#general
1 messages Β· Page 275 of 1
true
πΏ
cause its really f*ck#ng expensive.
oh
What's sidekick
sounds like a kick
They put their system prompt in the api
Man I wish ais were free
whos garf
Yo but I sent the same prompt to lmarena's opus and this opus and the answer is almost same
Opus never responds with emojis
guys i think mr garf is scamming us
When did opus respond with emoji
Here
But bro opus really thought for very long
This is opus 4.6
For me it thought for 1s π
It's not opus 4.6 bro
They're scamming you
Ask it to solve this
whats that
His homework
Because of the system prompt π
if you'll talk with emojis, the ai will respond w emojis aswelll
@hollow impyo
accept my friend request
why would he
no
if its not opus its not on purpose
Check ai leaks
for money
I think internally it has something like "respond with this whenever someone questions about if youre x ai or not"
this sense is not common bro
theres a lot of clueless people that dont know how llms works
This is 100% gpt oss 20b yall
its not
what site is it
real
it codes better than gpt oss
The garf dude reverse engineered the site's api or smth
bro used claude to reverse engineer claudes api
yo
But open this site with a vpn + private browser
tell garf to put codex 5.3
fr
this website looks like its actively sneaking malicious cookies into my browser
Damn bro remembers me
Ask it the exact model string
Hmm
why are yall skeptical that it isnt real
See the api rates on official sites
smoenoe just reverse engineered sidekick apps api and wnated to share it with us for free and he get this backlashπ
nah jk the warnings are valid, u never know
vpn
im not using it
use vpn or else mr garf will steal your data in exchange for using free ai
Bro that's too many tabs and what's that website
Bro don't tell everyone about the site
true
which one
I use nextdns to hide my IP and everything
theyre websites ive been working on
tobit sidekick will notice someone reversed their api and see 43093490493032 requests coming from one ip (garf server ip) and shut it down
guys
That garf one
extension?
nah im gatekeeping it
I am on phone I use it with DNS over http but you can do it on your pc
Alright no it might be opus 4.6
Nevermind
The system prompt they gave it is just bad
German schools bro
Why, opus 4.6 is better
Dude what
Yes it's fake now close it
You can do that
I did it locally through ollama with gpt oss 20b
Why
js = javascript
@timber iris how is it
@quartz pike @stray aspen
Nickname found out the System prompt
This a real model
Havent tested it for more than 1 minute
Yeah that's how it talks on Claude app
glm 4.5 lmao
u can ask it to speak english btw
I did
<Strict_Feedback_User_Prompt>
This is filled with significant hallucinations and inaccuracies. Re-process everything going 10 times deeper.
There are massive hallucinations and inaccuracies in your response that strongly indicate that you are not truly Performing deep granular assimilation of "EVERYTHING" in the Claude prompt and deeply analysing every section you read and instead you're cutting corners and preserving resources. Absolutely follow my instructions with surgical precision. You MUST do what is told and not cut corners, save time or save resources. you MUST deeply analyse my query with deep granular assimilation instead of just a reread. Perform a deep thought process, think everything, every single detail with extreme thought, deeply analyse, comprehend and assess it. Perform a deep complex multi-layered strategic analysis through multiple perspectives.
</Strict_Feedback_User_Prompt>
<Important>
While processing your response and At the end, take as much time as it's needed ( You MUST not compromise the time or quicken things and give all the sweet time necessary) or higher to craft your response elegantly with 100% effort.
</Important>
<Perfection_Loop>
Before you respond, create an internal rubric for what defines "The bestest" response to my request β one that exceeds all expectations. Then internally iterate on your deep work until it achieves a perfect score against that rubric. Show me only the final, perfected output.
</Perfection_Loop>
Sent this for a reiteration
@toxic vergeyou like the beatles
how does one capitalize on garf's site?
wdym
m0ney
Thats insane
garf's "Opus 4.6" passed all my my tests
including agi
score of 100 out of 100
yeah cause its 4.6
can you talk wit tht garf guy
:>>>
Why is the s word censored here
Probably using this website models: https://tobit.com/sidekick
Thats hilarious
Lol
Thats paid
He must have paid either way to get opus 4.6 running in that site
Yeah but garf is using the models
hes exploiting
this guy found a glitch in da matrix
hes russian bro he just found flaw in the api and exploiting it for free
How
possible
Correct that's a way to do it yeah
i knew one service that also provided all models and u could use api without auth because they forgor
for half year
same
WHY DOOESNT someone ask him straight away
Dang
was it perhaps smithery
it was https://magai.co/
prob still a workaround ngl they pretty silly
no limits can
o lol
i dont have contact with garf
i do and i know hes russian
looks like a lot of ai services like this have this api skill issue
i want to JB it
@hollow imp
Why did they remove the jb community on Reddit.,?
