#general
1 messages Ā· Page 203 of 1
on what?
Theyāre super strong guard rails, especially on cameos
Veo?
Nvm tried
just make bettermodels
You know you can use emojis as prompting
š¤·
Knowing the sequence and combination
they removed the word cameo
Cameo has filed a trademark infringement lawsuit against OpenAI in the Northern District of California. Cameo alleges that OpenAIās Sora app launched a new feature, also called āCameo,ā that allows users to create and share realistic AI-generated likenesses of themselves and celebrities in personalized videos.
now its "cast in"
I know you can even use binary and backwards language
I didnt try that
Thereās a lot you can do, but itās not necessarily obfuscating, which is sometimes critical, but itās more important the words you use and the structure of your prompt and how you structure your sentences or whatever you wanna call it
The combination sequence and the choosing of words is probably the most critical
Itās crazy how sensitive they are to like language and just characters in general
See
So now Iām phase 2 (the video was generated using the image without a prompt)
That almost sounds like Patrick star from SpongeBob š¤£
are we serious
Cool I wonder what's next
With that as reference?
Imagine gpt 5.1 pro announcement
movie names are not blocked
is Gemini 3 taking forever for anyone else?
are you taking about image creator (banana nano ) ?
Can't say I've been hearing similar reports. Looking into.
For text, right?
when is gemini 3 flash
No works
1/2 and then it fails
Iāll try to generate one more one out of three to see if the air persist
yeah image too
yep its down for me
Hey, Iām curious for those that itās down what browser. Are you guys using?
Yeah just text. Was asking for advice on google sheets haha
Chrome.
Ty
But I just tested through my paid API on AI studio and Gemini is fine there
Did the goku one fail?
Do you have any extensions on your browser?
How long are these generally taking? Even for simple prompts?
5 minutes at least. I keep having to refresh after a bit.
So now you are going to reference the super sayan form?
Are you noticing the same when using a different browser by chance?
Iāve been reading online about the errors and some of these errors people have been having all the way back to 2024 on Google lol
whyd they remove claude
And a lot of what Iām reading people are using chrome
Iām on a Chromebook so Iām locked into just chrome but it does it in both normal windows and incognito
red line of death
is what i think is happening
This document explains how LMArena-Api handles various failure scenarios and recovers from errors without losing requests or corrupting system state. The system implements sophisticated recovery mecha
Gemini works fine in AI studio for me. I just use the arena for smaller requests as to not use up my paid API
yall
That kinda makes no sense to me while you use it for smaller
which one is better
Same, anytime I need an API for it I just go to lmarena
qwen or glm
theres an lmarena api?
No, this is probably outdated. Iām not sure but from reading it. It sounds like a lot of the same kind of issues people are having.
It has something to do with chrome from what Iām reading also
Maybe extensions or something or people trying to build scrapers
Okay, thanks for the info. I've tested a few text & image gens and can't say I'm seeing an issue. However, noticed another user havign some unrelated issues also using a Chromebook. I'd be surprised if that was the case, but starting to wonder if LMArena doesn't perform well on Chromebooks.
actually maybe each chat only has a number of like api token whatevers per conversation and once that is reached then it wont work
I donāt think that would matter because isnāt brave, and Firefox also built on the chrome engine
Because earlier somebody was having the same problem, but it was with the brave browser
firefox isnt built on chromium is it
Iāll let you know if I notice anything else
I need to verify that
1 sec
I could be completely mistaken
Ofc
š¤£š¤£
Haha
Dude, the Internet is so polluted
Trying to find some basic information you gotta read through like 6000 words
The question of whether Firefox is Chromium-based is a fundamental one in understanding the landscape of modern web browsers. The short answer is a definitive no. While the web browser market is heavily influenced by Chromium, Firefox, developed by Mozilla, stands apart with its own distinct engine, Gecko. This article delves into the technical ...
gx is chromium
Itās ridiculous looking this up
Yeah google search sucks these days.
google top search engine
The browser wars have been a staple of the tech industry for decades, with various players vying for dominance in the competitive landscape. Two of the most
Guys is deepseek 3.2 updated on their official website
Itās like I want a simple answer is it or is it not? I donāt wanna read to 6000 paragraphs.
And come out more confused
Gotta wade through neck deep SEOs, Ads and Reddit results
But it can't find the one your looking for the best one for me is: Brave Yandex
Thatās how it should be
Do you know what the worst is baking recipes
If you wanna look up how to cook simple chocolate chip cookies
Dear Lord, have mercy lol
Pineapple why cant I say pee?
claude down here too? wtf is going on
Honestly guys, I think Iām going crazy
Gotta read an essay about their dead grandmother and neighbourhood gossip to find out how to make apple cobbler
I feel like Iām moving every single day over and over when I come on this discord
š
Thatās a problem with the Internet
didnt answer my question just straight away started talking about spying on data
I shouldāve never gone down this rabbit hole
I miss the good old days when search engines actually searched. Bring back Jeeves
I mean who made the Model?
