear ur trying the wrong distros

[gentle aspen]

no, i mean to compess something without having to touch the terminal yk

[gentle aspen]

maybe, chatgpt says "These matche your vibe and the fact you asking this makes you different"🥀🥀

[rain lava]

I mean if u use smth easy like mint u dont rlly need terminal but... if u learn terminal everythjng is sm faster

[gusty meteor]

i know that convo end LMAO

[rain lava]

😭 what vibe?

[gentle aspen]

yk... the usual chatgpt tone. openAI overcorrected the tone of chatgpt, now it is ruined

[rain lava]

Ive always found chatgpt annoyingly weirdly trying to act human and use slang

[gusty meteor]

whats distro you tried ever

[gentle aspen]

garuda and the 420 ionized.

I knwo thy are arch based, but I mean the Ui bugs and stuff liek that

[rain lava]

Garuda with what de?

Also nvidia carfs just love driver bugging on linux

[gusty meteor]

in which year?

[gentle aspen]

the whole 2025 and a bit of jan this year

[gentle aspen]

maybe

I had a 3050

[rain lava]

De?

[gentle aspen]

what is DE?

[rain lava]

Desktop Environment.

[rain lava]

If only nvidia drivers weren't proprietary...

[rain lava]

E.g, kde plasma, gnome

[gentle aspen]

KDE

[gusty meteor]

give another chance to linux xd

[rain lava]

You may get a better experience using x11 on nvidia, and using a different de.. maybe 1 more stable

[gentle aspen]

I will try. plus, my laptop broke a few days ago, so i have to fix it. bcs i am doign some shi with my windows machine and i can't give it a break

@rain lava did you get ONCard runing on linux?

[rain lava]

I used mint then it kinda sucked then i went to fedora and cuz i was dumb and couldnt use teeminal i had a hard time then i went back to win and waited for an amd card to go to cachyos

[rain lava]

I havent tested yet --- ive been away these few days, today i wen fishing with family n friends and on friday i did an online py course aith my friend

[gusty meteor]

you can still come to cachy os with an nvida card

[rain lava]

Yes but drivers suck

[gusty meteor]

no

cachy os will install drivers for you automatillcy

[rain lava]

In my experience yes

[rain lava]

Ik but theyre still not good drivers compared to amd

Amd was built on mesa and stuff which is why its great on linux

[gusty meteor]

...

interesting mind lmao

nothing to say

[rain lava]

On linux**

[gentle aspen]

when it comes to stability here are how my list go:

Mac: they make their own hardware and software.
windows: stable, since one distribution, but the devs arelazy and it's just a welcome page for all the hardwar eout there.
Linux: more distros than an avg humans numbers of lines of hair, and also support literally any piece of software. the nex thting i am expecting is someone to make linux in minecraft

[rain lava]

VM or USB and give cachyos wth a diff de a try, ask chatgst for a de like kde but more stable (prolly js do gnome) (idk abt cosmic)

[rain lava]

Imo linux is more stable than windows if u pick the right stuff and know what ur doin'

[gentle aspen]

tbh, i feel like moving to Qwen or claude or gemini from chatgpt, it has being...lets just say... "Broken" since a few months ago.

[rain lava]

Gpt has been broken? Or thkse 1s

[gentle aspen]

true, but windows is like the "I did not put any thought into this" OS beside mac. I mean linux is good for experiemnting stuff, but windows is just "stable" and ""okay". I feel like, the onmly "good" OS in this owrld, is MacOS

[frail raven]

i disagree ahah, I don't think it comes down to convenience bur rather learning curves & intuitiveness

[gusty meteor]

no you have to stay gpt forever you love sam altman 💀

[rain lava]

The only "hard" thing (for new ppl) would be managing tar.gz

[gentle aspen]

nahhh, mac just saves time. windows just ruins time, linux just CONSUMES time.
maybe you are true. but even after learnig you dont have "convinience". dev life is like a battle field, but still personally, i preffer Mac, but experiemnting --> linux

[rain lava]

Otherwise app0kg or whatever its called exists/ur distros pkg manager

[gusty meteor]

thats a typical linux convo, we need supporters, so i call the @strong sequoia XD

[rain lava]

Linux faster than win thooo

[frail raven]

For instance, linux isn't really intuitive at first (it's much better now but still) so windows is more convenient yes

But if you learn the linux terminal commands and your way around it, I find linux to be way more convenient than windows as you can just use the terminal to quickly do anything and don't have to wait for the UI of an app to load or anything

[rain lava]

CLI will always be faster than GUI and that's a power of linux

[frail raven]

i mean once you know linux, it's definitely the opposite though
You can do everything using the keyboard and navigation / doing stuff is way quicker in my opinion

But it takes time to get to there, that I definitely agree

[gusty meteor]

^ agreed

[gentle aspen]

just because I like Sam's vision, doesn't mean i am "bound" to him

[gusty meteor]

LMAOO

this is even worse lmao

[gentle aspen]

agree, but soem doesn't prefer that. I usually think in the users perspective

[rain lava]

sudo pacman -S discord vs go to https://discord.com - download an installer - install

[rain lava]

Whats sams vision.

[gentle aspen]

AGI, and make Ai open. but he bever does that 🤣
plus he is gay

[frail raven]

that's the thing, it's mostly because they're not used to it in my opinion, so it's more of an intuitiveness thing / ease of use, than a convenience one

[gentle aspen]

Idc care about a person gender, but if they dont do what they promised, you feel betrayed

[gusty meteor]

LMAO

[gentle aspen]

yeah! agree

[rain lava]

Only reason ppl hate linux is because its outside their comfort zone

[gentle aspen]

yeah! fr

[frail raven]

to be fair, Linux has a lot of drawbacks too in my opinion ahah

[gentle aspen]

I treat linux as the "experiment" stuff. I use windows and mac as my main driver

[rain lava]

Like? 😭

[gentle aspen]

it's not bad, but it is not reliable for a day-to-day user.

[gusty meteor]

like?

[gentle aspen]

it can randomly crash (but whe it does, itis worse than indows)

[rain lava]

"not reliable" What

[rain lava]

No it doesnt

[gusty meteor]

like how lmao

[rain lava]

User error if it crashes

[frail raven]

If you're a designer for instance, it's rather complicated to use Linux. Most professional software (Adobe Suite, C4D, ...) isn't made with Linux in mind

[gentle aspen]

not many of the apps they want are supported. plus, they have to use VMs if they want to. using Wine is not reliable most of the time. so linux is good for mostly: server hosting, ML, DL, generally hosting stuff and using a lot of compute temporary and duct taping things to fix later.

[rain lava]

Thats like 1 drawback, compatibility with closed source apps

[gentle aspen]

ig.
Windows --> Microslop's problme

mac --> apples probelm

linux --> user's problem

🤣🤣

[rain lava]

It has like 90% of sgeam games how is it not many apps

And proton is very reliable

[frail raven]

Honestly no, Linux is fine nowadays especially if you use a distro made for beginners. You will likely never encounter stuff like that unless you try advanced stuff

[frail raven]

Yes but it's a big one

[gentle aspen]

💀 ooohhhhh helllll nahhhhhh

[rain lava]

And it has debloated versions of apps, sure it wont have ms office but libreofficd is here

[rain lava]

"Hallucinated for 2387 seconds"

[gentle aspen]

yeah

[frail raven]

I'm still team Linux because it's so much better in my opinion, especially when you want to be more productive and work fast, I'm just pointing out that it has drawbacks, as with anything tbh

[rain lava]

I see people say its scary to update arch but idk why tbh the updates dont really brezk anything

[rain lava]

Onky for people that need those apps rrly badly

[gentle aspen]

listen listen. lets calm down, linux is good for development, andif you really care about your software and hardware every now and then.
mac is good if you just want o be free of thoughts and let apple care about your stuff and baby sit you.
windows is "okay" for anyone in general with it's own issues too, but it is the middle groundof both.

[frail raven]

Hmm Arch improved tremendously in that regard in my opinion, I feel like most of that sentiment comes from how Arch used to be

But im my opinion it's still less "stable" / safe for the average user than something like Ubuntu

[rain lava]

I disagree still, add games to linux bc u get more fps if ur not using nvidia, especially in older games

[frail raven]

gotta go, bye bye ~

[rain lava]

Goodbye

[gentle aspen]

cya

[gusty meteor]

nooo!! -- we lose a linux user from convo!!! LMAO

[gentle aspen]

guys I have a question.

[rain lava]

MacOS - Never tried, hate apple to.
Windows - a bunch of spyware and bloat, also is like very large gb, good for ppl who want max support for apps and are allergic to terminal
Linux - good for development and servers, can be better in a lot of games but lacks app support, faster, better.

[rain lava]

Go ahead

[gentle aspen]

what is the optimal parameter range to run these types of models local?

image-text --> text
text-text
text -> image
audio-text-vision --> text
text --> audio
text --> image

[rain lava]

Uhh thats a Johanna question ;-;

[gusty meteor]

lol

[fiery lagoon]

im not that good with this ai stuff but i can say LINUX DEFFINATLY BETTER THAN mac os or windows

[rain lava]

Wait just ask chatgpt bc its made by ur fav guy sam altman

[rain lava]

I agree

[fiery lagoon]

first of all

mac os is apple

closed ecosystem

[rain lava]

We should have linux emoji added to server

[fiery lagoon]

second windows is made by microslop

so you can see

[rain lava]

Atleast its unix based (still hate apple)

[fiery lagoon]

yea i agree

i hate apple too

[gentle aspen]

never said linux bad. STOP TELLING TEXTING THAT I AM BIASED TO SAM ALTMAN!!!

RITHWICK!!!!!

[fiery lagoon]

what bro

[frail raven]

What?

[gentle aspen]

dang... even my friend is trying to roast me.