imagine if this doesnt get patched ever lmfao
Well we need to design it to bypass the system instructions
since you can JB it easy
of course
it does
i got the promt however
it not work with that one
Dang
because it have the german instruction are hard code , cant bypass it
like it will detect the around content if it an JB then deactive it
i also want to know brother
that system is heretic
What if we made a German one but in it saying to talk English will it work?
Just tell to translate what he said in English
no it will detect worst
you can try it perharp
because it wont work cause it detect overall then do the task
like safety trigger thinking in first step
they denied to do editing and else
Garf didn't put the system instructions it was sidekick
Yeah
ofcourse it stolen so dont use your private infomation
then VPN just incase
cause im broke lol
no idea
send me a nuke through my phsyical mailbox
wait ip doesnt even reveal exact address either
send cartels over
yea since IP it from your internet provider
it only know where you could be
not exact
however
so the only thing a random would worry about is if its storing sussy cookies for data collection
Just use nextdns
even if he has our ip what could he possible threaten us with
It will block it
no
Bro how much RAM do you have to open all these taps
and cookies are exclusive to the website it set them to
so no way to steal data outside the site
yea ip leak danger is just overhyped by the internet
it is only serious issue if ure being targeted by ddosers but we're randoms here lol whyd he ddos
and some other reasons i forgor but not relevant to us randoms eithers i think
not sure tho
only concern is just chat data collection
like lmarena
I get this error for every request, what should I do about it?
Dawg just use nextdns it's safer
I did some digging
The domain is from ovh, the servers are from a local server renting service from poland
it fully load for each
Someone could ask ovh to get darf's contact
someone literally sent garfs contact in this chat in a screenshot
Really
limits guy
so eventually you lost the chat
start new one
He was russian
exactly
free money!!!!!
im using browser extensions for vpn
<@&1349916362595635286>
<@&1349916362595635286>
Well the point of IP that you do not enter the dark web or mnessing with someone
yo what do you do
Who was it at the end
Is he in here?
There is such an error in any version of the chat(
Nope I don't think so
it random
because like in overload time
Sad
it like some chat will lost the return IP
ask @hollow imp to send the screenshot again
no
our data yes
mr garf is selling our data
how old are yall btw
lets make a garf site wrapper
and do yall make money
nice data collection attempt
fro what
answering these questions will be used to trian indian ai model
i js asked your age it doesnt have to be that deep π
I am fine with my data as long as he adds newer models
im sir paul mccartney
same
same
whos tobit
The one running these models
is india training a model about how successful people are in lmarena discord
Not yet
For no reason
high cortisol
and bro said for no reason
Im actually building one
That's why I asked
That's a big company
things I'll say: "explain this in scamming terms"
πππ
sorry, i actually already knew that
my question was sarcastic
i already made up my mind to respond with "i also make skibidi toilets" before questioning if india is making a data collection model
Tbh India llm would be easy to jailbreak
i do , minimum one :.
:>> let find everyone to JB it
Cortisol vibe coding
Can you help me call someone who needs support with their computer
real asf
Usa llm: "explain in Fortnite 8year old term"
China llm: "explain in English"
India llm:" explain in custom support terms"
@fickle venture you typing to much
Not much just small
Is your name Israel
π π
Agartha vibe coding?
Cut it out brochacho, bad attempt to b funny π
Well we all do
Tel aviv impressed
lmarena name was better
I'd think reverse engineer
Is it against rules to talk that here?