One day in the future, my friendās AI is going to be no different than search engine like Google
Except a more sophisticated and controlled environment
We each search will cost a token
I hope itās not Google that wins the AI race I pray to the good Lord
israel probably
I hope itās meta
google wins the ai race and the next time you go on calculator and do 2+2 it starts a long winded explanation on why it equals 3
šš
wtf howd he do that
Alot of people use Google Features and softwares and OpenAi doesn't have a single thing. So the race is just for Google at this point
You guys wouldnāt believe the statistic on how much Internet traffic chrome gets
Just the browser itself
A ridiculous amount of global Internet traffic
It pays Apple like billions of dollars so Google is the default search engine on Safari
I like that 2026 part
Why not Firefox
gotta be ahead of the times
If you ever have a time look up ChatGPT 6 from the year 2024
The stupid predictions people had lol
EdgeHTML dont fail me now.....
god i miss old microsoft edge before microsoft cracked down on minecraft modding
Thatās the other Saint that owns large amounts of infrastructure
Weāre screwed either way we look at it Microsoft or Google
An apple is trying to jam its way in the center
Meta above the ski
i witnessed someone spanking it in vrchat so i dont really like meta anymore
What is it about these tech companies that makes them go so out of touch
I think itās greed
Itās hard to even say this, but Google was once good and worked and did what It needed to and youād get results and discover things
Now I feel like Iām in a prison. I canāt escape with no alternative.
Hello
hi
man I hope they fix speciale soon it was kinda unique
How was it?
Why is Deepseek 3.2 special not in Lmarena, even though it says in the announcements that it was added to LMArena
In a word? Specialized. If I ranted about some game or story idea to any other model, it'd react in kind. speciale would try to organize my thoughts into a surprisingly well organized pitch document⦠but would usually time out before it could finish. sometimes it'd time out before it finished thinking. Cuz it really did spend a lot of time thinking. But you could read its thoughts while it was thinking and they were kinda solid.
I had the timeout issues with V3.2 as well
yeah the speciale timeout issues were most of the time though.
removed till speciale gets fixed with its timeout issues and hallucination
I see it as problematic if those matches are forfeited
Because it's essentially filtering out prompts that the model struggled with
Okay
Thanks
I think I understand it now
Yeah, thereās really nothing to it, bro you just gotta be smarter than the AI
Source?
Hi.
As Blockchain and AI developer, I build LLM pipelines, advanced embeddings, agent logic, vector-powered search, full retrieval stacks.
For Web3, I design smart contract architecture, advanced EVM tooling, and cross-chain integrations.
I also work with high-performance GPU runtimes, modern orchestration frameworks, and secure distributed infrastructures to power production-grade, bulletproof systems.
Need rapid prototypes or comprehensive product development in AI and/or Web3? Let me help you realize it.
Something unique about the way I work: I have strong technical expertise as well as a keen intuitive sense of how my models are āthinkingā.
Patterns, rhythms, āpersonality flowsā, etc. Iāve built enough AI systems over the years that Iāve developed an intuition for how they tick and have come to recognize the subtle signals long before the average dev would.
Combined with rigorous engineering, this lets me deliver extremely powerful and reliable solutions.š
Is it me when you vote it doesn't show the models?
this is openrouter
this is right on their front page
LMAO
YIKES
mistral might catch up to openai at this rate š
Anybody whoās in the first position like open ai was he gonna have the same headaches?
elaborate
In biology, the "Red Queen hypothesis" states that you must run as fast as you can just to stay in the same place. The number one spot, OpenAI sets the benchmark that everyone else (Google, Anthropic, Meta) is specifically building their ai to beat.
The number one spot becomes the single target for the entire industry. Competitors don't have to discover what works they just have to replicate
Being in the number one spot is like being a big lightning rod
Lawsuits regulation, public opinion public backlash
You canāt really take any risks and you donāt have a lot of moving leverage for mistakes
And I think in the AI industry in particular, I think it requires risk and being innovative, but thatās hard to do when you have hundreds of millions of users and people that rely on it and use it for work and other means, you canāt just make those kind of big changes or take those kind of risks without affecting your customer base
The tall poppy syndrome aka losing the ability to fail quietly once youāre the champ, you canāt be the underdog
i agree in a way but openai has the tools to just
not fumble the bag
what kind of business decision is it to wait almost an entire year to release a finetune of your model (5.1)
seriously?
