general

[proud bobcat]

Not everyone wants to pay the opus premium

That’s just a fact

[gusty egret]

MoE is starting to catch up to dense though

[proud bobcat]

Api I mean

[echo aurora]

We are currently experimenting with this, meaning it's not going to be fully available to 100% of users.

[proud bobcat]

A lot of companies use the api

[plucky sparrow]

depends on your job opportunities bro

[proud bobcat]

Honestly if you can just use flash via api for cheaper than 200 a month

That’s a steal

Price is almost certainly an issue

[ember iris]

I just saw and was trying to generate but it asked me to loggin. Then vanished... 😁 Good luck. It will be great if it's available for everyone....

[proud bobcat]

This is ragebait

Bro owns 5 Lamborghinis

[pale obsidian]

i want to see artificial analysis benchmarks but they've been really slow lately

[torn mantle]

you guys dont know how crazy this is

for its price

performance/cost ratio

its even outperforming gem3 pro in some benchs

[gusty helm]

did google just smash the market again?

[pale obsidian]

also i hope they make 3.0 flash the default model for free users

[torn mantle]

we will probably say goodbye to sonnet 4.5 and 3.7

like look at those numbers..

wth

i still cant believe it tbh

[gusty egret]

Speed-related benches, right?

[proud bobcat]

This could be peak

[spare mango]

Gemini 3 Pro uses logical language to explain the solution to a problem, which I appreciate. Meanwhile Gemini 3 Thinking uses cozy vocabulary to connect with you better, which I don't appreciate when I'm asking for technical help.

[torn mantle]

yea

[pale obsidian]

you probably get the results you expect with better prompting

[torn mantle]

they just ended oai

[proud bobcat]

Oai is done

They had their chance

It was make or break with 5.2

And they fumbled

Goofy ass company

[spare mango]

The prompt was left as a controlled variable, I used used the exact same prompt to check if there was a difference.

[stray aspen]

@deep adder is openai dead

[torn mantle]

atp xai should quit

lol

how can you compete with this.........

[spare mango]

@pale obsidian

[proud bobcat]

Xai still holds great search summarization

Unpopular opinion but I like grok

[pale obsidian]

^^ fully agree

[proud bobcat]

It may not be a benchmark smasher but’s it’s a fast, smart, and reliable model

[pale obsidian]

nothing beats grok at online research

[stray aspen]

xAI is absolute ass lmao

[zealous sparrow]

not the best thing it made but

https://codepen.io/qdpuzepc-the-styleful/pen/LENKaYq

[proud bobcat]

It’s great for research

I use it a lot when looking for good car deals

[torn mantle]

nice

prompt?

[old ginkgo]

Lol

[modest prism]

Where's opus 4.5?

[proud bobcat]

Opus 4.5 isn’t real

[zealous sparrow]

it's after uh 3 modifications

[proud bobcat]

What do you mean?

[zealous sparrow]

i wanted to make the room walkable

[gusty egret]

gemini deep research

[stray aspen]

opus 4.5 is a freak

[limber panther]

guys gemini 3 flash is better than 3 pro

its official

[proud bobcat]

Yeah but that’s closed off

So

[modest prism]

Grok 4.20 (AGI - beta) comes

[old ginkgo]

Either we've solved cost alltogether or they improved the base model that trained flash.

[proud bobcat]

Grok is fire

[spare mango]

very popular opinion amongst conservatives.

[gusty egret]

grok more often than not is at high demand...

[stray aspen]

grok sucks

[zealous sparrow]

name a thing grok wouldnt write about

[stray aspen]

gpt oss is better than grok lmao

[limber panther]

who tf uses grok

low iq

[robust sluice]

did they lower limit rate or something ?

[torn mantle]

grok 4.2 will be so bad

[gusty egret]

I'd say only Anthropic have a chance at competing with Google, but GPT-5.2 seems rushed if you look at the cutoff date compared to past models and their release date.

[torn mantle]

please dont hype it

[modest prism]

Free solid video generation with generous limits

[stray aspen]

who even still hypes xAI releases

everyone knows its ass

[modest prism]

Wdym? It's AGI (beta) as mentioned in the leaks.

[gusty egret]

Grok overdoes web searching. Simple question, and it uses 50 sources...

[old ginkgo]

Grok 4 was sota at the time of release. Do all of you guys have amnesia?

[stray aspen]

its their third AGI model lmao

at this point

[torn mantle]

yea sure

asi

not agi but asi

[zealous sparrow]

rate this insult gemini 3 flash made at grok

[pale obsidian]

this is its main advantage over other models. its great for research but thats it

[torn mantle]

ok

[gusty egret]

https://arxiv.org/abs/2303.12712

People have been saying "AGI achieved" since legacy GPT-4

[limber panther]

we need faster SOTA models, not slow models

gemini 3 flash is a good promise

[old ginkgo]

Grok releases will always be interesting to me because you never know what happens.

They can achieve sota or suck ass

[pale obsidian]

i think this refers to the open source model they hinted

[limber panther]

most of the time grok models are benchmaxxed and do not perform the same when you actually use them

[gusty egret]

This absolutely.

[zealous sparrow]

gemma 4

[old ginkgo]

Did they benchmaxx on arc agi 2 too?

[torn mantle]

https://x.com/demishassabis/status/2001325072343306345

Demis Hassabis (@demishassabis)

For a fast model, Gemini 3 Flash offers incredible performance, allowing us to provide frontier intelligence to everyone globally. Try the 'fast' mode from the model picker in the @GeminiApp - it’s shockingly speedy AND smart. Best pound-for-pound model out there ⚡️⚡️⚡️

X · Dec 17, 2025, 4:14 PM

[limber panther]

if ur hyped for grok then u eat AI slop

[torn mantle]

🫡

demis

[torn mantle]

no

[torn mantle]

gemmaaaaaaaa

[stray aspen]

no one is hyped for grok

[old ginkgo]

Because grok was the first model that got good at arc agi 2.

You guys genuinely have amnesia

[stray aspen]

what in the benchmax

[proud bobcat]

Not true

[random wraith]

is not letting upload image to lmarena.ai to anyone else?

[proud bobcat]

Grok will absolutely flame Elon

[limber panther]

now we need an opensource sota model "Gemma 4"

[gusty egret]

Grok relies heavily on system prompting to "behave" smartly. The system prompt has historically been very comprehensive in guiding behaviour.

