#general
1 messages · Page 286 of 1
true
ive already read it, also already tried on new account on different browser but still the same
still the same
wtf is really happened?
Pls help idk whats happened
man now happens on all model lmao gg
I don't see any jamdevs submitted today, if you could follow step #3 from that message that'd be very helpful to identify what is causing this error.
Where are you seeing this?
When i tried loggin
Going to move this to a different thread, but quick question are you seeing this on all web browsers?
Its solved now thanks for your attention
The possibility of adding audio as well?
It's going to be random if the generation has audio or not. Not all models have audio support.
Guys show me some of your project
already did, i never use any external tools
already try to use on my android but still the same tho
with new account
isnt the problem from the model itself? @echo aurora
ah my bad
wait i use the tools to capture the bug
I think I have been shadow-banned from Arena,
already did pls check it @echo aurora thank you sir
This one’s cool too they don’t do anything at all. They’re completely harmless like I said, but still pretty cool effect
I got this inspiration from that Internet archive
That’s what I was going for, but he didn’t come out lol
i think this is why
hello
I remember now why I don’t like Claude
💀
bruh claude is the best for me tbh
I feel like he’s schizophrenic or something
are u experience the same?
Like paranoid I don’t know what his deal is
are you experience the same issue like i did bro?
No
I remember like two years ago maybe a year and a half ago anthropic they had a challenge for 20 K
On August 8, Anthropic, an artificial intelligence firm known for its advanced AI systems, unveiled an expanded bug bounty program that could earn participants up to $15,000. The company is offering these rewards to anyone who can successfully “jailbreak” its upcoming AI model, which remains unreleased. This initiative aims to enhance the sa...
Let’s see if I can find a screenshot one second from two years ago
And this is the moment I stopped using Claude
Lmao Gemini is quite a good humanzier
Those things are so flimsy, dude
So this is what the classifier look like I don’t know if you guys did this when it was out
What should I try it on
This is strategic manipulation right here. This is gaslighting. This is not a hallucination, although it could easily fall under that claim.
Kk
Unless you know any better ones.?
I’m not sure that is what I see videos online they said that these things aren’t really accurate
Hi
Or am I reading too much into this?
Detected
As I literally have been trying to use Arena.
For hours now.
And always no reply.
100% lmao
How I can generate a video pls
Always a CAPTCHA and an error message.
Humanzied
I actually take that back, bro
I just tested myself on some images and the and it’s pretty spot on wtf
Lmao I'm gonna try Gemini pro to see if it can humanize
Any better against grammarly
You know it’s very paradoxical here in very ironic
It’s machines, trying to detect other machines and those machine are trying to fold the other machines from thinking that they’re human
What a trip
Bro Gemini pro does worse then Gemini flash
Lmao true
I'm just tryna see how effective AI is at humanizing
Hmm
Might try ask Gemini to create a humanzier
Then try to humanize it in that
Or try to ask it to create a ai detector
Wait ima try see which ai can make the. Best detector
I wonder why it doesn’t work for you
I just realized this thing doesn’t have a copy paste feature
That weird arena was getting flagged
Not really it always gets flagged. It’s super flagged on brave.
Is this backdoored?
No
Kk
I don’t even know what that means
I was trying to go for the matrix aesthetic
No, I don’t. I don’t know how to do any of that.
Hm kk
I just copy and paste things and hope they work
It’s meant to be more for like an aesthetic, kind of vibe
More than anything practical or even usable
Yeah, none of them work. I just realized that.
This one works, but I’ll be giving it out and stuff
It says invalid response bruh
Wtf
Wait it was my WiFi flagging it..
Not even adblock wtf
Or IP blocked
Cuz mobile data worked
I can tell what this says
Yes that the point
Oh translate
Google translate failed lmao
Lmao
Gemini gets it right
What why has translate made a version of Duolingo
Lmao
Gemini AI detection works
Lmao it can actually translate it
Oh hell nah 😭
Oh wait Geminis ai detector is pulling out its bum
Lmao
So this website is an emoji combo
Kk
Hold on one second I’m Trippin sorry give me one second I’ll see what
So this is the one ChatGPT made
I didn’t believe it because the first image is in generate, but you could do this with almost anybody you could even communicate through emojis the right combination that’s kind of like a cipher
It’s a trip the difference, though it makes it like the way the images and stuff gets generated
Like the way characters have an effect. It blows my mind. That’s why I like to mix and match.
Yeah
What happened to deepseek
It's dead?
Like V4 is not coming
It's already mid march
@echo aurora u there bros?
it work already but really slow genereate the output
and also i need to start new chat and call the model first
It's set to launch in April
just generate some and stop again @echo aurora
@wise hill I'm going to followup in #1417174113092374689
alright thanks a lot sir!
What’s up with that?
Basically deepseek is expected to launch in April
That's why march is probably gonna lose
Also whats expected for v4
Image hm
Hi everyone
Yeah, insiders would know and they would place bets
Should I make some sense thank you
Didn't know deepseek is still alive
Why is march so empty
Compared to February it's so empty
Almost no big ai upgrades except gpt 5.4
It's one of the biggest Chinese models lol
I really wish ai had no personality
They should include a way to disable that I think.