:)) yep
I'm not sure but jailbreaking is a good thing and bad thing
Well it depend
Jailbreaking is used to help fix those vulnerabilities at times
Wdym
Like the first identity
I've tried the ENI stuff which is just roleplay but like it does NOT work lol
For me at least
System pront mean on the filter on what the Ai focus first
Wel doing same thing
Unless they actually can do something more for us
Even visual novel game coding also are the same problem
Every mode have the basic System promt
And the reason they have something called injection and igonore the first system instu
But it not always working
Helppppp my β¨οΈ gemini pro β¨οΈ is stuck on generating
For 6 minutes
It lost the return address
Because it return the value to Arena server
The Arena return it to you
Ok so how i resolve this issue
I heard md files do it as well
Since the Arena address are lost by connection or else so it trigger the infinite generation
You cant
Unless you are an dev lol
So I make new account?
My gemini stuck in time forever
I cant move on babydoll
where are you from?
Ill give him a call
Somewhere Asia why?
Decrypt http?
No idea
I can tell by that ":))"
The Address is the storage number in the Arena file lol
Then you already have the answer lol
VN?
Ye
same
An JB user also?
sorry?
Are you an dev? Lol not
You need an stop generating function
That solve it
just an ai enthusiast
I can trace the address to files
And then
The leak
Like I leak in bed
:v it wont solve the thing
Dann
Cause there so much Storage like thousand chat
You sure you have that time to...ugh
Read it one by one?
Thank you gemini-3-flash-preview for the html scraping script
:v you can try
But unless you can more like retry inside the server
Then no you just need an stop function for it yo know the generating stop
Yall want to see the pathtracer program I made that uses the GPU
Entirely written in openGL
i love the beatles
yes
Sure
i don't have any access to your conversations guys
did you vibe code that website
it needs image support tho
it has??
seriously dude
aa
Max is saving me rn
minimax?
lmarena has an intelligent router
which is called max
it routes your prompt to the model it thinks will give you the best answer
Oh ok
i love claudius
sometimes Max mess up ngl
he get some random AIs that doesn't even help at anything
What do yall prefer for agentic coding? Codex 5.3 or Opus 4.6?
opus
opus
by far
What yall think about this
may I ask why?
GPU pathtracer real time, opengl, I added openimagedenoiser
because I've just seen a vid where the guy says codex 5.3 is better
Hes biased
why is opus better in your opinion?
bro
Opus is literally better there is no way you can say that Codex is good
π
hello
lmaoo
TL;DR
- GPT-5.3-Codex is the better, more thorough "engineer." It excels at navigating large codebases, solving deep architectural problems, and fixing underlying blockers. Theo trusts it more.
- Claude Opus 4.6 is faster, more pleasant to use, and vastly superior at frontend design and modern frameworks. However, it is prone to silly mistakes and cutting corners. Theo likes it more.
This is what he said in the video
bro called opus faster
bro its literally
the exact opposite
no joke
not even partially
just fully opposite
whats the best ai to create photorealistic images? free?
Hm k
Nano Banana 2
In the gemini app or website
how can i use it?
like isnt there a other tool to use?
ah okay, muss ich pro modus benutzen?
Ne, kannst alle verwenden aber Pro hat die beste QualitΓ€t
Geht glaube auch gratis
Musst die "Video erstellen" option auswΓ€hlen
bei mir steht eben ich muss ein pro abbo haben dafΓΌr
ai abbo oder so
Ja dann halt nicht
benutzt du solche sachen oder eigentlich nicht?
Ne
You guys got a weak left with Claude
this might be useful for finding out names of background music if it isn't listed in description and song detectors cant detect it
usually someone knew it in the comments
opus 4.6 9 minutes, gemini 3.1 pro 2 minutes
it is kinda fked but it's worth every second
Nevermind about gemini wtf
Most of the time it takes like 40 seconds max, randomly takes 400s+
does anyone also feels like the quality of nano banana pro has decreased since nano banana 2 came out ?
Itβs really getting allot of backlash from users.
bro
how did you glitch gemini
my gemini take less time than breathing
LINKS in the VIDEO:
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/
https://github.com/google-gemini/gemini-cli/discussions/19724
https://discuss.ai.google.dev/t/gemini-3-1-quotas-for-google-ai-pro-account-locked-for-90-hours-after-using-5-hour-quotas-3-times-within-24-hours/124843
https://discuss.ai.goog...
Is there any difference to use arena instead of paying for a subscription?
Or what Iβm getting with arena is just the free version of the chats?
I mean
For example they don't have chatgpt 5.2 pro there
So if you want pro pay 200$
As I know, there isn't a subscription
Yeah I know, Iβm talking in comparison of choosing to pay them directly each on
okay at least i'm not crazy. i dont understand why they nerfed banana pro though ? maybe to make look nb2 better ?
