#general
1 messages ¡ Page 142 of 1
The end is coming đ
The one guy of the just ones 5 atives guy in lmarena banned
20% of ALL Lm Arena is... Dead...
R.I.P.
Greg
???? - 2025
He was a good man, and he always made people laugh with his funny humor. May God bless his soul.
All we have left of him is his floating soul to talk to usâĻ
Real
Remmber my nani banani and sodadream wifes
I'll take care of them while you're gone.
Hm maybe i will unbanned
Remember
Perhaps.
hope is always the last to die
The wise words of the dead.
Among us be like đđ
Quite literally.
Pineapple is the imposter, and we're the crewmates.
Ah.
Lmao
Yeah toothless is not toothlessing
@fiery gull hey sorry for ping
I have a question for u
I have said something of politic/NSFW... today?
We're still reviewing.
btw can I have the download for the extention again
Toothless is trying to Toothmax.
He's Toothmaxxing that jawline.
AlsoâĻ
No way!!!1!1!11!
LM Arena, and LM Arena 2!!!1!1!
@echo aurora WHATS GOING ON
I don't remember this, I love to be pinged đ
Huh?
đ !!
Hi
@fiery gull Greetings.
Lol it's two versions of the same bot
Thx u
Incase one goes down
Revive the greg pls
Krusty Krab 1 and 2 ahh moment.
Sorry but I won't be discussing mod actions publicly or with folks who aren't directly involved.
Just joke I know this đ
mm
But I want greg revived đĢŖ
Hi
Any news?
About what?
Idk
Anyhow, if anyone needs a reliable translator that can translate into any language with accuracy and slang usage, then I've got one here: #ai-creations message
Deepl
This one is better than DeepL.
Even DeepL sometimes makes mistakes, but that one is the most efficient.
GPT-5-High should show reasoning trace?
Yes
..
/video make this city alive
ok
GPT 5 Pro doesn't even show reasoning trace officially until it has finished responding. No matter how much I click details it won't show anything
..
/video prompt make this under water city alive
What ..
here you go
Bro sorry I forgot the extension
any idea on gemini 3 launch date frends
I'm 48 hours waked up đ
Make sure to read â how-to-video-bot to learn how to properly prompt the bot.
70% is in oct
how can I access the different modes in LMArena? (battle, side by side, and direct)
New mod?
i just bet some money that it is being released by 15th on polymarket
mostly losing the money
Yes
đ
gato is a new mod and already looks tired
didnt get you
You from south?
get polymarketed
no but why should i study đ
Hell nah bro đ
Because jee is in 3 months
just 70 dollars
250âš mujhe dede agar 70 dollars waste karne ko hai toh
Mera bhala hojaye
Is a max glm 4.6 plan
tu padh le bhai
Yooooo
Is there a prompt checker or rewriter, that makes a prompt lmarena terms of use friendly, with changing the prompt as little as possible?
Use Claude
just wait\
First time using Claude and it worked đ
how do i fix my gemini being depressed
turns out lmarena really hates the term 'bent over'.
đ i wonder why
Is that the only reason to bend over lol
maek a new chat
Is that ai studio or gemini.google.com
its google ai studio
and gemini decided it wanted to be depressed today
idk why
Increase the temparature
fool it into thinking it did it itself
huh
For my use cases it always disappoints
what are your use cases?
:/
No Gemini news?
i see
anyone know how to make consistent characters on sora?
gemini 3
gemini 4
gemini 4 is not testing stage
yeah i know
Gemini 5 when?
after gmini 3 and 4 come out
in the case we get that far without a nuclear war
its a bug or a not yet implemented thing?
Gemini 3 is out?
nobody is so dumb to fire a nuke at nuclear armed country
and if a guy was suicidal and tried it, the operators would reject it
(like Arkhipov and Petrov etc.)
as long as there are humans controlling the firing, it would not happen
Hey guys! Let's please keep the chat around AI topics. Thank you.
(sorry)
No problem
i heard, december, at the earliest
(for certain people or groups)
i wonder, why do people think, that GPT5-high is better in coding than Claude-4.5-thinking?
Oi gente
crack
Honestly, whatâs the point of vote craps made by Gemini 2.0 when thereâs apparently an advanced Gemini 2.5 (image gen)đ¤Ŗ10+ consecutive round of NO Gemini 2.5 results. Battle mode should have those kind of limit.
Guys is GenSpark GPT 5 Pro unlimited legit or fake?
I don't know
idk what the argument was about, but it sounds like to me like its talking about how reasoning models solve problems vs non reasoning models and how its inefficent?
eg. solve 1+1 the reasoning model does the thoughts and solves 1+1 =2, then has to say it again, meanwhile the non reasoning model just says 1+1 = 2?
which is?
Lot of people want Pro
yeah looking at the paper, sounds like to me they are saying exactly what I said
cost I assume is the problem
how did you use CLI with gemini?