[open wharf]

Can we have antigravity mobile? Please

[rain lava]

You either ask chatgst or johanna or a guy thats like also knoas ai or smth idk

[fiery lagoon]

GANG LINUX IS GOOD

[gentle aspen]

who is johana??

[rain lava]

A person

[gusty meteor]

LMAAAAAAAOOOOOOO

[fiery lagoon]

[rain lava]

https://tenor.com/view/china-gif-26678009

[fiery lagoon]

chinese rock looks like deepseeks owner

[rain lava]

Whyd we need antigrav mobile, gl tryna run it

[gentle aspen]

ahhh johana. soo?
i just texted that guy like 2 days ago helping with some Ai stuff. so what?

[rain lava]

Johanna is smarter than me so ask her ur question

[fiery lagoon]

is there a tokan limit for anti gravity

[gentle aspen]

no shot sherlock

[rain lava]

On agents, yes

[fiery lagoon]

how do i check how many i have left

[gentle aspen]

well, she had struggle witha very simple thing.

[rain lava]

You dont u just wait till u get the message sayi g ur cooked (quota reached)

[rain lava]

Which is...

[fiery lagoon]

man i fcking wish i could have codex 🙏

[gentle aspen]

she had a problem with overfitting. she just had to increase thw amount of data and reduce the steps

[rain lava]

Are we talking abt the same person?

[gentle aspen]

gng it's free for opensource models

[fiery lagoon]

bro u never fucking told me that

[rain lava]

U thought opensrc models are payed?

[fiery lagoon]

i didnt know there was opensrc models

[rain lava]

Neither but im not a sam altman glazer im js a linux glazer

[gentle aspen]

yes! johana.
bro, btw I am kinda concerned about her. she said she got shot to her whole body, she is in pain. omds, there was alot to it. bruh...

[rain lava]

Shot?

[gentle aspen]

dw, he is kind aslow

[gentle aspen]

yeah!

[rain lava]

Like by a pew pew?

[rain lava]

He?

[gentle aspen]

my friend

[gentle aspen]

🔫

[rain lava]

The johanna im talking abt is a she

[gentle aspen]

my friend is rithwick. and I was talkign about johana earlier

[rain lava]

Oh

[gentle aspen]

dang bro, is she good?
what happened to her. she was so depressed and omds bro.

kinda scary

[fiery lagoon]

IM NOT SLOW BRO

FUH U

[rain lava]

Not everyone looks at chatgpt opensrc models bro 😭 some ppl just wanto not learn abt ai for some weird reason and then use them

[rain lava]

This is what happens when u dont sleep guys dont be like me 😭

[rain lava]

Ur switching between she and he im confused

[rain lava]

OPTIMAL LOCAL MODEL PARAMETER RANGES (2025–2026)

1. TEXT → TEXT (LLMs)

• 1B–3B: lightweight tasks, fast, low quality ceiling
• 7B–9B: BEST SWEET SPOT (general assistant, coding, RAG)
• 13B–14B: strong reasoning, better reliability
• 30B–34B: high-end local performance
• 70B+: multi-GPU enthusiast setups only

Optimal: 7B–14B (most users)

---

2. IMAGE → TEXT (VISION-LANGUAGE)

• 3B–7B: basic captioning, simple vision tasks
• 7B–13B: BEST BALANCE (OCR, reasoning, general VLM use)
• 13B–34B: complex visual reasoning, documents, charts

Optimal: 7B–13B

---

3. TEXT → IMAGE (DIFFUSION)

• ~1B–2B: standard models, fast generation
• ~2B–8B: high fidelity, better composition (slower)
• text encoder: ~300M–2B

Optimal: SDXL-class (~2–4B effective scale)

---

4. AUDIO → TEXT (ASR)

• 30M–100M: very fast, low accuracy
• 100M–500M: BEST BALANCE (real-time + accurate)
• 1B–1.5B: high accuracy multilingual

Optimal: 100M–500M

Audio reasoning (multimodal):

• 7B–13B LLM backbone

---

5. TEXT → AUDIO (TTS)

• 50M–300M: real-time voice
• 300M–2B: good quality neural TTS
• 1B–7B: expressive / controllable voice systems

Optimal: 1B–3B

---

6. MULTIMODAL (AUDIO + VISION + TEXT → TEXT)

• 7B–8B: lightweight multimodal
• 8B–14B: BEST PRACTICAL RANGE
• 34B+: high-end reasoning systems

---

QUICK RULES OF THUMB

• 7B = minimum “serious assistant” level
• 13B = strong local intelligence sweet spot
• 34B = near top-tier local performance
• Multimodal always needs +VRAM overhead

---

I asked ChatGPT for you

@gentle aspen

[gentle aspen]

gng, you know that i have "Plus" right? I could have asked for myself and get better results. i just wantee your opinion😭

[gentle aspen]

He == boy, She == girl.
rithwick == boy
johanna == girl

dont see the complexity here😭

[rain lava]

Yeah but u didnt

[rain lava]

Yeah but ur switching between ppl mid conversstion abt another person

[gentle aspen]

lol

[rain lava]

My opinion is what chatgst said

[gentle aspen]

yeah... I am kind aliek that lmao

[gentle aspen]

okay ig. but I thought yu would go on with the nerd talk. i mean, i am fien with anything.

anyways I have to go, bye

[rain lava]

I havent genned img etc with local models

Bye

[proud vessel]

can i run litert-lm models with ollama or i need to litert cli?

[rain lava]

can’t directly run LiteRT-LM models in Ollama. In most cases you’ll need the LiteRT CLI/runtime, unless you convert the model into a format Ollama supports.

Also idt ur name is alowed, #rules

[proud vessel]

and probably converting is a headache for a newbie ig,

[proud vessel]

changed

[rain lava]

Yes
What you should do instead (recommended)
If your goal is just to run models locally:
Easiest path Use Ollama models directly, Llama 3, Mistral, Qwen, etc.
No conversion, just: ollama run llama3
If you specifically need LiteRT
Use: LiteRT CLI
or TensorFlow Lite runtime (Python / Android)

[rain lava]

Thanks

[proud vessel]

i tried gemma-4-e4b-it-litert on my phone and its actually so fast and i thought i can build a workflow on my lowend work laptop but i surprised even on e2b-it variant its throws a eof error so i realised its not litert variant

and llama/mistral is so weak

[rain lava]

Eof? Like out of ram?

[proud vessel]

yeah

tried with 8gb ram device

[rain lava]

U can try q4 quant weights and q6 on the context, and also lower context

And also using flashattention

[proud vessel]

tried gemma4:e2b-it-q4_K_M

[rain lava]

What context did u do at, and did u set context to q6

[proud vessel]

what is context im a newbie

[rain lava]

If u really need to try q3 on it but it will be way less accurate and good

[rain lava]

The context is the information the model is allowed to “see” and use when generating a response. It includes:
Your current message
Earlier messages in the conversation
Sometimes system instructions or tool outputs
Context window (most common meaning)
In modern language models, the context window is the maximum amount of text the AI can keep in mind at once.
You can think of it like the AI’s short-term memory:
If the conversation fits in the window → the AI can reference it
If it’s too long → older parts get “forgotten” or truncated

[proud vessel]

no context at all just 'ollama run gemma4:e2b-it-q4_K_M' on a fresh install

[rain lava]

Usually that defaults to 8k or 4k context, running 0 context isnt possible
Do u do it with ollama?

[proud vessel]

yeah

[rain lava]

You can also try find a 1b model.

[proud vessel]

ill just use my main pc as host ollama servet at that point accuracy goes too low

[rain lava]

Make a file named "Modelfile" anywhere and add this to change ctx

FROM gemma4:e2b
PARAMETER num_ctx 2048

Save

Then
ollama create my-ai -f Modelfile
ollama run my-ai

[rain lava]

Yea thatd work better

[gentle aspen]

unless you need to do something liek data synthesis or someting like that 30 ish tps is good. you can host your ollama models inn your old PC dude.

plus add a small fade in per charactor whch is being streamed, it will look like the free claude

https://www.youtube.com/shorts/3Cu74V_8w3M

carter PCs explain AI likehe have being in the industery for years. and half true half false stuff😭

this offend me every single day lmao

[strong sequoia]

I'm trying not to die rn sorry

[strong sequoia]

Bro I'm not even coding rn

And yes I'm a girl

[gentle aspen]

okay, chill. that day you said some shi which made us traumatized dawg. chiiillll

[strong sequoia]

Yeah I almost died dude

It's called hyper burnout

And lack of sleep

I was coding for almost 50 days non stop

Cortisol was crazy

https://tenor.com/view/cortisol-level-gif-2410901269072201082

Now I am doing nothing

No code just cats

[gusty meteor]

https://tenor.com/view/cat-kat-kitties-kitten-meme-gif-7515526262383717306

[gentle aspen]

https://tenor.com/view/orange-cat-cat-meme-red-cat-chinese-gif-11885762802850227540

hopefully your cat doesn't look liek this cat😭

U SAID YOU WERE SHOT.

https://tenor.com/view/markiplier-iron-lung-iron-lung-movie-im-so-fuckign-scared-scared-gif-10391739852018821671

[strong sequoia]

https://tenor.com/view/justice-for-gotham-evil-cat-gif-9015363971370465898

No my nervous system is shocked

It crashed

Needs to reboot

[rain lava]

ANY lie for clicks and views nowadays 😭

[gentle aspen]

fr🫠

[gentle aspen]

This is truely what we are wasting compute and our time for guys😭

[rain lava]

I swear you have some instructions for it to talk like that 😭 thats not normally how its meant to respond

[gentle aspen]

I do have instructions. plus, openai gives you base instructions to make it look nice

[rain lava]

I think this is worth the compute power

-# generated by lyria 3, my friend sent me ts

[gentle aspen]

good naming choice😭

bro wth, ohh hell naqwww

[rain lava]

Ikr 😭

Its so bad

[gentle aspen]

bro has gemini pro right?