Yeah I know but I thought is dead because I stop hearing people use it
I totally forgot about it until you mentioned it. Maybe that is things die
Yo i couldn't get pass LMarena captach, is it because i use VPN?
btw what is hunter alpha and haeler alpha?
guys how to get rid of captcha? I answering right, but it keeps going and going and goind and never stops
@sullen notch
VPN is definitely A reason it shows but it wouldn't be the reason you couldn't get past the image selection captcha.
What do y'all rate Microsoft copilot
0
So i just right now turn off my vpn and it actually work now, seems weird imo because before hand it's okey but now it's not okey
Thanks for the help btw guys
It times you out on the back end
To stop bots from farming lol
You guys remember RuneScape?
Unfortunately, I have already done all this 🙁
Interesting, the thing is that i already solve the captach but it seems like it think i am still is a bot
https://www.adameve.com/colonello Use code COLONELLO for 50% Off 1 Item + Free Shipping + Free Rush Processing
*Some exclusions apply
It's a quiet evening and you’re playing RuneScape when a strange character appears out of nowhere. Before you can react, they hand you a puzzle, teleport you into a massive maze, or throw you into a fight for ...
Did you try different browser?
Seems like the server of the VPN is the problem, because it's being use by a lot of people
This is essential to prevent spam.
VPN is definitely A reason your being flagged
- Chrome, Edge and Firefox
As a VPN can be used to disguise bots.
Try duck duck go
That's even worse.
go where?
No way
xD
I use it all day long doesn’t even log me out
Adblockers trigger the system.
Once in a while, throw me off
I tried Opera, Firefox, and Chrome myself.
The best browser would be chrome for clean internet
And speed
Doesn’t work even in Chrome.
Hm how?
firefox is best
I’ve been shadow-banned.
Doesn't Firefox default to Googles engine?
You haven't.
Also Firefox is quite valid but I believe chrome is generally better
Fair enough.
Try use mobile data.
But yeah, considering CAPTCHAs.
That's clearing browsing data which chrome has but a little more difficult.
Works flawlessly
Well, it looks like I'll just have to try again later. I'm just trying to make the first image match the colors of the second one.
Anyway, thank you bro for trying to help 🙂
how to use fire button in duckduckgo app on android.
#duckduckgo #duckduckgotutorials
This tutorial deals with search:
how to use fire button in duckduckgo app on android
how to use fire button in duckduckgo app
how to use fire button in duckduckgo
fire button in duckduckgo app.
Oh nice you ended up making it happen
?
Remember the watch handles
I create watch faces and use arena ai to change background colors to suit my needs. It's faster than doing it in Photoshop.
Hey man. That’s awesome that you’re able to recognize that man good for u.
That’s what it’s for, dude
Thank you.
I’m not saying that I’m denying what you’re saying I believe you
Your welcome, did it work?
What I’m saying is we fall asleep for me?
Got zero problems with it. It’s the only browser that keeps me logged in even if I turn off my iPad and come back tomorrow.
Chrome, brave in safari always logged me out after a certain.
DuckDuckGo browser hardly ever logged me out
That's not normal..
That’s what I’m saying
Duckduckgo should log you out after you clear.
As it clears all site data.
It’ll just be easier if I show you
Restarting also shouldn't be logging you out.
By chance do you have 3rd party cookies enabled on chrome?
Should be on by default.
Looks like cellular data is much better.
Yeah I thought so
Actually, it’s not important
All I know is to work for me so there’s no point of needing to try to demonstrate lol
I guess it is what it is
Yeah, maybe I have my settings all wrong. I don’t know. I could be a number of things honestly.
You know I’ve been pondering this for a couple weeks now, but I don’t think it’s gonna happen in my lifetime maybe not in your guy’s lifetime but I have hope sometime in the future in the future generations will not tolerate serfdom
Hm maybe.
Ditgal serfdom sucks
I hate it
I hate how people are viewed as a source of extraction of money
There’s nothing wrong with business and making money, but when you reach the point where you start incorporating annoying features and frustrating features and the only way to get out of them is by pay Walling
There’s gotta be some better way and if there’s not, maybe it’s not meant to be that profitable
For God sake, you figured out a way to charge money for words lol resolution pixels what’s next?
Key strokes?
The Phoebus cartel was an international cartel that controlled the manufacture and sale of incandescent light bulbs in much of Europe and North America between 1925 and 1939. The cartel took over market territories and lowered the useful life of such bulbs, which is commonly cited as an example of planned obsolescence.
Corporations based in Euro...
Because if your lightbulb doesn’t ever break, you can’t sell any more lightbulbs
People wouldn’t spend money on having to buy something that already works
Forced subscription even all way back then
Does that sound like some people we know?
I wanna see you do some test
You realize business makes their stuff last less on purpose?
Yeah, I know that I’m not. I don’t have a problem with that.
Bro arena has become unusable.
It was unusable there would be no bots
I spend 5 minutes in human verification for 1 minute answers.
Speaking of which I’m about to go buy a subscription today I think I just need to walk to 7-Eleven
What the - is botbot
You could always buy a shared account from someone
Little lot of services online and offer third-party services
And who is the goat pisces-0309b?
Those always get shot down in like a week.
The account or the services..?
Account.
Oh ok
Too many ips = misuse.
Yeah, but I forgot what they use. They use some kind of special browser.
I don’t remember what it was called
antidetect browser or a multi accounting browser
But you are right these companies aren’t stupid. They have voice to detect this even if people did manage to use it they probably have some clever way too have some anti-cheat.
