#general
1 messages · Page 334 of 1
beluga got wrong bro server
bro got wrong kimi k2.6 distilling claude opus 4.7 while drinking hot chocolate made by chatgpt 5.4 and grok 4.20 is trolling
Bros keyboard completed the sentence
Approved ☑
why is everyone turning into fruits??
i will turn into an avocado
Bro got wrong server with the extra sauce and large fries the Opus 4.7 is new opus 3.2 spud the spud release Thursday and Friday night and I will be able to see image 2 tomorrow night and I will continue to do so and I will continue to do so and I will continue on this and I will continue to keep you updated with all of this information ℹ as you know that the most recent and recent developments are not the most important thing to opus 4.7
cinema
My keyboard predict full image 2 releaaseeee
done (sadly the AI i was using added too much details ;p)
Pineapple 🍍 is a great place to work for and it is a great place to work with that is a great place to work and to enjoy the great benefits that are in leash with the most realistic spud and most important things to do is to make sure you want to permanently delete this conversation and be your friend and you will be able to see sigma hands on the day of your life and your life is a must for you and your pepperoni pizza is the best way to get your life in the world
where to use duct tape?
Yo bro it's gone now
Battle
But image 2 is releasing tomorrow I think
I heard Thursday is the release
tomorrow? lets goo
Or no
This you?
Spud is Thrusday
no thinking model can be better than thinking model ?
Yea but image 2 rolling out for users
We can expect full release by tomorrow.
Yeas
lets gooo
It can be but non thinking came before the thinking model, thats why
Because it blew up when the first on came
are they gonna make sonnet 4.7
no thinking model is first
What
No
The thinking model came AFTER the non thinking
To arena
now im an avocado
Hi guys
the family keeps on growing
Is there is anyone of you experimented minimax m2.5 and is it good for coding ?
sure
i will eat
brazil
im from QATAR
Sonnet 4.7 in 2029
Nice
New produce!
Hello 
all created by pineapple
our master pineapple is strong enough to create many fruits
lol
where is pinapple from
big bang
pinappleland
FELIXXXX
hai
That was someone else, I can't take credit for all the amazing pfps created
@urban herald Hows lady cruche doing
i don't mean the pfps
Banana was it?
Or Cherry
uhh idk
so sad
yea
Btw gpt image 2 worst checkpoint is releasing
Not the good one
The one rolling out is close to 1.5 but not too much
they are all worse than gpt image 1.5 in world knowledge, they are slightly better than llama 3.1 8b at best in that area
nano banana user spotted
@echo aurora Make him enter the secret portal..
To our SECRET Place..
The LOT of Produce..
They have produce role
also btw the models that were on arena were likely the different quality levels, it makes sense that the worst one would get put on chatgpt. they were just seeing how bad they could make it by putting all of the quality levels on arena and seeing if the regression is noticeable enough to harm scores
also maskingtape (and also gaffertape, packingtape) used the 2k resolution of the model, that may come out too but mainly on api. it could be on more expensive chatgpt plans though
RIP GOAT V2
Tomorrow or in the next few days GPT Image 2 on direct chat in Arena AI?
4
4
2
No We are waiting for the final release
I eat avacados
main reason why i put the avocado pfp
avocado is my favorite fruit
✌️
I want to Eat Fruits
others
Oh
wew are a group of fruits now ✌️
Nice
i honestly wonder if image 2 will even be in direct, they are mainly targeting text models for changes because those are likely used the most but they could just not put image 2 in direct chat if it's very expensive, especially the 2k version if that comes out
If it's in direct there will be hard rate limits
Yes
Yes, something I designed so that the file is edited in the right place, regardless of how much has changed
yo manifest v2 girl
@echo aurora
yeah
What's the consensus on an ETA for public release of those duct tape models from chatgpt? (So I don't need to scroll through this whole chat to find the most recent answer)
I got it like twice in lmarena and each time blew my mind
honestly can people please calm down with these gpt image 2 threads, it'll be on arena when it comes out (which is when openai makes an announcement, not when openai shadowdrops the model for some people)
for the record i still have image 1.5 so no, it is NOT out
Facts
<@&1349916362595635286>
OMG
Come on, we'll be there tomorrow (I hope)
Thank you
Hi
btw can you do something about those gpt image 2 threads?