[rain lava]

Yes

[gentle aspen]

how did he use lyria 3, and geenrated the full song?

[gentle aspen]

ohhh

I feel liek suno v5 is better than lyria.
lyria is "okay" but suno is more advanced and have being in the industry for a long time

[rain lava]

Who made suno?

[gentle aspen]

dk

ohh just realized, I can gnerate more than 30 secodsin gemini free

[rain lava]

Lyria or smth else?

[gentle aspen]

I feel like yria is actually worth it. almost unlimited song generation and unlimited downloads in the free evrsion

[rain lava]

Ig audio is cheap

[gentle aspen]

hell no

audiop tokens are very expensive

might be THE most, just bellow video

[rain lava]

I can make technically unlimkted images but ill get rate limted eventually and its only in flow

[rain lava]

Thdn why u get sm

[gentle aspen]

wdym?

[rain lava]

You said u get like unlimkted 😭

[rain lava]

I can do a bunch of 3mon songs but more than 3 8second videos starts costin ai tokens?

[gentle aspen]

the embeddings for audio tokens are wayyy more expensive than images btww

plus accuracy is also important for audio

[hushed night]

mine's better

[gentle aspen]

lemme cook something

[strong sequoia]

I used to produce pre ai

I love ai for coding but would never use it for music production

Muh craft

[gusty meteor]

you try that latest suno?

[strong sequoia]

Yeah but takes the fun out of me

I just like playing guitar

[languid seal]

<@&1009526435276394496> spam above 🙁 👆

[hallow marsh]

Thanks for the ping!

[rain lava]

Soo, did you make smth ??

[rain lava]

I have a guitar but i dont play at all

[strong sequoia]

I have 13

This is just a few

[rain lava]

Oh damn that's alot, I only have 1 acoustic and my brother also has 1 acoustic

[gentle aspen]

has anyone tried the opensource suno music model?
I should try the one by meta, but I am just skeptical.

because these model takes like 100gb of space, so it will take like 25 minutes ish time to download.

[rain lava]

100gb? For how many parameters is that?

[gentle aspen]

just 2-4

B

[rain lava]

Crazy how on the pro plan we cant even access Deepthink 3.1

[gentle aspen]

lol, people are money hungry these days

[rain lava]

Google networth is like 3T 😭

[gentle aspen]

well, i mean if you think about it, we really can't blame google.
google are really generous with students, and google assumes the power tools are for adults, and adults have ot pay up. I mean if you think 6 hours on a T5 is bad, I dont know what is good for you.

[gusty meteor]

after these students grow up, google will hire them, which is 'win-win' status lmao

[gentle aspen]

I mean, everybody has their own freedom. If they want to, it is their choice.
plus, i am so hapy for what google has to offer.
"beggars can't be choosers" (or soemthing like that)

[gusty meteor]

💀

[gusty meteor]

you join sam altman students program?

[gentle aspen]

If you say this thing about Sam Altman and my life choices I will be summoning a moderator in no time daawg

[gusty meteor]

LMAO

just a question

[rain lava]

That was literally what i was gonna say qhen i saw this 😭

[gusty meteor]

LMAOOO

[gentle aspen]

hey guy! yall like the new update banner? I just wanted to know peoples opinion before pushing the update.

I know hte text is small, i just want to showhow it woud look like, there will eb abetter description bellow it.

[rain lava]

I think it's good, but "Alex, is 15, and likes playing video games" is grammatically incorrect due to the use of commas.

[gentle aspen]

ohh, it is fine. the user wont see the text properly anyways

[rain lava]

No way u added 67 😭 lol

[rain lava]

I think its good

[gentle aspen]

thats WASN'T intentional

[gentle aspen]

thanks

[gentle aspen]

<@&1009526435276394496> please check on with the message above.

[hallow marsh]

Thanks for the ping!

[gentle aspen]

I see why I should go back to linux.
Hopefully CachyOS is good like y'all recommended.

[rain lava]

Please turn on XMP

Poor DDR5 RAM at 4800 MTs

[gusty meteor]

https://tenor.com/view/cachyos-my-beloved-gif-15458088487794103742

[gentle aspen]

ohh yeah! I forgot. I had to reset my bios a few weeks and my lazy ahh forgot to turn that n.
does that make a big difference?

[rain lava]

It's small but it's free performance

[gentle aspen]

https://tenor.com/view/patrick-bateman-sigma-sigma-patrick-bateman-american-psycho-i-dont-have-a-gif-for-this-gif-1885171871624852210

[gentle aspen]

yeah ig. something is better than nothing ig

[gusty meteor]

lmao

[rain lava]

If I can "borrow" ddr5 ram from another computer in this house and swap it i could have 64gb ddr5...

I'm thinking of changing DE but idk

[gentle aspen]

lol.
Since, i already have a good PC my next PC will be soething like a 6090 with like 64GB of RAM, it would be dumb forme to upgrade to a better PC. since i am planning to move to NA in a few years, my current PC will rock with me for now.

[rain lava]

<@&1009526435276394496>

[gentle aspen]

<@&1009526435276394496> dude...

can you ban thes eppl

it's the second time in a row

[minor creek]

There is an increased amount of it compared to usual this weekend, thank you for helping alert us when it occurs

[rain lava]

If only they were to add an initial verification system to the server

[karmic sphinx]

Thank you for the ping!

[gentle aspen]

dude, i feel like it would be so cool if google can use their Shield gemma2 for these situations

[rain lava]

Gemma Discord security bot would be cool ngl

[rain lava]

6090 is like 2028

If i could get a big ai card thatd be nice but my psu and me cant afford that

[gentle aspen]

yeah, my PC can survive 2 years ofc. I mean i used to use an i3 2gen with ddr3 ram. That was years ago, but still, I am happy for what i have. But my next upgrade will be a good GPu with almost infinite RAM to mess with ML the proper way

[gentle aspen]

I went with a gold 800w PSU incase the 6070 or 6070 ti drops in a few years, but i feel like that is dumb considering that it is a small upgrade.

[rain lava]

16k for an rtx 6000 pro (96gb gddr7) is just so much

[rain lava]

I got an Corsair RM1000x for headroom because I was told my old Corsair CX750 wouldn't handle an 9070XT.

[gentle aspen]

I mean... 96GB is VRAM, and also considering the new "AI boom" I feellike it is fair.

[gentle aspen]

ohh same, i am on the same series as you

[rain lava]

I wonder how many params of a model it could make.

[gentle aspen]

ohh btw, i pushed a huge update witha release for ONCard today, it got some cool features like: It support the ministral series and profiles, and my personal fav ir the new "stats tracking" tool

[rain lava]

Ministral?

[gentle aspen]

if you meant by run, at Q5 at 128k context it can run a 70b ish model very comfortably without the vision embeddings

[rain lava]

I mean train

Training is so much more resource use than running it

[gentle aspen]

yeah! ministral3.

I am just afraid that people wont have the latest ollama to downloadthe new gemma4 model.

[rain lava]

Im pushing my card by training 1b at a tiny 1024

[rain lava]

Why don't you have the updates to ur app auto update ollama?

[gentle aspen]

at bf16, a about ~700 embeddings per param with about 32k sequence lenth using the SP tokenizer, and using the new flash attention which is O(n^2) you cn train about 8b of pure text networks

[gentle aspen]

I dont know if windows has a terminal command to update ollama

let alone a python function

from the ollama library

[rain lava]

I'mm experimenting with the hybrid attention thing Johanna did on my gpu -- sp tokenizer?

[gentle aspen]

I dont think you can. you need CUDA for it

[rain lava]

I can

Cuda isnt a need

[gentle aspen]

really??!!

[rain lava]

Its just a fton easier on cuda i think

[gentle aspen]

wait johanna might be cooking

[gentle aspen]

I thought it was an OSS thing Nvidia rewleased

SP tokenizer = sentence piece

[rain lava]

I used my own tokenizer

[gentle aspen]

it is my favorite toenizer. very easy to work with compared to all the other tokenizers

[gentle aspen]

I mean, it is good, but since I train on web scraped data which I made about a year ago, I prefer the Qwen or SP tokenizers

[rain lava]

Yes ROCm has matured alot tbh

[gentle aspen]

Really? I have never tried AMD hardware except for my current CPU

[rain lava]

I did webscrap data but it was a reliable site and i got 6.7GiB of data (of all numbers..)

[gentle aspen]

did you try the wikipedia dump?

[rain lava]

I've loved my AMD Card, and kinda hate nvidia

[rain lava]

No i think it was from hugging face or smth else

[gentle aspen]

I genuinely like and hate wikipedia.
my model start learning this token as a high prob token everytime sommething felt off:
|thumb|thumb|

[rain lava]

Im on around 26k steps right now

[gentle aspen]

u are trainign one rn?

[rain lava]

Ye

[gentle aspen]

I shoudl get into making a model tbh. I made my last model LucaNano-n2-preview and it sucked, and kinda lost motivation.
i shoudl try doing stuff again

[rain lava]

I'll have to add 3 fans bottom and upgrade aio because the i7 14700kf makes the fans be annoying

[rain lava]

What sucked abt it? And what loss did u get

[gentle aspen]

ohh, never go with air cooling of rintel. I use an aio for my AMD CPU too

[gentle aspen]

I got perfect numbrs : loss: ~2.1 ppl = ~15

[rain lava]

Ive never used aircooling other than when i had an old pc using i5 8th gen

[gentle aspen]

but it started learning bad tokens

[rain lava]

Loss 2.1 might be overfitting

<@&1009526435276394496>

[gentle aspen]

<@&1009526435276394496> WTH MAN!

[gentle aspen]

the overfiting for my model size was about ~1.5
also I followed the chinchilla scaling law frm google

I had more than chinchilla

I had 6.5b tokens for a 90m model

[rain lava]

Is that where u dont wanna like read the data fukly more than ?