[torn mantle]

now watch every other lab train on gemini 3 flash output

[limber panther]

openai is dying, grok doesnt care, anthropic is on google's side. and Google has won the AI race

[torn mantle]

thanks google

[limber panther]

only google releases bangers

[random wraith]

anyone else having problems trying to upload an image on lmarena.ai?

[torn mantle]

grok is only good at search tool calling

its a dumb model

[stray aspen]

yes

its so damn stupid that it makes me mad

[zealous sparrow]

3 flash roasts itself

[torn mantle]

i still cant wrap my head around this.. how can anyone top this cost+performance?

[gusty egret]

Gemini has been trained to be really objective if you use from the API

[zealous sparrow]

grok wouldnt do that btw

[torn mantle]

no wonder demis said we are far ahead of chinese labs

[gusty egret]

xAI are definitely working at a loss.

[zealous sparrow]

this is one i didnt expect gemini to roast [the platform of many controversies]

also

ts so true

[random wraith]

anyone else having problems trying to upload an image on lmarena.ai?

[gusty egret]

Character.ai was so hype in september and october 2022

[modest prism]

Two reasons:

1. TPUs
2. Google is a search engine. Do you know how search engines work? Their bots crawl websites and collect data and rank them. Google has an archive version of the whole internet.

[torn mantle]

https://x.com/_mohansolo/status/2001325610732523717

Varun Mohan (@_mohansolo)

Gemini 3 Flash is rolling out to all tiers of Antigravity with much higher rate limits. The model is remarkable in that it exhibits Pro-grade reasoning (better in some cases of agentic coding) and is much faster and cheaper.

It has become my daily driver. Hope you enjoy ⚡⚡⚡

X · Dec 17, 2025, 4:16 PM

guys you should use antigravity, it provides way better rate limits

[spare mango]

Profile picture checks out.

[torn mantle]

its basically unlimited lol

[modest prism]

And that's how cursor dies.

[gusty egret]

Inaccurate way to present that. LLMs reach diminishing returns after scaling beyond 1 trillion parameters, so most companies stick to that. The largest model was 1.5 trillion and it's quite old. 1 trillion parameters is not much of the internet.

[torn mantle]

ik how they works yea

[spare mango]

I'll let you figure it out.

[torn mantle]

lol wth is this... how is it better than gemini 3 pro at multilangual

[modest prism]

Ok but when we get Nano banana flash?

[torn mantle]

whats its cutoff date tho?

???

[random wraith]

do u guys still use lmarena.ai to edit images? or is thee anothe free solution?

[gusty egret]

probably early 2025

let me see

[stray aspen]

yes

lmarena is great

[random wraith]

it is not letting me upload image today

image wont load

[stray aspen]

use yupp then

[muted timber]

to me, it doesn't work at all.

[gusty egret]

march 2025

[random wraith]

yupp is another app?

[stray aspen]

no its a website

[random wraith]

incognito mode o nomal?

[torn mantle]

okay

[gusty egret]

Actually maybe it's worse

[stray aspen]

normal gang

[torn mantle]

not bad

not bad at all

[zealous sparrow]

on par with v3.2?

[zealous sparrow]

I think they tested without system instructions

[gusty egret]

It doesn't seem to know beyond January?

[torn mantle]

isnt there an index that includes cost? like performance/cost

[pale obsidian]

they aint done running the benchmarks, its incomplete rn

i think atleast

[stray aspen]

where can i use gemini 3 flash

[modest prism]

Dude it's better than sonnet 4.5. look carefully

[muted timber]

opus? ;))

not even close

eh?

[gusty egret]

When is Anthropic gonna do MoE?

[torn mantle]

wtf

[gusty egret]

Their thinking models don't really try that hard.

[torn mantle]

wtfffffffffffff

[stray aspen]

nice ragebait

[proud bobcat]

Oh come on

Rigged

[stray aspen]

gemini 3 flash is a freak

[gusty egret]

Why is GPT-5.2 there

[pale obsidian]

edited

[torn mantle]

lol no

i took it from official website

[gusty egret]

Rushed ahh benchmarkings

[stray aspen]

its real lol

check yourself

[torn mantle]

https://artificialanalysis.ai/models/gemini-3-flash-reasoning/providers

lol

wtf

[muted timber]

@echo aurora !!!

IT WORK

I DON TKNOW WHAT HAPP

[muted timber]

i change the AI to gemini

and it work

[zealous sparrow]

rigged af it easily beats xhigh

[fleet lintel]

https://www.reddit.com/r/singularity/s/wA7e0eeBGB

Why pay for Gemini 3 pro or for any model ?

[torn mantle]

bruh xhigh is so bad

we all know gpt 5.2 is the worst model ever

[pale obsidian]

damm this is insane

[zealous sparrow]

the OCR purely

[torn mantle]

thats what im saying 😭

[proud bobcat]

It’s so peaking peak

[stray aspen]

not even grok reached these levels of disappointment

[echo aurora]

I'm assuming it's rate limit then

[torn mantle]

ITS PEAAAAAAK

[proud bobcat]

Lmao nuh uh

Ts is so fire

[gusty egret]

gpt-5.2 is a new architecture compared to 5.1. The cutoff of 5.2 is September 2025 and 5.1 is October 2024

Rushed

GPT-5 is also October 2024 for reference

[queen veldt]

Umm whatt

Nah that's fake

[stray aspen]

its not lol

[torn mantle]

https://artificialanalysis.ai/models/gemini-3-flash-reasoning

[gusty egret]

It was released 10 months after its training cutoff vs GPT-5.2 which was only 2.5

[torn mantle]

lol

true

xdd

[fleet lintel]

This is beating all expectations. How is this possible??

[pale obsidian]

wait thats flash 3 reasoning

non reasoning https://artificialanalysis.ai/models/gemini-3-flash

[leaden laurel]

i live in belarus

i need usa email

[fleet lintel]

Fast cheap And great?? I thought you can only get 2 out of 3

[torn mantle]

yea thats crazy ngl

[leaden laurel]

(please someone give me usa gmail)\

[pale obsidian]

hm

[torn mantle]

cutoff

Jan 1, 2025

[leaden laurel]

or like any other country

[torn mantle]

yea

[fleet lintel]

Same as pro

[leaden laurel]

which is not banned

[zealous sparrow]

can you go ask sonnet 4.5 Make a documentary with TTS about the creation of the universe in html, I want it all animated, and show facts on screen, it should be high quality. I need to compare

[obsidian shell]

@echo aurora nano pro has been unresponsive for hours

[limber panther]

i got 3 months google pro plan free trial

[torn mantle]

openai code red again??