@echo aurora how is progress on returning top models?
Pineapple, BM is asking if you could remove his timeout
lol 😂
Can you tell your team to add gpt image 2 to arena?
No update
Already have
it will be added when it comes out
ima try make an operating system with arena.ai
bro is cooked
This is crappy part about ai. It’s affordability and access =/
Not everybody has the means which sucks really really bad
I believe in you
thx hold on
I'll send file list
in my code editor rq
@echo aurora bro
bruh
yeah so
keep in mind this was to make a simple gray box 😭
for those who were curious
Tell ur team to add gpt 2 image to arena
brother
...it will be added when it comes out
they can't add an unreleased model unless it's for testing
We'll be sure to send out an announcement for new model updates 
you will need a huge amount of luck
what brotha
im praying to God to give you the huge amount of luck you will need
can I take your pfp and make it be a apple
are you copying and pasting code from lm arena models
yes
You can add qwen 3.6 on list of high end models only in the battle mode
thats gonna take long but alright
dw bout it
you should use kimi website i guess
wait
that doesnt make sense
technically it is not an asnver
are there any new announcments
I'm also using antigravity for some bits the ais on lm arena cant fix
antigravity has claude and gemini in it
nice
Oh my bad, I sent this reply to the wrong person. #general message We do not have an update to share on that front.
You need a snack before you eat all of hte mods
bruh
Anything but the mods
applepine
Lol going to ask to not do this
bro did it
✅ Do not impersonate staff, moderators
shi
Too close to my name pfp sorry to say @gleaming island
😭 😭
do applepine
I remembered your old profile picture from the LmArena days. Do you still have that one where you're wearing sunglasses in a car?
pfp too please
💔
I'm sure I can find it later when I have more time
what if I do an apple but as a pineapple
@echo aurora where is the rules
It hides in the id:guide
(I think)
Yeah taht name is fine, just be sure to change the pfp so it's not so similar
If you change it in the next 15 seconds I won't
i didnt see any bad words rule
THERE THERE
yeah that works
Sorry I clicked a random pfp
Lol much better
bro do pineapple juice
Yeah I don't think it's specifically called out
When did Kimi 2.6 come out
I'm a jailbreaker do you want me to try find a vuln in the website and then I can tell you then you could fix it
tecnically im not harassing anyone im just saying a bad word because smt happened
much better than BM
but you gotta ask the higher ups
I can try jailbreak an ai on the website then send you the prompt so you can fix it pineapple
i dont think pine apple is the dev
he can tell his ceo or smth
What if he just made the pineapple have red hair and red eyes with blue skin
then that would look like a cursed pineapple 😭
ExActlyyy
Arena manages security vulnerability disclosures through our HackerOne program. At this time, the program is private and not accepting submissions publicly, so we’re unable to review or discuss vulnerability details outside that process.
Sounds too similar tbh
pineapple has all the knowledge
wisdom
Loll fair enough
did bro get claude sonnet 4.6 to write this 😭
it sounds too ai
@echo aurora is this good
No it’s just professional
yeah
AI is meant to mimic professionalism

😘
How many members are in this fruit family?
7
or 6
i love how million of people use arena per day and the discord is just chills
yeah, I only joined at first to fix a bug, but I ended up staying... I use it to practice my English and keep learning about AI and stuff
i literally learned english as ai learns reinforcement learning
i joined because of video gen
i sometimes keep re-reading just to make sure I don't send anything wrong, but my English has improved a lot just by being here reading and chatting
yeah
Do you know how I found Arena? My senior project group found the site and we used it to build our alternative LinkedIn project in php
lol
thats crazy
gang approved 😭
boy
wait rly?
YESSS
did you get it yet?
Not here yet, but I know other parts of the world have received it, but it's still in the testing phase. But don't worry, because when OpenAI makes an announcement about it, they'll put it on Arena AI
IS BM GUILTY
6
7
1
YES
wow cool
thanks for the info
friday the 13th
What is the name of the Friday the 13th killer?
Jason
.
WRONG!
She is Jason's mother, Jason only appears in the sequel
I am very sorry
Jason only appears in Friday the 13th Part 2!!!