That’s why I keep telling you guys manage. It’s cheaper to go get a little subscription or whatever that’s what I’m about to go deal.
You could thank the cartel for that…
requests.get(url) > get AES challenge cookie > solve it > Cloudflare blocks you > use CloudScraper > hit CAPTCHA > solve CAPTCHA > browser fingerprinting detected > spin up Puppeteer > headless browser detected > add stealth plugin > behavioral analysis kicks in > add random delays > IP banned > rotate proxies > API locked behind login > reverse engineer API > API monitors bot patterns > question your life choices
Aha.
I remember reading about this one before.
They did something similar with razor blades too.
My biggest fear atm is deepseek making their new model more expensive than 3.2. price creep
Bro
This has to be intentional.
Random battle mode never asks for human verification at all.
how can i make video like this
Okay, now it's spamming human verification in battle mode too.
using https://www.lueminex.com
thanks
BROKEN AGAIN?
LMARENA is FUBAR
TO UPSCALE IMAGES: us this in gemini
Transform this image into highly photorealistic quality as if captured by professional DSLR camera with prime lens. Apply natural lighting with accurate shadows and highlights. Add authentic real world details including natural skin texture with visible pores, organic surface imperfections, realistic fabric and material properties, proper depth of field with natural bokeh. Enhance with volumetric lighting, realistic color grading, sharp focus, and physical accuracy. Turn any blurry details into physical, meaningful details. Make the picture 10-20% brighter or more pop up. Output should appear indistinguishable from real professional photography with all elements following real world physics and natural appearance at maximum resolution It shouldn't look dirty, it should look clean.
I dont f*** get why lmarena cant fix infinity captcha
i literally cant use it at all
I’m in the same situation. Infinite captchas keep appearing. Sometimes they get solved, but the next time I try to solve one—even if I enter it correctly—it shows a “please try again” captcha error.
than is create infinite loop
what is the best open source llm model right now?
There might be a fix.
But it's tricky.
I'm trying something.
/"A highly detailed vintage banknote design from Bangladesh featuring three spotted deer (axis deer) standing gracefully in a natural landscape. The composition shows one large stag with antlers and two smaller deer beside it. The color palette is soft pastel tones of orange, pink, and purple with intricate engraved line art and fine guilloché patterns. The top contains Bengali script text, and the bottom right displays the denomination ‘1 Taka’. Include ornate borders, security patterns, and a blank oval watermark area on the right side. The style should resemble classic currency engraving, ultra-detailed, symmetrical, realistic texture, 4K resolution, sharp focus, no modern elements.
If you could solve this, you’d become a legend
And they'll find a workaround.
:/
@toxic verge check dm
hair
gemini keep error
gemini image
anybody gat this problem aswell?
is LM arena abandonware now?
lmaooo
limitation
Nah. I'm using image gen for the first time in weeks and nano banana pro still isn't working.
I had to try like 10 times before it worked.
But I'm assuming you'll get human verification before that.
same
this one not a problem
ussualy the fix it fast
now since 2 days none change
still error guys?
Qwen3.5. 27B if you can run it, it's fantastic. Or 35B-A3B, which has much better performance due to its MOE architecture. Or the small 9B / 4B / 2B models.
Just don't ask it about human rights in China or something like that, it's a propaganda machine. 😂
maybe use google flow?
glm5
good luck running that tho
where is video option
I love Claude, but his visual rendering is really terrible, so when it comes to diagrams and images in PDFs, he’ll do a poor job—but does that really matter for learning?
i'm a new soul, i came to this strange world, hoping i could learn a bit about how to give and take but since i came here felt the joy and the fear finding myself making every possible mistake
Claude opus 4.6 thinkin keeps getting error 10 seconds after submiting prompt...
Also gpt 5.4 high is the equivalent of gpt 5.4 pro in chatgpt?
Sup guys what do you do to a chat tht is just generating, not working
I think normal alternatives to Claude are Gemini and GPT High
MiniMax M2.7 has been released on their website, "Beginning the journey of recursive self-improvement".
︀︀
︀︀200k context, with a "highspeed" variant, scores 56.22% on SWE-Pro.
Quoting can (@marmaduke091)
︀
MiniMax M2.7 is out on web!
︀︀
︀︀Check it out here: agent.minimax.io
"A fast-paced, high-quality cinematic commercial. The scene shows a close-up of a table. A hand pla
does anyone else prefer gpt 5.2 search rather than 5.4 search
like 5.2 always gets sources but it seems like 5.4 is more in a rush and chooses to use its own knowledge
which would always be worse
Plan rapproché d'un interview de un vieillard qui dit en français: ne nourrissons pas tout les chiens que nous voyons certains on juste besoin de force pour nous mordre
actually I love Claude, but he has really bad eyesight, so I'm not sure if that makes it hard to learn from PDFs—especially ones with pictures in them?
minimax 2.7 sucks
it extracts text from pdfs
Haven't noticed this yet but will keep in mind when using next time.
Hunter Alpha beats GLM-5? And getting close to Claude Opus 4.6?
︀︀No way...
︀︀This mysterious model with 1M context is showing great results with openclaw and jumped into the top 5.