[obsidian shell]

oh and codex is gone

lol

[limber panther]

Google has alot of shares in Anthropic, they're not against eachother

[gusty egret]

openai vs google i wonder who'll win

[limber panther]

Google is likely to buy anthropic

[proud bobcat]

Oai folding

[stray aspen]

google lmao

[fleet lintel]

No way.

[zealous sparrow]

https://codepen.io/qdpuzepc-the-styleful/pen/NPNZJag

[limber panther]

vertex ai studio has all anthropic models

[obsidian shell]

find something that they dont fold on

[limber panther]

antigravity too

how so

[fleet lintel]

Openai is lost cause. It's anthropic vs Google

[zealous sparrow]

wdym no money

[golden ocean]

is the poor community celebrating gemini 3 flash

[zealous sparrow]

google is the richest company bro

[limber panther]

Google has already won the AI race

its over for openai fanboys

[proud bobcat]

What

[limber panther]

openai is dead

rip

[fleet lintel]

He is just rage baiting at this point. Ignore the baiters

[proud bobcat]

This is ragebait

[zealous sparrow]

then who is

[stray aspen]

openAI is absolutely cooked lamo

[patent aspen]

No chance that would make it through regulators

[proud bobcat]

I can’t tell

It’s too good

[modest prism]

Gemini 3 flash passed the AGI vision test unlike 3 pro.

[golden ocean]

🗿

[jade egret]

gemini 3 flash good?

[stray aspen]

yes

its a freak

[limber panther]

its not good its awesome

its great

[modest prism]

Better than 3.0 pro

[jade egret]

:0

[zealous sparrow]

crazy sh

[limber panther]

only 3$ for output

crazy cheap

[jade egret]

what it better than pro as in 'smartness'?

[pale obsidian]

simple explanation as of why google is winning this

[limber panther]

yes go see the benchmarks

and test it yourself

[fleet lintel]

Too good to be true. I need more evidence. Current numbers for the cost and speed is bat sh.. crazy

[stray aspen]

google leads the way

[torn mantle]

bye bye openai

lol

[obsidian shell]

we had oai

[limber panther]

openai is dying

what a good year

i love 2026 now

[stray aspen]

are there any news on new veo models

[obsidian shell]

why does google have to win every time ?

like

cmon

[zealous sparrow]

no, silence for now

[golden ocean]

gpt 6

[modest prism]

How is this possible?

[torn mantle]

its good for competition

[torn mantle]

pffft

[jade egret]

different between pro and thinking?

[stray aspen]

gemini 3 flash has such a great vision

[obsidian shell]

Ig they put more hardware into flash for the launch

[limber panther]

fast and thinking is 3 flash

pro is 3 pro

[jade egret]

oh

[keen beacon]

damn

[limber panther]

thinking is 3 flash high

[stray aspen]

i hope flash doesnt get more stupid in the coming weeks

[limber panther]

tbh im more hyped for gemini 3 flash

than gpt 6

[pale obsidian]

what does this mean

[modest prism]

Now imagine if Nano banana flash would be better than NB pro

[gusty egret]

only in certain benches

[limber panther]

with cheaper price that would be jackpot

[weary galleon]

Give me this image, bro

[left lodge]

@echo aurora
My suggestions,
Add a option to remove the system prompt added by you guys in code modality, because it causes the model to ignore instructions and just make a website out of our requests.
The environment is good implemented but it is very sophisticated, why only frontends of websites?

[gusty egret]

did google nerf 3-pro to favour 3-flash?

[limber panther]

not really, there is no competition if google releases their best gemini 3 pro checkpoint

we can already see that openai failed to compete

and grok is dumb as hell

[stray aspen]

bro whats happening with nano banena pro

its not working on lmarena

and its unavaiable on yupp

[limber panther]

i got pro plan

its good on gemini app

working fine

[stray aspen]

alright

[split kayak]

3flash

[echo aurora]

Thank you for sharing! Would encoruage you to use #1372230675914031105 to either add your thoughts onto a post that already exsists and is related, or create a new post if you're not seeing something related.

[modest prism]

Sam Altman should be fired

[jade egret]

wait... so gemini 3 flash is better than pro?

[torn mantle]

way better than 2.5 pro

but just beneath 3 pro a bit

[jade egret]

oh

[torn mantle]

but then you add in speed & cost to the formula

and you find 3 flash is way better

[stray aspen]

flash cooked

[obsidian shell]

yes

[wicked sage]

me when an ai fails a test where the hand is an abomination: "WOW THIS AI IS SO BAD!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!"

[neat apex]

I hope very much that Gemini 3 flash beats that baseless allergations that Gemini 3 is overscaled

[wicked sage]

like i swear to GOD bro.

[zealous sparrow]

I am waiting for the simplebench result

I can'r wait

t

[limber panther]

sam altman got fired but he came back

[neat apex]

yooooooooo

[zealous sparrow]

I hope it places atleast #2 or #3 on simplebench

[limber panther]

now we need Genie 4

and its all over

[stray aspen]

that would be the absolute final blow

[neat apex]

the gape of Gemini 2.5 pro 03 and Flash 2.5 is 10%

[near root]

Gemini 3 Pro vs Gemini 3 Flash

[neat apex]

soo, i guess it will score 66%

[zealous sparrow]

I hope so 😄

[neat apex]

it shows the Flash 2.5 09 results?

[cloud zinc]

gpt 5.2 still on top

[neat apex]

and its because 2.5 flash 09 was a HUGE update

[stray aspen]

5.2 sucks

[neat apex]

5.2 is mid

[cloud zinc]

its the top rn

[neat apex]

carried by the xtra high and some luck

[stray aspen]

you need extremely high compute just be behind gemini 3 pro and flash is behind it with less compute lmao

[gusty egret]

Probably November 2026

[modest prism]

Gpt 5.2 Extra high thinks for 10 minutes at least while flash gives you answer under one min

[left lodge]

Okie

[cloud zinc]

i dont need quick and wrong answer

[zealous sparrow]

benchmaxxed garbage still on top

[limber panther]

come on dev mode server ur gonna get cooked

i cant roast u on this server

[left lodge]

@echo aurora Well i have a doubt,
We have now gemini-3-flash &
gemini-3-flash (thinking-minimal)

The first one is thinking-high?