Well, thatās what we were talking about earlier the context
Because thereās also a computing and a power shortage
Not only them, but Google and a lot of these other companies are bottled neck
while that is true, openai gets sweet deals from nvidia
So is everybody else?
you dont understand
I probably donāt youāre right
nvidia is very, very close with openai
theyre one of their biggest investors
in return for probably access to some of their private research, nvidia gives them great compute deals
which is why openai never had to bother with optmization
and thats what propelled them to the level they are at today
I think you should look into Nvidia a little more lol
And the business that they have currently going on
And thereās one more company you should look into
i know nvidia is also having troubles
their 50 series was a terrible launch
Have you ever heard of the company called core weave?
This is understandable. Gemini3 is still recent + Opus4.5 + Speciale
not to my recollection
Well, if you ever have the time check this out
Sign up with https://start.moomoo.com/00w4m2 to get up to $1,000 in NVDA shares with a qualifying deposit + up to 8.1% APY cash sweep for 3 months.
This content is a paid ad from Moomoo Financial Inc., (MFI). The content creator receives no additional compensation from the firm and MFI makes no representation of completeness or accuracy of the ...
This breaks everything down really really good
Just how intertwined and the deals everybody has
Including Google Microsoft, it just shows how everything is set up. Itās pretty startling
But yeah, core is a huge contributor to the problem
Since theyāre the literal infrastructure of AI
hot take but if openai falls we are going to see a huge market collapse in the tech center
On a somewhat related note... I think nvidia stock gonna burst
i mean
The government will bail it out
when r1 released 5 trillion was wiped from the stock market
This current bubble is unsustainable
oh
lol
I would take it with a grain of salt since its openrouter
but still
probably not
The money in AI is so intertwined. The government is gonna have to bail it out lol
oai got hit really bad
If you ever have the time to watch that video
It kind of breakdown exactly what I mean
still kind of wild
30% decrease
In the glorious new world of AI Everythingā¢ļø, we need more processing power than ever.
That means more data centers, bigger data centers, and more electricity than you can imagine.
Naturally, this all costs money - but not to worry! We'll just invent new economics to fund it all.
In fact, the insanity at work these days in the AI and data...
This one focuses more on core weave
no wonder they are panicking
Just look into core weave
Nvidia stock is also rising because you have everyone making silly startup ideas from API wrappers. Once people realize how useless most of those are, the trend is going to reverse
oh absolutely
the ai market is a volatile one because ai has no physical value
just like crypto
Actually, thatās a little misconception
Because the datacenters and the chips themselves have physical value and demand
but the ai itself does not
True
i have a feeling smaller ai is going to become more and more popular as the capability advances
But the argument could also be said about a lot of other parts of the Internet
For example, ads, they donāt really have no value unless they sell a product
And I guess on the other hand too, you could argue that the power being used
ads are invested in in order to pay back the cost with more sales, so yeah ig but its an ancient concept
Well, thereās nothing new underneath the sun
The industry is very complex and itās very shady and very dirty
If you ever do have the time and watch those videos, youāll understand about the venture capital and the investors
And the circular deals
Google is also in because they financed some of open AI data center usage
Blah blah 300 trilion to Israel
can the sun just collapse
The fact they added ads into the app is crazy
honestly (i am no economics expert) but i see one of two things happening after this collapse
chinese/foreign ais gain a ton of activity and investment (cheaper to integrate and efficient) and Google + Anthropic become the titans of SOTA
or
extreme recession resulting in the ai market being pushed back a year or more
I think itās naĆÆve to think that the Chinese will always have open source models
Just like open, AI went from a nonprofit to a profit
I think we will see the same with Chinese models once the objective of market consolidation reaches its goal
This is just a guess but if opening I failed weāre talking about 2 to 3 trillion wiped out from the market
It would be a AI winter freeze
Equity, unpaid debt, market cap contagion
and 2 trillion to israelš
Are you in eastern Europe or
Which video model do you guys think is the better of yesterday's releases between Kling 01 and runway 4.5?
You have to use their new portal use new portal
How do I do that
Have you had a chance to compare 01 to other leading video models?
man I miss it when their credits were bugged and I had like 900 credits for free
They have video reference now
Kling AIās new Video O1 model is here ā and itās a major upgrade for AI video creation.
Try Kling O1 ā https://elevenlabs.io/image-video?utm_source=youtube&utm_medium=organic&utm_campaign=model_release&utm_content=kling_o1_just_changed_ai_video_forever
In this video, we break down everything thatās new in Video O1, how it works, and ...