They still might do it, just with even more limits then opus
yeah the hybrid model seems to be just a better version of the autorouter, which isn't that complicated since I thought of that a while ago too.
Wasn't there a Singaporean paper on this
Might have been this one: https://arxiv.org/abs/2504.13837
Reinforcement Learning with Verifiable Rewards (RLVR) has recently demonstrated notable success in enhancing the reasoning performance of large language models (LLMs), particularly on mathematics and programming tasks. Similar to how traditional RL helps agents explore and learn new strategies, RLVR is believed to enable LLMs to continuously sel...
When they increased the samples, the base model started match or exceed the RLVR model.
"Create a video for me where two 6-year-old children, dressed in panjabi, pajama, and wearing caps, walk from opposite directions along a lush green rural path and meet each other. Upon meeting, they begin a conversation in Bengali: Dialogue 1: Assalamu Alaikum Dialogue 2: Wa Alaikumus Salam Dialogue 1: Do you know that the first right of a Muslim is to greet another Muslim with Salam upon meeting? Dialogue 2: Yes, youâre absolutely right. Itâs also mentioned in an authentic Hadith that those who exchange Salam abundantly are destined for Paradise. Dialogue 1: Subhan-Allah"
brother what
and also that lol
Do you think the Sora 2 app could fall into irrelevance now that uploading images of real people isnât allowed and copyrighted characters canât be used? Honestly, this is what almost everyone wants it for.
Hello!
Greetings
Greetings, Greg.
Same.
Hru
Doing good, and you?
.
Me too thx
My pleasure.
Sapeva che ho creato un traduttore molto avanzato che utilizza un LLM per la traduzione?
Translation: Did you know that I made a very advanced translator that uses an LLM for translation?
Yes i had seen
What do you think of it?
I didn't tried it
Ah.
You can use an HTML editor then.
Oh
Damn its awesome
Nice.
Nice job
You can type in Italian, and it'll translate to native English.
Seen
You can even toggle slang usage which will incorporate common slang that language uses.
Bht qait, u used the google translate service and then you uograded hin whit Gemini?
Nah, it uses Gemini from the Google AI Studio API.
Oh only?
Yeah, it's built-in.
Damn
please help me how to use sora 2 on lmarena
You can do so in #video-arena-1, however, it's random.
nah it did not works
Bruh đĨ
What did you do?
i see other videos
He did /nuke at least of /vudeo generate
Heh.
Bro u know how discord work?
Did you use "/video" before your prompt?
yes
And what message did you get?
i get without audio video
thanks
Average iq
My pleasure.
In ur opinion who is the best ai video generator
sora 2
Either Veo 3 or Sora 2.
Yours?
I have also tried hailuo 2.5 and the other of dream machine
I dont remember the name
Hm.
anyone need sora 2 invite code?
Nah.
Me
Anyone tried 2.5 pro deepthink?
Nah, considering I don't have Ultra.
Plenty of people, I'm willing to bet.
What's the best/newest
Maybe you could try consulting the possibility of adding it within the #1372229840131985540 channel.
Smartest
The smartest model right now would be Claude Sonnet 4.5.
Grok 4 but it takes a life
No way, ultra users only get 3 a month
Right, but with LM Arena, they could probably negotiate something.
Also how is LM Arena profiting from adding all models for free?
Tried the thinking model, not impressed
That's something the mods aren't able to disclose, but apparently, they make quite a bit of money off of it.
What about it didn't impress you, just curious?
What "hard tasks", specifically?
Cryptography
guys how can i access sora 2?
Like encrypted messages
Either through the official app or #video-arena-1.
Interesting.
but official app ask for code
Are you asking them to write code or to solve it directly?
We have Sora 2 Pro on LMArena
To just solve the encrypted text directly
Shouldn't be so hard for a robot
Oh yeah it probably can't do that
Correct, but you can always ask someone on Telegram servers for a code.
Maybe try word unscrambling first
Like for instance, I'm in a Telegram server that gives out that stuff.
Some can, with long explaining
LLMs are a bit weird with characters due to tokenization
C5472W sora 2 invite code
i wanna join
See? Just like that.
Can i have one cyphertext prompt pls
That's why I am looking to test the latest models
You've got a code right here: #general message
oh thanks man
What's the best so far? Or all bad?
Try using DeepSeek then.
gpt-5-high is the best so far
DeepSeek is very good at thinking out complex problems.
It contradicts itself, which is efficient in improving its answers.
I was shocked about this but actually microsoft copilot solved very well
Hm.
Couldn't get it to work recently
yes but its always give me error when i use it
It just thinks for a long time then errors out
Try using it on the official website then.
Do they have 3.2 exp thinking?
Possibly due to high usage the servers can't handle this much request.