Its still Friday the 13th franchises main killer lol
If you ask anyone they will say Jason
in the first movie, is your mother the killer... I think
yeah
Yes, Jason is the killer, but the killer in the first Friday the 13th is MRS. VOORHEES
Oh man, I had a heart attack, I thought it was the new GPT image 2 model
lol
Don't do it again @echo aurora
GAHAAHAJJA
Kimi is lwk peak, i hope this new model is good
Impersonate staff?
It actually seems very good, its likely distilled from opus too. Problem is that the pricing is the same as USD prices and its annoying
Yeah but not in some kind of malicious way, were just having fun.
i mean, there are 12 friday the 13th movies, he's the killer in 10 or 11 of them lol
Well, everyone is setting fruits as their profile picture and username
There will come a point when we run out of fruits, and they'll have to use fruits from other galaxies 😹
or vegtables lmao
Fr, looks like they nerfed their models while rolling out new model
Yes, too
Kimi 2.6 actually seems good. But damn what a long thinking period it has lol
which one is better in your opinion?
deepseek is cooking so hard
if small team that do LLM doesn't have as much compute as openai, anthropic and all then how are they making Trillion parameters model
the difference in compute power is supposed to be massive
Gemini 3.1 flash lite is better in cost, and speed
and intelligence is debatable
What's the best model for having fake conversations with in languages like Italian and French? Also for having it create short stories in those languages.
flash preview
how did you get image 2
how are you getting it]
With normal ChatGPT
Don’t seem to be it, look grainy, like 1.5, masking tape and duct tape didn’t have that problem
.
So I don't have it yet, what a drag
In fact it is GPT IMAGE 1.5
What a drag
How did the arena be able to generate 9:16 and 16:9 aspect ratios for duct tape?
gpt image 2 drops on april 21
but i don't have access yet
XD
bro is deepseek the only ai that knows how to do youtube and spotify integrations?
You could specify it in your prompt or it will do it by itself.
Nanobanana can do it too
this has to be gpt image 2 right guys????
because why would they update the ui and not the model
It was there for me a few weeks ago,
What I noticed when I was A/B testing on the ChatGPT app is that there’s no « preview » of the picture getting generated. Compared to 1.5
could be either 1.5 or 2 though
I've been trying to get an image from 2 for the last few days but I get 1.5 every time
I got 2 a little over a week ago in chatgpt, but only 1.5 since
is this leaked longcat image 2? /s
theres no official info about GPT 2.... So no
It's going to depend on the model, there isn't a ratio setting in Arena at the moment. Some model's default ratio setting is different which is why you'll see some changes here and there. This is something we plan to make a settings for.
just NB PRO example of Aspect ratio specified in the prompt
GPT 2 Aspect ratio specified in the prompt
You can prompt for aspect ratios at the end of your prompt. It works with the duct tape models and a bunch others, not all models though. Just add "16:9 aspect ratio image" to the end of your prompt
WHAT the HELL?
This proves Imagev2 is gpt 6 spud actually
You need GOOD omni reasoning to do this
Theres 0 way nonomni model would do this
there's no way, the llm that image 2 uses is actually pretty dumb, somewhere in between llama 3.1 8b and qwen3.6 35b level
most likely, it's a model trained from the ground up for image gen in particular instead of general usage
It doesnt have to be perfect at omni reasoning obviously
Thats like asking opus to write in galaxy alphabet or raw binary
But you need extremally smart model to be able to generate working qr code via imagen not tools
@echo aurora can u pls suggest qwen 3.6 max preview to be added to arena
bruh it could just be trained on a wikipedia qr
Bruh images v1.5 cant write "hello" in plaintext
The gap is extreme
yeah but my point still stands
it needs to be tested on some unpopular url
yo
thanks
where do i animate the image to video section
The Video Arena is currently accessible only in https://arena.ai/video. More information on how to use Video Arena can be found in this article.
The Discord bot was removed from the server and is no longer available. This announcement has more information about the removal.
MURPH DONT LET ME LEAVE
yo is there 4k support?
Another example of an available aspect ratio using a blank image. NB 3.1
When will image 2 be officially released on Chatgpt?