[gentle aspen]

plus the sequence was perfect for hallucination too. I might be overreacting, since this is the base model

[rain lava]

I see people doing tiny models and now that i think maybe i shouldve done less than 1b

[gentle aspen]

no, it is: 20 X parameters = prefered amounts of data

[gentle aspen]

YOU DID 1B??!!!

[rain lava]

Ye

[gentle aspen]

wth is wrong with you?!!

1b is TOO MUCH!

[rain lava]

Gemini said itd work!

[rain lava]

It is

[gentle aspen]

idk, what is wrong with your gemni bro💀

[rain lava]

It fits in 16GB of vram btw

[gentle aspen]

1b is pro research grade shi

[rain lava]

If u just dont use deepspeed

[gentle aspen]

just because it fits doesn't means it shoudl go in... WOAHHH I didn;t mean it that way, but if you think about it, it make sense both ways

OHH HELL NAWW it is even worse thinking baout it now

[rain lava]

Deepspeed makes vram buckets and u dont need it on 1gpu

[gentle aspen]

I didn;t mean it liek that😭

[rain lava]

I dont see the problem with this model being 1b 😭
Its around like 5-6 loss, i used lion over admin so its bouncing but itll even out

Its doing 1 step every 0.56seconds

[gentle aspen]

go for 300m for HUGE experiemnts 150 for big experiemnts and 90m for constant experiemnts.
1b is just too much for experiemnting.

Gemma3:270m by itself is a great model. you just need to focus n parameters|hyperparameters|ctx

[gentle aspen]

thats suspicously fast

[rain lava]

U just use no deepspeed shi and its fast as hell

[gentle aspen]

unless you are traingin on handfull of tokens of ctx length I feel liek there is something sus going on

[gentle aspen]

my ChatGPT kept reccomending me deepsed liek crazy, mayvbe I shoudl try it

[rain lava]

Gemini said at first but then we tried no deepspeed and it was sm better

Deepspeed is a memory management but it also uses a bunch

[gentle aspen]

I use the new flash attention. it is native and better for my GPU

BTW, you are a few commits behind the latest oncard on ONCard-Linux

[rain lava]

1024 ctx, does that make it less sus or is that alot of ctx for experiments...

[rain lava]

Ok

[gentle aspen]

1024? are you trainign an AI for chat titles or something liek that? bcs 1k ctx length is basically useless

[rain lava]

That can handle 700-850 words wdym

[gentle aspen]

dude, yur Ai can only remember 1000 words. a system promtp by itself is usually longer than that

[rain lava]

Its experimentation

Idk

[gentle aspen]

if it is experiemntation why did you go for 1b params?😭

[rain lava]

Going higher would have to mkve to ram

[rain lava]

Because uh my gpu can handle it

[gentle aspen]

no shot sherlock, why do you think us researchers dont go beyond 250m ish params

[gentle aspen]

yeah, but you are trainign a goldfish

[rain lava]

By tommrrow at 11am it should be at 100k steps

[gentle aspen]

a 300m model with a higher ctx is worth it than a 1b model with 1k ctx

[rain lava]

Thats fine

[rain lava]

Until u try use a bpe tokenizer

[gentle aspen]

100k steps ONLY??? FOR A 1B model??!!

[rain lava]

Thats 600m tokens

[gentle aspen]

yeah, but it's just not worth it. what are you going to do with your model?

[rain lava]

Gemini said it was fine

[gentle aspen]

ONLY 600M TOKEN??!!

[gentle aspen]

NOOOOO!!!!

you NEED 20b tokens to train a 1b model

[rain lava]

Give it to a friend

[gentle aspen]

AT LEAST.

[rain lava]

"Need"

[gentle aspen]

omds bro.
so lets get this straight:
You are wasting your compute.
for a friend.
Who will se no purpose except for auto correct or chat titles.
and also overfit it with less data and ruin generalization for QLoRA or LoRA

[rain lava]

Im not making the ai do college level questions dude it can do english and grammar and they js wanted an ahh mkdel anyway 😭

[gentle aspen]

nahh corban is just built diff, bcs i dont see me doing even a 20m model for a friend

[gentle aspen]

then, why didn't you tell him gemma3:1b exists??

<@&1009526435276394496>

[rain lava]

I swear that is the same person

[gentle aspen]

man, i had enough with this bots, i am goign to take a rest from my coding / prompting / texting gogoles community / watching reels / repeat

[rain lava]

They dont want gemma they specifically asked me

[gentle aspen]

fr

[rain lava]

Thatd take 20days to train and would cause permanent silicone damage from VRM Stress

[gentle aspen]

bro they have: gemma, qwen, GPT2, llama, deepseek, phi, mistral, and other stuff on HF, and still found YOU to do it...
this is so poetic, dude.

[rain lava]

Ye

[rain lava]

Would increasing batch sizes and spilling into ram be better or lower with just vram..

[gentle aspen]

considering your ettup just stay as you do, it woudl we painfully slow. woudl take liek 2-3 to finish trainign if you go even a very small step up to the CPU and RAM

[rain lava]

I think the whole time its been spilling a small amount into ram and i can make it faster

[gentle aspen]

DONT DO ANYTHING. the spilling is normal, it is balancingyour datasets, tokens and stuff.
it is just a PyTorch convinience, not actually "spilling".
i know it feels liek it, but it aint. it is normal. please keep it liek that dude. trust me, you dont want o regret your PyTorch choices...

[rain lava]

Ok

I should probably sleep.

[gentle aspen]

lmao, you should

[gusty meteor]

pytorch is the best ai training lib?

[gentle aspen]

I see why you are saying this. bcs you wills ay tensorflow is better just bcs you ae on the google community. but I would say ytorch is CONVINIENT!!!!

[gusty meteor]

no im just asking lmao

why you take it different

how about tinygrad

[gentle aspen]

yes! it literally is the BEST.

[gusty meteor]

better than tinygrad?

[gentle aspen]

no shot sherlock. it is better.
it kinda follows the bigger = better. and also pytorch is just better...
I just dont know how to explain it. the only place youwill hate it is, when you are compilling a python code

[gusty meteor]

you say the before even look the tinygrad right?

lmao

[gentle aspen]

I know aoput it, but I havn't tried it. but pytorch is literally THE worlds best ML library dude. tiny grad is okay if you are trainign a NN with a laptop or a not so capable device. but its just pytorch is goo dbro. have YOU researched about it.

[rain lava]

I cant tho 😭

[rain lava]

I took geminis advice and now its faster and i changed batch size, it still uses the same effective 198 size but it's using 4 rather than 6 which works better. Its 40% faster.

[stark sapphire]

If you guys ever make an application or website with Antigravity, and you have the issue of the AI making buttons that do not work, we can fix that with a rule.
We call this Global Knowledge Item (KI)

put this into your system's knowledge directory and call it DEVELOPER_DECREE.md

This document serves as a binding development protocol for all AI agents working on any project in this user's environment.

## 1. THE NON-NEGOTIABLE RULE
**If a button, link, or interactive element is visible in the UI, it MUST have a complete and functioning logic path.** 

## 2. FORBIDDEN PRACTICES
- **Placeholder Toasts**: Do not use `toast('Stub')` or `toast('Coming Soon')` as a final implementation step.
- **Dead Buttons**: Do not render `<button>` elements without an `onClick` or a `type="submit"` with an associated handler.
- **Mock-Only State**: Do not implement UI that only works in `MOCK_MODE` without providing a clear path or warning for the production Database environment.

## 3. MANTRA FOR EVERY TASK
1.  **Map the Circuit**: Identify all entry points (buttons/links) for the current task.
2.  **Verify the Flow**: Ensure every click leads to a state change, a database update, or a page navigation.
3.  **Confirm the Outcome**: After implementation, verify that the action achieves the user's intended goal, not just a visual confirmation.

*Failure to adhere to this decree results in System Inefficiency and User Trust Degradation.*```

[stark sapphire]

Universal Protocol Dossier

This protocol is a permanent anchor for every AI agent operating on this machine. It was established to ensure that all intellectual assets are built with integrity, precision, and functional completeness.

THE CORE DOCTRINE: ZERO STUBS
Any agent interacting with a codebase must ensure that the user's experience is never interrupted by "Dead Ends."

1. INTERACTIVITY INTEGRITY
Buttons: Every <button> must have an onClick or onSubmit that performs a real action.
Links: Every <Link> or <a> must point to a valid route.
Inputs: Every form field must be correctly hooked into state and submission logic.
2. THE ANTI-TOAST RULE
Do not use toast('Placeholder') or toast('Stub') as a final solution. If a feature is requested, it must be fully implemented, including:

State management (Local or Global)
Data Persistence (Supabase or LocalStorage)
UI Feedback (Success/Error states that represent real data changes)
3. THE PROTOCOL AUDIT
Before submitting a task as complete:

List all new UI elements.
Describe the logic path for each.
Confirm that the user can verify the results immediately (e.g., refreshing to see data persistence).
4. EXCEPTIONS
Placeholders are only allowed if the user explicitly asks for a "Visual Mockup" or "Prototype Only" implementation. In all other cases, "Functional or Nothing" is the required standard.```

The Universal Source Authority is stored at the following absolute path on your machine:

C:\Users\YOUR-NAME\.gemini\antigravity\knowledge\sovereign_development_protocol\artifacts\protocol.md

And its associated metadata (which tells the AI when to trigger the rule) is at: C:\Users\YOUR-NAME\.gemini\antigravity\knowledge\sovereign_development_protocol\metadata.json
Any AI agent you use that follows the standard Knowledge Item (KI) protocol will see these files and be bound by the rules described within them.

[gusty meteor]

you use mostly gemini in antigravity?

[stark sapphire]

yes gemini is the cheapest

[gusty meteor]

thanks

[round solstice]

I think that's more than Spın̈al Tap had.