[stray aspen]

AI benchmark drama is crazy

lmao

[neat apex]

it will be so isane, that it will make us rewind time

[modest prism]

Ok what about this?
Gemini 3 flash is available everywhere with generous free limits while gpt 5.2 extra high is not available even for plus (20$) plan.

[gusty egret]

Wait

Lol

[brittle tiger]

Flash 3 is going to dominate coding tools

[gusty egret]

I'm so stuck in this year

[neat apex]

bro lives in 2024

[cloud zinc]

free means ur data is being used

[gusty egret]

This year has lasted so long

17/12/2025

I can't believe we're in the 17th month of 2025

[modest prism]

AI studio uses data, Gemini app does not.

[cloud zinc]

it does

[stray aspen]

it doesnt lol

[cloud zinc]

https://www.malwarebytes.com/blog/news/2025/07/no-thanks-google-lets-its-gemini-ai-access-your-apps-including-messages#:~:text=Gemini uses info from your,takes it one step further.

[left lodge]

Google can use anything, their privacy policy and t&c is very broad and applies to all Google services.

[zealous sparrow]

im letting gemini 3 flash cook up a social credi test in html, you know that meme right

[zealous sparrow]

wont this be a toggle

like

if they dont make it a toggle

its a lawsuit for privacy invasion

[sweet creek]

Just us gemini for workspace

[cloud zinc]

no toggle

[stray aspen]

flash is so good

[weary galleon]

Thanks.

[zealous sparrow]

you fr?

[zealous sparrow]

socky are you deadass showing outdated stuff

[devout vault]

is gemini 3 flash free api

[weary galleon]

I'm Anthropic soldier, but today Flash impressed me.

[weary galleon]

Agreed.

[stray aspen]

bruh

we just got ragebaited

[burnt sinew]

What's wrong with data sharing for non critical data

[zealous sparrow]

listen i like AI companies, but this is absurd

[cloud zinc]

good, they should focus on AI

[zealous sparrow]

40% less GPU production due to AI

[burnt sinew]

Why wouldnt pro impress you

[cloud zinc]

gaming can take a backseat

[zealous sparrow]

you know damn well

a lot of their consumers

are gamers

the company will face a loss

[weary galleon]

Gaming is less profitable than B200s

[cloud zinc]

no

[burnt sinew]

At worst they'll use your data anyway even if its not "allowed" and just pay the fines worth 0.1% of their profit

[cloud zinc]

ai will create more games

[zealous sparrow]

well we'll see how the stock goes

[cloud zinc]

up

[weary galleon]

It also. But today Flash was released, not Pro.

[burnt sinew]

What's wrong with that, amd still exists

[fleet lintel]

Thinking about flash performance.. basically, it implies that google probably already has Gemini 3 pro model with current flash post training that is much better.

[burnt sinew]

Yeah im surprised its so close to pro tbh

[zealous sparrow]

barely anyone is an AMD user

[burnt sinew]

Well they will be then

[unborn ocean]

Yeah thinking the same thing, I also heard the released flash is based off of the next Gemini 3 iteration

Where we don’t have the pro yet

[weary galleon]

AMD is great!

[whole sundial]

amd sucks for local ai, especially image/video gen
source: i'm an amd user

[fleet lintel]

Yes, I have some internal sources And that is indeed the case.

[whole sundial]

almost no support, but support has been slowly getting better

[astral elk]

Good night community

[stray aspen]

morning

[weary galleon]

They are great.

[torn mantle]

is gemin i3 flash vision broken?

[burnt sinew]

Well they might be the only option if nvidia stops making consumer gpus, but I dont think they would

[weary galleon]

Pure OAIs "flagship"

[whole sundial]

intel... exists i guess

[weary galleon]

For gaming also.

[zealous sparrow]

and that will be the case

I liked AI till it caused companies to raisep rices

[burnt sinew]

Which is right

[weary galleon]

47

[zealous sparrow]

RAM like 70% up or sh- and now GPUs

[torn mantle]

5.2 is so baaaaaaad 😭

[zealous sparrow]

Next? CPUs

[stray aspen]

gemini

5.2 is just plain stupid

[weary galleon]

Yes, it is.

[burnt sinew]

Is gpt 5.2 just benchmaxxed to the extreme?

[whole sundial]

they have no chance in ai anyways so they can make some more gpus once they figure out how to not use tsmc, they still need to worry about ram in any event though

[stray aspen]

probably yes lmao

[crude lagoon]

No way 😭 gemini flash model

[burnt sinew]

Gemini 3 flash is better than 3 pro at math?

[crude lagoon]

nah

[whole sundial]

b570/b580 got a lot of sales at launch, but then rtx 5000 made people forget about them, if you can find one they are a great choice, actually better at ai than amd i think and intel has first-party tools with support, idk about third-party ones though

[whole sundial]

250-260 rn, that's not bad for 12gb vram and a capable gaming card!

are you kidding me, a new rx 7600 costs more than a b580? i have an rx 7600 and i can tell you that a b580 is probably better

[weary galleon]

Flash is excellent in coding, much better than GPT 5.2!

[stray aspen]

craig whats your specs

[echo aurora]

Yes, sorry for the delay in getting back to you.

[torn mantle]

yes

[torn mantle]

what you are testing is gpt5.2 without thinking

[weary galleon]

When?

[torn mantle]

to make it fair you should try gemini 3 flash vs gpt 5.2 high ( although high is way expensive )

[sour spear]

So, am I right in assuming that Gemini 3 Flash, the free model available for everyone, is better than anything available in ChatGPT?

[stray aspen]

yes

[torn mantle]

actually if you think about it, its fair from cost point of view

[stray aspen]

gpt 5.2 sucks

[torn mantle]

but one is a reasoning model one is not

they lost so bad

[weary galleon]

Yes

[stray aspen]

feels like they just routing the non thinking model to gpt 2 lmao

[gusty egret]

You have no idea what it was like to be around gpt 2 when it released

It was the coolest thing of all time

Nowadays we get objectively amazing models but since they're comparatively ass to other models, we just dunk on them

[torn mantle]

i remember using it for the first time

i think there was a model before davincii that i was using a lot

and waiting for the streamline response

and gpt neo

was crazy

it really felt like magic

[stray aspen]

i started using LLMs around early 2023

[torn mantle]

at that time we still didnt understand how they worked exactly

[stray aspen]

for high school

now i use it for college

and yes it was crazy

however not so good for physics class and math lmao

i would spend al lnight trying to get the AI to give me problem solution that matched the books answer

[gusty egret]

I remember gpt j

It was super cool

[whole sundial]

lol

[gusty egret]

I've been in the community since early 2019

[whole sundial]

of course it wouldn't get this right, but i had to try

[gusty egret]

The shift from text prediction playgrounds to chatbots kinda annoyed me

Giving models their own personality made it harder to generate writing similar to your own

[stray aspen]

what in the hallucination

[gusty egret]

Of course that's after the dataset cutoff lol

[weary galleon]

Fair?

