general

[zealous sparrow]

MILES IS A MEASUREMENT UNIT!

Very little LLMs will say Miles is a human and not a unit of measurement

[compact sleet]

If you add Morales on the back it might think otherwise.

lol

[lofty frigate]

Was reve-v1 removed in side by side mode?

[zealous sparrow]

likely pulled from imgarena

[hushed gyro]

Yo is there video arena now?

[zealous sparrow]

Yeah its a rollout on the website

[hushed gyro]

For everyone now???

[zealous sparrow]

I have it

Not sure

After changing Miles to Macy and specifying 2 seperate bikes
ghostfalcon got 2/3

[queen veldt]

I mean miles could be a trick question tho

It thought it's a trick question

It's a riddle so

[latent crest]

U guys how can I use Sora from Europe? Like downloading opera and using the VPN? 😩

[compact sleet]

I think it's all a trick question or just possible facts done in the most anticlimax way.

Both answers are kinda true in some sense

Gpt 5.2 seemed got less imagination in this case?

Lol iunno, it's either a win for 3 pro and 5.2 high depending on the user preference.

Surprisingly it was a good bench of telling llm scope of thinking. Thanks for the prompt. @zealous sparrow

[zealous sparrow]

I observed a ton of LLMs just always think Miles is measurement, even though its also a name

[latent crest]

Where’s the video arena?

[zealous sparrow]

It will be in the homepage if your part of the rollout

[latent crest]

Im def not

[latent crest]

Show a screenshot of that

[half mist]

I asked to be a part of the feedback program, but haven’t gotten any response

[zealous sparrow]

a film reel will appear

[zealous sparrow]

I aint a part of the program, if i was. I wouldn't be able to show this as i would be under a NDA

its just a rollout

[half mist]

Show me if it actually works

Because it could’ve easily been faked

[zealous sparrow]

I couldnt fake this

[half mist]

Wait actually, this could change everything. Is there a rate limit like the discord bot, or no?

[zealous sparrow]

prob

i didnt test much

[zealous sparrow]

2 videos per 14 hours

video models are expensive

trust me

This is a limit because it has a video model direct chat

[queen veldt]

Sora 2 pro 😮

[latent crest]

But u can choose the model or its a battle types stuff

[zealous sparrow]

In direct chat you can

[queen veldt]

What country are you in?

[zealous sparrow]

Im in the european continent

but

it's not a country rollout

trust me on that

[half mist]

Gonna say it, if that’s for all video models, that’s very disappointing. I understand 100% if Sora 2 needs it since it’s probably very expensive, but some models aren’t too expensive to run at all

[latent crest]

What models are there ?

[latent crest]

True some models are crazy cheap

[queen veldt]

They're not a charity tho

[zealous sparrow]

but you dont realize

that

LMArena has

thousands of users

and most of them would use those models

[half mist]

Well, at least have a higher rate limit for those cheap models

[queen veldt]

Saying "disappointing" for free stuff is crazy 🤣

[zealous sparrow]

still

[queen veldt]

Well you have fal.ai go use it if it's cheap?

[left iris]

Does anybody know what could be the best ai video generator for fast paced or high intensity action scenes ? I have this one scene i am trying to make with Kling01 or Kling 2.6. It has three archeologists running towards the camera while mummies are chasing them, but it only creates it in slow motion, tried multiple times with different prompts but it still shows the running motion in slow mo. Is it that by default Kling models recognize mummies in movies and overall being slow and it gets stuck wanting to create it only that way ? Tried with other video models in lmarena discord and worked out fine. Motion was fast as it should be and told in the prompt but Kling yeah creates the same slowmo thing all over again, i wonder if there is some workaround for it ? Like to totally force it to be in fast pace

[sterile tartan]

Gotta Love LMarena and Yupp

[latent crest]

When will it drop for all of us?

[zealous sparrow]

eventually ig

also not sure which format uhh LMArena is usin for sora-2-pro but this is the pricing for one SECOND

so basically

[sterile tartan]

8 seconds

4$

[half mist]

So around 3 to 5 dollars for a 10 second Sora 2 pro video. That’s less than Claude opus though, and that one has a higher rate limit, and more expensive

[zealous sparrow]

the videos arent even 10 seconds

they are 8

thats still 4$ per video

[half mist]

Claude opus is 25 dollars per message I believe

[zealous sparrow]

and multiply that by a lot of users

[latent crest]

Where are u from

[zealous sparrow]

it doesn't matter, the rollout is luck based

not region based

[queen veldt]

Lmao

"please" and "gift" doesn't go together

[zealous sparrow]

joins discord today, joins this server and starts his convo with spam and "gift"

[latent crest]

Can I dm u

[queen veldt]

What gift you got bro?

[zealous sparrow]

i ran out of my videoarena usage

[queen veldt]

Elon musk crypto?

Or is it

[latent crest]

Didn’t wanna ask that bro ,

[zealous sparrow]

why do you need to dm me tho

[half mist]

Is there a rate limit for battle mode on video arena?

[zealous sparrow]

its shared

[latent crest]

Im Italian, I’d like to use sora. How can I do it? Is downloading Opera and activating the VPN good?

[queen veldt]

Yes bro

Use VPN for countries that suppor sora and that's it

[half mist]

Video arena is cool, but I hope the rate limit is higher when it’s released, or it doesn’t take as long for it to reset

[latent crest]

If I use the VPN in Opera, the VPN will be internal to the app right? The whole phone with the other apps won’t be, am I right or…?

[queen veldt]

No bro

[sterile tartan]

Only for opera yes

[queen veldt]

It doesn't work like that

[queen veldt]

If it's extension then yes

But if you downloaded the app on device than your whole device is vpn'ed

[latent crest]

Grazie 🙏 do i need an invitation link or a code for start using sora ?

[half mist]

There isn’t a code to just unlock Sora 2 Pro. You only get Sora 2 with an invite code, but not pro

[queen veldt]

Also this is too long to be any code

[zealous sparrow]

It's unlikely, LMArena is already paying for running

1. All the Text Direct models
2. All the Image Direct models
3. All the code Direct models
4. Videoarena

[sterile tartan]

I think so

[latent crest]

Oh damn! Does someone have an invitation link perhaps ?

[queen veldt]

<@&1349916362595635286>

[sterile tartan]

https://tenor.com/view/spider-man-we-one-gif-18212100

[zealous sparrow]

he got banned

[queen veldt]

Nah

[queen veldt]

I thought messages are deleted too when he gets ban

[quasi atlas]

He is banned.

[zealous sparrow]

its a toggle

[queen veldt]

Thanks

[sterile tartan]

Do you guys prefer LMarena or Yupp Ai?