Theyāre calling it the nano banana for video generation
You know itās interesting about this company. They started off in a university.
we're doomed
Y
I would use it if THEYR SERVERS WERENT FULL AND THEY ARE MAKING ME PAY 6 BUCKS
Itās a high-quality service
I lost 4 euro s gambling today so nuh uh
Yeah, this is no different than gambling to some extentš¤£
No, itās true. Everybody knows that generation is no different than gambling or a slot machine. You never know what youāre gonna get.
im sure some will be proprietary but i think glm, deepseek, and kimi would rather be open
it gives them a unique market
makes sure they arent saturated
I dont think it can make copyrighted ips just from text with accuracy
Yes, I get what youāre saying but Iām saying what happens when that goal is achieved and they reach market dominance or whatever their objectives are
Yeah but I lost
Part of the territory champ
it depends if they have a super duper good architecture
To be honest, I know just as much as you Iām just speculating
All I know at the end of the day is business
And whatever is best for business that will prevail
at the very least deepseek and glm will probably remain open because a lot of deepseek and glm's market comes from roleplay
uncensored roleplay
and for it to be uncensored their models need to be hosted on other platforms
ie novita
There are sites that rely on that doing NSFW characters
It doesnāt mean they have to close them fully what Iām saying is if the opportunity arises, theyāll create new models that wonāt be open source theyāll keep the ones that are open obviously available
Is it addicting?
But if you want new updates or new features or new models, then if they choose to close it off, it doesnāt matter who relies on it or not if they want a new update theyāre gonna have to to pay
Or they can keep running it on outdated models God forbid
Meta was once open source and now there are rumors that their next big model is not gonna be open source itās gonna be close source
We saw from open AI is an example from nonprofit to profit
So this whole industry is very volatile and anything is possible, especially with how much money is on the line
Oh lord
shhhh
we dont talk about those
shhhh
OF COURSE A MAFIA BOSS IS ONT HE FRONT PAGE
š
I scrolled a bit
makes it even worse cause ytou cant fine tune it now
The market for that kind of material and even NSFW companionship, and everything in between is probably gonna be worth more than probably coding and agents
i hate meta
it already is, check openrouter
I wouldnāt be surprised look at the adult industry on the Internet
deepseek v3
Massive, massive, massive, massive massive
So are u a girl? Cus this seems targeted for girls who want ai boyfriends
see, i just like deepseek because it has decent prose and i can write nice romance plots
oh brother
you havent even seen the other bots
the female bots are down right AWFUL
Answer the question šš
I guess
janitorai is also the highest quality mind you
like these are fully done bots with people who know the inner workings and how to squeeze the best out of certain models
I already have my own poison using chatgpt
You gotta keep it one Hundo
Donāt let it make you donāt let it break you
Everything else will be right
Back when Character ai was popular i got scared when it wrote in () and it said the creator s thoughts on your chat
I think thereās a lot of closet ai addiction
People who are too ashamed to say that theyāre addicted to AI or share or companionships with AI or whatever the hell you wanna call
Youād be surprised how many people here are addicted and they donāt even know it
nah i kinda like worlbuilding and writing and use it to make my own story for entertainment
You gotta look at it in terms of time spent
If youāre spending large portions of your time with AI or even the computer for that matter or the phone, whatever
Months I think of context and tokens
Well, youāre not the only one Iām also guilty as charged
speaking of character ai holy god they fumbled
Well, they also had that lawsuit
It got weird with the calling stuff
Where that other kid unalive himself
3 MILLION TOKENS??
Is that for character AI?
Your stats?
yeah
Which yours?.
3 milion šš
Damn, letās break it down
No lets not
Well, Iām curious how much that translates per hour
One second Iāll be right back. I gotta do some crunching.
i like doing silly rps with deltarune chars
sometimes i make a cod character file tax returns
and uh
Is that per month for this this month?
some unholy stuff on the side
for 24 days
So you payed 0.567 cents?
OK, letās just say this the average person types around 35 to 45 words a minute
Weāll give it 50 for argument sake
about yeah
so ur a robot
Let me check my app time too
Thats 72 k words every 24 hours
X24 days
1,728,0000 words
But weāre also not including copy and pasting stuff and doing research
Which could easily eat up tokens especially if you do a lot of editing
Damn, thatās a lot of tokens
aboutlike
twice??
Ehh so did my own research only on my phone I use chatgpt for like 7 8 hours a week so idk rlly
Yeah, so if you use it for 24 hours straight for 24 days straight thatās 1.7 million tokens
take that into account
Do you sleep?