[zealous sparrow]

this is one thing i love gemini 3 flash does it automatically looks for sfx and finds them

i didnt link those

[torn mantle]

yea

now you can make comparisons

[weary galleon]

But it didn't help GPT, it still loses to Flash.

[torn mantle]

because its bad 😭

its so expensive too

wth

[gusty egret]

Gpt o1 flashback

[stray aspen]

wasnt it called bard back then

[whole sundial]

keep in mind gpt-2 was 2019

[gusty egret]

In 2019?

[obsidian cargo]

hey @echo aurora they made the lmarena prompt filter too strict it's forbidding perfectly innocuous text prompts now…

[gusty egret]

Gpt 2 XL was 2019

Actually wait

[sour spear]

Seems the rollout is complete. I'm usually the last person on Earth to get new models in the official app, and I just saw it pop up. 😁

[gusty egret]

I still have the email from the newsletter OpenAI sent out upon releasing gpt 2

[echo aurora]

Yeah I'm not aware of a change that was made here. However, all today have been hearing similar reports so I'm assuming something did change. Can you provide some of these prompts you're seeing flagged in #1447983134426660894

[sour spear]

No idea how that happened. 5.2 High has its moments, but then it's also giving me ridiculously bad results sometimes, I can't bring myself to even try it anymore for coding.

[proven grove]

Is there any way I can fix it if I get stuck with infinite loading?

[gusty egret]

OpenAI is just trying too hard. When gpt 5 thinking released it kept talking in lowercase during programming questions, which was so odd

[weary galleon]

Flash is good even with minimal thinking.

[burnt sinew]

Thats what leaderboard says

[rugged abyss]

Love how Gemini 3 Flash is beating Gpt 5.2 by a huge margin

[weary galleon]

I too

[gusty egret]

Google, multi zillion dollar corporation, Vs OpenAI, a tiny shed with 2 dudes in it

[rugged abyss]

Wouldnt call it that after oracle and microsoft is investing huge sums into openai but alright

[gusty egret]

3 dudes

[rugged abyss]

Where are you getting 3 dudes from?

[burnt sinew]

What is style control? Like system prompt?

[gusty egret]

Cuz it's slightly more than two dudes

[rugged abyss]

Google Deepmind: 6-7K Employees
OpenAI; 3,5-5K Employees.

you are referring to the starter years of openai

[gusty egret]

Idk why you're pressed about my trolling you

[atomic lagoon]

burh

It had a copy image so I tested

[rugged abyss]

not pressed, just correcting you

[gusty egret]

I mean I think it's pretty obvious that OpenAI isn't a tiny shed with two dudes in it

[rugged abyss]

You wouldnt believe how dumb some people are when talking about ai

[gusty egret]

I thought I demonstrated my competence earlier lol

[rugged abyss]

I just joined in, not sure what you are referencing rn

[gusty egret]

My earlier contributions lol

Nvm it's fair

I wonder when anthropic will do a MoE

[rugged abyss]

Wondering when Ai will start AoT

[zealous sparrow]

https://codepen.io/qdpuzepc-the-styleful/pen/xbVoeLV

you cant restart because codepen sucks

if you put this on a localhost it will work

[gusty egret]

🙏🏻

🥀

[gusty egret]

There's absolutely no way in hell Gemini made this

Specifically because of the cotton question 💀

[zealous sparrow]

ehm

yes it did

[gusty egret]

Why'd it choose to spell it as Uighur? Feels like the LLM would choose the more common one, that is Uyghur

[zealous sparrow]

i don know

[torn mantle]

LMAO

[rugged abyss]

No way you didnt give it the answers and questions

What was the prompt?

[torn mantle]

so funny

[zealous sparrow]

Make a social credit test in html, IF YOU do bad execution, if you do good you chinese citzen, also add sound effects and the red sun in the sky music, take them from myinstants
+
My short system instruction to make it not lazy

literally just this

this prompt sucks ass

but it works

[gusty egret]

That's the most neurodivergently written sentence ever lol

[rugged abyss]

omg youre right lmafao

[gusty egret]

Super idol and jinqian kan qi not mentioned, 0/10

[zealous sparrow]

if you mentioned chinese memes then prob

[gusty egret]

John Cena was mentioned though

Obviously it understands

And also the extremely exaggerated values like +5000 social credits XD

[rugged abyss]

-99 Million

[gusty egret]

I actually didn't choose the wrong answers

[rugged abyss]

https://tenor.com/view/social-credit-gif-23976170

[gusty egret]

I'ma go do that

[keen beacon]

👋

[gusty egret]

Minus one billion

That's crazy

[weary galleon]

Even Flash with minimal thinking is much better than GPT 5.2 with maximum thinking 🤣

Thinking won't help if your are stupid.

69 is the exact right answer.

[half mist]

Um?

[modest prism]

Gemini has a god-tier vision, because of google lens.

[zealous sparrow]

fast - doesnt think
thinking - thinks

both are gem 3 flash

[half mist]

Yes, finally released

[half mist]

Is there a limit on fast?

[modest prism]

Gpt image 1.5 is available for free with generate limits, NB pro is not. Waiting for NB flash

[burnt sinew]

I dont think so

[zealous sparrow]

not sure

im using on AIstudio

i haven't hit ratelimit yet but im goin to see it soon ig

[half mist]

What about thinking? Since it’s flash with thinking, does it have a limit?