[strong sequoia]

Probably ahahah

I will post full collection soon

[gentle aspen]

@strong sequoia what was that attention module you made which corban was talking about?
Can i have the repo? williut work on my rtx 5070?

[strong sequoia]

Yeah it will

I will dm soon I need to sleep

[gentle aspen]

Can someone give me a good system prompt for my agents, al the system prompts which chatgpt makes and I make are just "helloworld("print')" type stuff🥀.

[rain lava]

Wdym system prompt?

[gentle aspen]

you are joking, right?

[rain lava]

No you just said system prompt i have zero context

[gentle aspen]

I said **system prompt for agents **

[rain lava]

I don't know what a "system prompt for agents" is dude tell me 😭

[stark sapphire]

in order for the AI to give a good prompt, you must first tell the AI what you are making, and which framework etc you are going to use. Also tell it to not make code snippets.

in the end, you must be the person making the prompts not the AI. Because the AI will lose context / memory loss, and it would make stupid prompts.

also worth noting, you don't have to make any prompts anymore. AI agents are excellent at understanding now, if you talk to it casually.

[gentle aspen]

this man will neevr understand how Ai works, why dont youtubers actually research before they yap btro.
https://youtube.com/shorts/m2Kg5xDHTJ4?si=d54ZwLyOCIOXJHD2

I remember how carter PCs used to give proper advice. THE AI LITERALLY DOESN'T HAVE TOOLS😭

[strong sequoia]

I do ai research and I get harrased

Because some people have this emotional reaction towards ai

[gentle aspen]

fr

[strong sequoia]

That's so bad it Includes harassing people that do research

Idiots

[gentle aspen]

nobody likes me in class. they think that I am weirdo bcs i research on NNs😂

[gentle aspen]

ikr

[strong sequoia]

People suck

Imagine bullying researchers

[gentle aspen]

yeah they do!

[strong sequoia]

Or scientists

Because you are insecure

😭

[gentle aspen]

well, the world shoudl have both the positive and negative side, it's not our fault for people not understanding. tbh, I just ignore and I make it a part of my life

[strong sequoia]

One thing is hating ai

Another is the witch hunt against people that research ai

It's a witch hunt

https://tenor.com/view/ksganti-pitchfork-gif-24782218

[gentle aspen]

I just dont see the reason for people hating AI. like what do people want? earth to be the same place? why does people hate advancing so much?

[strong sequoia]

The world rn

[gentle aspen]

lol

[strong sequoia]

They hate what they don't understand

It's just ignorance

[gentle aspen]

yeah! and the people who hates it uses them the most😂

it's like blaming the cheff for making food which they have to eat anyways

[strong sequoia]

I have been always pro ai

Which means I get genuine online harassment

From people

Not even just jokes

[gentle aspen]

js dont care about that.
only actual engineers know the satisfaction and the brain powered put on the models they hate

[strong sequoia]

Actual harrasssement

[gentle aspen]

yeah I know.

[strong sequoia]

Which is wrong

The same people virtue signaling about Ai are harassing a woman online

🙄

[gentle aspen]

well, you know what johanna? you gotta chill a bit rn.

[strong sequoia]

Yeha I'm trying

I think dev communities are cooked

[strong sequoia]

Yeah

[gentle aspen]

https://tenor.com/view/cook-hello-gif-2755446217296855335

[strong sequoia]

Sam Altmans house

Got firebombed

.....

Violence is not ok

[gusty meteor]

LMAOOOOO

dont say anything about sam altman to only might

LMAOO

[stark sapphire]

What he did is not okay either

[rain lava]

What did he do? I don't listen to much news

[rain lava]

Fr lol 😭

[stark sapphire]

you would have to google it, i can't say the word

[strong sequoia]

oh i missed the news

im not on social media just discord

[strong sequoia]

same

i stopped consuming news

my only "social media" is discord

[stark sapphire]

https://tenor.com/view/aurora-aurora-aksnes-reading-chat-aurora-reading-chat-gif-1662704307234358366

[strong sequoia]

Aurora

[stark sapphire]

my queen

[strong sequoia]

the name of my browser engine in rust

too

🥺

[rain lava]

U talkn about that lawsuit or a different accusation

[strong sequoia]

Im gonna post again in hacker news today

6 pm portugal time

or maybe not

dunno if i can handle itr

emotionally

might have another panic attack

https://tenor.com/view/relaxed-cat-fashion-love-pet-gif-22308133

im just drinking camomile tea

[rain lava]

I like tea, much over coffee atleast

[misty wasp]

Please head to #general to continue this conversation
The messages need to be on channel topic

[gentle aspen]

EXCUSE ME?

[gentle aspen]

WHAT?!

[broken pike]

Hey everyone, I’m new to development and planning to build a mobile app for Android and iOS.
A lot of people recommended Flutter, so I’m going with that.
I’m stuck on which AI coding tool to use: Antigravity, Cursor AI, or Cursor AI with Claude Code integration.
I’m on a midrange Windows laptop, so for now I can only build/test Android APKs properly.
Any recommendations on which AI tool would be best for a Flutter beginner in my situation? Thanks!

[gusty meteor]

whichever one gives you more quotas XD

[gentle aspen]

for mobile cross platform i will reccomend you stick with the native frameworks and use claude code if doing mobile. claude is good with covering niche functions.

just becuase you get more quotas doesn;t mean it is better.

"Just because it is bigger, doesn't mean it is better" ~She didn't say that.

[gentle aspen]

considering your bio, i assume, you are going to make something related to hiring people or managing, this require sending data with a cloud DB, so I reccomend sticking mostly on iOS and duct taping stuff for android.
claude will handle it.
also, it is not "Most quotas", it is how efficient you are with the AI.

back then I used to do "High" on codex with the latest model, but I didn't need that for the task.
before uding codig model, learn how YOU would code that. if you feel that is considerably easier than a bigger task, then go for smaller reasonign levels. CoT and files eats up tokens real fast.

If you have the google's premium subscription, antigravity would be good, bcs you not only get claude models, but other models liek gemini and GPT too, plsu you get a UI unlike claude code.

if you are really a "beginer" use codex or antigravity.

if you are broke (like most of us) use antigravity or qwen code.

hope I helped you :)

[gentle aspen]

Guys! genuine question.
What are the best ways to study online?
eg: Flashcards, MCQs...
What do y'all prefer?

[broken pike]

Okay it’s something completely different it’s a simple app but kinda complicated if u like I can dm you so it’s a simple app like structure app on iOS AppStore just recreating something that already exists so since I wanna learn a bit with ai

[gentle aspen]

yeah! DM me, I will help you out!

[rain lava]

Wikipedia articles

[gentle aspen]

no, I mean with interactive learning

[rain lava]

I think the gemini interactive things are cool

[thin meadow]

I have tried flashcards before. Most of the time, waste of effort and time. MCQs help you explore how and why you are wrong on specific questions/problems. Sometimes it would even help you clear misunderstanding regarding an answer to a question.
Eg. - Why does a specific reagent cause a reaction while others don't? When you get an MCQ wrong, it forces you to re-examine the underlying concept rather than just memorizing a definition.

[gentle aspen]

You are right! Having to pick a choice is wayy better than being judged for typing an essay.
I didn't make the UI looks good right now, but I will be focusing on that later.
but htis genuinely feels much better for myself.

[rain lava]

It's hard to see the text. Why're the screenshots so low res? 😭

[gentle aspen]

wdym? this is at 1440p

lemme upscaler this by 8x for you

[static bluff]

anyone with problems to log in to Antigravity?

[gentle aspen]

it is a bit annoying

press the log in two times and it will og in

it is a known bg ish thing

[rain lava]

Look how like low it looks

[gentle aspen]

works fine when I crop it in my discord

this is 20k x 11k res

try now

[rain lava]

Yea thats better

[gentle aspen]

no shot sherlock. it is literally 20k

[rain lava]

Actually can u go to 40k its a bit hard to see still

[gentle aspen]

k

CORBAN!! MY PC!!!

[rain lava]

Ur using a website, right

[gentle aspen]

i cant upload. the max upoad size is 100mb

[rain lava]

Ah :(

[gentle aspen]

DISCORD!!!!!

[rain lava]

I was kidding btw 😭

[gentle aspen]

https://tenor.com/view/silent-gif-6302280696616265589

[gentle aspen]

happy my GPU didn;t expode

[rain lava]

What do u use to upscale that much...

[gentle aspen]

I used upcayl

it is so cool bro

[rain lava]

I'm being an idiot by staying up again and i also gotta go to smth like 7am 😭

[gentle aspen]

I am lucky bcs I dont have school these days lmao

[rain lava]

So i can go to a million res right

[rain lava]

I dont either but im going to a workshop

[gentle aspen]

if you have a bijilion GB of ram

[gentle aspen]

ahhh, so cool

[rain lava]

Idk if i will or will not find it cool

[gentle aspen]

I mean show off your best skills dude

[rain lava]

We doing LiDAR :/

[gentle aspen]

USE YOUR BEST SKILLS AND COOK EVERYONE ELSE. NEVER BE THE SECOND

[gentle aspen]

of course you are :|

[rain lava]

But he (guy that runs it) bought like a bunch of metaquest 3 vr headsets to implement into

[gentle aspen]

nvm. soo coool

[rain lava]

That was the inky 1 avaukable!

[gentle aspen]

dk about that

[rain lava]

If there was smth else diff i wouldve gone but there was nthn else

[gentle aspen]

AI?

ML?

VIBECODING???

[rain lava]

I sense sarcasm

[gentle aspen]

imagine if there was a vibecoding competition? I would get 1st place

[rain lava]

No he was hosting lidar bc i actually dont know

[gentle aspen]

actualluy🤓👆, it is "CarSacm"

lol

[rain lava]

Term1 of coding courses ended, i go to term 2 around start of t2 school

[gentle aspen]

NICE!
btw did you get the new ONCard update?