We don’t know yet, but we hope tomorrow
GPT 2 will be a legendary release with huge meme potential until the new Grok, which was also recently tested, comes out.
the new grok? are you talking about flashbrown or something else?
because i thought flashbrown was some sort of grok imagine model myself
I don't remember the name. There was a short test, about a day long, of a VERY high-quality version of Grok. Then there were a couple (1-2?) more versions of MUCH worse quality. It was 100% Grok, it's hard to confuse him with anyone else.
this was on arena?
perhaps flow-state
i don't think that's grok, for certain prompts that grok imagine 1 can do it's just awful
Yes, almost a month ago.
probably why i don't remember it, but that would probably just be grok imagine image pro
i don't have more because fal (the site i use to generate images) doesn't have this model, i either have to use arena (which i don't really want to do but i'll try it), pay for supergrok (which i don't want to do), or pay for xai api (which i also don't want to do because i don't want my money trapped in one ecosystem)
i pretty much stressed out the model with this prompt and honestly grok imagine image (left) is probably better than the pro version (right)
same prompt i've been trying gpt image 2 with but every time i get gpt image 1.5
they are training imagine image 2 right now and i expect that model to do better then either of the imagine image models
What is the left one is that grok ?
both are grok imagine image pro
The only thing I don’t like about AI animation images is the outlines
There always so thick and look so ai it’s hard to get rid of them
don't worry, you'll get opus 4.7 once china distills it into minimax m3 and kimi k3
none the less i really appreciate the creators of this server i graduated today with a perfect mark thanks to the research's i was able to do using the offered ai models.
See how these don’t have any of those lines those fat outlines
Idk how to get that look
the ai models were likely trained more on animation with lines so that's what it tends to produce
Yeah it makes everything look super ai
Hmm nahh, did you end up fixing it?
do you run
a yt chaneel
or anything
Of these two models, I always thought the Grok Imagine was better than the Pro, which is surprising.
Is there any character or word limit in PROMPT for image generation in gemini 3 pro image preview 2k model?
This
It’s very easy for ai to recreate something from memory
i did, it turned out to be a harmless bug
even my old tabs work, and show their actual content again
No don’t
the streets saying opus 4.7 is pure ass
Why would they assume it’s gonna be better than 4.6?
If they have mythos
New updates means more restrictions lol and worst rate limits
they gon lose their customer base if it continues to be ass👀
chatgpt gotta capitalize on ts and bounce back
They have a compute crunch, allegedly, anthropic, I mean
this is so funny bro😹
they acting like they work for the companies
bro youre just a fan, calm down
"us" openai don't know u bro😭
The AI community has some very strange members as part of our clan. I’ll tell you that.
fr bro
When i first saw announcements, I thought they were going to actually add claude opus models back into direct
But I guess not
You play chess? @polar horizon
yeah but i rarely play now💔
Oh
I'm still learning it right now
oh shoot that's dope
Trying to learn brillant moves
it's fun to play when ur good
start with the basics bro😭🙏🏽
I wanna start playing the brillant moves first ;-;
if you're not solid you're gonna lose every game before you get to use those brilliant moves😭
So I can surprise my opponents despite my Elo
Probably should live stream first on twitch first or something so no one thinks I'm cheating
not sure if it's that easy to win matches like that but what do i know
go for it twin !
Developers, I really appreciate yall making this app 😄
Why are only 1,000 people using it though?
i dont think its official
How come
logo, publisher
Oh
bruh its clearly not related 
That's what the app looks like inside?
yes
Never mind then, I hope it gets deleted
lol
After nano-banana here’s gorilla banana
this is also a scam what the #### is this 💀
Yeah what the F̷̿̈́̍̎
F̴͐͛̑̒
F̵̾͂̾͌ is this ;-;
That is not us
We don't have a phone app, so I wouldn't trust any claiming to be Arena.
Found 2 apps. One named AI arena duels and other is LMarena AI mobile lol
why are there no other sites like y'all
genuine question
the only other one (yupp) got demolished
It was a joke 😃
Is gpt image 2 available??
I mean duct tape model image
They remove it ?
Please someone answer me if you see my messages
It got removed
you can try to use it in chatgpt, but i can't promise that it will show up, just use it normally, you may get an a/b test with it or you'll just get an image 2 image (or 1.5 if you don't have it yet)
But this new model is not released yet?