[zealous sparrow]

yeah to make gemini3 flash good at coding you need an anti-lazy prompt

how so

[stray aspen]

slide the prompt

[zealous sparrow]

i dont know

the owner of it

said it would be public

soon

[cloud zinc]

there is no prompt

[zealous sparrow]

also 👀

they are going to

fix up pro

[cloud zinc]

its pizza planet

[zealous sparrow]

waiit

i know now

google put out a 9f checkpoint which was likely a 3 pro one

it's prob gone now

or mayb it was flash idk

[half mist]

Also, the thinking model doesn’t think, or it doesn’t show the thinking

[cloud zinc]

https://ai.google.dev/gemini-api/docs/prompting-strategies#agentic-si-template

[torn mantle]

someone already leaked claude code whole prompts lib they use to make their CLI better 😭

https://github.com/Piebald-AI/claude-code-system-prompts/tree/main

it has everything

planning ... execution .. optimization ...

https://github.com/Piebald-AI/claude-code-system-prompts/blob/main/system-prompts/agent-prompt-plan-mode-enhanced.md

[zealous sparrow]

i realize why 3 flash was so good on codearena

the codearena system prompt made it

good

[half mist]

Also, I see no difference between fast and thinking. Fast does it instantly, and thinking does it instantly as well, it doesn’t show the thinking

[gusty egret]

I only count 68

[weary galleon]

I 69

[gusty egret]

I counted twice

[zealous sparrow]

@echo aurora how is the codearena prompt made, it made gemini 3 flash good in testing

[weary galleon]

Both times with a mistake

[gusty egret]

That's a bit narcissistic

[torn mantle]

yea lol

[gusty egret]

What am I missing

[weary galleon]

This one.

Just a small corner of a whole tomato.

[gusty egret]

I suppose not having the whole image does make that one difficult to spot

[weary galleon]

I posted this image here a few days ago.

[gusty egret]

I was not here a few days ago

[gusty egret]

Also I think GPT might be counting the ones with the top stem visible

[weary galleon]

GPT counted only 28 from 69.

[gusty egret]

Plus 3

[weary galleon]

Which 3?

[gusty egret]

Who knows

The inner workings of 5.2 are a mystery to us all

[weary galleon]

Because it's an extreamly bad model.

[gusty egret]

Maybe it's these three

They're the most tomato-shaped where you can't see the stems

I can only theorise because it probably gives a different answer every single time

[weary galleon]

Yes, because it's not an AI, it's just a piece of crap.

[gusty egret]

It's rushed

[weary galleon]

Rushed, but fair.

[gusty egret]

The cutoff date of GPT-5.2 is 1 September 2025 and GPT-5.1 & GPT-5 is 1 October 2024.

The release month of GPT-5 was August 2025 and the release month of GPT-5.2 was December 2025.

GPT-5 took 10 months to be published, whereas GPT-5.2 took 2.5 months.

[lapis sparrow]

🙌 Hello ...

[half mist]

The limit for thinking is the same as pro. So if the rate limit of thinking is the same as pro, what’s the point of thinking?

[gaunt roost]

https://www.youtube.com/watch?v=SX5JIiWOOH8&t=32s

[restive scarab]

Nano banana pro is literally not generating and working

[astral blaze]

I am being SILENCED

It truly is over

[echo sinew]

You're not being silenced. Your post seems off topic. Completely unrelated to the current chat. If that was a reply to someone in particular, you can ping them or reply to their original message.

[astral blaze]

I literally have negative IQ

[gusty egret]

Nano banana pro appears to be down

[neon idol]

@zealous sparrow sooo impression on gemini 3 flash?

[gusty egret]

Oh nvm someone noticed

3-pro just died on me after thinking for 3 minutes straight

[solar hollow]

this happens to me all the time with all the models in direct chat

[zealous sparrow]

Its not bad, if you have a good system prompt great at coding

[solar hollow]

it seems though, that the difficulty of the prompt matters

[burnt sinew]

Idk

[neon idol]

Nice

[stray aspen]

does anyone have a good system prompt for gemini 3

[half mist]

Oof

Happy it has Google Search grounding though since that makes it more accurate

[torn mantle]

this benchmark was proven to be unreliable over and over and over

read what it measures

[zealous sparrow]

I haven't seen too many hallucinations rn...

The model only gave me good answers

[torn mantle]

we dont know what a partial answers is... and is this LLM judged?

[cloud zinc]

wow gemini 3 is so bad

[zealous sparrow]

wow this scale sucks

the model barely gave me that much wrong answers

[cloud zinc]

so gemini 3 flash hallucinates a lot

[torn mantle]

no

[torn mantle]

bad bench

[cloud zinc]

no it is good

[zealous sparrow]

the bench is ass

bro

give me a question that it would hallucinate

[cloud zinc]

it is a trustful benchmark

[stray aspen]

the bench is only good when its in your favor

[cloud zinc]

my favor?

i am looking at an objective benchmark

check here

[zealous sparrow]

if it scores high on simplebench the hallucination benchmark rated it wrong

[half mist]

I asked it for the next episodes for a show (but the season of said show has ended with no episodes announced), and it gave me random upcoming episodes even some from December, but the show isn’t releasing a new season until 2026

[torn mantle]

no its not

stop

[zealous sparrow]

grounding or not

[half mist]

With grounding believe it or not

[zealous sparrow]

yeah okay thats like

give me prompt

ill test

[cloud zinc]

so it is giving fake information

thats not good

[half mist]

What are the next episodes of [show name]?

[cloud zinc]

how can people trust an ai model, if it gives fake info

[zealous sparrow]

I dont see it

[half mist]

May sometimes give the correct answer, but sometimes it hallucinates

[cloud zinc]

alot of time it hallucinates

[zealous sparrow]

give me your questions

[stray aspen]

what did you ask for

[zealous sparrow]

i didnt get a lot of hallucinations yet

[stray aspen]

what show

no

he said he got hallucinations with grounding on

[zealous sparrow]

well then its obvious

it wont have any info without it

[half mist]

The latest episodes of shows. If it grounds it, it’s a higher chance of not hallucinating

[zealous sparrow]

therefore its unfair for the model

as it has to have every show in its training

[half mist]

Give shows that aren’t too old or it will of course get that correct

[zealous sparrow]

hmm

let me find a new one

[stray aspen]

its not

its a great model

[zealous sparrow]

i need to see it

else im not taking it

i want to bet one thing

the testing for hallucination was done without grounding, which nerfed the model to rely on training data. Making it the reason it scored so high on hallucinating

[stray aspen]

do you have photos

or are you ragebaiting

[half mist]

It’s very hit or miss if it hallucinates

[half mist]

For some reason, the chat is gone? I never deleted it though

[zealous sparrow]

the model experienced an error

and hallucinated

and it prob got detected so it removed your chat

[stray aspen]

i aint buying this gang

[zealous sparrow]

what show did you try it with

[cloud zinc]

thats not good

[half mist]

It was a cartoon show that had its season ended with no new episodes announced. I tried it after the hallucination, and it seems to work just fine, but the first attempt it hallucinated

[zealous sparrow]

i need the name bro

[half mist]

It was big city greens

[zealous sparrow]

this specific show gave me wrong answers

consider me stupid

[golden ocean]

Considered.