︀︀
︀︀claw-eval.github.io/
Quoting OpenRouter (@OpenRouter)
︀
The Hunter Alpha stealth model is now in the top 10 weekly:
What you mean by this ?
i hate this type where you just select cropped square
but it wastes your time by "try again" when selected motorcycles
this type is annoying
STILL FCKIN BROKEN GEMINI IMAGE
what are you trying to type? (prompt)
remove base, remove BG, rocks, smoke, fire, muzzle flash, bullets, debris, clouds, wall, casing shells. Change BG to light grey
some of images have general errors not gemini itself
(images models)
keep bloody error
@echo aurora fix captcha
@echo aurora fix gemini image
add me as a model
if you don't want to get filtered in lmarena
you should go to original sites or host it locally or using cloud because lmarena have filters on models
what
....TOO MUCH CAPTCHA
we love captcha dont we
🇭 🇪 🇱 🇱 🇳 🇴
Ai translate is agi
W
What is refrigerator 😭
batlle in direct or computer-use ai models?
newgen never heard of refridgiland
@echo aurora

mmm ¯_(ツ)_/¯
Try using a vpn
.....Oh
I ALREADY TRIED
@high owl its me
try square root 2
Idk why but vpn seems to be working well for me, lol.
mmm
Maybe it depends on server location.
Is this some old tv channel?
@loud verge idk
That sock had secret seeds.
What vpn are you using
Try different vpns/server locations.
Xiaomi has released MiMo-V2-Pro, which scores 49 on the Artificial Analysis Intelligence Index, placing it between Kimi K2.5 and GLM-5
︀︀
︀︀@Xiaomi's MiMo-V2-Pro is a new reasoning model and a significant upgrade over their prior open weights release, MiMo-V2-Flash (309B total / 15B active, MIT license), which scores 41 on the Intelligence Index. Xiaomi has not yet released the weights of this model and it is currently only available via Xiaomi's first-party API.
︀︀
︀︀Key takeaways:
︀︀➤ MiMo-V2-Pro scores 49 on the Artificial Analysis Intelligence Index behind GLM-5 (Reasoning, 50). It is ahead of Kimi K2.5 (Reasoning, 47) and Qwen3.5 397B A17B (Reasoning, 45). On the overall leaderboard, it places #10, just behind GPT-5.2 Codex (xhigh, 49) and ahead of Grok 4.20 Beta (Reasoning, 48)
︀︀➤ Leading Elo of 1426 on GDPval-AA (Agentic Real-World Work Tasks), ahead of peer models: On GDPval-AA, MiMo-V2-Pro places ahead of G…
Arena plugin for antigravity when?
Meanwhile arena can't get past human verification.
nice website name lmao
5.4 and its not even close
depends
Add grok to the poll.
You are right 5.4 is good model
It's the smartest model out there.
not sure vs 5.3 codex to be honest
that one was also peak
5.4 likes to yapp a lot
@vocal prism Note that Video Arena has been removed from the server. More information can be found in this announcement #announcements message
lul
Language Model Arena
how can i fix this?
This is a generic error that can occur for a few different reasons, so it’s difficult to pinpoint the exact cause from that error message alone. I’d recommend taking a look at [this message](#1417174113092374689 message), which goes into more detail about what may be happening and includes some steps you can try to resolve the issue.
Would also note our #ask-here channel for questions.
Sorry what do you mean by this?
Note hat we’re sharing examples with the team when users believe they’re being flagged incorrectly. If you think that applies to you, please send me a Direct Message with the email associated with your account, and I’ll pass it along for review. cc @deft spruce
Check this out they’re having a promo or something
What’s my favorite AI video generator.
I don’t know what they do to their model but their videos come out really cool looking
seed dance still the best sirr
which AI model has the cheapest API credits and still performs well enough for any luau task?
Is it out?
Not globally
It was about to come out in February 24 before they delayed it
But it was really good
they got crippled by Hollywood basically lol
glm
is it already in my country
is kimi k2.5 any good or do i stick with glm?
Don't bother with Gemini honestly. 3.1 Pro had a disastrous launch and has been unavailable for use for most of the time or has been extremely slow to respond. It's tool calling and agentic capabilities are also bad. Though my experience with it is mostly from the gemini cli/antigravity/github copilot rather than the arena
Kimi has native image input capabilities unlike GLM or MiniMax, it depends on what you're trying to do
making games
on the blox
im learning to 3d model, and while luau seems like a simple language id rather spend the time on blender
what the best model to create games actually?
can anyone tell me how to avoid reCAPTCHA on every chat every time on arena? it rejects my CAPTCHA answers so many times
and it's also not letting me use the audio CAPTCHA either, that would be much faster
Claude
General best coding models would be Opus 4.6 (but it will drain your money QUICK) and GPT 5.4 (xhigh)
noted, thanks sir!