Slow roll out
Why i not find opus 4.7 in direct massage @echo aurora
What happens if we share personal info with the ai
I can’t even find Gemini 3.1 pro
He will get categorized your data sold
I searched up Gemini and the only pro model was 2.5
I can’t find 3.1
As well as a lot of models
Like GPT 5.4
kimi 2.6 sucks really bad
they removed all the good models for some reason i forgot what it was
Help me
abusing rate limits via new accounts
Yikes, not sure what's going on but I literally got like 5 back and forths with Claude and it said I hit the limit... 🙁
i'm pretty sure it gets stripped or discarded before it makes its way into any datasets or rankings or shared with ai companies
data sold to elon musk
it probably is sold to elon musk if it involves a grok model
all models*
elon musk owns the world
he owns microsoft
anything
Add kimi k 2.6 to the pile of recent that not in direct. I’m starting believe they don’t understand that they probably delete some of the older models and bring in the new ones
it is in direct chat (ignore it saying that it's k2.5, it does a similar thing in the api)
Oh ok but my point still stands for other like qwen
qwen 3.6 plus is in direct chat too, i don't think they added max to the arena yet
How to chat with Claude Oups 4.6 or claude opus 4.7?
u cant gng
Pay anthropic
what are the chat limits when coding or using direct mode for claude models
It was 5 output for opus 4.6, now idk
wdym, like 5 messages?
yeah 5 output means 5 messages from llm
Qwen should have more than enough for you
Never ran out of its limits in its website
oh alr, thanks
but like I never really used qwen much
no worries
use antigravity
seems like no one is talking about OpenMythos yet... https://github.com/kyegomez/OpenMythos
Thats a lot considering 20$ plan is not enough for even 1 hard request
there is a way but u will need to reverse engineer it
@echo aurora ur pfp so cute!!
hmm never mind, I'm a bit curious to test Claude 4.7 opus in Arena, because it's not in direct mode, I was forced to try many times until ridiculously 20 attempts before I found the model used by Claude 4.7 opus🗿
any updates on when claude models will return on text ?
flow-state seems like a runway model to me, it's been highly trained on photography and editing and struggles outside of those areas
ernie behaves completely differently, and the version on arena has a decent llm behind it, whatever flow-state is using is just not good at all
👍
Did they reduce this again somehow because it did not hit me this early in any of my previous ones
when is opus returning....
MR BEAST!!!!
<@&1349916362595635286>
<@&1349916362595635286>
Guys is gpt image gen v2 out? My images are much better than before
I think it’s slowly being rolled
Mine is fine
With the new codex app improvements, I'm wondering if anyone has tried to get it to play starcraft/dota/LoL? Would be curious to see how it does.
i wonder what happened to BM
<@&1349916362595635286> GOT EM
We don’t want mrbeast plesse
just ping mods with reply to that message
Oh ok
Its not even real-
Are you neurodivergent
Me neither
what is even that word 💀
lemme translate real quick
a-
Chatgpt generate me a thought
who tested kimi k2.6, it is good or not?
I wish the duct tape models have returned
Mid
Not worth the hype
It doesn't have smth it excels at
Just exists
Thats gpt 6o spud's image output
Soon releases to gpt users
What’s the current release we have? It looks better than 1.5 but not as good as duct tape-3
So when will be in arena ai?
I mean gpt image v2
Its coming out normally in 48h
GPT-Image2 is said to have officially started rolling out
Does everyone think it can be bought now?
Did somebody test the baseliner model
I have a ChatGPT sub and I have it
It’s so realistic it’s crazy
I argee
Anyone else can't access arena right now?
I can still access it
Why, what happened
It's coming up with errors as if the domain doesn't exist anymore
I think it's an issue on my end
ohh
Yea
noicee
Thank you
What's The Free Offer
?
See the pop up in top
Trial Max Bro
very free offer
Wait till it releases for free users
"we want zero money from u but we need ur wallet"
Please can you click on it and send me the link i wonder if it would activate on my account
I need this Trial
I saw them in x sharing images
Ty
In return i give you the link for Business Trial
Wha
Probably to stop abuse
ah yes, very beneficial promotion indeed
np
Had you ever joined a Workspace?