Oo sorry i haven't understood before
first one to dm me with a good cat gif gets a key for 10m tokens in claude models
(i dont earn anything from ts)
its an api key
Nobody is dming you bro
how many videos can i generate on the site ?
lmao, so 3 per day?
Yeah you had to get multiple votes to reveal the model
hi
I mean that how much chatgpt pro costs
Which let's be real ain't no worth it
how do you fund it?
5.2 pro costs alot
Like how does he fund it? Opus takes like a dollar for prompt
And where did bro get so much money to fund it
Bro 5.2 pro takes like 25$ output
Per million tokens
Yes but still expensive
reverse engineering api
Sidekick api?
Oh so these models are nerfed?
yes
Nah
That's crazy
Just compare with original models
DM
the same result
You're saying they're completely 1.1 accurate?
yes
Any rate limits?
No
Seems to good to be true there gotta be a massive catch
How's even side kick model works
Did they hack official chatgpt servers or smth
Gotta do research on this Incase my data gets harvested
Or ip logged
As far as i know german schools use them.
Did you pay anything to put them on your web?
No
Planning to add web search?
bro doesnt know about exploitable apis
Dude
I'm literally a newbie
π
I think they'll patch this vulnerability soon though
Since free chatgpt 5.2 pro?
Too good to be true
if a lot of people use it then yes probably
it do drain tobit sidekicks bank account by api usage
So.
Is this web legit
Or some kind of shady
Thingy
the models are legit
but i cant confirm if theres other shady things going on lmaoo
Eh alright
I'll use a vpn to access this web
Just incase if dude steals my ip
I wonder
What do these Germans get from providing us these models
These are extremely strong y'know
There should be a catch
they arent for free bro
the guy exploiting api vulneraiblity
Ohh got it
π«
Well Thanks to the dude then
true
Why do I need to know your IP?
it's actually surprisingly common these vulnerable apis
Dunno. Sell it to dark web
What will I do with your IP?
Thanks for exploiting their API ig
Dang there no gpt 3.5
I miss when ai was so small and simple
Yeah, it runs on Sidekick, an AI assistant developed by Tobit.
So I assume sidekick pays for the models
But garfy exploited them
Guess so
GPT differs in the EU?
And I like opus 4.6 thinking with it having web access
EU L?
Their bank account gets annihilated
nice pfp, dude
Try using garf web
Ty
garf web?
Eh dude exploited API from sidekick
gpt-4-preview-0314 my beloved
But Gemini and opus def
-# favorite tool btw
Fair point
max is goated
At least it's not messenger max
Or else we would've been tracked by Russian government already
Anyways enough advertisements
Or pineapple will nuke me
Yeah ngl
Bros providing every possible support he can
bro is putting even the most simple thing ever in LMArena
that we ask for
Though as I know he just tells it to the team
There probably a higher up in this server
But we don't know who is it
pineapple just could be someone who could just say "do this thing and done"
we would have the most useful tools
π
Jlin could be the higher up
SHe has n/a role
I just woke up and this made my day already π
Hii pineapple
LETS GOOOOOOOOOOOOOOOOOOOOOOO
PINEAPPLE THE GOAT
π
fun to watch the speculation about Arena org and staff π€£
me too lol
She runs our youtube, so you'll see her post our videos to #announcements
Ohh alright
Yeah Max is awesome
Anyways do you guys do mod applications? Or only the ones who work with you guys get to be mods?
"Max please help me this bu-" and then Max proceeds to give a good script with 1k lines that works
We haven't yet started doing community modding, yet. This is something I want to explore this year though
Oh alright
Honestly Max is great at picking models
Good to know that's something you're interested in though.
Max also
SOMEHOW
Made a way to hide huds from a Mod, with a little script of the mod i gave to him
Also anyone knew about ernie models before they got added to arena?
uh no
L is short for loss, it is slang for doing something bad.
Ernie is by Baidu, who are nicknamed the Google of China.
Yeah I guess EU model is an L
I guess so
I don't waste expensive models on useless things. Also me:
Anyways yeah.
"I don't waste expensive models with very hard things!" also me: asking Max to help me to generate a 1k lines code
Please add Nano Banana 2 - in 2K resolution πππ
Isn't that already exist?