Yeah, M.AI seemed pretty decent
ah yeah
I've yet to try gpt-5-pro thinking
Does it tend to error out more the harder the prompt? đ¤
same for me
but probably just a bug
should be a loss imo
because it can just timeout for stuff it doesn't know how to solve
Not just an error
i dont vote if its timeout
it's not counted anyways
but i think it's theoretically possible to cheat that way...
coz it'll answer on prompts its confident in but crash on ones its not confident in (high reasoning time)
it's like being able to ignore questions you don't know on an exam and not losing marks for it
Is the gemini 2.5 flash lite preview any good?
it's fast but otherwise not really
About the same as the old one.
glm 4.6?
Oh yeah, no way it can help me
it's actually quite good
āϰāĻžāϤ āϤāĻāύ āĻĒā§āϰāĻžāϝāĻŧ āĻŦāĻžāϰā§āĻāĻžāĨ¤ āĻļāĻšāϰā§āϰ āĻļā§āώ āĻā§āϰā§āύāĻāĻž āĻāĻžāĻĄāĻŧāϤ⧠āϝāĻžāĻā§āĻā§āĨ¤ āϰāĻžāĻšāĻžāϤ āĻĻā§āĻĄāĻŧā§ āĻāϏ⧠āĻĒā§āϞā§āϝāĻžāĻāĻĢāϰā§āĻŽā§ āĻāĻ āϞ āĻ āĻŋāĻ āϏāĻŽāϝāĻŧā§āĨ¤ āĻā§āϰā§āύā§āϰ āĻā§āϤāϰ āĻĒā§āϰāĻžā§ āĻāĻžāϞāĻŋâāĻŽā§āϞāĻžāύ āĻāϞā§, āĻāĻžāύāĻžāϞāĻžāϰ āĻŦāĻžāĻāϰ⧠āĻā§āϝāĻŧāĻžāĻļāĻžāĨ¤
āϏ⧠āĻāĻāĻāĻž āĻā§āĻŖā§āϰ āϏāĻŋāĻā§ āĻŦāϏāϞāĨ¤ āĻšāĻ āĻžā§ āĻā§āϝāĻŧāĻžāϞ āĻāϰāϞ, āϏāĻžāĻŽāύā§āϰ āϏāĻŋāĻā§ āĻāĻ āĻŦā§āĻĻā§āϧ āĻŦāϏ⧠āĻāĻā§āύāĨ¤ āĻŽā§āĻā§ āĻšāĻžāϞāĻāĻž āĻšāĻžāϏāĻŋ, āĻāĻŋāύā§āϤ⧠āĻā§āĻā§ āϝā§āύ āĻā§āύ⧠āĻā§āĻĒāύ āϰāĻšāϏā§āϝāĨ¤ āϰāĻžāĻšāĻžāϤ āĻāĻāĻā§ āĻ āϏā§āĻŦāϏā§āϤāĻŋ āύāĻŋāϝāĻŧā§ āϤāĻžāĻāĻŋāϝāĻŧā§ āϰāĻāϞāĨ¤
āĻŦā§āĻĻā§āϧ āĻŦāϞāϞā§āύ, âāϤā§āĻŽāĻŋ āϤ⧠āĻāĻ āĻā§āϰā§āύ⧠āĻāĻ āĻžāϰ āĻāĻĨāĻž āĻāĻŋāϞ āύāĻž, āϤāĻžāĻ āύāĻž?â
āϰāĻžāĻšāĻžāϤ āĻāĻŽāĻā§ āĻāĻ āϞ, âāĻāĻĒāύāĻŋ āĻā§āĻāĻžāĻŦā§ āĻāĻžāύāϞā§āύ?â
āĻŦā§āĻĻā§āϧ āĻļāĻžāύā§āϤāĻāĻžāĻŦā§ āĻŦāϞāϞā§āύ, âāĻ
āύā§āĻā§āĻ āĻāĻ āĻā§āϰā§āύ⧠āĻāĻ ā§ āύāĻžāĨ¤ āϝāĻžāϰāĻž āĻāĻ ā§, āϤāĻžāϰāĻž āĻāĻŋāĻā§ āĻā§āĻāĻāϤ⧠āĻāϏā§âĻâ
āϰāĻžāĻšāĻžāϤā§āϰ āĻļāϰā§āϰ āĻ āĻžāύā§āĻĄāĻž āĻšāϝāĻŧā§ āĻā§āϞāĨ¤ āĻā§āϰā§āύā§āϰ āĻāϤāĻŋ āĻŦāĻžāĻĄāĻŧāĻā§, āĻāĻŋāύā§āϤ⧠āĻāĻžāύāĻžāϞāĻžāϰ āĻŦāĻžāĻāϰ⧠āĻā§āύ⧠āĻĻā§āĻļā§āϝ āύā§āĻâāĻļā§āϧ⧠āĻāύ āĻ āύā§āϧāĻāĻžāϰāĨ¤ āϏ⧠āĻŦā§āĻāϤ⧠āĻĒāĻžāϰāĻā§ āύāĻž, āĻā§āϰā§āύāĻāĻž āĻā§āĻĨāĻžāϝāĻŧ āϝāĻžāĻā§āĻā§āĨ¤