I use it mixed

[plucky sparrow]

same mixed

LMArena is still better for a lot of things

plus rating isn't mandatory

I guess it's not on yupp either, but it is sort of is

[left iris]

Does anybody know what could be the best ai video generator for fast paced or high intensity action scenes ? I have this one scene i am trying to make with Kling01 or Kling 2.6. It has three archeologists running towards the camera while mummies are chasing them, but it only creates it in slow motion, tried multiple times with different prompts but it still shows the running motion in slow mo. Is it that by default Kling models recognize mummies in movies and overall being slow and it gets stuck wanting to create it only that way ? Tried with other video models in lmarena discord and worked out fine. Motion was fast as it should be and told in the prompt but Kling yeah creates the same slowmo thing all over again, i wonder if there is some workaround for it ? Like to totally force it to be in fast pace

[ocean ferry]

both

[queen veldt]

Sora 2 pro is best

For everything

Any fast movement is hard for ai video gen to do

Just speed it up in capcut or something

Kling avoids fast movements because it will show wierd stuff if it's too fast

[zealous sparrow]

micro-mango was addeed to codearena btw

[plucky sparrow]

is it any good?

[zealous sparrow]

https://019b17a9-da44-7c68-8d79-50cdde12a0da.arena.site

[ocean ferry]

what model u think it is?

[zealous sparrow]

idk

[ocean ferry]

slop

very very slop

[zealous sparrow]

oh wait

i know

[left iris]

Thank you very much ! Will try to speed it up when i get home later will check how it looks that way, also will try with sora2 pro

[zealous sparrow]

it like slipped once

and said its a MiMo AI model

or Xiaomi

[ocean ferry]

lol what xiaomi

[rancid oxide]

maybe that is gemini 3 flash (but it is highly unlikely)
google had nano-banana but only time will tell who made micro-mango

[zealous sparrow]

yes they got into AI model creation

[zealous sparrow]

MiMo AI most likely

it told me that itself

[rancid oxide]

yuh

[sterile tartan]

Yeah yeah boi

[zealous sparrow]

here is one fiercefalcon made
https://019b17ca-d9e0-7eb8-ac4c-fba86021b204.arena.site

[south cave]

hi

[zealous sparrow]

hello

[south cave]

you from

[zealous sparrow]

you should reword this

worst anticipated model

[small sluice]

Create a 15–30 second Facebook/Instagram video ad for ZaFEx Express, a new local delivery company based in Bouira, Algeria.
The goal is to generate messages and phone calls while building local trust and credibility.

The video should be modern, fast-paced, and professional, showing real-life scenes: a delivery problem → ZaFEx Express as the solution → fast delivery → satisfied customer.

Key points to highlight:
Fast delivery
Reliable and secure service
Local coverage (Bouira and nearby areas)
Professional customer support

Expected deliverables:
A scene-by-scene video script
Facebook Ads copy (primary text, headline, description)
3 ad message variations (general, short/direct, merchants-focused)
A clear call-to-action (Message / Contact)
The final output must be conversion-focused and ready to launch in Facebook Ads Manager.

[weary galleon]

Agreed, but I can't change.

[hollow ivy]

ah i should have given Alibaba instead of TML :/

[weary galleon]

GPT 5.2 is terrible, but GPT 5 is more terrible.

[golden ocean]

-# ah i should have given Alibaba instead of TML :/

[sterile tartan]

It will be Deepmind and Alibaba

But Deepseek is the underdog

[stray aspen]

Wdym alibaba

They suck

[golden ocean]

steal a brainrot

[weary galleon]

Anthropic is the best

[sleek phoenix]

wtf

wait

i forgot im in the lmarena server

[hollow ivy]

Does Qwen3 (a downloadable model) suck for you?

i know of a guy who created himself a $10k rig to run the big version of Qwen3

(on YT)

[weary galleon]

Good. But there are better.

[zealous sparrow]

this is fire https://019b17f1-351b-7a0e-bfdc-ba98482a71dc.arena.site

fiercefalcon made this

[hushed gyro]

What is fierce falcon

[zealous sparrow]

google model

[hushed gyro]

In code arena?

[zealous sparrow]

yes

[queen veldt]

https://x.com/i/status/1999682480006427101

Diego | AI 🚀 - e/acc (@diegocabezas01)

After 44 minutes of thinking -> GPT-5-2 Thinking -> Make a PowerPoint, great results, the presentation was amazing, regarding the time it thought, doesn’t bother me, the result was actually great.

Excel below

X · Dec 13, 2025, 3:27 AM

[zealous sparrow]

i feel like kimi can do faster

unless im wrong

dont take my word

[weary galleon]

GPT 5.2 is bad. Period!

[hushed gyro]

Can't find it 😔

[hushed gyro]

There's very little difference to gpt 5.1 and 5

Google with G3 has taken over OAI's monopoly on AI chatbot

[weary galleon]

Difference is BIG! 5.1 is much better!

[hushed gyro]

One problem with kimi slides x NB Pro is the model doesn't use the images I provided

[hollow ivy]

Which locally runnable model is better than Qwen3?

DeepSeek V3.2-Exp? glm-4.6?

[hushed gyro]

Also one irritating thing is

All OAI models take ages to generate my code

[weary galleon]

EXAONE 4.0

[hushed gyro]

@zealous sparrow is sonnet or opus 4.5 better

[queen veldt]

Wait can gpt 5.2 THINKING create powerpoint presentations?

Why'd it took it 40 minutes?

I thought only the pro version actually takes it's time

[lofty frigate]

Where is reve-v1? Was it removed??

[crude lagoon]

40 mins just to get 2015 ahh style presentation format

[queen veldt]

Hey I didn't know models could create presentations that look somewhat good

It's doing it like a begginer but still it's something

Imagine companies one day and boss just says create me a presentation for our profits and what products did the best and why

And in 30 mins DONE

[weary galleon]

You can't run any of them on your Integrated GPU🤣

[hollow ivy]

lol? what integrated gpu?

[compact flame]

It's like GPU but integrated

[lofty frigate]

@echo aurora Why was reve-v1 removed?

[hollow ivy]

i dont use such critters ^^ lol

[weary galleon]

There's no button on your keyboard "to run LLM locally", it's not easy for very begginers.

[hollow ivy]

LMstudio: "Hold my beer..."

[narrow comet]

igpu is good

it lets your real gpu do the work

[weary galleon]

But bro doesn't have real GPU, he has only iGPU and he is gonna run 500B+ parameters models.

🤣

[narrow comet]

halo igpu is good

ion really understand the benchmark but this was strix halo igpu

[torn mantle]

whats fiercefalcon

flash?