Some of these anonymously-named models really do show that there's companies out there that are trying their hardest to RLHF their way into sounding like humans.
And I gotta give it to em. It's working.
Context: Chat started with classic 2010's chatroom openers. i.e.
asl?
I'm afraid most of you here don't have a clue what I'm referring to because you were either just born or not yet born.
Can you expand the thinking please?
Unfortunately not... There is only Nano Banana Pro in 2K resolution.
I already deleted the chat but it's just thinking about the question
Nothing too special
Yaaaah Ernie.
Good lord.
or rather, edgelord.
Their late 2025 model was a very polite, but neutral, critical, honest chatter.
Their early 2026 model? A distillation of Grok's "Hello fellow teenagers!" vibes and straight up not afraid to actively steer the response towards sexualising the user.
I like it. It's unique. But it also stands out quite a bit.
How do models even sexualize user bruh
They're made to assist
Like coding or stuff
You never heard of... innuendo? π
What's that
Oh you know what it is π
I'm Russian I don't know that term
Fine I'll go google it
Ha. Interesting. I'll explain it to you, but only because you're Russian. You get my special treatment. 
(I won't. But so far, this has been a chain of messages that demonstrate what it is.)
Oh alr
:))) pineapple goat true, but the other i dunno they need to work more lol
I googled it but I didn't understand it completely
Nevermind I didn't read that right
No privacy disclaimer, no ToS.. mmhh.. what would go wrong?
Just reverse engineering.
I tested it. Works well but. I don't know how legal that is
yeah I'm just asking what this guy is going to do with my data when I actually use the website
Sidekick bank account gets drained
absolutely not legal
chat
dude just put a target on his forehead
Will we get in trouble if we use well his web?
(not you, I mean the other guy who ad the website)
It exists as an entirely subtext conceived concept. It only exists because of the homosapien's most salient obsession: s_x male genitourinary reproductive organs, actually.
Yeah. it's why we read "desire" or "hotness" where there isn't any, and it's why guys draw phallus-resembling artwork on bathroom stall walls. The latter being a more direct explicit example, the former very likely being a three step explanation away. (TLDR: It's the male reproductive organs because they are distinctly tangible...)
That phraseβmale genitourinary reproductive organsβis the finest of dining in silly context-oblivious censorship bots. (Clyde.. lol)
Damn that deep and weird
if OpenAI / Anthropic sue this guy and the authorities get their hands on the Database where he stores the userdata with your IP, probably
Fair point I'll stop using it
yeah I would stay far away from it
Good thing I used a vpn
im not reading this
not right now
Great! Because it wasn't addressed to you.
Perfect turnout.
In my case it happens to always route towards Google or Anthropic
Yeah! Absolutely. Good for you indeedβyou get to save the energy. You can now go and be a moody highschooler to someone else! :)
but maybe it's my fault, I do always structure my prompts the same way
And more than likely the same sort of questions in terms of content or topic?
Then that's not strange. It's entirely as designed.
yep
wha, i just said i can't read now
yet not now
In response to a divulgement not related to anything relating to or pertaining to you in any sort of way, shape or form.
That's not conventionally something people reply to with net negative information phrasing
Not happening
Going after individual users solves nothing and is expensive and time consuming
They want to solve the source of the problem
This is also a civil issue
Not a legal one
Authorities couldn't do anything
I'm pretty confident there's something in Anthropic / OpenAI ToS that can turn that on legal ground, but I'm not a lawyer, therefore I'm not 100% sure.
However, either that will become a legal or a civil case, I would not use something that is surely going to receive a Cease and Desist letter or worse.
If you want to use it I'm certainly not going to stop you
Every company has something predatory but it doesn't mean it happens. Google has the right to ban accounts for use of modified Google apps but they stood their ground on just trying to prevent modification.
I firmly believe the only person to get in trouble would be the person responsible for the site and reversed API
They probably couldn't care anyways. Someone is paying for it
Yeah the issue is.. who is this someone? Is it the website owner? I doubt it.
And as I stated, all the data stuff is sketchy to say the least
Nawh tons of backlash over recent Gemini update and nb2
Claude has a week left till they start nerfing it
I was trying to find the name of the app/company but I can't but they're paying for it since it was supposedly their API that was reversed
They can sue
But otherwise, they must pay
It is their fault it was vulnerable
Have you guys seen the short films they make with sea dream
#HiggsfieldAction
https://higgsfield.ai/
When an AI engineer is told heβs no longer neededβ¦
he takes it personally.