āĻŦā§āĻĻā§āϧ āĻŦāϞāϞā§āύ, âāϤā§āĻŽāĻŋ āϝāĻž āĻšāĻžāϰāĻŋāϝāĻŧā§āĻ, āϏā§āĻāĻž āĻĒā§āϤ⧠āĻšāϞ⧠āϝāĻžāϤā§āϰāĻž āĻļā§āώ āĻāϰāϤ⧠āĻšāĻŦā§āĨ¤â
āĻ āĻŋāĻ āϤāĻāύāĻ āĻā§āϰā§āύ āĻĨā§āĻŽā§ āĻā§āϞāĨ¤ āϰāĻžāĻšāĻžāϤ āĻā§āĻ āĻā§āϞ⧠āĻĻā§āĻā§âāϏ⧠āύāĻŋāĻā§āϰ āĻŦāĻŋāĻāĻžāύāĻžāϝāĻŧ, āĻāĻžāĻŽā§ āĻā§āĻāĻž āĻļāϰā§āϰ āύāĻŋāϝāĻŧā§ āĻļā§āϝāĻŧā§ āĻāĻā§āĨ¤
āĻāĻĄāĻŧāĻŋāϤ⧠⧧⧍āĻāĻž āĻŦāĻžāĻā§āĨ¤
āĻŦāĻžāϞāĻŋāĻļā§āϰ āύāĻŋāĻā§ āĻĒāĻĄāĻŧā§ āĻāĻā§ āϏā§āĻ āĻĒā§āϰāύ⧠āĻāĻŋāĻāĻŋāĻâāϝā§āĻāĻž āϏ⧠āĻŦāĻšā§ āĻŦāĻāϰ āĻāĻā§ āĻāĻŋāĻāĻĄāĻŧā§ āĻĢā§āϞā§āĻāĻŋāϞāĨ¤
āĻāĻžāĻ āĻāĻžāĻāϞ⧠āĻāĻŽāĻŋ āĻāϰ āĻāĻāĻāĻž āĻā§āĻāĻāϰ āĻŦāĻž āĻāĻŦā§āĻāĻāύ āϏāĻāϏā§āĻāϰāĻŖāĻ āϞāĻŋāĻā§ āĻĻāĻŋāϤ⧠āĻĒāĻžāϰāĻŋāĨ¤ āĻā§āύāĻāĻž āĻāĻžāĻ? đ
Brokk? I'll google it
I found it, but I don't see GLM 4.6 on the leaderboard: https://brokk.ai/power-ranking
What's the difference?
ÎÎŦÎŊ ÎĩĪΚĪÎĩΚĪÎŋĪĪÎąĪÎĩ ÎŊÎą δΡÎŧΚÎŋĪ ĪÎŗÎŽĪÎĩĪÎĩ ÎŧΚι ĪĪÎŋĪĪÎŋĪÎŽ, Î´Ī ĪĪĪ ĪĪĪ, ÎąĪ ĪĪ Î´ÎĩÎŊ ÎĩίÎŊιΚ ĪÎŋ ÎēÎąĪÎŦÎģÎģΡÎģÎŋ ÎēÎąÎŊÎŦÎģΚ ÎŗÎšÎą ĪÎŋÎŊ ĪÎēÎŋĪĪ ÎąĪ ĪĪ.
Their graph is a bit crowded, but I found it đ
GPT-5-Mini seems surprisingly high
(even higher than GPT-5, lol)
The cost graph is good
It isn't quite updated
Where is an updated lb?
GM, anyone know which app to use , to recreate existing videos with your avatar ? Like video cloning but with your reference image. thx
Two good models are Qwen's S2V model, and Sora 2.
Link?
GPT-5 consistent top 1
Not sure what tasks they tested it on, but it might be a bit competition-programming (CP) heavy
Since o4-mini is scoring higher than Sonnet
Also it's scoring higher than o3 for some reason
That happens on LiveCodeBench, but I thought that was due to CP-style problems
Quite interesting, because on one Gemini models are on the Pareto cost frontier, but on the PowerBench it isn't đ¤
@echo aurora
āϝāĻĻāĻŋ āĻāĻĒāύāĻŋ āĻāĻāĻžāύ⧠āĻāĻāĻāĻŋ āĻāĻŋāĻĄāĻŋāĻ āϤā§āϰāĻŋ āĻāϰāĻžāϰ āĻā§āώā§āĻāĻž āĻāϰ⧠āĻĨāĻžāĻā§āύ, āϤāĻžāĻšāϞ⧠āĻĻā§āϰā§āĻāĻžāĻā§āϝāĻŦāĻļāϤ, āĻāĻāĻŋ āϏā§āĻāĻŋāϰ āĻāύā§āϝ āĻāĻĒāϝā§āĻā§āϤ āϏā§āĻĨāĻžāύ āύāϝāĻŧāĨ¤
Translation: If you were trying to create a video here, then unfortunately, this isn't the place for that.