@patent aspen

what about an updated gemini 3 pro

[stiff coral]

hi Can I copy the link to my favorite chat bot on the lmarena.ai website somewhere? Is there an lmarena.ai app?

[ocean ferry]

i hope the upgraded gemini 3 pro or the "full" release use the best checkpoint and not lobotomized

[torn mantle]

yea

[proud bobcat]

SOTA!!!!

[torn mantle]

did they lie or what

because they clearly stated this model has way less hallucination

[narrow comet]

has anything closedAI ever said been true?

[torn mantle]

no

😡

[hushed gyro]

Chat are they still doing a/B testing for video arena?

[neat apex]

Who voted in ExaOne?

@wraith bobcat it was you?

[proud bobcat]

Looks like they did lie

Artificial Analysis has always been very reputable

Their tests are independent

And not disclosed

Unlike GPQA and other ones

[twilit delta]

what the best model for programming?

[torn mantle]

what are the chances that this model has false positives in this test

[neat apex]

They biggest lie was saying o3-mini had 80% allucination rate to make gpt 4.5 looks better

But many times, i not sure but maybe most times, the o3-mini allucinated less than gpt 4.5

[twilit delta]

what the best model for programming?

[neat apex]

Opus 4.5, whatever if its Coding or scheuldering

Or scpriting

But Glm 4.6 is a close call

[hollow ivy]

just click on the votes, to see who voted what

[neat apex]

Ah yes, of course it was Sandy

[golden ocean]

-# just click on the votes, to see who voted what

[ocean ferry]

Opus 4.5

Gemini 3 for frontend

Opus 4.5/GPT 5.2 for backend

[hollow ivy]

but only the "thinking" version

[golden ocean]

artificial air conditioning intellgience

[narrow comet]

opus is good if you already know how to program

[golden ocean]

fr

[sterile tartan]

Gemini for Front
Opus for Back

GG

[brittle temple]

Why does the video arena suggest I've generated 5 times today. I've generated once???

[golden ocean]

gemini for vibe coding slop when u cant code or too lazy

opus 4.5 if u already know how to program and make super good high quality apps

[proud bobcat]

5.2 officially most censored ai

No wonder 5.2 sucks

It has a massive safety tax

[narrow comet]

I wonder how slow this would be with offloading https://huggingface.co/huihui-ai/Huihui-GLM-4.6-abliterated-GGUF

[twilit delta]

this is good for programming?

[vivid coral]

Well they haven't censored the sports and prediction markets like some of the Chinese models have. So they still pass my benchmarks. But I'm sure they politically suck though. Or something like that.

[narrow comet]

do you work for scam altman or are you just trolling

[proud bobcat]

This is bait

There’s no way

[narrow comet]

I didn't try it yet lol it is kinda big

[sterile tartan]

China: Hold My Gun...

[neat apex]

Gpt 5.2 Xtra High looks to be on pair with Opus 4.5 non thinking at coding

[stray aspen]

this benchmark cant be fr

[cloud zinc]

whats this

[devout vault]

livebench is dead

[cloud zinc]

gpt 5.1 codex is # 1

[viral cedar]

"trust me bro" benchmark

claude 4.5 opus is best in my opinion for software engineering

and grok is best for doing research and planning

[plucky thunder]

Ai

[lunar glade]

No gpt5.2 model on direct chat mode?

damn it why you leave it on the bottom??

[queen veldt]

Have you guys even tried gpt 5.1 codex??

It failed to create a working website than i had to paste the error 2 times for it to make it work again

[stiff coral]

hi everyone Can I copy the link to my favorite chat bot on the lmarena.ai website somewhere? Is there an lmarena.ai app?

[native yarrow]

why did they remove NBP 4k from yupp ?

[sullen quest]

wdyt of gpt 5.2?

[weary galleon]

I hate GPT 5.2, but this picture is clearly fake.

[low fiber]

Can I not access my previous chats in this website

?

[hardy crow]

Hey is Lmarena down ?

[weary galleon]

It works great for me.

[sullen quest]

lol

[proud bobcat]

its from AA

you can check the site

[weary galleon]

Give me a link to see this

[limpid nymph]

Please tell me why I always get a message about limits, how it works, where can I read about it?

"You have reached your rate limit for claude-opus-4-5-20251101-thinking-32k. Please try again in 44 minutes."
I just sent my first long message and he immediately wrote to me about limits. How can I fix this?

[split topaz]

How does this even happen? Like why I see -9 results before research completion in Grok? In no other AI it happens but just Grok whilst searching web.

[proud bobcat]

openai website:

artificial analysis is a reputable source for benchmarks

they dont post this stuff without verification

[thorn path]

I don't think 5.2 is nearly as bad as the consensus here is alluding to, however these OpenAI benchmarks are unverifiable and most likely fabricated

[proud bobcat]

actual openai bootlicker bro

chatgpt hasnt been clear sota since 4.5

these benchmarks are clearly exaggerated

and dont forget oversaturated as well

[thorn path]

Tell you what, if 5.2 gets #1 on the lmarena once the numbers are out, you can request anything from me frfr, I'm that positive it will come in 2nd, 3rd, or even 5th

[proud bobcat]

human preference is key

just because benchmarks are good doesnt mean its practical

yeah until they realized it was ass

lol

[thorn path]

Yes but choice is anonymous till you vote no? Therefore people who hate 5.2 can still unknowingly vote for it to the top

[proud bobcat]

https://tenor.com/view/patrick-s-patrick-patrick-star-bottle-dumb-gif-1741625486363070335

[thorn path]

And human choice is our best metric in a test to be human

[proud bobcat]

and what is it now

it got accurately benched by humans

and put lower than meta 3

[native yarrow]

talk to me dirty corn dogs

[proud bobcat]

or because they made new strides in reasoning??

you are impaired dude oh my god

dude

this is like talking to gpt 3.5

istg

[native yarrow]

bro he's just ragebait it's jsut been like this ever since i've joined

[proud bobcat]

REAL

[native yarrow]

especially with the lunatic politician as the pfp cmon bruv

[proud bobcat]

ts bait

[golden ocean]

ts = typescript

[native yarrow]

is ts a tuff edit?

[neat apex]

Many comments are saying that Gemini 3 does that and better

[zealous sparrow]

google is testing some checkpoint on AI studio

[weary galleon]

It's not a link. It's AI-generated image.