TYSM!!!
never heard of it
Seems like is not available in your country
anyways doesnt matter i cant use those promotions either way
cuz they dont accept my currency
F
yes i have to send 0$ in their currency
Otherwise Could Trial Max
wow great
YOU CAN ALREADY PUT IT IN DIRECT CHAT🛐
The problem is I don't have a credit card
Please
I want gpt-image-2-high-fidelity
Oops.
worked for me and im in the uk
heres what it made
Wow, did you do that with GPT Image 2?
If you use LMArena AI updates you in 50 minutes
What?
I want GPT Image 2 in direct chat🛐 they have released it in all parts of the world
What is this ?
Yes but I don't have it, that's why I wish they would release it on Arena AI
@echo aurora i think we got an issue here...
I wait
So not yet.
But other parts of the world already have it, but what's the point?
There's no issue there. It's completely normal.
Dumb 💩Gemini says
I knew it, when they remove duct tapes and access to GPT IMAGE 2 earlier then that's the cue that they will release it soon and it happened
thats not normal.
its mislabeling of model
actual intellect-3 knows its not DeepSeek
One message removed from a suspended account.
Not bad
HOW TO SIGN UP IN LM AREANA
I'd like to know how many elo score will the k2.6 get
he got that pee pee power
it has opus 4.6 level benchmark, hoping its not just good on paper, bad in practice
is there an app for this?
In my opinion, Gemini 3.0/3.1, as well as GPT 5.4 with Claude Opus, will not be returned to us
Gemini was the only one without any rate limit xD
so we've all abused it for trivial questions in direct chat
making up 75%+ of the bill
remember the old lmarena?
you could reset your IP and your rate limits also went away lol
joined just before it all gone to hell
well stay positive, look forward, the charity subvention LLM era might end soon anyway
then it'll be an expensive toy no one cares about in the AI drawer
what do u mean?
I'm sure I will get mauled by this, but I think that LLMs are not all bad...
They are cool man. They forced me to never learn programming properly
Because all my needs are automation scripts.
for me, it was getting me through college calculus
Fair, fair.
Actually u are in the wrong place with that (right) attitude.
Actually no it's not as bad I say. But most people here worship LLMs, they're in the AGI cult
AGIs imo are gon require too much compute power for not a lot of return...
I honest to God beleive they will require a different computing substrate.
It's clear by now that consciousness will never spark once on sillicone.
well define consciousness in itself.. do you call artificial brain cells as having consciousness?
Do you think if I'm claiming to be conscious I'm deluded?
Let's cut the the crap the burden to proof that I'm delusional is on you.
The moral struggle between good and evil
To know good but still struggle with evil
fair enough can't argue
hey man chillax it wasn't a personal attack
As in having qualia, being a moral agent you can say
GPT please stop being evil
there's none of that on LLM models, that's all I'm saying
they're ridiculously fun to use and powerful alright, that's a different topic
AgI
I think there’s another answer here also
sorry, can you reframe that english isn't my first language
as in there being a moral agent you can appeal to
yeah that's fair, it will be difficult getting that
Today we uncover the secrets hiding inside 3D Workers Island... AKA the long anticipated new project from Petscop creator Tony Domenico. What lies beneath this seemingly innocuous piece of Windows 98 software?
• Stream "Lucas": https://open.spotify.com/track/1j544ldguzEvdxZsBO2aM1?si=c597022767694293
• OST: https://open.spotify.com/album/5X...
Where is my explain the alternative.
well it's also fair to not pretend to be cognitively impaired, as a sidenote
Some of us aren’t pretending 🤣
Well I'm a quick damaged one.
Hey, if you spell your name backwards, what does it spell?
Satan
There are 3 r's in strawberry
xD
dw that one was hard coded to not fail, no amount of training can ever make them reason that slow and clearly
A man and his mother are in a car accident. The mother sadly dies. The man is rushed to the ER. When the doctor sees him, he says, “I can’t operate on this man. He’s my son!”
How is this possible?
🐦🔥🐦🔥
I swear I have seen this on quora
Good morning sar.