Robot?
Seems able to write long code
@lusty vortex If you're trying to use the video bot please review the information in #1397655624103493813
GLM 4.6 seems to be drop 7 places after style control for some reason
āĻāĻŋāĻĄāĻŋāĻ āϤā§āϰāĻŋ āĻāϰāϤ⧠āĻāĻĒāύāĻžāĻā§ #video-arena-1-āĻ āϝā§āϤ⧠āĻšāĻŦā§āĨ¤
Translation: You'll wanna go to #video-arena-1 to create videos.
Maybe because it can write long đ¤
There's no good way of doing that.
I wish there were more with multiple languages
Models will usually tell you that they're a previous version of themselves due to the knowledge cutoff.
LMArena could help with that if they could make it easier to run code on the website
Seems counter productive
In a a way such that even a layman can run C code, like Web Dev Arena
Like, for instance, if you ask GPT-5 what version of GPT it is, then it will respond that it's using GPT-4, even though that's not the case. That is due to its knowledge cutoff being somewhere in January 2025.
If people had to run the React code themselves on a NextJS instance, I think the number of votes on Web Dev Arena would be 95% lower
I have absolutely no idea. Probably because it would use up too much money trying to keep up with every day events.
I feel like nobody asked that question before
Then again, I don't see how that could possibly be an issue, considering these types of companies tend to make billions upon billions of dollars every single day.
Well if there was a Copilot-like plugin (LMArena had one for code completion), then you wouldn't need to create the benchmark yourself. It'll be real codebases from the real world.
Well, no crap.
Because it's on Twitter, so of course it's going to be updated every single day.
I'm sure companies like Copilot, Windsurf and Cursor have a ton of internal stats
Possibly to encourage the usage of later models?
The later the model the later the knowledge cutoff.
I think votes are trustable as long as we know it was executed
Perhaps they use that as a persuasive method to get you to use new models because they have the most "recent" information.
Thought so yeah, then again why wouldn't they update it to actually be latest
Also there are basic telemetry that can be reported such as number of errors from the linter, etc.
Because greedy corporate companies like those always like to make a good buck off of scamming people.
Well, not really, you can assume that the voters are voting correctly if they're executing the code.
LM Arena is the only reputable one
You can get basic metrics like number of compile and runtime errors per language, to start with.
That doesn't charge $200 a month to use AI lol.
True.
They actually care about the customer experience, unlike other AI companies that are dishonest.
I would actually support this site rather than them
True.
And we already are, in technical terms.
By voting for which model we think is best.
Yeah I guess data is worth a lot
Indeed, it is.
Well it's real human preferences ig
Companies spend billions on market research đ
Rather than simulated ones.
Makes me wonder if I should create a real-time updating bar graph showing the preference of different AI models on different benchmarks.
Whenever they get a new score, they smoothly update by moving up or down.
What granularity?
And whenever a new model gets introduced, a new space is created for that model, and it is assigned a color.
Hey guys their used to be an icon on the left panel which allowed me to go to a page which showed all the content I had created here. Its no longer there. How can i access the page now please!?
Wonder what that is.
I support this idea!
Which "page", exactly?
Like a live score website
SO all the photo to video content I had preformed. It would all come up in a window in a row.
It's possible ig, but they're using MLE to fit the Bradley-Terry model which is slightly different from chess ratings systems since its basically trying to solve a global optimization problem on the entire dataset rather than updating turn-by-turn, so might be a bit inefficient unless an online ELO system could approximate it
That way lmarena doesn't have to update the leaderboard manually.
Like one place to see all the videos one has created.
You mean #video-arena-1?
They could add web-integrated video creation
It looks like that however it is only your works and no one esles!! You can scroll back and redownload all the content you have created
So the DM bot...
See on how the left panel now there is an icon that looks like a bin with the number 4 on it? It was above that same icon!
I'm not quite sure I understand.
I dont know where it went đ I cant access all the previous videos i did.
Take a screenshot.
Is there a way to see the history of photo to videos one has created?
You can access the Discord Bot's message logs.
Here icon on left with red 4. There was one that used to be above that exactly the same and I was able to access the history of content created
Thanks! How do i do that please.
Thank you so much!!!!
No problem!
What's the best AI for coding which is not claude and free?
Either DeepSeek v3.2 Exp or Qwen 3 Max.
Glm 4.6 might be a good start
Z ai?