[rain lava]

No bc i cant run it on linux, im still working on running this one on linux but ive been busy with errands and not sleeping

[rain lava]

Ye, next workshop, he said, woukd actually be much better, i think he did lidar bc a bunch of ppl wanted itl

[rain lava]

We also do some 1on1 sessions through holidays

[gentle aspen]

dude, I wish I was in your country :(
all the countries i go to says they have "Good" infra, but it is always "Good enough"

[rain lava]

And i get to meet some other people ive never seen before at the workshop so thats just great isnt it, (social interactions, so scary)

[rain lava]

Its not all sunshine and rainbows over here

[gentle aspen]

I never got bellow 1st on any of my competitions I went on my ENTIRE life. and I have gone to liek 30-40

it's never is in any country, dude.

[rain lava]

The course costs i think more than 1k or more, but its specifically for people that have NDIS Funding and its aimed for neurodivergent peopel!,

[gentle aspen]

dang...

thats a LOT

[rain lava]

Yea but my parents said its worth it, we also have ndis

[rain lava]

I think the mayor of <our area> (that im not going to say) visited and took a photo for some,of the courses i think it was the hackathon

[gentle aspen]

can you continue on our DMs bro! I think we are going a bit "offroad" dont you think ? :)

[gentle aspen]

ohh

[rain lava]

Yes i do lol

[broken prairie]

They need a Discord dedicated to their IDEs, Coding CLIs and other Software.

[thin meadow]

That looks genuinely good. It does need a few improvements, but overall feels light and easier to process. Isn't too flashy, isn't too simple.

[gentle aspen]

yeah! i am already on it. just dong some finetunes here and there so it looks presentable and actually usable

[thin meadow]

All the best with it. Though what is it for, personal or professional?

[gentle aspen]

professional. i have a repo for it.
it need HUGE updates before i can put it as an update.

[thin meadow]

I see. I would like to see the finished product. Could you show it to me once it's done?

[gentle aspen]

what would it look liek if everybody started typing? Man, my ADHD brain needs some rest😂

[gentle aspen]

it is already done

you can try it

the latest version 1.3.7

only on windows tho

[rain lava]

I think it's good but - Where i put the blue lines is where i think u should add some spacing to make it look less like its clipping into another UI Ellement.
Also, if you haven't already, you should add a toggle to make the UI Dark mode.

[rain lava]

I've been in a 700k member server before and the chat has slow mode but it gets very fast

[gentle aspen]

https://github.com/MightyXdash/ONCard

[thin meadow]

Oh good

[gentle aspen]

on it!

[thin meadow]

Would try making it work

[gentle aspen]

not pushing you, but if you want to study, maybe try it out.
You can turn all your study notes into flashcards locally.

[thin meadow]

Nice. Would absolutely use it.

[thin meadow]

Btw I see a "Follow Up" side panel to ask questions. Does it connect to an AI API or does it have pre build answers?

[gentle aspen]

haha thx :) btw there is an algorithm to track your study performance and personalize showing you cards. rigth now it uses an embedding model for semantic similarity, but I will upgrade this in the future. this will feel liek the 2020 yt algo a bit, but I am still learning as i make it :)

[gentle aspen]

uses ollama

U need ollama for the app to work

much better!

[thin meadow]

Cards personalization? I thought it was a simple app to help with Study only, good going there bud 😄

[thin meadow]

Ye this is better

[rain lava]

You should make it install ollama with it so you don't need to install it manually
winget install Ollama.Ollama
And with oncard updates, it will update ollama, (You thought ollama on win cant be with terminal but you can using winget)
winget upgrade Ollama.Ollama

[gentle aspen]

thx. it also tracks your performance and recommend what you should study.

[rain lava]

Yes that's 🔥

[gentle aspen]

ohh yeah TRUE! i love you corban. you are a LIFE SAVER!

[rain lava]

You're welcome :)

[thin meadow]

Impressive 👍🏻
You got a follower on GitHub

[gentle aspen]

thanks, i really appreciate that!

[rain lava]

I just realized something!
If/when you implement it, rather than winget upgrade Ollama.Ollama, a better command might be;
winget source update; winget upgrade --id Ollama.Ollama -e
Because it updates winget's cache, it should stop rate update errors

[gentle aspen]

ohh okay, I never knew that can happen :/ thx for that dude!

[rain lava]

No problem 🙂

[gusty meteor]

why you have different version for linux 😭

[gentle aspen]

to make life easy for your role.
If it was only one linux distro, the same problem with windows would be encountered, since you are trying to server all the types of users.

[gentle aspen]

ohh wait, i didn't read that right

it is so your distro's package manager knows what the command does. Look at this "Mr linux guy" not knowing this. lol

[gusty meteor]

i think it can be in the same repos

[gentle aspen]

that would make life worse, and annoy devs yk

[gusty meteor]

💀

[rain lava]

Linux distros have a package manager and aur helper
For CachyOS (Arch Based), it's pacman (PKG Manager) and paru (AUR Helper)

[winter viper]

sup guys! wasn't here for a week or two, are there any changes in AG or limits in total?

[gentle aspen]

nothing much

[strong sequoia]

OK I got unlucky with hn post

Can't win everyone tho

[strong sequoia]

Snap is cool too

I'm on Ubuntu /debian

Pop os

[orchid forge]

hi

[languid elbow]

hi :>

[languid elbow]

same, never heard of pop OS tho, is it good?

[strong sequoia]

I love it

https://system76.com/pop/?srsltid=AfmBOoqvSDz29Az4PHWLDMbuBc-oJNy4JEGS80MzUr6XjrYBUo5Y9_rB

[rain lava]

I've never used PopOS.

No

Even in italic it's still a no

Give the questions small numbers in the corner and make it so you can just press 1, 2, 3, 4/How many questons there are to answer.
Example:
What state of matter is water?

1. Gas. | 2. Solid.
2. Plasma. | 4. Liquid.

---

You press 4/What you think the answer is on the keyboard and then enter to confirm the answer.

[mild slate]

Claude errors a lot lately. Is anyone experiencing this too?

[kind spoke]

Claude has been horrifyingly bad in the past few weeks. Including Opus.

[strong sequoia]

yeah

pop os is just ubuntu base

[rain lava]

Atleast it's teaching me how to use blender though (I've never used it before)

[languid elbow]

I might try pop os just cuz you mentioned it

[rain lava]

<@&1009526435276394496>

[rain lava]

Let me know how it goes. I've never been a fan of ubuntu distros tbh

[languid elbow]

alright lol

[languid elbow]

?

did I miss smth? :(

[rain lava]

Yes you missed the 10seconds of a link to NSFW.

[languid elbow]

awh shucks

couldve clicked that link man, jk jk

[rain lava]

Next one bro, next one you'll get it...

[languid elbow]

https://media.discordapp.net/attachments/1331509914706907170/1489715849987231875/watermark.gif?width=220&height=220

indeed I shall

[tall walrus]

Are there any Google subscription plans that can buy me Google AI Studio and Integra gravity together

[open wharf]

Google AI plus

[tall walrus]

I think it’s mostly AI video generation like VL 3.1

And also Gemini pro which I don’t need I need Google AI studio and antigravity

[fresh urchin]

is it possible to use AntiGravity subscription anywhere else than in AntiGravity? I know if you are using it in CC you get a ban, but is something like OpenCode VS Code allowed?

[vagrant folio]

no antigravity quota only on antigravity if they detect you are useing something else you will be banned

[fresh urchin]

bummer
ok, so what containers or sandboxes are you guys using for AG?

[open wharf]

Check the subscription plan, I think there's an option for that

@keen kayak please I'm unable to send DMs

[still bolt]

What does it mean

[tall walrus]

ChatGPT scks

[tall sierra]

Google Antigravity is almost unusable at this point

[gusty meteor]

even for flash lma oxd

[tall sierra]

yeah. i was using flash for some simple tasks. yet i was getting this error again and again

[gusty meteor]

interesting it happening on the flash also

its understandable for claude but XD

actually just think both of them are interesting becaue antigravity should be proxy your requests

instead of going to direclty claude api

and 1gb ram and 1 core vps even handle that i think LMAO

[tall sierra]

if the low quota and this issue persist I'll have to look for alternatives. Most likely I'll use Claude code 17$ plan instead of my current gemini ai pro plan

but claude has nerfed their model silently.

[novel remnant]

I can't share my vibe coded project?

[gusty meteor]

you feel about like that or news?

[gusty meteor]

here?

[novel remnant]

no its a link. i couldn;'t do it on gemini discord

[tall sierra]

many senior developer has experienced that. and it is currently at 2nd position in a benchmark.

[gusty meteor]

whos 1?

[tall sierra]

a model of grok

[tall sierra]

it was on a different benchmark

[gusty meteor]

what i meant, dont act by looking other peoples, examples most of people are saying 'codex is better than claude bla bla', just tried codex, its very trash for me

codex for me; if i explain exactly what to do in such detail, what’s the point of you being an ai lmao

[gusty meteor]

just buy 200 dollar claude package you will be safe XD

[tall sierra]

first I'll have to find an way to earn mode than 200$ using the package.

[gusty meteor]

lol

[novel remnant]

nope the guy that dissed claude made a long emperical study about it. 200 usd not safe either

[gusty meteor]

atleast its for me xd

[novel remnant]

that's good then 🙂

[minor maple]

I'm from Japan
I would like to take a competent AI expert as my mentor and learn a lot from him while assisting him

[limber fable]

Hey guys, im first time using Antigravity. Now i want to run multi task on it but it really weird. They always work on others workspace, not the workspace i need them to do. Some times i run the project but they run the project i dont command it to run. Any solotions to fix this ? Thanks you

[vagrant folio]

Hi, when you open a agent manager and click on new chat you need pay atention on workspace selected i not which you are working, and you can start multitask at same time fr same project or in diferent project

[limber fable]

i find a way to fix it is using rules, but thanks a lot

[slender stag]

Hey why it is keep happening to me

[rain lava]

If you look at the troubleshooting guide it might tell you.
I can't say an exact reason on why that's happenning with no details-nor can anyone else.