[zealous sparrow]

people say it makes a nice 3d iphone model

the google AI studio A/B's are stupidly rare

[queen veldt]

Bruh

They added

Nano banana pro in antigravity

It says it's rolling out but idk

[zealous sparrow]

wouldnt trust literal AIs

[queen veldt]

https://youtu.be/FB6HO7CZHWw

[proud bobcat]

what.

its right here

what???

actually

no

im not falling for the bait

[neat apex]

Gpt 5.2 high is trash

[neat apex]

Cuz Xtra high exists

[weary galleon]

Maybe Artificial Analysis is too harsh. Like to me their tests don't look accurate.

[weary galleon]

OpenAI is trash in general.

[neat apex]

But its clearly xtra high

[proud bobcat]

theyre quite reputable

[sterile tartan]

There should only be one benchmark

Real World Practical Test

[proud bobcat]

yes

i agree

[sterile tartan]

These kind of benchmarks are just complex for normal people to understand and Choiceless for Companies to Benchmax

Normal people can only understand the graph the higher the better that's all

💀

These Benchmarks have ruined the progress of Ai

Companies instead of Focusing on Real World Progress are Forced to Benchmax

[proud bobcat]

lmarena ig

[sterile tartan]

LMarena and Yupp is Good

That's how it should be

The World Tells which ai is good or bad

But i feel like they might be manipulated too

[tawdry widget]

Holy sht that's insaen

[native yarrow]

gemini makes the best powerpoints

[ocean vortex]

Yeah unfortunately I will somewhat agree. It's benchmaxxed and also kinda broken lol

timeout/no response. Cause it exhausted the entire context and returned malformed cut response like one of those open-source badly made reasoning models

[zealous sparrow]

technically you are wrong

googles notebooklm can make presentations

how so

btw

i even saw a notebookLM presentation in my school

[devout vault]

which ai is like #1 rn

[latent crest]

Getting to spot AI vids and images is
Getting harder and harder…

social medias will become THE weapon of propaganda

im scared

[left lodge]

Well here's some ai news,
Lmarena have backed for image attachments in search modality.
https://discord.com/channels/1340554757349179412/1449482775823515688

It's just been disabled/locked.

[burnt sinew]

Gemini 3 preview

Though I suspect it has been lobotomized

[grave plaza]

pain

[left lodge]

But halucination is also sota. Better in search but still sota halucinator.
Its good because scaled up a lot.

[torn mantle]

it is benchmaxxed indeed

[left lodge]

Gemini 3 pro , gpt-5.1 high and opus 4.5 are on the same ground.
gpt-5.2 is improvement in some cases but shows depression in others.

[torn mantle]

i noticed this issue from a lot of models, which is the trade-off of improving in a specific benchmark

they tend to get bad at other benchmarks

gemini 3 pro has that too

although on paper it may seem they improved on all of them

for example gem3 pro got bad at multilangual

[left lodge]

Claude is concise & a great coder and overall nice performant.
Gpt-5.1 high is best for learning & education.
Gemini 3 pro is best in multimodality

[hushed gyro]

@zealous sparrow how is vid arena going for you?

[unborn ocean]

afaik coding mostly increases reasoning

the other stuff is basically unchanged

• the nested learning thing has been around for a bit longer at google (with a somewhat different name and concept) Mixture of Nested Experts 🤓 👆

but the real question that i wanted to ask for some time: @patent aspen when do we get gemini 3 flash?

alright, feels like they have just started to train it after doing 3 pro

bc it is taking unusually long

oh, maybe so they can release it as a joined thing with nb2 flash (or what ever it is called)

or 3 pro was just pushed forward so that the stock can ride the ai wave a bit (bc of combining time lines of anthropic-google deal with it)

but that was the same with gemini 2 / 2.5 was it not?

and there they released the flash model earlier

[elder solar]

theres rumors that it will drop on christmas

[unborn ocean]

so i have heard

just feels off

[elder solar]

i just hope 3 flash is not dumber than pro

pro goes kind of dumber when more tokens

[unborn ocean]

be prepared to be disappointed then

btw brian did you also hear about gemini 3 pro being the size of gemini 2.5 ultra?

[elder solar]

why?

[unborn ocean]

flash < pro, for 'smartness'

you sure, bc i did not hear it from you..

[elder solar]

i still feel pro kind offset

they say its smart but this 3rd version looks off

[unborn ocean]

ok i honestly don't know where i have this from, so your story might as well be true

yeah that was also my understanding of the whole thing (though obv not the numbers)

[ocean vortex]

You mean like benchmarks they haven't tested or advertised for? I think that's the most apparent with 5.2. Their goal seems to have been with it to literally oversell it to the public, after Gemini stole the show.

While 5.1 was focused on response style, so not exactly making the model smarter either

[elder solar]

5 high was enough life changing

but too expensive

the first week of release, was really cheap

[unborn ocean]

honestly 50% seems like a lot, like a lot a lot

is thought the rule of thumb would be 1:1 in really high context scenarios only

[ocean vortex]

that 'xhigh' reasoning effort sounds good on paper, but...

it seems the same as high. Except when it breaks and exhausts entire context, unable to output response

[unborn ocean]

sounds good, probably wont understand them though, lol

[ocean vortex]

and lighting your money on fire lol

[elder solar]

sounds like scam

[stray aspen]

lmarena got verified

crazy

[sterile tartan]

What happened to @swift oyster haven't heard from them in a while

[zealous sparrow]

Craig said banned from openrouter discord, and no response followed

wonder what happened

[sterile tartan]

Strange

I wonder too

[torn mantle]

yea

[zealous sparrow]

https://019b198d-a906-7e1b-98d2-afe0d71d05e9.arena.site
I feel like ghostfalcon got a bit better

I like the document design

[torn mantle]

its the same checkpoint

[zealous sparrow]

Yeah but

[torn mantle]

i dont think they are changing anything

[zealous sparrow]

it was removed

and readded today

That happens if it's modified

[torn mantle]

i would care less about one-shot testing tbh

we saw this issue with gemini 3 pro

it decrease exponentially after 1st prompt

[torn mantle]

i would like you to test it on multiconvo with many prompts and ask it for many changes

see if it follows that correctly or misses something

[zealous sparrow]

cannot

it's a battle mode model

[lilac dawn]

Any guesses as to the upcoming GPT 5.2 position on text?

[zealous sparrow]

Less than 10

[lilac dawn]

I mean it surely can't be worse than 5.1, I just meant like whether it was gonna beat Gemini 3 or not

[manic pike]

hi

[torn mantle]

let me test it then

[swift oyster]

We are cooking !

[sterile tartan]

That's more like it i have been curious what's going on!

[weary galleon]

How many GPUs do you have?

[swift oyster]

We a

@sterile tartan

beta testing

[sweet tinsel]

They supposedly use "MPUs". Their own chips.