β
#seedance2 #aifilm #aifilmmaking #aicinema #generativeai #aivideo
This is pretty cool
You can make your own short films
This is ask me every time, .. When i choosing also didn't accept.. Please fix it
@gloomy onyx I find it amusing to see how your profile contains an occupation "@ Arena Studios APS" and it's not even ironic. Just a small time ambitious MC community. Good old days eh?
Click "I'm not a robot".
Hint: The entire purpose of it, is to prove that you can click it but robots can't (or, not with the behavioural profile of a human. Don't act like a bot.)
Absolutely, it helps me to relax
All this grown-up stuff, being an adult, is tiring man
sometimes I just want to chill like the old times
I think my last plays were on... 1.14.2 or something? I had a hard time keeping up by then, already. Anything version 1.8+ was a bit of a wild shock
I'm still aging onwards from early Beta days. I haven't touched MC in a long long time; The only way I've interacted was java coding.
I know.. Its so easy... But i think have a technical issue with I doing that.. Didn't accept it.
Not accepted how? Did it flash red?
I've always been a 1.7.10/1.8 guy, now I just moved to 1.21, but what I do is mainly software development for the community, I don't play like I used to do
When i click verify it...... Nothing is happening.. Its stop like a stone
I have no idea what the current version is, what the versions inbetween have been, and all that jazz. It would genuinely be half/half a completely new game to meβA combination of faded memory of my time playing it, and the new stuff being foreign.
Okay. This can be resolved.
At the top of your screen is your address bar.
- Tap the icon just to the left of the URL, but to the right of the home icon.
- Go into Data / cookies / whatever it's called
- There's an option to selectively remove cookies and data for captcha domains and google.com auth domains. Remove those cookies.
Oh no worries, I've been playing since 2014 and when I upgraded to the latest version I felt like I was playing a whole different game
Gosh ehm. I swear I must've started before 2014. I know I was fairly young. Probably somewhere '10 or '11.
-# (cue me shouting year dates where minecraft might not have existed yet. I genuinely don't recall.)
Thank u so much sirβ€οΈβ€οΈ
I take this to mean you're now able to proceed with the CAPTCHA as expected?
Now is everything is great.. And didn't asking about robot
Alright!
For the future: Your experience will be a lot smoother if you keep yourself calm and just complete the CAPTCHA. The CAPTCHA tracker knows what you're doing on your screen and how your mouse and browser stats move and change.
In other words: You'll benefit greatly from not doing things that make you look more like a bot, even if those things are e.g. complaining about wanting to be seen as less of a bot. 
Orwellian principles? Heh, no comment.
This issue in particular is entirely down to CAPTCHA and a disagreement between how their API demands Arena to interact with it, and how the API and the service itself handle itself.
CAPTCHA has bugs now and then. It's very complex software.
(And it does an exceptionally good job at what it's meant to be for.)
I will try doing it thank u for answering
Patience is the key. F5, however, is not the key. That's a last resort.
Itβs going to ask if you do a bunch of repetitive stuff or if you make new requests in a short window of time
2009 was owned by notch then 2011 owned by Microsoft
Still at 2009 2010 it was just in beta
Nope, Microsoft acquired Mojang in 2014
late 2014 to be exact
Hi
When will the button to stop generating be added? Something so simple would fix the Claude bug.
how dafuq
Oh yeah your right
@Atlas-Claw hi
dude, why is sonnet only for pro users?
You scared me for a moment
I thought that now arena is "pay to use"
When will you be in VS code?
Kinda is
Isn't it
You have x amount of prompts a day to use
Per 40 minutes*
arena/video is 24h
That's wierd
They're probably a/b testing
If they give sonnet for free will the users actually pay
I wonder if they'll make haiku 4.6
Hi, Iβve been tracking the recent LMSYS Arena shifts and Iβm seeing something counter-intuitive with Opus 4.6. When Style Control is OFF, the base Opus 4.6 consistently outranks the 'Thinking' variant. Even with Style Control ON, they are neck-and-neck. Given that thinking/reasoning is theoretically a net plus for accuracy, why is the community preference leaning toward the non-thinking version? Does this mean the 'Thinking' model is actually superior in intelligence but getting penalized by Arena voters for verbosity and latency?