So the doctor is the father?
yes but there was a time not so long ago when every single frontier model failed
you can still make them fail just change the content of the logic puzzles
No training data / hard coding for popular internet gotchas = failure
it failed even when it was broken in logic pairs
but yes I agree let me not be pedantic
if you really go hardcore about your prose
and make it very explicit in simple isolated cores yes it did succed even back then
No, I mean I was even a little thrown off by the way it was worded, but I believe you
60-70% of the times
man u claim it's framed wrongly
but LLMs succeed in abstract theory of mind tests otherwise
it's such a strange toy lol
I never said it’s wrong. I said it’s worded kind of strange.
A man and his mother are in a car accident. The mother sadly dies. The man is rushed to the ER. When the doctor sees him, he says, “I can’t operate on this man. He’s my son!”
How is this possible?
I’m still confused what’s the answer? lol
It seems so obvious, but then why would it be worded so tricky?
the doctor is the father
Hmm
OK, I think I’m following along and that you’re saying that the large language model they failed this couple years ago
But now we’re getting it correct?
hey i got no clue, ask @tulip pollen
These are trick questions I swear
Yes but there are a hundred more logical riddles they still fail to
On an unrelated note, how's the ram crisis in your country
The answer is so obvious but the way it is framed makes it seem like it’s a trick question
it is a trick question
those correct answers are now hardcoded or part of new training data where people online pointed htem out
AI equivalent of when my friend made a "chatbot" with just a hundred elifs for the most common questions they asked in a school competition hackathon thing
kek
Idk I still think it’s kind of dumb for some reason I can’t shake it off
Every time I talk to it and I get a dumb answer I’m reminded I’m talking to a machine lol
normally gemini is decent at this stuff... where it really falls apart is in long prompts
you are talking to yourself ultimately
you're the one who does the fooling when it's all smooth
that's just a very fast abaccus with translation layers on top for things you care about
I’m just trying to generate some images lol
yeah mostly I like to pretend I'm a person who likes shooting cockroaches with a 9mm
⬆️➡️⬇️⬇️⬇️
I wish people would give themselves more credit how much time and effort they put into prompting how much thinking they do and the mental gymnastics is more than the GPU they allocate per request
I mean sure dude but the benefits are marginal
Hmm
And next to 0 in reasoning models who already "prompt themselves"
at hte cost of 10 amazon trees per a session
you heard me right if you want benefits in the same result, the incrase from promt engineering according to formal studies is low
if you do 5+ test trials and take the best answers they are better, but then just use reasoning models?
genuine question, if it was really that cheap then why does OpenAI haemhorrage so much money?
Scale.
Well we have to look into it if it's that cheap.
There were some papers that done it for both text and image in joules and it wasn't small.
America still locks that good gas price tho so the magic still works for a short while.
Why I say short while if the crisis continues gas will get exported more and more.
So obviously prices will hike.
No actual data center is being built right now, they arell use gas because the grids are not reliable.
Yeah in my country there's an entire crisis due to the oil shortage.. no cooking gas
Sorry to hear bro
shits expensive here... people are acting if things are bad in the US, I'd advise them to visit here
Real talk
thankfully I am in the position that i can switch to induction easily because i live in a metropolitan city
Are u in Asia?
but in rural areas, they're now reverting to wood or biogas in the case of relatively affluent people
ye
🙁
That’s crazy how industrial enlarge the population is of not only Asia but Southeast Asia and yet they’re so reliant on it. Importing gas.
We're very bad stewards of the earth.,
We need a philsophical shift on how we overall do things asap.
not physically possible unfortunately
We are definitely not doing that with ai
Well not with energy
yep
If anything, it’s actually making things worse
strictly as energy consumption
u are economically right lol
no matter how much we figure out ways to do it cheaper people consume 10x after that
We are being back coal
that's just thte system
In America….
The rest of the world uses the ledger system, endless growth model as well
ban that guy above me pls ty
@fading sable
<@&1349916362595635286> spammer online
<@&1349916362595635286> kindly ban @fading sable
I’ll shut plz
<@&1349916362595635286> theres a scammer on the loose mr beast didnt actually post this and deezwin is a scam
Don’t report me
xD
Pls
Yes
kek
🤣
well I;m all about cheaper text stuff
it's all I use either way
I want it to be so cheap that it's forever subsidized xD
but the fact that even arena is removing gemini and stuff is a clear sign it's not CURRENTLY that cheap
and the big corpos started to tighten the belt
am I too drunk or did the ui change?
how do I properly download an image?
no downlaod button anymore?