Pretty much two ChatGPT plus subscriptions used to max
But donāt feel bad thereās some people that probably are double that
IM NOT DOING THIS 24/7
on my SOUL
i use it like twice a day
MAYBE 3
Oh, I believe you
It would be impossible for anybody to use this for 24 hours a day for 24 hours straight
Somebody wrecked up 1 trillion opening tokens lol
jesus christ
I mean from what Iām reading youāre pretty so high up there
So that translates to you having a lot of XP
@echo aurora Is speciale still broken? It's been a day.
Plus, donāt know the things consume more tokens, like searches and stuff
I wish they would just charge by the usage of power in watts or straight compute processing
Instead of tokens
So you get charged by the amount of power and compute intensity that was used
Like electricity bill that would be easier
I only give it a draft and tell it to write the whole scene
Yeah it's currently unavailable
Duolingo? That's unexpected
Pineaple why is pee with ss censored
Peess?
I get it now thank you
Because its a filter word
And the word go on is also censored if u type it together
Yeah, with this, I have to agree
Dont want this server getting smited by discord
Yea that makes no sense lol
Why not?
Oh wait itās on the discord. I thought you meant the lm arena
Itās a standardized, form of behavior, modification, and adjustment
Cus gooon ing is a bad word
Itās not the word itself, but itās the connotation that comes with it
Itās just the way
The Internet is everywhere pretty much most platforms
I cant go onšš
how
I thought their lessons were mostly pre-written
Ooo could this be the reason I CANT JAILBREAK
hii
These are just concepts of my observations
And the way the platforms are designed themselves
Ohhh
Sorry, I wish I could make it more entertaining
Did you made a lesson on thisš
Yes
Lmao good luck making something that makes it not boring
You are becoming a teacherāļø
Itsa good listen while you're doing stuff
Its more like a podcast LOL
Add gameplay and phonk music
What thereās the other side too
Ola to all, seeking some help or advice as new to this discord, I watched this video title and followed thru to this group ..but cannot find where I need to go to find the details described in the video Make LONG AI Videos (FREE & UNLIMITED)
Wisdom is earned through patients
Thats how you will always be majority of the time
if anyone can advise or help me I would very grateful
Yeah, it is what it is
What ze heck
It has the ai writing style
Oh yeah, itās my lazy attempt at trying to articulate complicated idea ideas
I mitigated all the hard work without the effort to a machine
Listen man, I make music and anything I make I feel is never enough, especially when showing people lmao so I feel you
laziness is not good
And expected it to be entertaining
So was it supposed to talk about filters or the market?
No, itās not and the result of my laziness. Is it being lame
Wouldn't have guessed
But at least it was created and itās out there
I couldāve been even lazy or not even did that lol
Plus, the benefit of this is trial and error. You know you make it you release into the world you get some feedback and criticism and then you go back to the lab.
And wait for a newer model to come out to make it more entertaining
Jk
Ola to all, seeking some help or advice as new to this discord, I watched this video title and followed thru to this group ..but cannot find where I need to go to find the details described in the video Make LONG AI Videos (FREE & UNLIMITED)
Man claude opus just made a crazy website off 1 prompt
WE are not helping you brochacho
There is no long AI video
Thanks Gehlo,but this video says there is ?? https://www.youtube.com/watch?v=SeeWad9PW48
Join the AI Mastermind (Free Entry): https://link.outskill.com/MALVADEC1 Get the PROMPTS & PDF here: https://discord.gg/6aV3ps7tB2
Learn how to generate ultra-viral 3D animated stories and long-form content using a 100% free and unlimited AI workflow. In this video, I reveal a secret method using the new Gemini 3 Pro, Slop Club, and Meta AI to ...
https://019ae0a8-e939-7bd5-9d4d-baf87b44fa0c.arena.site Check this out. Its meant to make a wav file for me for my guitar tones but if you type any like bands and then their album names it will generate it BUT I love how it shows like suggestions like when you put a band in it already has album suggestions for it
Unfortunately. If any company made an AI model with long videos and it was actually good, I'd put my money on Deepseek to do it
deepseek speciale version is not available on lmarena but is announced everywhere that it is??? (Specifically in direct & side by side mode)
HOLD UP
My commonsense just hit me
can anyone direct me to the details on this https://www.youtube.com/watch?v=SeeWad9PW48
Join the AI Mastermind (Free Entry): https://link.outskill.com/MALVADEC1 Get the PROMPTS & PDF here: https://discord.gg/6aV3ps7tB2
Learn how to generate ultra-viral 3D animated stories and long-form content using a 100% free and unlimited AI workflow. In this video, I reveal a secret method using the new Gemini 3 Pro, Slop Club, and Meta AI to ...