[swift oyster]

Shocking results from out internal model

[torn mantle]

its not bad at instruct following

it actually feels better than gpt 5.2 thinking

lol

[keen beacon]

Someone made good video about 5.2

https://youtu.be/hofOTpBgLOA?si=_3OIGDhVXhbZbtO0

[hushed gyro]

An experiment?

@zealous sparrow

[zealous sparrow]

So discord has experiments too

It's features that are not done

But are rolled out to people for testing

[ocean vortex]

Passable, but seems very surface level analysis. It's like he only looked into it while already in the process of making this video.

[shell oasis]

@echo aurora I will always be very grateful for my selection into the chosen ones to try the video arena on lmarena website...
I mostly fall into unlucky ones, but this time I really want to thank the algorithm or manifestation or whatever it was that I got to test this feature 🙏🏻

[keen beacon]

It’s one of the biggest ai focused channels on YouTube

[ocean vortex]

In the process, he kinda misses the point on some things

[weary galleon]

Can it achieve SOTA level of models?

[keen beacon]

Just one person interpretation

Everyone is opinionated and has a opinion of their own

Myself included

[ocean vortex]

He literally argued his reasoning based on the assumption that user backlash was based on superficial things like OpenAI's naming for xhigh or their focus on longer reasoning. When in fact data shows that on average it doesn't even reason more than the previous 'high'

Me at least, I couldn't care less for how they approach this naming or marketing and all, and longer reasoning is always good provided it actually translates to better performance. And you know... it actually reasons more

and doesn't break while doing it, among other things

[ocean vortex]

"The immediate reaction is suspicion"

that's what I'm talking about

completely off the mark

for me at least

immediate reaction was indeed excitement

then I tested it

and... 🥀

[lucid geyser]

Send a photo of a mpu

[ocean vortex]

what do you think about 5.2?

[ocean vortex]

5.1 wasn't anything special. But I would still use 5.0 over both of them tbh

cheaper, more reliable/stable

[torn mantle]

ghostfalcon is good

follows my instructions very well

[keen beacon]

More proof 4o was the perfect model

[torn mantle]

and fixed so many issues

[keen beacon]

But now it’s abut money

API

[ocean vortex]

5.1 is essentially...

5.0 with different output style

[keen beacon]

No they do not

[ocean vortex]

Yeah but you can't really argue with stats like tokens generated

[keen beacon]

Yes

[ocean vortex]

it is supposed to

but doesn't in practice 🤷‍♂️

[keen beacon]

Practice? lol

[ocean vortex]

#general message

[keen beacon]

What’s it supposed to practice?

[ocean vortex]

he probly isn't

🗿

[keen beacon]

https://tenor.com/view/jim-carrey-funny-good-one-gif-10820434279148874810

[torn mantle]

hmm

[ocean vortex]

@keen beacon In practice means people are referring to how a given thing actually performs IRL. Or in this case on average over huge number of diverse prompts. And I can't believe I have to spell it out for you.

The first time I managed to make it output more than 5-high, was also the time it consistently reached hard cap failing to output a response

nuking my API credits into negative $30

🗿

@deep adder

[ocean vortex]

It didn't help that I did parallel 10 requests, ran that 2 times. And somehow every single time was 100k+ with majority of them capped out no response 💀

yeah but like

the model is broken

it isn't supposed to do that

I've tested similar/same questions with 5.0 and 5.1 variants

not a single time I saw reasoning beyond 50k or for it to timeout

funny thing is that those few times it managed to not exhaust the context, the answer didn't look better than what 5.0 manages to do lol

[torn mantle]

ghostfalcon is really good

[neat apex]

I thinked gpt 5.2 was only gpt 5.1 with xtra high option

But saddly no, its a ahh benchmaxxed model

When you ask to it make a pagoda garden it does a marvelous work

But invent to ask anything else...

[torn mantle]

gpt 5.2 is so bad

[neat apex]

I will give a try to Qwen3 but ask to it reason way more

Here is any method to that especifically?

[ocean vortex]

Now this I can call a good/interesting video ✅
https://www.youtube.com/watch?v=tR2M6JDyrRw

I somehow did miss that QA at the time

[queen veldt]

Imagine we get gpt 6 in 2 weeks 💀

"code red" ahh

[neat apex]

It would be fire 🔥🔥

But if they manage to release it, maybe Deepseek v4 also manages to be launched

I am very optimist, since Deepseek v3 was technically only a upscale of Deepseek v2.5

They will get all Deepseek Especiale data and train in only 10 days the Deepseek v4, it will have 4 trillion parametters

[full shore]

Hey any one here to answer me?

[neat apex]

Why Grok 4.1 is worse than Grok 4 fast 🙃

Like, since always

[keen beacon]

Old but yet it is

[keen beacon]

He said it wouldn’t be a year wait

But idk

https://www.cnbc.com/2025/08/19/sam-altman-on-gpt-6-people-want-memory.html

[golden ocean]

"in the next 5 to 6 months"

[valid jetty]

Hello

[queen veldt]

5.1 was released on 13. November, 5.2 was on 11. December

[native yarrow]

seems ngl like yupp is slowly scraping its models off

[cloud zinc]

why

[native yarrow]

removed 4k, removed claude 4.5 online

a few others i believe too

[cloud zinc]

cuz its expensive

[keen beacon]

I'm experiencing a bug. I'm generating a website with Claude, but after it returns after the 50-minute limit, it doesn't create the site and gets stuck indefinitely on the "creating index.html" page.

[native yarrow]

yeah, not hating

just observing

[keen beacon]

help me anyone

[thorny schooner]

Why on Earth did this not register

[sullen quest]

@deep adder according to openAI, the api costs to use their latest model 5.2 are going up and are higher than 5.1, but subscription users get the same deal, I know you use the api, so wdyt?

[thorny schooner]

😓

If anyone Minds can they verified I'm not crazy that verified verify stuff is hyper inconsistent or sensitive https://discord.com/channels/1340554757349179412/1449553712618930187

[torn mantle]

re-designed lmarena

i will probably add more later and fix some bugs on my side

[thorny schooner]

https://tenor.com/view/tommy-boy-comedy-chris-farley-stressed-upset-gif-15396052

Every time verification comes up it decides to be stupid / every time

[hazy forge]

bro gpt 5.2 is so incredibly slow and annoying, all the gpt-5 iterations are grannies, for medium-caliber results

[tardy plover]

This website should have like a stop generating button like chatgpt

[native yarrow]

this just for you for now or do you have a link?