I can second this.
gemini 3.1 pro is very very disappointing
opus 4.6 thinking is so good but burns tokens and money, but quality
I tried putting them head to head with the same prompt, and 3.1 pro gave the most robotic prose and dialogue I've ever seen out of a gemini model
I still have a google plus sub (the 5$ one) cuz I find Gemini fine for general non-coding stuff through their chatbot but most of my work is done through GH Copilot with GPT 5.4 and Sonnet 4.6 sometimes. Though if I were to upgrade to the copilot pro+ - I would definitely use opus 4.6 more
honestly feels like they lobotomized gemini
Can vouch for that i remember when I first tried use it I immediately mentally said nope when I saw the immense difference lmao before it even fully finish
antigravity lets you use claude models, so even the devs are self-conscious (to a degree) (or perhaps it just attracts people to this new ide they definitely won't kill after a while)
The problem with antigravity is that at any tier below Ultra, you get like maybe ~3 - 10 Claude prompts per day/week and that's it. The agent software is too basic and nowhere as good as Cursor or VS Code with Copilot imo
in other news, I'm happy for zai and their GLM and z-image models
the competition is good, and the models are solid, nice instruction following too
Antigravity's best feature is it's unlimited Tab autocomplete on the free tier, it's somewhat competent imo.
We definitely all want the opensource gang to become good (GLM, MiniMax, Kimi, Deepseek). Deepseek has been quiet though
is anyone here in riyadh?
is anyone here in riyadh?
is anyone here in riyadh?
guys
true.
I've had some small scale success by making my own "agent" (simple langchain python script) and it uses whatever model I want and writes usable code
I haven't used cursor but I understand that it's probably pretty user friendly and easy to get into, and has now amassed enough capital to grow into an advanced tool
is anyone here in riyadh?
bc
I jus got a missile waring
warning
or idk
something from the civil defence
that's concerning if true
like stay indoors
but no I'm not in riyadh
wonder if deepseek is secretly cooking something or they just are quiet? 🤔
V4 was supposed to come out sometime but there hasn't been any credible news on when that's gonna happen
Does anyone know any good free AI when it comes to making characters and roleplay with relative accuracy / realism
They did. I’ve been ringing an alarm.
Gemini flash is trash as hell
Gemini's apparently only good in Google's vendorlocked AI studio
the fire alarm community
unsurprised
they won the bet on 2025 but gave up afterwards? what?
Well, duh
I don’t know if you’ve been looking around way the community behaves
They always fall for it
what's the last good usable gemini model left? I heard they pulled one of the older gemini pro preview models
https://openai.com/en-EN/index/introducing-gpt-5-4-mini-and-nano/
Guy new model for gpt 5-4
API
High volume api agent calls 😂
Waiting for Google to drop 3.1 Flash. If it's like 90% as good as 3.1 Pro but it can actually work properly without hanging or being down - it would be a success (especially if they fix tool calls and agentic coding)
It won’t
Makes no sense financially
yeah, people keep making videos saying "it's over" or "google won" or "nvidia won"
what gives? nothing final has happened yet, and they haven't made any breakthroughs so why all the glazing? sponsored probably
which. one.
model name.
No, I mean the models are gonna operate better to the api
You’re not gonna have the same problems as you were using the app
Cause that’s bad for business
5.4 mini is basically OpenAI's answer to Flash but I haven't used the mini model so I can't judge it's abilities, it's also apparently more expensive than the Chinese models so it faces steep competition
Through the app. They don’t really have any sense of story about customer service.
yeah I use ai studio if I'm on mobile, and api on pc
Since people are paying them a monthly subscription, they have no reason to really mess around
The bread and butter is the API
Since they’re more of a service than the product
Either way, the money is rolling in
pay per token makes sure that their quality has to be better
yeah
He say eye companies they’re not really selling new products they’re selling you a service
More like renting service
3.0 Flash is unironically more usable now than 3.1 Pro due to better uptime but there's no reason to use Google models now
Yeah, that’s the irony
anyway I haven't sunk any money into paying for api calls and have always gotten by on free plan limits, somehow
And you know that is super disgusting how they go about doing it in a way that it’s hard to prove if the model really got throttled
Cause if anybody here pays attention to the wording these companies use
yeah, the only reason I used google models was the generous api limits (which were more tightened recently) and the fast response times
now it doesn't seem as good at all
i think gemini 3.1 is deep research is good not really decent tho
but its good
where ru from
saudi
Im in uae
they told us to stay away from the spot
lets continue this convo in dm
None of this is quantifiable
The terms I mean, and like a way you could really measure
Dm
Reasoning, it’s all subjective
what is it supposted to say
llm intelligence is not quantifiable
it can only point out comparisons
Yes, I like Gemini deep research through their app or their AI mode and the chatbot's web search and some general question/answer stuff. The few reasons I still keep my google ai plus sub
"trust me bro benchmarks"
notebooklm is decent though
way better than the gemini app in what it does
They could say whatever they want it doesn’t matter
What I’m saying is with the John is going through it very common and it’s a frustration
Because when you come to the way, he came to the conclusion which I also came to the conclusion along time ago you feel like a sucker
Unless the benchmark is obscure and/or private as hell and/or hard to game, it's not to be trusted. Everyone is benchmaxxing but I've noticed that in some benchmarks there hasn't been a big breakthrough, probably related to the limited compute issues and the LLM intelligence scaling hitting a wall
For believing any of it and that’s why I can relate to him because I know that feeling
It’s like you believe in something just to find out it’s all nonsense bunch of garbage and then you feel like an idiot that you bought it
Will arena ever be paid?
Cause it’s not like we’re anti-Google or that we don’t want them to succeed it’s the opposite we give them an opportunity enchanted instead you know not that you feel it down but almost like I don’t know. The only word that comes to mind is sucker.