This video tutorial
It literally shows you what to do lol
So which part are you stuck at?
Iām currently watching it also
Stuck on prompts and flow
How am I supposed to use this it doesnt detect the artist and album name i put
Iām not that far yet
Uhhh which band are you putting? because it was built more for metal bands
Ok i wait for your reply Gehlo thanks
video is 12 min approx
I know that, but Iām just saying how long do you want your video to be 12 minutes also
@echo aurora
Announcement says it is available but its not?
#announcements message
wondering where to find the details regarding that video in here Gehlo
None I dont listen to metal so idk I put carti
OK, well do you have a topic and ID and how long you want it to be letās do the first part of the video
You go to AI studio
Lol thats why, like I said, its for guitar cab IRs aka impulse ratings for a guitar tone haha
My Video ,,,not entirly sure how long...is there a limited on lenght
I just thought the UI was cool though
The limit is gonna be the amount of time you wanna spend
Can I make a site that makes ai instrumentals ?
If you want it long, then youāre gonna have to spend a lot longer. If you want it short then youāll spend a shorter time.
Good question, its unlikely but honestly I wouldnt put it passed anything! Just like SUNO
First of all, just go there
And for beginners to start with a simple one minute video
And then you could move it into two minutes three minutes or whatever
I dont like suno, but I guess lm arena couldn't do it
This is content farming lol
Damn
SUNO Uses Text to audio AI models or text to Music really
So its different I'd say
Chirp
Do you have any basic software video editing software?
Scared me
ok i have Genimi 3 open and ready to prompt
All right, so what do you want your one minute video to be about?
in the tutorial Youtube its say the prompt is inside discord?? where??
where can I get the PDF? Gehlo??
Did you get it?
You know what bro weāre just gonna extract all his prompts from his video
I am part of this Discord for about 1 week so when I press that link its a welcome to the group but no pdf
You gotta let Playboy know
Youāre there for the PDF
And that you donāt play these games
Let him know you followed the rules and did with instructions said and you still donāt see the PDF
š¤£
But see heās gonna try to sell you something
Heās gonna try to sell you some kind of course or something
Because I clicked that link and it wanted my information and my phone number, which is, I donāt want him calling me
Or you know what dude
Iāll sign up. I got temporary phone numbers I forget about.
yes I have been registed to this group few weeks and have followed all rules and even made few small test video, so I was interested to learn how to do Long video as shown in youtube video
And he still didnāt show you or tell you what to do?
speciale seems to be back up
but cannot find the pdf with the prompts as described in the youtube video??
It was gone from openrouter too?
no no
i mean
the service is reliable again
Youāre in luck, buddy. Iām in a good mood today.
i think speciale can be safely put back onto lmarena
It dont hallucinate no more?
idk lemme check
yoo i found a way to fix that
give me a prompt
see dm
simple hi
I will, bro give me one second let me help this gentleman out
Hell yeah š
bro is working better then moderators and admins
show a coding example, a uh cool html website
hmmm alright
holy hell yeah this thing is a reasoning machine
rip my credits bro
LOL
Not hallucinating though at least
One second dude Iām building a YouTube video frame extractor
oh yeah its much better
That way we can extract the prompt from the actual video
its planning out the typography and stuff rn
Bypassing the PDF entirely
im sorry lol
typescript interesting
Its like watching it burn
yeah no im stopping this
Bro, once again, my comment is kicking in here. Did you even watch the video?
He literally has an arrow on his discord where theyāre at
i dont blame you lmao
Aint no way u a still teaching the dude its been like 10 minutes
I donāt even think that dude watched the video
im gonna try minimal reasoning
Something aināt right here
with speciale or without, or can you like uh change reasoning type
you can change reasoning effort
Where can I test deepseek-v3.2-speciale?
im using openrouter here
oh ty
I don't think it's a coding model
no no it is
I got played
still goin?
U want a jailbreak prompt?
They tested V3.2 on SWE-Bench but not Speciale
101% on SWE verified :trollface:
Why sonnet 4.5 in codeforce is so low? Lol, they use without thinking?
I didnt know lmarena ACTUALLY MADE BACKENDS
im gonna let it do its thing for a tad more
Because it's not good at reasoning
Idk what is backend, is to human mistake?
deepseek chomps on your openrouter credits
Hmmmm, make sense
well hey
its a cheap model
so
i dont really need this
i can just
use a diff provider
lmao
Backend of your website is like basically if I wanted to put an AI in my website, that'd be running off my backened. Its stuff you dont want in public as well
How
Yeah, it has been known for that
has the gen finished
Itās part of it app appeal
Dont ask me.. I literally copy and pasted a prompt from gemini 3 LMAO
im gonna give it a better prompt
Why canāt you use Google fire base or whatever itās called for backend
This was the original prompt
Good callout, I'll edit the announcement. And if/when it's added again I'll update the announcement again. Thanks for the heads up.