Yes
Okay, thank you, i will try deepseek and z again as i am using qwen now.
Another qus, Is lmarena working fine now?
Yes.
I'm using it right now and it's working fine.
Ah.
GLM-4.6 without search and thinking, good or bad?
Hello
Crazy, just like me.
Greetings.
Sora is this good?
It's very good.
It's able to mimic different shows very realistically.
/video prompt make the city sink under water
Why do you want the city sinked?
for a short movie about atlantis
Deepseek has a knowledge cutoff of 2024 january
It's unusuable for my projects
Why so outdated knowledge?
Well, if you wanna use the most recent knowledge, just enable its searching mode.
anyone who wants 150 images of nano banana a month?
No need, considering Dreamina already offers a free, rate-limit-free version of nano-banana.
what about seedream 4
Darkness, which model is best for cryptography would you say?
Try DeepSeek.
If you'd like.
Seedream 4-2k, please return đ
Which model is best to use for this? Code Llama?
You can always just use Dreamina to use Seedream 4 2k.
It's free on there.
And there's more customizeability.
i make a game guys!
http://localhost:8158/__unsaved_file__
And surprise, surprise, it can't be used.
I think the link is broken
I'm not hacking you guys
Because a localhost network requires hosting one directly from the user's terminal.
In order to actually publish your game, you'd need to buy a public domain.
where is this dreamina-ai.com ?
Lemme gladly direct you.
Here you go: https://dreamina.capcut.com/ai-tool/generate
I remember this website months ago, couldn't get in
It's available here.
GLM4.6 Air and V ?? Some models with s đļ
Sora is insane in few years we can do anime with vedio generation models
True.
This is nothing
See this
Yes ik
Did you see what I sent
No i saw other ones i can't open cuz here tiktok is banned that can u download and send me
This is real scene from bleach animw
I think sora is just edit
Wait a minute
Bro you cannot do video to video in sora 2
Does it have editing capabilities?
Holy moly this is insane
Yes
This is nothing
Yea I can understand fully what you mean
Editors gonna lost their job if sora release the editing capabilities
Realize what?
That is exactly as the real clip in the anime, sora made it exactly 1 to 1 as the anime + with that badass edit, sora doesn't have access to anime clips ofc
Bro sora don't have editing capabilities but It can recreate scene from anime what if it got editing capabilities
Soooo real we can't tell difference
Well, it can also do the same for certain sound bytes, like the San Andreas intro part with the protagonist saying "Ah s**t, here we go again.".
But this anime clip it was handrawn every frame every drawing by animators and like soo much stuff
It's much more complex than some game
Wait a min what if people give models to ability to view video analyst the video content we are totally done
Well yeah.
@prime mulch can you help me find which ep is this from
It's hard to create a anime scene if you compare with games
This is from the turn back the pendulum arc but which ep
Can u find that girl name
It's hiyori
It can be, yes, but then again, there are already image generators that can do that efficiently too.
The man is urahara
This is the real clip
Should I use deepthink or search when building my llm and asking about info
@hollow imp @robust yoke
DeepThink and search.
You can do both.
Yeah, I can see that.
It's pretty crazy.
Can only select 1?
No?
What if they give ability to view videos to models like chatgpt
Oh nevermind yeah!
Regive bro, you cut the first 10s
Videos to models?
Does deepseek have a memory option?
Sort of like gemini watching yt videos
I didn't cut any i got this from YouTube
Greetings
Maybe.
They can see images and tell what in those images like that what if they can see and analyse the video
Saluti, amico mio italiano.
Hm.
@prime mulch wait a sec don't go offline
Is anyone else having issues with LMarena today? Every few turns I get an error message, I refresh the page and it asks me to verify Iâm human. This happens to me but usually itâs on occasion and it only asks me to verify every other week or so.
But today itâs asked me to verify about ten times now. Ping me if youâre also having this problem
I believe that's normal.
Sometimes, it'll ask more frequently, and other times not.
Unsure as to why.
I face this everyday i thought that's a normal because of website traffic same it's ask sometimes and other time not
Bro tells me to wait and goes offline......
Prime rage baiter.
đđ
Well it's not downloading
I'm downloading episode 207 dvd iso
Why u can use anime websites to see
Or use telegram to download
Where do u download
Do you know torrenting?
I use torrent for download movies
@prime mulch what is the anime you're currently thinking about watching
But i only got 1080p links
@prime mulch
Let me give you a blu ray release of that if it's available
I don't have anything to watch so i switched to manhwa and manga now i plan to read light novels
What do you mean
You don't have pc/tv to watch now?
I have and i watched more than 500 anime in 3 years
Did you watch Naruto?
Weeee
Then i start read stories in manga
@prime mulch
Ofcourse yea
Bruh that is nothing
I know
Dal momento che tutti i moderatori sono offline, posso esprimermi in un italiano fluente.