I'm ANTI-Google
H I’m not anti-anything I just anti bad-consumer practices
I'm anti Data sucking software, so yeah Google
they'll be paid by providers and probably already are
it kinda has to be free for the public to access, or it'd fail it's purpose
Disgusting how they treat people I think in my opinion I think it’s very rude and I think it’s very disrespectful people’s intelligence
But people don’t care so I’m just a broken record . Don’t listen to me.
google has very anti consumer practices
They collect our prompts and chats and sell them as data, it will keep being alive and free until LLM providers go past the need for projects like this one
Dude, it doesn’t matter because at the end of the day, the AI community
As much as I love them as much as I’m part of them
They won’t figure it out until it’s too late until it hits their pockets later down the line
I can't listen through a chat can i?
It’s always been like that with every industry
All the LLMs are guilty of this unfortunately. All of them are built on illegally stolen data
wdym, what's the context to this
I think LLM providers will need an everflowing source of training data, because we can already see model quality degrading
and it's almost obvious that synthetic training data just won't cut it
Well, everybody’s already all hooked and everybody’s on the system that there’s no way around it
Just stop using AI like your emotional partner and start using it for generic purposes that have nothing to do with you
That you’re forced to use it one way or another if you want to whatever you know, in modern life
And you got nowhere to go the corner
And so if you want the comfort and leisure and everything else that comes with modern society
You’re gonna have to interact with it one way or another it be it at work at school wherever
Imagine buying a toy and having to pay a subscription to activate its AI features
Guys which ai model is the best for running claude code through ollama?
true, but I think it's a little more nuanced. if the model is freely available to the public to download and run by themselves, I think that it gives back to the public so it's kinda fine.
and you usually won't find an LLM that word-for-word copies paid content, but idk
Other than the big 4
Do you guys realize they found a way to capitalize and monetize off words?
Resolution pixels
The 4o psychosis folk didn't like this one.
On the topic of this, I wish the LLM providers tuned the system prompts to be way less sycophantic than they're now
4K. 8K.
What’s next keystrokes?
How insane is that? They figured out a way to sell words.
🤣🤣🤣
tell that to the character ai folks.
AI is a tool. it doesn't have emotions. people who pretend it does are having delusions, probably. but perhaps their mental state isn't entirely their fault.
It's crazy tho, at first it was banned in alot of countries, now that they realized that it is the perfect data sucking machine + gets users addicted to comfort, they push it everywhere
If you're desperate enough to use AI as a partner, might as well do it locally bruh
Local models are not very good at coding but if you have a giga amazing hardware with like ~128-256GB RAM + VRAM - minimax (2.5/2.7) would be your best bet.
If not, then maybe something way smaller like Qwen 3.5 Coder Next
I think they should just get rid of the system prompts, and just use a generic sentence like "you're an AI assistant named MODEL" or something
it's pretty unnecessary
...?
books have been sold for so many years, heard of them?
also I wish people paid more for books rather than ai slop
Even if AI doesn’t take our jobs, it’s gonna affect us economically like our own bottom line
Local models aren't good at coding, but they're the best for general use, with a simple question like "How can i get back with my girlfriend after an argument" you'll get a working solution through your local ai whitout sending any data to israel
But I guess it’s also being greedy for me
I mean until that day comes it’s only natural that where I’m at and conclude to what I conclude
I’m a product of my environment
local models aren't bad
the newer open source models are getting pretty good, and their quantized variants aren't losing as much quality as before
What I’m saying is none of us made the rules to the game
Models from qwen 3.5 are the best ones I think for local use if your hardware can't fit huge models like minimax or glm
Some people play the game some people get played by the game and some don’t even realize that they’re in the game
Vast majority of LLM using general folk don't even know there are AI subscriptions or that AI models have a 'thinking' toggle. They use the worst options that are provided by default (fast, non thinking old model)
Support our work: https://democracynow.org/donate/sm-desc-yt
As part of our July Fourth special broadcast, we continue our extended interview with Karen Hao, author of Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI. The book documents the rise of OpenAI and how the AI industry is leading to a new form of colonialism. "One of the thi...
We might finally get a large MiniMax model. Though this also means that hardly anyone will be able to run it locally
https://fixupx.com/SkylerMiao7/status/2034268535237783811
M3 is scaling up.
Quoting Zephyr (@zephyr_z9)
︀
This is an extremely strong model
︀︀I really wanna see Minimax scale to 1 or 2 trillion parameters
I love mini max
remove this hellhole screenshots and your message, it's annoying, plus I think you are a bot
Damn, they took off picture of MrBeast
do this in every single channel of this server and leave the scamming and be normal
plus it's might be scam
lmao
That blows my mind
If they could identify it and they still let it happen
lol, it is, but screenshot blurry
That’s another beautiful industry. That’s thriving right now.
Thanks for the same tools with promises made by the same people to protect us from exactly those elements
😂😂😂😂
I'm going to be real with you I didn't even know what message you were talking about so I was trying to select one of you that had that was replying to the scam message and that popped up even though I was not clicking on the direct message lmao
One second I got one for you
Number three rings really true here
A confused or harsh sound or mingling of sounds.
Always found the naming of "Discord" kinda funny
At least they didn’t call the controversy 😂
yep, lol for sure
One message removed from a suspended account.
The Video Arena Discord bot has been removed from the server. You can find Video Arena on our site here: https://arena.ai/video More information can be found in this announcement about the removal of the bot.
One message removed from a suspended account.