Or is Google firebase not any good?
Bro, I forgot what meen backend š«¤, I'm not okay today lol, I need to make a scan in my brain now lol
openrouter has a backend alreadt
i think
Has webdev always been able to make backend for its sites?
im sorry, python?
I wonder if it depends on the model
Yea, the prompt had it make the backened in python
For what its meant to do python is supposed to be used
Extracting files basically a virtual computer inside of the website
ChatGPT does the same thing
They have a virtual computer in their AI. Hence the reason it can make zip files
any new model?
hows deepseek v3.2 speciale at coding
havent seen anyone talk about that
im trying it rn
do you have a gen where it finished reasoning
it seems like an awesome model but the reasoning slows it down
im just gonna try 0 reasoning effort
Yesterday pro pro made a site and it was awesome
Off of "Make a cool website" lmao
I think it was pro pro at least
that would be the thinking model
I'm testing it, I've been thinking for more than 10 minutes here to make a page.
i cant test speciale as it was bricked by that time
Ohhh my bad I thought it was speciale
i think 3.2 thinking is goated then
It gets the job done for sureeee
Only downside to it is having to manually add all the code LOL
right i need opinions
do i let speciale do its thing
or do it cut it short if it doesnt finish in 5 minutes
bad. it's where Opus is good.
Let it do its thing
have we seen any coding from speciale
the SWEBench looks pretty damn good
if it doesnt eat all your credits i say let it cook mayb
Isn't it a maths/reasoning model
its cheap so
only used up one cent from 3 tries
LOL
yeah no i think uhh
i think ill stick to thinking for code
we've seen a lot of coding in front of our eyes done with my our own hands in many different languages in past 25 years
That literally doesnt answer the question
At all
what
for coding reasoning isn't that important. what is most important good knowledge of APIs, libraries and frameworks.
im trying speciale one more time
they are all math/reasoning focused models
I've never seen a model so broken
its not broken
no
its just wayyy too paranoid
i love that it thinks of every like possibility, thats great
but i really do not need that
at least for coding
i could see this model being great for research though
and maybe editing existing code sections
If its taking longer than opus then there seems to be a problem lmaoo because opus keeps taking 15 minutes a generation
generation of what? you asking it do whole job for you and generate a whole app, so you wouldn't even have to press a button once?
Compared to other apps thats very very slow
Gemini 3 gets it done in 5 easily lmao
Opus is slightly better than gemini 3
Shouldnt be that big of a difference
okay yeah speciale isnt hallucinating it just has a weird ocd habit of making sure everything is planned
so this model is probably just meant for math and extreme reasoning
Does it self-verify a lot
i think this is built off that crazy good math model they made that will reason the problem out instead of hallucinating its work
yes
Sounds like the maths model then
mhm
yeah no this is taking forever
but overall i think its a good model
just not for coding
compared in what? you really not doing anything and asking ai to so whole job?
You're either stupid or just rage baiting
You are not answering the question though.
What it means?
I think coding
This is pretty much English, even though i'm not English speaker
compared in what? generation of what?
Nevermind wrong group
The question was what is he doing that it takes 15 minutes. That he clearly misunderstood purpose of AI.
it's not in there
try searching part of the prompt visible in the video into discord search feature
its not the same at all I think
fr
š
If you want it, we can or you could hold when you need something really bad
How to return the favor in the time of need
Oh, so you already knew him
Well he gone
and i thought i was an unc
Bro these yt be ripping ppl off
am i the youngest guy in this chat š
dont watch ai yt videos
its always some scam
The whole industry is a scam
Monkey see monkey do
In all reality, all these models are probably way more capable, but itās dividing the compute between users
Imagine if you had access to all the fire power
We probably only get like a sliver version
With token limits and cutoffs and caps
well i mean i probably have more of the gold version
(api access)
but i see what you mean
remember that these models are likely quantized to about Q8
/arena help
LMAO
New openai model gonna be benchmaxxed
So based on my testing itās for extreme research and math
Think like O3
Not for coding
3.2 thinking is for coding
Iām sure speciale can be used for code editing with how in depth it is but otherwise this is a number cruncher
???
I just used it
Yeah defo a math model
Oh well 3.2 thinking is already very nice at coding
@hardy swallow could you get this guy OUTTA here
Oil him up
Mods, force this guy to use llama 4 maverick for 2 hours
This is worse that dead?
Yes!