[torn mantle]

i will share it when its finished

[native yarrow]

bet bet

[torn mantle]

if u have anything in mind let me know

[native yarrow]

inspired off yupp definitely but yupp has an insanely good UI so no shame

[native yarrow]

if there is a way you could proram a stop response button in, thatd be amazing

also a way to see like how many usages you have left of the model before you hit rate limit

[thorny schooner]

That would be useful if it's actually damn consistent and not stopping randomly for no real reason of the entire chat

[torn mantle]

Yea i can do that

[thorny schooner]

hi

[torn mantle]

Mm idk if its possible, i will see if there are any clues in their js

[quartz light]

nice

[torn mantle]

Yea i used gemini and yupp as ref

• perplexity

[quartz light]

this is real btw

[native yarrow]

totally not ai

[thorny schooner]

Anyone know how to log into yupp i keep getting hit with a non-authorized warning thing when I'm using the Google Auto login

[quartz light]

im dead serious. it disappeared when i logged in.

[native yarrow]

would be sick ngl, but it wont be on the website for sure

[quartz light]

is everyone just going to ignore this?

[quartz light]

im actually being honest

[native yarrow]

well its gone now so like

[quartz light]

bruh

[native yarrow]

¯_(ツ)_/¯

[quartz light]

it means its in testing

[native yarrow]

i dont see it LOL

[thorny schooner]

What is with the wave of hand wave emotions

[quartz light]

yeah i dont even see it in private tavs

[native yarrow]

that would be very cool if they had video on there

[quartz light]

bots, admins dont do anything about them

[native yarrow]

expensive but

prob HELLA rate limited

[quartz light]

so basically, when i tried to send the prompt it forced me to log in

[thorny schooner]

I'm referring to the AI website

[quartz light]

and when i tried to, it obviously had to refresh and disappeared

so yeah its going to be limited

if its linked to accounts

[native yarrow]

oh so you have to

LOG IN to use it

[thorny schooner]

I just realized i can't even send a prompt when not login because it won't accept me hitting the accept button since I have been able to use it without being logged in and before anyone says as usual this is already after I did hard resets developer mode all That Jazz

[quartz light]

yeah, sorry i didnt capture the model list i forgot it was going to refresh 😔

[thorny schooner]

For anyone confused it happened around the same time I reported/ probably connected to not being able to log in glitch

[quartz light]

the fact that it has sora 2 pro is insane though

[native yarrow]

GAHHH, they prob got the stuff thats already on the leaderbaord for video

so i wouldnt stress lolz

i ran it thru synthid, ur image, it didnt see anything and honestly looked too clean so yeah i believe you

[native yarrow]

does it show it if you log in?

[quartz light]

i was expecting to see people already talking about this here lol

[quartz light]

yess nice

[native yarrow]

oh boy

so it is there

[thorny schooner]

Does anyone know how to log on to this i keep getting error messages that basically say Authentication failed.

[native yarrow]

did this JUST drop lmao?

[normal abyss]

it only comes up about 5% of the time, i refreshed alot

[native yarrow]

oh that is weird

so just spam refresh until you get it

i wonder if it even outputs anything though or if its just a gui code so far

UI*

[quartz light]

i got it first try so idk if its just chance based

[native yarrow]

if this works for me i mean crap honestly no reason to use yupp

cus they removed nbp 4k

[thorny schooner]

Is it weird I don't know why thought the repeated handways are triggering me and I don't know in such a weird way like I'm not even actually mad about them but it feels like instinctually repulsed by them does that make sense i have no idea what I even say?

[thorny schooner]

@native yarrow i'm just going to pin you since it seems you know what you're doing and no one is answering if that's all right

[quartz light]

whats the modality url parameter?

can you copy the address bar?

is it "https://canary.lmarena.ai/?chat-modality=video" ?

[normal abyss]

it took it away from me when i logged in

il have to try get it back

[thorny schooner]

Fair enough for me though cuz this entire website is basically killing me in trying to find anything of actual interest because of how glitchy it has started to become for me

[quartz light]

@normal abyss ya shouldve read!!!! 😭

[native yarrow]

well id start with refreshing everything opening a new tab, trying a new account, or using a different browser

tries those three

i feel like now that people know abou this they kinda have to release it soon 😭

[thorny schooner]

Well new browser is not changing anything so new tab is not going to probably not going to work new account here we are ( give me Merc and pray for me)

[quartz light]

im glad i was the first to spot it lolololol

[native yarrow]

alright yeah new acc, google is like

also ahving issues right now pretty sure

cus it also wont let me sign into lmarena with my google acc

to see if i can get this video thing

[quartz light]

just got it again!!!

and yes, its https://canary.lmarena.ai/?chat-modality=video

[granite heron]

**Mod break: if you see users spamming the 👋 emoji, please give me a ping **

[native yarrow]

does it actually LET you make anything though

[native yarrow]

you just keep refreshing or sum?

[thorny schooner]

And does anyone have any actual advice when it comes to this website's verification the picture one/Only only one that has no real right to exist

@granite heron

[quartz light]

too many :p

[native yarrow]

i wonder if its actually working

or if the code is just not complete yet and its a UI aspect

[quartz light]

oh youve deleted a lot

[native yarrow]

and then you have to wonder on top of it if there is any potential difference in censorship especially with sora

[native yarrow]

u got it to make anything for u?

[normal abyss]

unfortunately currently if your not logged in, it its 0 use, and when logged in, it vanishes

[native yarrow]

huh

well we could prob expect it within a few days

thats cool

much better than the video format here, which is complete bs and you cannot even do side-by-side the way that is most practical

[normal abyss]

hopefully, i was so suprised to see video models just pop up out of nowhere 😭

[native yarrow]

its gonna be hella gas when it releases

[robust sluice]

why 2 models of Flux always come up in Battle mode ?

[burnt sinew]

i think they unlobotomized gemini 3

[native yarrow]

how do

so

[quartz light]

have been trying to reverse the url for lmarena auth, sadly it requires supabase key directly in the parameter so we'll have to stick to the annoying method

what you have to do is

you open an incognito window

then

[native yarrow]

mhm

[quartz light]

it doesnt matter if it doesnt have video st first

if you log in and it has video its gud

[native yarrow]

why specifically incognito)

?

[quartz light]

im not sure how tormmane got it from just refreshes but im p sure it needs new cookies etc

maybe just cache

[native yarrow]

oh does incognito prevent cookies from getting tracked

/ cache

mmm

[quartz light]

yeah both cookies and cache

ok ill keep tryin

[native yarrow]

even then

[quartz light]

afk

[native yarrow]

he said you can't use it

so, i'm gonna assume we're just trying to get it long enough to analyze each model

[keen beacon]

Open ai keeps patching Sora stuff wow 🤯

They need to really protect Disney IP from being misused.