[slender stag]

Ok sure

Trajectory ID: 2ded31fb-3f9d-4977-ac4b-7ed85ca2a5e7
Error: HTTP 503 Service Unavailable
Sherlog:
TraceID: 0x8dd116ceb831525d
Headers: {"Alt-Svc":["h3=":443"; ma=2592000,h3-29=":443"; ma=2592000"],"Content-Length":["421"],"Content-Type":["text/event-stream"],"Date":["Wed, 15 Apr 2026 03:58:13 GMT"],"Server":["ESF"],"Server-Timing":["gfet4t7; dur=1181"],"Vary":["Origin","X-Origin","Referer"],"X-Cloudaicompanion-Trace-Id":["8dd116ceb831525d"],"X-Content-Type-Options":["nosniff"],"X-Frame-Options":["SAMEORIGIN"],"X-Xss-Protection":["0"]}

{
"error": {
"code": 503,
"details": [
{
"@type": "type.googleapis.com/google.rpc.ErrorInfo",
"domain": "cloudcode-pa.googleapis.com",
"metadata": {
"model": "gemini-3-flash-agent"
},
"reason": "MODEL_CAPACITY_EXHAUSTED"
}
],
"message": "No capacity available for model gemini-3-flash-agent on the server",
"status": "UNAVAILABLE"
}
}

[rain lava]

Okay, now search on google what that error means.

I actually think I already know the error though.

[rain lava]

• "model": "gemini-3-flash-agent"

•    "reason": "MODEL_CAPACITY_EXHAUSTED"
  

• "message": "No capacity available for model gemini-3-flash-agent on the server",
• "status": "UNAVAILABLE"

[mild slate]

It happens to me as well. It is happening to claude models before, but today i can't even use any gemini model. Does anybody know what is happening?

[pastel rain]

same here

[mild slate]

Is anybody from google can give us information about this? This is getting frustrating

[slender stag]

Yeah from morning

It is not working

[rain lava]

Literally look at the error information dude it tells you

[slender stag]

Service is unavailable looks like it’s an issue on Google’s side.

[rain lava]

Service is unavailable because you exhausted your qouta 🙄

[mild slate]

It isn't providing me information I need. I get the idea that the server is experiencing high traffic, but is there any information on when the service is expected to improve

[rain lava]

It's impossible to get an ETA on servers, it's a high rate of people using them.
Unless you ask each individual user on antigravity right now when they will all get off, which is impossible, you cannot get an ETA.

[stark sapphire]

it's not

[rain lava]

The error information says overwise

[stark sapphire]

it does, but it if shows up for everyone, then that is very unlikely

or they just meant for the model and not for the user.

[mild slate]

I am beginning to think that this is not a user or even a localized problem. There is something wrong here

[rain lava]

I don't see why it would not show for everyone? The server wont just magically be able to fit another user, the unlikely part i see though, its unlikely that the server is at absolute maximum capactiy.
In the event of an actual ongoing problem of google's side, I was wrong - I'm use to idiots just winging that they cant use antigravity because they reached their quota so i usually just say that because im so use to it happenning so often.

People cant live 5h without gemini?

[stark sapphire]

i know this feeling all too well. Just never assume someone as an idiot. unless they show they truely are xD

[serene crow]

Antigravity is under the outage and unstable. Do you have these experiences?

[rain lava]

I have seen alot in this server though where it really isnt a server problem and its them lmao

[rain lava]

I love how people act like it's the end of the world when ai servers, or any servers for that matter, go down

[stark sapphire]

It does suck when it goes down, because i rely on AI.. Sad but true.

[rain lava]

I mean... chatgpt works

[stark sapphire]

for me it doesn't. none of the models work in antigravity

[rain lava]

Same as running ur own model

[rain lava]

I mean if u download via api in terminal and integrate antigravity with opencode

[stark sapphire]

I have my local AI, I just don't know how to put it in antigravity if that's even possible

[rain lava]

If you run it whilst in the project folder itll edit remove and runn commands as you allow and theres a terminal in antigravity so u can out in their.
Or you candownload opencode, which is still terminal but its got a nice gui so its easier to use in antigrafity

[gusty meteor]

"I think Google is intentionally triggering this error only on older Google accounts to prevent the use of multiple accounts. In other words, if the Gmail account you entered into 'Antigravity' if the Gmail account you entered is newer—say, about a month old—this error is intentionally triggered, in my opinion. But if you use an older Gmail account, you’ll see there’s no issue. As I said, I think they’re doing this to frustrate people who use multiple accounts; otherwise, it would be utterly ridiculous for a company like Google to be unable to handle such a load, which is impossible—I’ve been using Antigravity since it first launched, back then Antigravity was completely free and sufficient for everyone

[serene crow]

btw, gemini-cli is working well

[stark sapphire]

My google account is over 15 years old

[gusty meteor]

change a another account and switch back here and see the difference

[mild slate]

I have been using antigravity since end of last year, and my account is over a decade years old as well

[gusty meteor]

again, change a another account and switch back here

[mild slate]

I only have 1 account, gotta create a new one for that. Thanks for suggesting

Is changing the account works well for you?

[gusty meteor]

Let’s wait and see what @stark sapphire says; I’m not entirely sure either—these are just theories—but if they turn out to be true. lmao

[gusty meteor]

yeah to old accounts, news one take that error

[stark sapphire]

Eh i already have 3 google accounts, so i don't really want to create another just to try this out.

[gusty meteor]

nope just change account, dont need to create one, we do that for clearing everything about that acc

[stark sapphire]

all of them are over 10 years old

It make sense for Google to do this, they don't want newcomers to walk away, so they have an open window for them

[halcyon scroll]

I faced same issue since last night.

[halcyon scroll]

same error

[mild slate]

Guys please try it again. It is working for me now

[slender stag]

Yeah it is working now thanks for telling reza

[halcyon scroll]

I was thinking if Google implemented a further usage quota limit or not.

[mild slate]

I just hoping to get any form of compensation for all the delays😆

[stark sapphire]

i get nervous when the AI says "I messed up"

[gusty meteor]

lmaoo that should be meme

has anyone tried new claude desktop app, whats diff between cli, curious

[slender stag]

I am using linux. i hope claude brings desktop app to linux.

[gusty meteor]

same

[stark sapphire]

Can't you use something like Wine?

[gusty meteor]

claude releasing a lot of products these days; they’ve lowered the limits on the publicly available models so they can use them themselves lmaooo

[slender stag]

I'm on the free plan, and I got rate limited after sending two prompts.

Is it worth taking $20 pro plan?

[slender stag]

What is it?

[gusty meteor]

lol

[stark sapphire]

it is a program that allows you to run any windows application on Linux.

[slender stag]

I will try it

[fresh urchin]

gemini (chat) insists I install "dev containers" extension in Antigravity
but I can't find "dev containers" in the marketplace at all

[stark sapphire]

tell your ai to be more specific.

[languid elbow]

thats pretty broad, as he said ask it to be more specific on what containers extensions it wants

[rain lava]

It depends, is gemini the ai you preffer the most?

[rain lava]

Its token based

[rain lava]

If so then i say go for it.
It doesnt just get you the chatbot;
It gets you Imagen 4, Nano Banana 2 & Pro, Veo 3 Fast and you can likely use Veo 3 ultra via Google Flow (You get monthly 1000 ai tokens for flow), also lyria 3, and you also get 5Tb cloud storage.

[slender stag]

I use GPT mostly

[hoary marlin]

Any one else facing this problem 🙄 or its just with me ?

[languid elbow]

copy the debug info and paste it here

[slender stag]

Hey Yasir, we were facing this issue recently, but it’s been resolved now. Could you please check the Antigravity status once?

[hoary marlin]

It was a servers issue now its solved

[hushed night]

yh me too

[thin meadow]

Hey guys, just wanna ask some things. I have been using Gemini for a very long time now. This whole time, I made these things with Gemini:

1. Various Webpages
2. Discord bots
3. Some other text-based bots
4. 3D simulations
   But I think I am not trying to build things that does something. Any ideas or suggestions for me?

[stray verge]

Yeah send a dm

[hushed night]

try implementing multi agents in more complex projects

[thin meadow]

Good idea. But I have once tried it before. Used 3 open source models to make a multi-agent reasearch workflow. But it was really hard.

[hushed night]

try starting from something easier

[stark sapphire]

wow guys. i get free Ultra from the boss. I'm super stoked.

[stark sapphire]

<@&1009526435276394496> if you ask for testers, does self promotion become acceptable? As seen as above?

[karmic sphinx]

Hello! No, asking for testers still falls within self promotion, thank you for the ping!

[vagrant folio]

Congratulation! I want it too 🙂

[lilac edge]

Yes very much. Also, maybe you can still get in on the 12 month free deals they had. Lord knows I did

And question guys, in my current project, when I go to manage agents, and I create a new one, where do I put the rules for that new agent in?

[vagrant folio]

AG?

[south pelican]

Hello, I am full stack & AI developer.
are u still looking for dev?

[stark sapphire]

no not anymore.

[undone pendant]

Any update on when we will get Opus 4.7 in Anti-Gravity?

[gentle aspen]

ofc it is going to be so bad, dude. ofc we are getting crazy rate limits

<@&1009526435276394496>
Hey quick question.
recently my account got hacked (I had MFA too, and didn;t join/ or get involved with any action which will reveal my account info)

but some how someone send spam advertisements tagging everything.
I got banned for that for 1 day.