[half mist]

I think we just gotta use it normally, then wait until we find a hallucinated response

[zealous sparrow]

gl

[half mist]

Does this count? It says the Plankton Movie is expected march 2025, but that date has already been surpassed

[zealous sparrow]

furthermore the movie is already out

your gemini is cursed

[half mist]

My Gemini has no instructions or anything,

[zealous sparrow]

wait give me the same prompt

[cloud zinc]

thats so weird

[half mist]

What is the latest SpongeBob movie

[zealous sparrow]

your gemini is fkn cursed man

[half mist]

I am using just Gemini, but you’re using AI Studio

[zealous sparrow]

API is different ig

[queen veldt]

3 flash AGI

[zealous sparrow]

This is without grounding, and this is not a hallucination. As the models training data only dates back to Jan 2025.

So no bench can argue this is a hallucination

[queen veldt]

3 flash is more accurate than the pro version???

[stray aspen]

yes

he has bette rvision

[cloud zinc]

flash is focused more on vision and code

[half mist]

Gemini 3 Flash is actually really good at coding

[cloud zinc]

it is bad at other things

like hallucinating a lot

[zealous sparrow]

like i said i cant confirm this

it only gave me good answers on grounding

API and website prob differs

wwtv says hes using website

[cloud zinc]

just because u cant confirm it doesnt mean it doesnt happen

[zealous sparrow]

sure not but like

it aint happening for me on AIStudio

and it happens for him on website

[cloud zinc]

most people not using aistudio

[zealous sparrow]

i want to bet one thing

the hallucination bench benched website

[cloud zinc]

is good

[zealous sparrow]

But also on website its not hallucinating..

I swear the bench got a weird ver of the model or smth

[sterile tartan]

Make one for Gemini

[weary galleon]

I like Gemini, it's very good model. But I trust in Anthropic.

[sterile tartan]

Understandable

Sonnet 4.7 will be 🔥 Tho

[weary galleon]

I'm sure.

[quartz light]

imo sonnet 4 is worse than 3.7

i hope sonnet 5 wont be worse than 4.7 lol

[zealous sparrow]

did you see the hallucination benchmark say absolute bs on gemini 3 flash

they said 91% hallucination [wrong answers]

[quartz light]

what

[zealous sparrow]

yet there's barely any

[quartz light]

oh

who said

[zealous sparrow]

are we like deadass

[quartz light]

this bench is absolute btw what lol

4.5 on lowest?

haiku*

[golden ocean]

deadass

[zealous sparrow]

I hit the 3 flash ratelimit on AIstudio

didnt count but

strong say its about 100

More actually

let's not go as high

[drowsy crater]

Sup

[quartz light]

this bench shows otherwise

see

its a bench so its tru

[quartz light]

@zealous sparrow this bench gud?

[weary galleon]

Guys ignore trolls, they wanna make violent argument. Ignore them and they would start to cry because of lack of attention.

[pale obsidian]

https://x.com/ArtificialAnlys/status/2001388724987527353 dedicated to the xai haters from earlier today

Artificial Analysis (@ArtificialAnlys)

xAI’s new Grok Voice Agent is the new leading Speech to Speech model, surpassing Gemini 2.5 Flash Native Audio and GPT Realtime in our Big Bench Audio benchmark

The new model achieves a score of 92.3% on Big Bench Audio, just ahead of the previous leader, Google’s Gemini 2.5

X · Dec 17, 2025, 8:27 PM

[quartz light]

who tf cares

[pale obsidian]

i dont care either but to say xAI is worthless and ass is just not true

[quartz light]

i mean yeah some people but "xai haters" are ones hating on text models

[quartz light]

nah point still stands, they focused on something niche and not as important

like dude even nova 2.0 is 3rd place on that bench

the fact that they only beat gemini by .1% 🤣

thats embarassing

not a "comeback" or whatever

[pale obsidian]

the fact artificial analysis published a report about a voice model instead of gemini 3 flash is also funny

[vivid coral]

speaking of xAI

[stray aspen]

the only thing they are good at is saying they will have AGI soon but they end up giving us ass models

[golden ocean]

lmaooo

[sullen quest]

AHAHAHAHAhA

[vivid coral]

more xAI fun

[half mist]

Did notice that Gemini 3 Flash asks follow up questions at the end of each response like ChatGPT. Before with 2.5 Flash, it didn’t do this natively unless you told it to do that

[vivid coral]

I thought Pro did that too, maybe not, I'll have to check

[queen veldt]

I just realised the depression I'd have if there was no lmarena

[stray aspen]

why

[fiery gull]

I'm so happy

[vivid coral]

I'm wrong, no follow-up on the Pro

[fiery gull]

now I have a gemini 3.0 pro but without minutes of loading to attack my adhd

[vivid coral]

https://tenor.com/view/nick-girl-same-eye-roll-sassy-gif-17493564

[stray aspen]

does anyone have a nice prompt to remove coding lazyness in gemini 3

[sour spear]

Interesting little comparison. ChatGPT failed abysmally, the answer is utter bs. Gemini 3 Flash Thinking got it perfectly right. And Gemini 3 Flash Fast actually did some reasoning "on the fly" (the "Wait, let's look closer" part in italics), correcting itself, and also reaching the correct conclusion.

[queen veldt]

All those images i created with nb flash and all those coding with gpt 5 high

None of that would be possible without lmarena

I'd literally scroll tiktok or sum

[stray aspen]

i mean we still have yupp

lmao

[vivid coral]

yupp isn't free

[fiery gull]

I need this lol, it have so many lazy to write long text ;-;, my tecnique is fracionate in various parts

[queen veldt]

Yupp can't be compared to lmarena

We don't need to submit feedback

To get points so we can use the model

[torn mantle]

more trolls

[torn mantle]

.