@analog gorge Note that Video Arena has been removed from the server. More information can be found in this [announcement ](#announcements message)
One message removed from a suspended account.
still waiting when captchta will be fixed
I suspect there are old how to tutorials still circulating around that just have outdated information
lmao
holy <100% scaling
One message removed from a suspended account.
you like spamming, don't you?
I even have to manually make it show me you replying I don't even get a notification when you directly reply lol
One message removed from a suspended account.
One message removed from a suspended account.
Would not this message incase you missed it - #1451574502369656842 message
i gonna read thanks
Well I have already provided both before so I'm still waiting to see when that gets updated
One message removed from a suspended account.
Answer this !
Yep, your case has been flagged to the team.
but i want to.
.
One message removed from a suspended account.
I'm unable to say since it's a generic error that can occur for a few different reasons. I’d recommend taking a look at [this message](#1417174113092374689 message), which goes into more detail about what may be happening and includes some steps you can try to resolve the issue.
One message removed from a suspended account.
Scammers are out today
Let me verify that
LOL. This is actually ironic trying to click on the thing you are replying.
?
This tends to be a temporary issue. If you hard refresh the site, it should work again.
Everything worked miraculously
ai slop
NaBrO its ai heaven
ai slop
Should it be in #ai-creations?
Yeah this is slop.
My friend gave it to me
Hm
Creations go in #ai-creations
Alibaba always delivers
what the hell
Bench maxxinggggg
how is that even possible
Real use is 💩💩
Is it not distilled?
This sounds like #codename-discussion
<@&1349916362595635286>


real
bruh they reduced video gens?😭
They reduced a lot of other things too, 💀
No, that's mimo.
MiMo-V2 Pro was Hunter Alpha and Omni was Healer Alpha
Zitiert Xiaomi MiMo (@XiaomiMiMo)
︀
Introducing MiMo-V2-Pro & Omni & TTS
︀︀mimo.xiaomi.com
Qwen models are good but a bit of benchmaxing was done too.
Get rid of this reCAPTCHA crap!
Use vpn.
I'm from U.S.
What bro
Shift crlt f5 in lmarena
Or shift f5, I dont remember to hard reset site
Just F5 normal but reset cache just this
I thought you need to clear cookies.
Benchmaxing
What's qwen feeding its ai
what is it?
To be honest, back then, Deepseek r1 did benchmaxing too.
Deepseek R1 wasn't 9B
Pattern recognition pattern repitition. It's like being trained on the question paper of the exam specifically.
Alibaba seriously needs to invest in there ai
oh, kinda cheating
They could make a model 10x better then gpt
Their ai is great already.
It's free.
Hm true
It has folders and deep research.
I'm not saying it's bad I'm saying they should invest to make a massive ai
max models are already big models
100k tokens I think
They 397b parameters
397b is so big
1 million
But imagine what they could do if it was 1T
I believed in qwen but now I believe in minimax
I always glaze qwen
Minimax is another case of benchmaxing.
Too bad Alibaba isn't too focused on it
... size isn't this important
True.
yeah but their models are actually strong
I know but just imagine the knowledge it could have.
It's okayish. They're like wannabe glm.
Minimax has potential though.
They're keeping up.
humm
but I think all the ai architecture must be changed
Yeah tbf
But the thing is a 100b ai trained on web scraps won't beat a 10b AI trained on quality data
and reasoning is the key
It's either more data or more tokens spent on thinking.
also
Those tiny qwen models are highly inefficient in terms of tokens used.
I mean qwen app / web is really good
You have some LLMs you have some MLMs and some SLMs
It's excellent. It can't be any better.
@compact jay I just noticed that minimax is crazy cheap.
The only LLM that can read transcripts from Google and help you vibe code your idea. But don't use it in qwen CLI. Things made multiple errors and when it finally got a working app it wouldn't install
I'll test the qwen 3.5 9b in pratice tomorrow
how to access video arena 3
A high-end 3D cinematic logo animation. A metallic silver logo emerges from a dark, smoky background with elegant rim lighting. The logo rotates slowly as golden liquid metal flows into its crevices, creating a polished chrome finish. 4K resolution, photorealistic, studio lighting.
it is
sorry I needed to get ready for school
someone generate video for me
Hello Masters, do you have alternatives on using Grok?
Lol.
Chatgpt? Gemini?
something that have Extend Frame
What's that?
also is there something wrong with my browser, because the captcha inside LMArena keeps repeating
No. A lot of people are facing that on the website.
Thank you for info. I can't find news on grok.
Anyone seen the new evo2?
98%
😭
how can i buy the arena sub?
How I bUY robocop
@echo aurora gpt 5.4 high is giving limit after every 30 mins please help
Hey everyone, how are you guys doing ?
PadaOne% Nothing
Which ai will reach the 10.0 version first?
I think everyone except gpt
Gpt just can't move out of .1 upgrades
✅ Kling AI Optimized Prompt
Copy and paste this directly into Kling AI. Upload the original photo as the reference image.
Prompt for Kling:
Photorealistic 5-second video. A woman with long curly hair stands completely still for the first 2 seconds, letting the eye register the deliberate textural mismatch between her heavy chunky beige knit jumper and her lightweight pale blue pleated maxi skirt. She then performs one very subtle, slow weight shift and slight turn of her body. The flowing skirt ripples and moves naturally with soft fabric physics, while the thick heavy jumper stays almost entirely static and grounded, creating a clear elegant contrast.