Disney got a billion invested in open ai

[native yarrow]

they prob wont be happy abou this new lmarena mode, but i actually assume a deal mustve been arranged. it will prob come with like 2 videos with sora basic and just 1 with sora 2 pro

[keen beacon]

Well this deal has been long in the making

[cloud zinc]

what did they patch

[keen beacon]

They rejected Sora one

A while back

[native yarrow]

lmarena?

[cloud zinc]

new lmarena mode?

[keen beacon]

https://www.telegraphindia.com/world/hollywood-split-but-ai-sora-aims-for-the-stars/cid/2070027

[native yarrow]

leaked, it'll come out soon pretty sure

[keen beacon]

No, I’m not aware of any lm arena mode

No when Sora 1 came out

Disney rejected the deal at first

[native yarrow]

ah yeah

well disney is fully invested into the ai slop now

[cloud zinc]

source? this fake?

[native yarrow]

just scroll up i nchat

i was talking about w a fwe other people

[quartz light]

m e

[native yarrow]

yes creds to jolly lad over here

[cloud zinc]

anyome can create fake screenshot these days

[keen beacon]

Damn dude

[native yarrow]

i ran it thru synth id

all of it

[keen beacon]

They been having law suits for Sora since the first one

https://biz.chosun.com/en/en-it/2024/12/12/GETK7B2XRNFEVK7NQPKDBFQP2A/

How fast everything changes in. YEAR

U guys should check out

https://www.designarena.ai/

[quartz light]

site is garbage

doesnt save anything properly

unless you vote

only recently started sending EMAILS for generations

not always works

plus videos dont render

[native yarrow]

doesnt even look like you can choose your model too

[keen beacon]

I like it.

[quartz light]

and input/output is severely limited

[native yarrow]

got potential tho

no way you can offer that many servics tho

[keen beacon]

Damn they really patched a lot of the stuff I was doing

Holy moly quick lol

[ocean ferry]

it looks like Yupp

[cloud zinc]

photoshop exists u know

[quartz light]

bruh

[native yarrow]

💔

[quartz light]

https://cdn.discordapp.com/attachments/1444765554815533126/1446189367927504947/togif-4.gif

you rn

[native yarrow]

barely anyone here would know how to use photoshop because theyre so reliant on NBP

[keen beacon]

True lol

[cloud zinc]

u can even do that in microsoft paint

just cut and paste

[native yarrow]

this is literally an ai server bro nobody is smart with image editing skills independently

bro 😭

[quartz light]

gng visuals arent even the main point

[cloud zinc]

or inspect element, thats super easy

[quartz light]

its the fact that it specifies things like "(input required)"

obviously legit

[native yarrow]

u cant inspect element logos vro

[keen beacon]

It’s not that it’s even smart. It’s that Photoshop is behind apaywall

[quartz light]

er you can but-

still

[keen beacon]

Subscription based

It’s a lot easier to use AI and cheaper than getting a Photoshop subscription

[quartz light]

LOL

[native yarrow]

and that was when i was 8.,.,

sooo

[quartz light]

btw its "rob ux" thats filtered im p sure

yeah

lo

[native yarrow]

huh

weird

https://tenor.com/view/silksong-spoilers-silksong-hollow-knight-lace-hornet-gif-1369488558298384028

[quartz light]

yeah "rob lox" is too

[native yarrow]

i can see why people on here are reliant on NBP i mean

this is photoshop level editing in 25 seconds
"change yellow accents to pink"

like whats even the point of photoshop unless for very specific things or graphics

[cloud zinc]

and u can do that in 10 sec

[keen beacon]

Photoshop will make there own

[quartz light]

my friend 😭

[keen beacon]

Photoshop will make its own powerful image editing ai

[native yarrow]

pfft

yeah pay $30 a month for it though

[keen beacon]

They have the most image editing data lol

[native yarrow]

NBP is essentially free, if you know how to do so

[quartz light]

nah

they didnt start collecting everything that long ago

google on the other hand

[keen beacon]

Allegedly

Yeah, that is correct

[native yarrow]

jeremiyah love didnt win the heisman, this is a fake graphic. just to change the text in photoshop of the award name would take HOURS

this took like 13 seconds

[keen beacon]

Still are industry standard

Adobe is huge with fat pockets

[native yarrow]

adobes a PSO

POS

[quartz light]

affinity photo is p cool

they do limit ai features but

basically photoshop but free lol

they recently made it free

[keen beacon]

The thing is ai integration is just expensive hell

Especially on scale

[native yarrow]

as with everything, AI will eventually get less expensive as other industries burst up, meaning by at some point while other WAY more impressive models/purposes will be out, things like NBP now will be completely free

just paling in comparison to either new models, or perhaps video EDITING models on the same scael of NBP

video editing in the ai world has not been dabbled in yet bcause of how insanely difficult it is, though its predicted to be the next step

[cloud zinc]

https://www.theverge.com/news/841369/chatgpt-apps-adobe-photoshop-acrobat-express

[quartz light]

in 5 picoseconds we'll have all ai models runnable in browser via webgpu

[keen beacon]

They been saying that

It ain’t got cheaper

[wicked mason]

Holy peak

[native yarrow]

its true, ai models that were very impressive back then are pretty much completely free now

look at midjourney

[keen beacon]

Cause they got meta paying them

[native yarrow]

that is my point with NBP

okay, still it proves my point

google will eventually just throw out NBP for free one day once higher quality models come, and or dabbling in other areas of AI like video editing

video editing is what people i feel are kinda waiting for in terms of hte next step

but thats really hard

i mean obviously; the video editing models we have now are complete dogsh

[thorny schooner]

@native yarrow Update trying a new account change nothing ( referring when I was I pinned you earlier)

[native yarrow]

hm

clear cookies/cache maybe

[thorny schooner]

Well at least I have the account now because god damn that took longer than it should have

[native yarrow]

not exactly sure what authentication failed is but it usually resolves itself

[thorny schooner]

Well it said nope to your solution thank you though I just damn i 😓

[keen beacon]

Damn I’m spitting out 100% movie scene lol

And they still can’t patch it all the way lol

😆

I’ll tell u now once Sora makes Disney available the characters

You’ll be able to reproduce the movie scene if you prompted properly

Sora has a huge flaw

[native yarrow]

bro this mf i got a phat crush on better come back 💔

[limpid nymph]

You have reached your rate limit for claude-opus-4-5-20251101. Please try again in 50 minutes. help me please

[thorny schooner]

https://tenor.com/view/skeleton-hell-play-gif-12684788351502543434

My will when it comes to trying to get AI to work properly it's draining my will day by day

[keen beacon]

That will be interesting to see if they’re gonna do actors

Or just , cartoon characters

[cloud zinc]

how

[native yarrow]

huh?