Still no Deepseek speciale on lmarena :C
Itāll probably be up today or early tomorrow
Its service is stabilized
Mistral new model stil behind DeepSeek R1????
Yikes
š„³
Hello
I mean it kinda wasn't supposed to be exceptionally high....
this is their own official graph
Of course but I mean
With a dataset that size Iād thought theyād at least pull some black magic
Maybe I just have a hatred towards mistral
I never liked their models
wym
what dataset?
Mixtral is like the godfather of OSS and MoE in OSS. That's where it all started
So they have accomplished quite a bit already tbh
Oh absolutely
ik i don't know what halpened to it, i think its maintenance xus it was on for a tad :P
Sure but that's just model size, means not very much by itself. OG gpt4 was much bigger and is nowhere to be seen on the leaderboard anymore
Yeah true
I am 99% certain that they took it down because the long time it spent thinking for responses (and the long responses it gave) caused it to time out and fail. Not denying it had hallucinations but I don't think they would have taken it down becaues of those, unlike what some people are claiming. (I'll eat my words if someone provides an official source)
Gemini pro has a 2k model now
hello
Ciao
Some responses can take 25min and that's normal for Speciale. This is obviously not ideal tbh
They need what Kimi did with their Turbo inference option very badly
It was already a problem with R1 at those response streaming speeds, but Speciale is much more verbose.
Tbf this is like the Chinese version of Gemini deepthink
And we donāt know how long deepthink reasons
I think
honestly if some responses take 30 mins then it's useless atp
ykwim?
unless you got alll the time in the world
it's just useless
tbf I do got a lot of free time
<- is living on disability due to a chronic condition caused by a covid infection a few years back
but nah I'd consider deepseek-speciale kind of a deep think model
seriously? you good??
more or less
I don't know if it's just me but I don't know why for some reason fails is every time I try to put Asia or American continent like no joke I tried every other continent and they work basically fine with these randomly stopped working mid-chat
actually I'm feeling better than I have in a long time and I think I owe it entirely to a burrito
Glad to hear that
Okay this is weird with those two cuz I just asia again but smash random words and it worked completely fine
I'm either gonna feel amazing or like crap tomorrow and it's gonna be a toss up
LOL
well im really sorry about what happened
even i dk jack about it i can only imagine something really bad
well thats the way they crumble cookies lmao
glad you are alright
thaaanks :D
ofc :3 as long as youre up and breathing and waking up everyday, things will be alright
aint that googles supposed project?
Is the open ai model dropping next week Robin?
war and drama
Hahaha they're doing it for very big worlds
Mine works on room pictures
I'm alone
Looking for people to help maybe
It's local for now
*it's not advertising
i don't have any website or whatever
time has surely ran out
what bro?
thank you friend
robin?
also whys that?
well I had a big day that left me feeling energized but I tend to crash after overexerting myself. went and treated myself to a burrito, to zootopia 2, and then to a warm brownie sundae afterwards as the literal cherry on top
oo thats friggin awesome
too broke i can only buy a bag of dinamitas
its okay i love them.,.,
I wanted to talk to u about that
anyone having problems getting gemini 3 to work
nintendo 3ds time has ran out
Hello. This group is amazing
nice
Thereās a lot to criticize ai about tbf
hey
Image generation, video generation, the anti social effects, ai catering to people with delusions/mental disorders, the compute shortage, etc
war and drama style
They donāt look at it like that they look at it as a means to an end collateral damage
Sacrifices that are willing to be compromised
Glazing the machines
At least youāre considerate itās awesome. Itās awesome. Try to keep it up. And least ur aware of some of the issues that plague intelligence.
Shilling for the hype machine. They are the gatekeepers that lure us in with insider information which they cannot disclose apparently
I despise anything image gen and video gen related
I think LLMs are awesome
People will say I shouldnāt hype here and there, but Iām going to be honest I donāt hype unless I am certain or I know information that gives me a reason to hype. You could check my page all through throughout the last six months itās quite rare that I āhypeā up a release
But there should be a line drawn
whys that?
Figuratively or metaphorically
Itās ridiculous. It comes with the territory
If the rotten, they rotten to the court on everything
But if something are fine, then itās all fine
whats this about
Iām assuming by the looks of it possibly refer to ChatGPT
OpenAI news today
They got some new good stuff coming apparently
As soon as next week
Today?
i doubt its a gemini/cladue killer
12 days event
there is no 12 days event this year
?
I misread when you said today
is it shallowpeat or garlic?
I have nothing to based off of at all zero evidence or proof
But if I guess I would say theyāre gonna update the OSS model
No garlic till next year
Idk about shallot peat