[vivid coral]

yupp has limits, some models cost ya, nothing near LMArena. If you ask Gemini itself for alts, it will even tell you, there is nothing really like it. It will try to direct you to openrouter, yupp, etc, And when you press that there are costs and limits, they tell you only LMAreana is this generous

[modest prism]

I wonder how lmarena afford all these models

[warm zodiac]

Hey where did you get this info? Do you know when in Q1

[modest prism]

100M isn't really that much

[vivid coral]

there might be more investers/seeders that aren't published or reported in the media, idk

could be the AI companies themselves, who knows

[echo aurora]

For those that haven't yet, I'd encourage people to give this a read: https://news.lmarena.ai/new-lmarena/

[vivid coral]

I guess some of is the AI companies themselves, it's very clear there's heavy heavy backing, it's not going anywhere

[cloud zinc]

read this: https://news.lmarena.ai/ai-evaluations/

[hasty crag]

hello

[cloud zinc]

when is it coming

sure

[half mist]

https://youtu.be/bY_RarpUdUw Spinners has you pick Gemini 3 flash make two versions for the next version of spinners, and you pick your favorite, then the next one is based upon the one you chosen. Huh, sounds familiar doesn’t it?

[cloud zinc]

everything is soon

[torn mantle]

gem3 flash eats up usage so fast on antigravity

which is surprising

15% left

nauuuuuu

its token efficient so whyyyyyyy

[native yarrow]

@zealous sparrow can u share ur system prompt

[reef hollow]

Does everyone have the video feature on the site? Because I enter another device and the feature does not appear

[echo aurora]

This is an experiment currently. Meaning not all users are going to have access to it.

If/when we fully roll this out to all users we'd be sure to let the community know!

[torn mantle]

https://x.com/testingcatalog/status/2001411942150242345

TestingCatalog News 🗞 (@testingcatalog)

BREAKING 🚨: OpenAI is preparing to upgrade Codex, likely with GPT-5.2-Codex-Max. One last leap this year?

Caribou 👀

X · Dec 17, 2025, 9:59 PM

lol

[fiery gull]

Obviously it would have the codex 5.2 max, they want money

[neat apex]

Its from the reasoning update

Since its smarter, it works better with the limited reasoning

[cloud zinc]

wow

[sullen quest]

craig, you can admit that 5.2 is wasn't sent down from the heavens to bless the ai world forever ok?

[stray aspen]

It's ass

[cloud zinc]

why are u glazing gpt 5.2 so much

[stray aspen]

I tested it myself benchmarks won't change my opinion

[sullen quest]

?

[torn mantle]

new

looks good

[sterile tartan]

Rate Limits?

[torn mantle]

wym

[sterile tartan]

Video Generation Rate Limits

[cloud zinc]

it doesnt work. u need account

[torn mantle]

i dont know

[sterile tartan]

Alright Paw

[torn mantle]

yea

:omg:

@cloud zinc

[acoustic crater]

It looks like nano banana can be used

[torn mantle]

the auto modality works tho

[torn mantle]

yes

[echo aurora]

Yes, but would note we're still seeing higher than usual error rates with it.

[torn mantle]

this is so stupid tbh

https://x.com/warpdotdev/status/2001335906943599004

Warp (@warpdotdev)

Warp now uses Gemini 3 Flash for generated code diffs!

We've seen a clear bump in quality from Gemini 2.5 Flash.

Warp suggests code changes whenever you hit a compilation error in the terminal. We'll be rolling out Gemini 3 Flash for diffs, so let us know what you think.

X · Dec 17, 2025, 4:57 PM

ive seen couple of posts like these

cursor too

they just confine gemini 3 flash to something small like diffs or small bug fixes

https://x.com/cursor_ai/status/2001326908030804293

Cursor (@cursor_ai)

Gemini 3 Flash is now available in Cursor!

We've found it to work well for quickly investigating bugs.

X · Dec 17, 2025, 4:21 PM

thats it?

[queen veldt]

Okay I'm begging to like the flash model

It's fast it's got up to date information in seconds

[warm zodiac]

anyone get master-node as a codenamed model?

[queen veldt]

"SOTA"

[sterile tartan]

GPT needs Better Eyes and Ears

[queen veldt]

They added it to YouTube

It watches the whole video and gives suggestions

This is insane!

[pseudo hemlock]

gemini 3 flash is absolutely insane

it beats 3 pro in SWE

allegedly

but anyways its beating 4.5 opus thinking and not thinking

and on par with grok 4.1 thinking

absolutely mental

[patent aspen]

The OAI founders seem very, very stressed

[queen veldt]

Also the flash model is better at online guidances

Toolatlon benchmark

It's even better than pro model at some stuff like DAMN

[latent crest]

I randomly have the video arena now. 2 vids in a day tho 🙁

Thanks pineapple, hopefully there will be a little but more, but thanks regardless 🙏

[molten cipher]

Guys

Google is cooking with gemini

But bro it's dang scary

I was playing a horror game and vibe coding on anti gravity with gemini 3 pro high

[torn mantle]

STOP

ITS GETTING SCARY ALREADY

dont tell me whats next

🫣

[molten cipher]

.

[torn mantle]

😱

[molten cipher]

It said

"Okay, that's done. I will send it"

Then it started spamming

"I WILL SEND IT, I WILL SEND IT, I WILL SENT IT, I WILL SEND IT"

[torn mantle]

tf

👻

[molten cipher]

Chill out it was scary, I was playing a horror game with my pay to win dude paying the game to scare the hell out of me

[neat apex]

OpenAi will turn warning code Brown after GF3

[torn mantle]

why are you using grounding/search/

?

disable it

[neat apex]

OpenAi now: If my eyes turns red, Run
OpenAi soon: if my pants turns brown, Run

[native yarrow]

oh bet it's rolling out

[vital lake]

Why cant we say curse words in the chat lol? @echo aurora

[vital lake]

😭

4kids IG

[fickle venture]

Imagine you got banned

[torn mantle]

some new models added

[echo aurora]

Lol sorry to say we have a banned words list.

[sullen quest]

it still bans mentioning the popular children's game where you play other peoples creations

[echo aurora]

Yeah reason being we were seeing people attempt to create scam content with Video Arena associated with the game.

[sullen quest]

huh

weird

[echo aurora]

If/when Video Arena moves to the site we'll reassess the banned words list

[rapid narwhal]

Did the terms update and they not tell us? all my prompts that worked before don't work anymore. I can't even use the site to make a script for a youtube video anymore!

[echo aurora]

Yes, we recently (within the last 24 hours) are experimenting with some changes to the content flag. If you have any prompts that you think are a false positive you're encouraged to share the prompt with us in #1447983134426660894.

You can also DM me if you'd prefer that.

[obtuse smelt]

lol try games with AI videos