[cloud zinc]

actors are not part of the deal

[keen beacon]

Look

100% from movie downfall

[wicked mason]

But how

[keen beacon]

You can then use the same exploit to generate other characters but I can’t figure out how to do it without adolf it insta blocks

[wicked mason]

Pls dm me it

[keen beacon]

I have seen this with other movies

Like sonic

Some tv shows

Major flaw lol

[cloud zinc]

u got anything beside downfall, sopranos?

[limpid nymph]

You have reached your rate limit for claude-opus-4-5-20251101-thinking-32k. Please try again in 50 minutes.

Why do I get limit restrictions on all my models at once?

[cloud zinc]

its all about hitting the right words

offbrand?

[keen beacon]

Kinda been busy with other stuff, but they keep patching it so I gotta use it while I can lol

[cloud zinc]

lol

[native yarrow]

WOW

yupp i think is going broke

they just removed nano banana pro from manual selection

[languid plank]

is there anyone who is looking for professional and reliable engineer?

[cloud zinc]

i can access nbp in othersite for free

[native yarrow]

what site?

[cloud zinc]

i still see nano banana

[native yarrow]

thats nano banana

NBP

[cloud zinc]

yes google released it

[native yarrow]

if you look it is greyed out

[cloud zinc]

i cant access it

[native yarrow]

yeah

so

yupp's antics caught up to them

[thorny schooner]

Of course now I have a whole nother issue of it continuing to load indefinitely for no reason 🥲 and if I just reload the prompt disappears and then when I post it it starts out over again

( sorry if I'm neasley complaining just kind of at my wit's End with how this whole website has become glitchier over the past like)

[cloud zinc]

where can i access nbp

[native yarrow]

lmarena

[cloud zinc]

4k?

[native yarrow]

not sure

the only plaec i could use that was yupp and now yupps probably gona go bankrupt or sum

[cloud zinc]

hopefully lmarena is next

[pearl jacinth]

Does anyone else have this?

[cloud zinc]

nah

[pearl jacinth]

yeah, I log in to try it and apparently it disappeared too.

[cloud zinc]

there is no https://lmarena.ai/?chat-modality=video

[whole sundial]

it's an a/b test

[cloud zinc]

i have tried on incognito

[turbid ether]

Hey can i get access to the champions chat my friend @surreal creek told me about it 💜
<@&1349916362595635286>

[surreal creek]

#announcements message

the link to the form is here 🙂

[old garden]

does lmarena have video yet on site

[cloud zinc]

it never did, its on discord

[quartz light]

wow really? they removed a lot of models recently, they removed online opus (which was better and was the only one with working thinking AND IMAGE INPUT) and they had a lot of nbpro models like 2k and 4k but they reduced it to just nbpro normal

and they started adding free models

:(

[native yarrow]

yeah they're running out of money, so they're removing expensive stuff

[quartz light]

😔

[native yarrow]

their gist wasn't gonan last long, you can't upkeep that many expensive models

[quartz light]

damn i used it a lot

[native yarrow]

HUNDREDS of dollars per hour

it's okay, sometimes you just gotta take what's given to you and not take stuff for granted

[quartz light]

plus they literally allow ppl to cash out the credits

[native yarrow]

NBP 4k is already really hard to find anywhere

[native yarrow]

yeah they shouldn't do that

imo

[quartz light]

they dont benefit from it at all

[native yarrow]

yes it felt like a charity company if anything

[quartz light]

hmm

maybe theyre not running out

arent they backed by ycombinator

or somethin

cuz they havent taken down opus or anything

only nbpro

[native yarrow]

imnnot sure but unless you're a big corp no way you can afford that money to provide all those services

they will, watch

[quartz light]

maybe its just down, its the only "greyed out" model which isnt just gone entirely

[native yarrow]

also lmarena is already pulling big strings providing opus and NBP i'd imagine

[quartz light]

arent there investors or somethin

[native yarrow]

i wouldn't know ngl so i might have just said some bs ><

[quartz light]

lol "Yupp AI has raised a total of $33 million in seed funding, which was announced in June 2025. The funding round was led by the venture capital firm a16z crypto (Andreessen Horowitz's crypto arm). "

[cloud zinc]

hopefully lmarena go bankrupt

[native yarrow]

why??

[quartz light]

[cloud zinc]

unfair for paid users

[native yarrow]

ain't no way bru

[quartz light]

ragebait btw

[native yarrow]

lmao

[quartz light]

https://cdn.discordapp.com/attachments/1444765554815533126/1446189367927504947/togif-4.gif

[native yarrow]

i just thought he was being serious for once 💔

[cloud zinc]

free user exploiting the service

[quartz light]

https://cdn.discordapp.com/attachments/1444765554815533126/1446189367927504947/togif-4.gif

[cloud zinc]

they should lock it behind some kind of ID

like driver license to verify

[quartz light]

least obvious ragebait

[cloud zinc]

nah

[native yarrow]

chatpgt adult modes ID verification is the stupidest thing i've ever heard of

[cloud zinc]

free users exploiting is drying up the seed money

[quartz light]

no

[cloud zinc]

sooner or later, they will go bankrupt

its a matter of when, not if

[quartz light]

thats like every company

in existence

[cloud zinc]

u want kids to have access to that? lot of kids use chatgpt

[native yarrow]

no

there are several other better ways to prove your age other than a literal photo ID with your personal info on it

[cloud zinc]

what other way

they also have ai age prediction system, thats why they delaying till 2026

[native yarrow]

database/background checks, credit card verification (don't gimme any sh cus you already put your credit card online to buy stuff online anyway), facial age estimation, etc

[cloud zinc]

driver license was just an example. ur other option also include personal info

[keen beacon]

Google does it too

I got hit with Google verification id

Like 2 weeks ago

[cloud zinc]

they sold ur id info

[native yarrow]

well yeah, youtube, but that actually just predicts based off your activity and i was able to easily bypass that

[keen beacon]

For Gemini

😭

[keen beacon]

Winner take all

[native yarrow]

i believe im right

photo and ID is not necessary

[sonic flare]

Investors for what exactly 😕

[native yarrow]

yeah almost like i don't want my ID card willingly given to a major corporation

[keen beacon]

They give u three options to verify

Id, selfie I forget third

[native yarrow]

ok lolz, still don't want it

[keen beacon]

Thats the problem

[native yarrow]

it's js going on the defense

[keen beacon]

That they don’t care

They care enough to obtain it and store it in some database which will get hacked sooner or later

[ocean ferry]

what an irony😂

[cloud zinc]

the irony is that u are following him

[keen beacon]

Exactly why people are resistant to AI