My Myanimelist shows that I've watched 1000+ animes
I watched 1000+ episodes of onepiece
@prime mulch
Where are you from?
I read more than 1000 including anime
Tamilnadu,india
Do you know hindi
No
Are you jee aspirant
What about Indonesian?
I'm a BE computer science graduate
Third tier college
Do you plan on giving gate
I don't fail in any subject still i don't have job
No
I believe in pg not give high salary
That means little bro
@prime mulch Thampi I'm only 16 years old
I'm 23
Just call normally
Anyways, coming back to the 500 anime you've watched, is code geass, monster, banana fish, gurren lagann in there?
Doorbell camera POV â ultra-wide fish-eye lens typical of a Ring or Nest smart camera. The image should have subtle barrel distortion and mild digital grain, resembling real home surveillance footage. A quiet suburban porch at night is softly lit by warm overhead lighting and the flicker of jack-oâ-lanterns with carved grins. A large wooden bowl overflowing with Halloween candy sits invitingly on a small table near the doormat. The setting looks calm â but a tall, shadowy figure dressed as a ghostly scarecrow (stitched burlap mask, glowing orange eyes, long straw-like fingers) lurks just off-center, half in the shadows near a column. The camera catches faint motion blur from the scarecrowâs head turning slightly toward the lens, as if itâs aware of being watched. Optional faint timestamp overlay in the top corner: â10/31/2025 â 08:42 PM.â
Visual Tone:
Nighttime ambient lighting
Cool blue-black shadows contrast with orange pumpkin glow
Slight chromatic aberration from lens edges
Realistic doorbell-cam framing and vignette
Atmosphere: tense but still, cinematic realism
I didn't watch banana fish i watched other animes
@prime mulch You can't be serious you missed out on all the good anime
Code geass?
Yes
What happens in the ending
He dies
He have the power to manipulate
I think i watched only first season
Then watch the second season this week
We can speak about anime in dm
W.
that confidence to claim "best model in the world"...
better than claude?
it probably just came up
no im in canada without vpn
or........
i lied and made it up
with inspect element

you guys got baited
Rage bait of the century.
@robust yoke @leaden sun đ
true...
u guys wanna see my system prompt for coding claude
well not the prompt
the results of the prompt
āĻāĻŽāĻžāĻ āĻāĻŋāĻĄāĻŋāĻ āĻā§āĻāĻā§ āĻĒāĻžāĻā§āĻāĻŋ āύāĻž
āĻā§āύ āĻāĻŋāĻĄāĻŋāĻāĻāĻŋ?
nothing
Well, it must've meant something if you pinged me just to see it.
I don't see what significance that message holds to you.
I've recently seen the Mudahar video on it.
I've since gotten rid of it.
ok
is it better to use claude 4.5 thinking with perplexity pro or lmarena
im not going to pay for claude.ai
a perk with perplexity is that the model can search online
Either-or is fine.
Good, and yours?
ok
As in it's "okay", or is that simply a response?
Why? đđ

No need to apologize.
Anyhow, glad to hear that.
thanks
My pleasure.
hello
Greetings.
Al momento non si hanno ancora notizie, purtroppo.
Guys
Hm?
I feel like there's a bunch of models missing like qwen flash and the newer gemini flash lite version
We are done
Yeah.
What does "hard prompts (english)" mean?
Hard prompts?
hard prompts for AI in english
but Idk what is a HARD prompt đ
Something very detailed, perhaps, heh heh.
like some complex scientific problems etc.?
Sure.
Is there any other way to access gemini 3 than A/B test?
I mean the X28 model or something like that
People call it gemini 3 pro
I agree.
So am I.
Gimme those gamma rays
I'm gonna test it and see if it can code up a game.
That's quite true.
2.5 pro can code a game
Opus is the server melter
Not a very good one, though.
In 2-3 years like 50% of new games will be made by ai i think
For sure.
50% of new games will be ai. Genie
100% by AI is what you meant, right?
The way we make games now is so inefficient
Gta6 costing over a billion
Silly stuff
It's a bunch of pixels
Stuck to a physics engine
With some voice acting thrown in
Ai will humble our notions of creativity
Ai slop will be more creative than peak human creativity
I heard the same line when gpt 5 release
I hope this turns out really well
Then just wait don't over hype then sad
Is lmarena down again?
Nah.
uhm, when gpt 5 arrived we saw what happened
Gpt5 is good at coding
Change the browser
Well, you know the drill.
Not that good
imma wait again for 10-20 minutes
Sounds good.
But yes it was overhyped
Gemini 3 is the one
Don't doubt it
Don't let Sama's failure make you doubt Demis
If we got gemini 3 I'm sure they will release free api in Google ai studio
They didn t release zennith and hide it
Hm?
it worked when i turned on my vpn
What did
Majiq.