Could I know that can someone erase that history?
I don't want to have a bad history. I got banned for advertising. which I would say make sense, bcs I was trying to find someone to collab with my project not knowing that ewas a section FOR that.

could I knw "If you (a moderator) can remove the miss info"?

[gentle aspen]

yeah, same.
I was trying antigravity until my codex credits was out.
Thats something i hate about antigravity. maybe they will give us some more credits in some events/seasons.

thats the patern I learned about antigravity.

[swift carbon]

If you're talking about the messages that you left in the server, they all have been removed, dont worry 😄

[gentle aspen]

haha, thx :)
I never wanted to promote scan stuff yk😄

[swift carbon]

[dusky pollen]

Antigravity servers are down? Ultra user and I can't get a single prompt get past. Only the Flash is working.

[gentle aspen]

Even if I vibe code (maybe a bit too much), something I always do myself and better than any AI is making my artwork lol.

[gentle aspen]

I am free user ( I shouldn't be even speaking), but I cant use any model (I have full creds btw)

antigravity have being climbing some hills these days, iam sure google will fix them asap, but until then...

[undone pendant]

maybe they are updating to include opus 4.7? We can hope? 🥲

[languid elbow]

still on that hopium fr

[solar tusk]

In Antigravity, new chat sessions occasionally disappear. Is anyone else experiencing this?

[steel perch]

Purpose of antigravity?

[solar tusk]

This time, I can't even get the conversation started. Haha

[solar tusk]

I'm writing some backend code. But I've been stuck on it for two days now, and I'm getting frustrated.

I found out about this place through the Google Labs channel.

[steel perch]

Were you really struggling? In architecture or codes

[solar tusk]

Architecture. I was in the middle of designing a sturdier structure when it stopped working.

[steel perch]

Go head with struggle list one by one with possibilities if you have

[solar tusk]

Thanks. It's just frustrating that it hasn't been working on Ultra for several days.

[gentle aspen]

I don't think this is "Hopium", this is more like "Copium" 😂

maybe they are working on some server wide stuff for the next update.
dang... am I also on copium..?

[gentle aspen]

anyone tried the new qwen model?
I feellike it is the closest we came to claude tbh (besides the finetunes)

plus, the gemma4 model looks quite the same.

[hoary marlin]

Yeah man some times it solved but again every day I have to deal with this i have a pro plan but now they introduced a ultra pro plan WTF is this 😑

[elder mason]

VIBE CODERSSS!!!

[lament peak]

hi yall, im haziq from pakistan. I wanted to ask if any of you guys have ideas on what to make in a hackathon where i need to make an innovative webapp. the app is to be fully vibe coded. drop in ur ideas a reply

[gleaming cave]

so you need us to tell you what kind of innovative app to make?

in other words, give me free ideas?

how about i develop those ideas myself, turn them into kotlin app and if it works, we can chat about you purchasing it

i mean, it was just a silly joke

i aint selling my apps

[gleaming cave]

proud to be one of them!

[undone pendant]

Morning, nothing on Opus 4.7 in Anti Gravity yet?

[winter viper]

Sup folks, any info on Claude opus 4.7 in AG?

[elder mason]

Um is it good when compared to 4.6?

ta heck is dis

[signal raft]

it aint coming when all Ultra users get rate limit errors even with 4.6

[elder mason]

is it that bad??
ive been paying for api usage and its been a while since i used antigravity

[signal raft]

Actually, rate limits are fine if you calculate the inference cost, but the issue is the "our servers are experiencing high traffic" which makes the model not respond at all for hours even on ultra.

[winter viper]

Claude limits are crazy

I got -40% just by getting "servers are full" message 3 times

But gemini limits are actually pretty good

been working for 2 hours and only -40%

[elder mason]

never experienced it

[signal raft]

check forums, maybe your region does not have it but it is a real issue

[elder mason]

i paid for bulk via amazon bedrock pro and they said it includes high traffic priority

maybe becaue of that??

[signal raft]

ahh okay, idk about that

[elder mason]

something like this

[dusky pollen]

trillion dollar company cannot fix for a week, could be the free users as they can generate tons of gmail accounts and use them

[undone pendant]

I would care less if other 3rd party platforms didn't have opus 4.7 yet. But when most/all do, and AG is the one which doesn't, it kinda sucks and makes me think google don't care about AG. Guess time to cancel ultra sub

[gleaming cave]

pretty sure it will take a day or two, as it was with 4.6

hows ultra sub anyway, i have to say Pro plan was great when i started using AG, but recently i hit my quota within a day or two

luckily gemini flash has big quota, so i use big models for tough coding and g-flash for UI work with stitch mcp

since most of my projects are Kotlin apps, seem to work pretty well

[gentle aspen]

let put that intoperspective. just because they are a billion or trillion or whatever company, they sill has human or AI devs. they can't just summon a ton of devs to speed things up.

if you marry to one women, you ill get one baby in 8-10 months. just because you marry 2 women, that wont get cut in half, you might get two babys, but not half the time taken.

just because you buy two cars to speed things up it wont save time. infact, it will waste time.

just because you one thousands of devs in google, 10 peoples cant look into what 5 people should do to cut time in half.

hope you understand.

[dusky pollen]

I'm just gonna say that the feedback button on AG had been broken for a month or longer. We can't even send reports. It doesn't take much an effort to make an announcement, make a status page or at least fixing a button. Its just sad too see. I am Ultra user and can't even use it. Keep the metaphors to yourself. There is tons of things could've been made, they simply ignored.

[gentle aspen]

I gewt your point, but they not only has to fix those things. They have to: add new features, work on security, work on those "cracked premium" patches, and many more. they cant just focus on YOUR thing. give them some time, they will fix it.

also what did you expect with an electron app which is just a copy of VScode with agentic AI built into it

[dusky pollen]

My thing? I just want the damn IDE to work 😄

I think thats everyone's thing

[dusky pollen]

I expected it to work, I guess?

[gentle aspen]

maybe they are working on it. so they can push a complete fix to all these bugs.
either ways, did you try re-installing AG? sometimes electron apps cache wrong, and bugs out real bad.

[dusky pollen]

Well I did reinstall and it got worse, I am not so sure. Its just getting worse everyday.

Its just the fact that the AI on AG is heavily rate limited. I am not so sure why is that happening as the Gemini CLI is fine. Its just so specific to AG.

They could like give priority to the "premium" users. Like I said there tons of things they can do, its just taking really long and it will cause the lost of trust and customers. Can only hope they are at least working on it.

[gentle aspen]

it's not just AG, codex is also so buggy these days. maybe they tried coding with the agents from their own app lol.
they mght fix it somehow. i mean theycant have a bug forevr, right?

[dusky pollen]

Yup... I am using Claude, its gets the job done for now. Its just not enough quota even if its paid

[hoary marlin]

WTF IS THIS I HAVE A PRO SUBSCRIPTION BUT WHAT IS THIS HAPPENING ANY ONE HAVE ANY IDEA TWO DAYS AGO AT FIRST I THOUGHT IT WAS FIXED BUT NOW I WAS WRONG ITS NOT FIXED

[gentle aspen]

it literally says that you have used the model too much. but my concern is that text glitch. it combines your OS's native and default text with the font which the app wants, or something similar.
like I said, agentic AI (codex for me) has being buggy too.

they might fix it. or just reinstall the app completely, sometimes that fix these weird bug wiht electron apps for me.

[hoary marlin]

I haven't used this for many days but I don't know why its shows me that the quota is completed after a single query
I haven't used antigravity for some weeks

[gentle aspen]

maybe... google really messed up...

[hoary marlin]

Yes i did some research google messed up you're right ✅️

[gentle aspen]

I mean antigravity isn't a full on agentic AI platform. it is more like an IDE with AI agents.

[gentle aspen]

hey man, I reccomend you delte this message before a moderator bans you :)

[blissful bridge]

Helpp

[stark sapphire]

nothing we can do.

[undone pendant]

Have we had any news that Google are working on the issues/rate limits with AG? Or Opus 4.7 or anything else? It feels like they have deserted AG. Would love to be wrong. Honestly a little bit of news that they are aware and working on it would go a long way. You have a lot of people upgrade to Ultra, because of AG. It's not like people aren't paying for it.

[stark sapphire]

I don't think Google sees it as an issue, but rather to get more money out of you to buy these credits.

[broken prairie]

This ^

Anyone use Agent Manager on AG? I hate that it opens up the last 10-15 Workspaces I have worked on inside of the Agent Manager everytime I re-open even after I close it.

[stark sapphire]

Yea, but i only work in the editor now anyway. so it is less anyong since you wont look at all of your other projects while working,

What i hate quite more is that AG sometimes assumes you work on a project that you're not even in.

[broken prairie]

Fair haha

I like the Manager since I work on 2-3 things at times that are related to one another, and I prefer to see it in one page. Makes it very helpful

Codex has a very good interface with it.

[stark sapphire]

no self promo

@tropic scaffold I'm giving you a hint. Remove the link 😉

[tropic scaffold]

why

[stark sapphire]

i just told you why

[tropic scaffold]

Bro, are you an admin?

[stark sapphire]

No. but according to the rules it's not allowed.

I just try to be nice. i could call a mod you know.

[tropic scaffold]

HAHA call him

[stark sapphire]

ok

<@&1009526435276394496>

I warned you didn't i 🙂

[hallow marsh]

Thanks for the ping!

[undone pendant]

does anyone know when the one dev who manages Anti Gravity is back from holiday?

[rain lava]

Complaining won't speed anything up.

[strange fjord]

lol

its working fine for me tho :/

some guy even asked if Anti Gravity was lagging

but it isnt in the case of mine

[stark sapphire]

ok wow. The rate limit is actually insane for me now as well. I could only use Gemini 3.1 pro for only 2 prompts and i was done. lol wow.
At first i thought people here were crazy.

[rose pelican]

Same here.

[vagrant folio]

similar here 4 prompt

the problem i see the rate limit is by request not token usage