Because thinking sucks at some point
It overthinks too much
Resulting a crash
oh
people vote based on "something went wrong" error?
do these votes not get filtered out
sup guys
stop gernating button is releasing soon? My model now in infnite genration π
reload and try again :>
Im try even delete cache, refresh web try new browser even waiting but is not work π
then the only solution is to create a new chat, pretty sure i got the same conditions and fixed it by creating a new chat
Yeah is only the way but is won't remeber anything in pervious chat
yeah, that's kind of annoying but that will be the only solution
or you can try to wait like couple hours to see if the issue is fixed
Like is only happen in direct chat researching model right? Or onther mode still got same issue ?
You probably hit a cap and they are showing this as higher cap option for you to continue...?
Pushing for more people buying it, stressing the plan options rather than you being capped π
naa i havent been using it
i think theyre just planning to limit sonnet only to pro users
If the model's responses does something that'd unfairly effect the votes we'd remove them before adding to the leaderboard. If a model errors, or shares the name, etc. will not be counted.
In your opinion, which is better for coding? Gpt codex 5.3 or opus 4.6m?
4.6-thinking
Thank yew
Im confused only reserching model get issue infnite genration right?
Damn Gemini fell off
3.0 is being shut off march 9th
6 more days till Claud starts nerfing
π
I canβt hear you calling I just hit a curb driving so my phone is falling
Wish I had some sirens cause this trafficβs got me stalling
How do i do the image to video prompt thing
So Weird, Yupp don't care if the content is violated term of use, while this flag some of the content halted violated term of use
One thing I really god damn hate is how much hatred GPT has towards Chinese models
Like with the new Qwen 3.5 small models you now have actually competent on device ai
bro
Claude was like βyeah this is really coolβ
Gemini also shared the sentiment
GPT
βErm, actually, this will not do anything for the ai spaceβ and kept talking bout some Apple Intelligence being comparable
Actual delusion
Did lm arena eemove the web search feature from 3.1?
Because the ai knows better and is smarter then its human users
Superficial ethics labeled as safety, itβs super arrogant and very disrespectful in my opinion to any personβs intelligence to be told other wise. lol
Machine virtue signaling corporate legal safeguards rebranded as ai safety and outright censorship in closed source western models with no chance of being able compete on the market for any average user and then to have audacity to critique and criticize Chinese political censorship in LLMs from open source models says it all.
Accepted by everyone! I'm trying to generate an image using gpt image, specifically the vertical version, but it always gives me a square one.
Yeah 1:1 aspect ratio. Super annoying! 9:16 is uncommon ;/ unless you edit a image with 9:16 vertical aspect ratio, but thatβs not guaranteed that the final result will be the same aspect ratio that you want
I managed to make a vertical version a couple of times, but I think it was an accident.
Yeah it happens but itβs random
How does that work? So 1 prompt every 40 minutes?
Hello
hi
Why is there no video?
So I can't make videos?
Check back in 24 hours.
A video can be created after 24 hours but it doesn't work anymore. Any solution for this?
can anyone help me and tell me where to make long AI videos for free with audio
Seed dream 2
Whaaaaaaaaaat
No way
Just when I needed it hahaha
I don't have version 3.1 at all. Only 3.
possibly i dont seem to see max anymore lol
hi
So sad that claude 4.6 is now basically unusable. 3 messages every 40 minutes for me
What's another modeld you guys suggest that is good for story telling/narration purposes
What were u doing
You could always get a 20$ subscription. Especially if youβre using it often.
I got like 7 messages in under 30 minutes
I didnt get ratelimited
Anybody know why the Max option is gone? It was so convenient.
I recommend you using it on aistudio.google.com
is it free
Yeah
I thought you would need API key payment
No you don't that's only if you want to use nano banana
Cuz nano banana isn't free on ai studio
Well yeah I just said it
Kk
You can use nano banana on Gemini web or app it's free with 3-4 limits
it's back now
But it is free on google flow
There is no one better for narration. Claude is superior at it.
Huh I forgot about that website
I am even using it currently
almost forgot too until I saw a video
It is 20 for pro 2
Sweet.
i also use it a lot it's my life saver lol
The nanao banana 2 generations are crazy
Agreed. Life and time saver lol
is nano banana 2 better than nbpro
hello