LM Arena.
maybe a prob on my router or sum?
Could be.
Oh true
i tried it on another device without vpn, still the same prob
but oh well ig this is the fix for it
Neat.
UHM
I'm constantly on VPN, never seen any problem.
Eh?
Which AI is the best for custom LLm building?
idk wtf is going on
Petition for qwen3 flash
IT NOW WORKED WHEN I DISABLED VPN
and refreshed it
now its even faster
????
Er...
XDXXDDXXDXD
Curently trying ollama 13B
-mild confusion-
has anyone here fine tuned a model?
?
Nah.
Currently doing
also, what's the best ai model out there for coding? grok? claude? gpt?
whats the process like
Sonnet 4.5 according to lmarena.ai
LFM 2 2.6b >>
and what are you tuning it for
Claude 4.5 or DeepSeek v3.2.
im now comparing the secret a/b gemini 3 checkpoint in ai studio to gpt-5 codex high, and i think gemini 3 is waaay better than gpt-5 in landing pages but im still dissapointed from gemini 3 result.
Trying that next!
Claude and GLM4.6
trying ya'll suggestion for my next project
LFM 2 8b a1b I never tested, but I think is good too
I wonder how one can access this secret checkpoint.
a/b test
Well, that's not exactly very descriptive.
hello
How do you get gemini 3 results?
hello friend
Is it just random in text?
just send requests to 2.5 pro until you will get like which model is better and sometimes you get gemini 3
Ah, okay.
lets do a blind test,i will send two sites and yall will tel which is better and then i will tell. prompt was: "Code visually appealing landing page for cpu manufacturers, no gradients. In one html document. It should be $10K design."
Zero-Shot
no iterations
The AION one is A, the other one is B
Index 9 immediately.
Compare 2.5 pro to 2.5 pro?
No, to Gemini 3, apparently.
Not available to select
B is cleaner, more efficient but A has more features and looks cramped
The Index 9 was 2HT which apparently is Gemini 3.0 Flash, and Index 10 was GPT-5 Codex high.
and the gpt-5 didnt listen to me and it did add gradients
i think flash won here
How did you get gemini 3 in compare mode?
also in gpt-5 there is too many things that are unnecesary and colors are not great like white text on light green buttons and icons dont work in gpt-5 response.
i told like 5 minutes ago
in aistudio
this
2hT in console means flash and X38 or 28 means Pro
you will get rate limited to next day until you will get a/b test so it isnt practical
Hello.
Greetings.
hello
I new here
Hello and welcome!
I am new here
Well, welcome to the server.
Thanks
We hope you have fun here.
Yes.
Sure.
has the qwen3 4b 2507, but Idk you need a speed AI
I need a deep reasoning AI
hmmm, for agentics uses?
for agentics use the gpt oss 20b still the better
No for my crypto project
hmmm, how many RAM do you have? gpu or cpu? what is your pc config?
normal
Lemme check.
Messages not going trough
Seems like it's doing good.
for me it's working
But you need a fast AI, right?
Fast but also deepthinking and reasoning
My name is Ammar and I'm a beginner on the Discord app and I want to learn artificial intelligence with you.
hi everyone
good
Why dont we intergrate
Greetings.
LLMs reasoning with video
Thank you very much my friend
That's already an integration.
oh it is?
Yeah.
Did you know that AI can create HTML files that you copy and paste into Word and the Word file is 100% professionally formatted? I discovered this recently
Can someone help me learn artificial intelligence?
Should I use Llama-4-Scout-17B-16E-Instruct
for what specifically do you want to do?
llama 4? đ
No?
just use the gpt oss 20b, have deep thinking and very good agentic use
I am still a beginner with you in artificial intelligence
or qwen3 30b
I love designing and editing photos very much using artificial intelligence.
You should learn some basic photoshop
Does anyone know, when Claude-4.5-Opus-Thinking will come out?
(should be the top coding model)
hmmm, talk to me, what is your biggest headache with AI, with it I can start with you
GLM 4.5 Air is a gem because it fits perfectly on 128gb of system ram
maybe never, don't exist opus 3.5 for example
bye everyone
I also want to learn artificial intelligence and the English language as well.
a tip, the qwen image 2507 It's 100% free
Does anyone here speak Arabic?
Just use my translator.
I cant speak Arabic cuz its broke the rules
I am very weak in English
use gemini learning guied to learn english, is a very good teacher
relax, I'm weak in english too, but I use gemini to learn english
;/
hi larry đ
Hi :)
but in this server we just can english okay?
Yes
đ
I think he is resident of Egypt
Why isn't deepseek working for me
anti-bot
Unfortunately, everything here is in English and I couldn't find anything to translate here.
try delet the cookies from the site
Refreshed multiple times
Would that delete conversations
Don't want that
don't do that
I think it's just simple anti-bot error
What his happening here