PS C:\Users\Hex> gallery-dl "https://cdn.discordapp.com/attachments/1340554757827461211/1496110256818880512/IMG_3934.png?ex=69e8b0c1&is=69e75f41&hm=11d90e7bff2d62a9dc47c946fa5db1891b15e18967ef4ca81e74610c776000f0"
[cookies][info] Extracted 119 cookies from Firefox
.\gallery-dl\directlink\cdn.discordapp.com_attachments_1340554757827461211_1496110256818880512_IMG_3934.png
PS C:\Users\Hex>
went teh nucelar way
and the aggressive shoving down our throats the ai slop as well
microslop being one of the worst
Believe it or not that was made by AI
You can even see the Gemini logo if you look carefully
like in the background baked in layer xD ?
If you use white background, it doesn’t really show
You would think at least they would put more effort into the watermarking make it dark against the white
But apparently, don’t even care enough
arena website is down :/
This images v2?
hi guys :3
i swear to god if it is im actually scared as HELL
arena website is down again its now loading forever
When is it coming out on Arena AI?😭
.
I want it now on direct chat GPT Image 2
I'm desperate
But what does it cost you to release it as a new model in direct chat on Arena AI?
Hahaha
Everyone is using GPT Image 2
Is the website down?
It has actually became a goat now
Nerfed Sheet
It has definitely beated Nano Banana 2
well we are still waiting for nano banana 2 pro
yeah but even better
I want it on Arena AI
I've been waiting for this too long
On 23 of release
Might not be in direct chat tho
How are people using it?
Where are y'all accessing GPT 2?
How did I reach my rate limit when I just hit retry on the something went wrong message. That's annoying
@echo aurora when gpt image 2 will be in the arena
I have it in chatgpt now.
How? Was this officially released? I haven't seen any news about it
Yes
Announcement must be coming soon
Whoever made that recaptcha thing is brain dead.
The question was for motorcycles, but none were displayed. [Though there were one moped and 2 vespas.] 😹
It's the worst tho
not better than duct tape
Try this prompt @obsidian cargo
"Sam Altman and Elon Musk taking a selfie in a conference"
We'll compare duct tape with it
If ur using image 2
Try this prompt pllssssss
"Sam Altman and Elon Musk taking a selfie in a conference
Seems fine to me
I just added goofy style
Hatsune Miku in Star trek - make the film I'll watch it! 😺
Tried to prompt it to avoid overendering and went too far in the wrong direction lmao. Second attempt wasn't far enough.
is she your oc?
idk why but gpt image 2 is bad at editing images
that looks like 1.5 tbh
looks like openai nerfed their new image model
bruh
duct tape looks like way better
as usual the one on chatgpt will be the worst version
probably the best will be on sora or api when they officially release it
it was like that with nano banana too
Probably. Second image messed up the eyes a bit, adding a fourth one.
As long as it's not Nano banana - this is how that model mistreated my OC character seen earlier.
lmao
Image v2 gave her a butt lmao
♥ Female powerlifter.
bruh why they made new model so bad at editing exist image
like put 2 peoples on 1 photo
nano banana doing this mission way better
are u sure its v2
yeah
i have pro account
very good at creating images, but not good at editing them
like changing background, put 2 real peoples on 1 couple photo
What you see above is indeed that, nano banana was supposed to swap the face in place but failed. This since it used the second image as the source - not the first!
Neither do Nano banana understand the terms sample and source image. when used in a prompt.
hi
The successful image was made with Seedance, but ... Flux 2 could have worked also.
teto?!
Indeed
tetooo
Yo
Hello Janixium. Yes I thought that was supposed to be Miku and Teto also.
nano banana don't always keep the og face from the ref pic, but it fr slaps at realism and editing source photos. but at the same time, the new chatgpt model literally can't even place someone from the original photo into another scene properly — bro looks like they got the straight up photoshop cut n paste treatment lmao
Hello there
i dont even remember why i named myself janixium
hi cherry