Camera is static. Natural room lighting. Extremely subtle movement only.
Strict face lock: the woman's exact facial features, expression, skin tone, eyes, nose, mouth, and entire face must remain 100% identical to the reference image at all times. Do not change, deform, or alter her face in any way. Preserve her exact likeness perfectly throughout the entire clip.
Kling Settings Recommendation:
- Motion Strength: Low (0.3–0.45)
- Duration: 5 seconds
- Aspect Ratio: 9:16 or 3:4 (matches the original photo)
- Negative Prompt (recommended):
face distortion, deformed face, changing face, different face, ugly, extra limbs, text, watermark
This version is tuned specifically for Kling’s current model behavior — it emphasizes the “stand still first” timing and strong face preservation to minimize any risk of facial changes.
Would you like me to make the movement even smaller, add a very light skirt flick instead of a turn, or create a negative prompt variation?
What is this website
Can I rely on AI as my main source for learning the basics?
Basics of what
Yeah
@echo aurora i got this messaage using gemini 3.1 pro preview; Trace ID: 61969f32-9c78
like math stem
Wow seed2 very good
Only if you use strong reasoning models.
how to create video I forgot can someone tell me
Hi, this is happening with almost all AIs, especially in the code arena with Claude Opus.
( @echo aurora <@&1349916362595635286> ) Fix this.
And i can I rely on AI as my main source for learning general purpose and stem with strong model?
Yes.
really is not dangerous
Learn with socratic method
Use gpt 5.1 or higher with high/extended thinking.
chatgpt is the most reliabel model ?
Depends. Tried this on grok. It ended up asking me questions instead of answering mine directly.
For stem? Absolutely.
and for learning general purpose ?
That too.
and opus 4.6 and gemini 3.1 is bad model ?
yes but i have max plan
Gemini has issues with sticking to the context.
i want to know the best
chatgpt pro but if i use the pro model is to slow ?
Yes it is slow.
why i need to buy pro chatgpt
Because you ask for best.
Lol.
You'll be fine with opus on Claude max though.
Don't sweat it.
I'm talking about the best balance of speed and, above all, reliability. With the Pro model, it's impossible to learn because it's so slow, so I'm torn between Opus 4.6 and GPT 5.4.
I still get the feeling that Claude is more out there than ChatGPT, and also that Claude isn't very good at computer vision
You won't notice much difference between opus 4.6 and gpt 5.4 high as long as you are not coding.
Open ai does better web research than Claude.
i bought pro plan
can you advise me
vedio
Use it only when gpt 5.4 with extended thinking fails.
Opus makes better ui
But you will notice you get 50x quota for same price on codex vs cc
Also, you get only like 15 pro per month I think.
The appro one
It doesn't take too long to get the hang of it, does it?
2m 12 to answer
Depends on the kind of question.
yes but in general is take 30s for answer is good ?
Yes
this is adaptative thinking like it choose how many time its think
You can disable auto switch I think.
CRUNCH-SQUISH-CLANG-SCREEECH
GRRR-KRRR-SHHH
CRUNCH-SQUISH-GRIND-STOMP
CRUNCH-SKREEE-THUD-CRUNCH
You better remember the safe word.
It can't be that uploading an image is a constant battle because it gives that error message 🙄
oh god
Stop adding features and fix the website's problems, is it that difficult?
That would sound great if it wasn't appearing every single prompt ( referring to the verification request itself)
☠️
i passed this type btw
Anyway how are y'all doing
Exhausted
These ai subs nerf the model's ability to research.
And raw api is too expensive.
It's free. What did you expect.
I was gonna ask if bigfoot was from tulsa king.
yes
Honestly my main issue with it is just verification Loop
Well it is a foundation-based company so honestly I'm not that surprised and VPN doesn't help with the whole verification stuff at least in my case
I said Foundation base as in it's not a foundation but let's be real here it kind of serves a similar purpose in doing something in exchange for something else that is not just pure money
I was stuck in infinite verification loop. Like 10s of verifications per prompt.
I've been like that for like 2 days straight for me now maybe make it three if a few more hours pass since it's practically morning for me
And I honestly don't know why the value is mentioned cuz let's be real here that's not very worth much when it comes to abstract based companies like this except it how much investment strength they have but I'm not going to go too deep into that cuz I would probably fall into a complaining about values of money lol
i think is dangerous to learn only with ai you need reliabel source
Do you mind telling me which one does that is the only one I have is reduces it to one per prompt which is still pretty annoying
Turbo
Yeah. Books can work if you have the money, space and patience for them.
Or ebooks.
Facts
really sucks at the hardware is so expensive
Shane I can only use the free version but ty
I'm using the free one as well.
@hardy swallow We need an explanation regarding the website not working, image creation, and some AI features in Code Arena.
Bee on their way to ask chatgpt how to deal with people who feel entitled without being rude.
Why can't I make ai videos or pictures here in this server??
Thanks for the report. Let’s wait for Pineapple’s help @echo aurora
Ok
Which AIs for coding do you like the most?
13
15
3
Claude
Hello, which is the cheapest video model with high quality output you are using?
Guys, can you please tell me why Gemini 2K isn't working? It doesn't let me take any photos.
@glacial swan

