general

[alpine coral]

although.. that chatgpt-4o-latest API endpoint is explicitly dynamic.. it never has a datestamp (it's just whatever 4o is powering chatgpt at the time )

so ig that makes benchmarking challenging..

[ocean vortex]

it is, but no one is going out of their way to do it. Especially given that it's just a pointer and it will become irrelevant once they make new version and start pointing chatgpt-4o-latest there instead

[cedar tide]

https://x.com/ArtificialAnlys/status/1905563427776651344?t=Ade7EDjFb3DDumNIqnvwtw&s=19

Artificial Analysis (@ArtificialAnlys) on X

Today’s GPT-4o update is actually big - it leapfrogs Claude 3.7 Sonnet (non-reasoning) and Gemini 2.0 Flash in our Intelligence Index and is now the leading non-reasoning model for coding

This makes GPT-4o the second highest scoring non-reasoning model (excludes o3-mini, Gemini

X · Mar 28, 2025, 10:11 AM

.

[ocean vortex]

like 0129 chatgpt-latest is now as good as non-existent. You can't access it

[cedar tide]

Yes

[alpine coral]

i dunno.. i use 4o (via the OAI desktop app) more than any other model in terms of general day to day usage - it's not mind blowing in terms of reasoning, but it has noticeably improved over the last 6 months in ways that more than just stylistic or tying to game the Arena

it's better at instruction following and just gets stuff more than it used to, and provide working code way more often

[keen beacon]

Chatgpt 4o latest is primarily human preference based but still since December they've been continue pretraining the base model to match up with other models

The model is notably stronger imho

[ocean vortex]

that's a misleading game to be playing relying on impressions from different ever changing prompts and tasks tbh. The only way you can test it properly is side by side with clearly defined correct answers

[alpine coral]

i mean it's either helpful or not

it doesn't need to be scientific

[keen beacon]

U can see the latest 4o leapfrogged in benchmarks by artificial analysis anyway

[cedar tide]

every time i ask spider his exact model name he tells me llama 3.1 405b

[ocean vortex]

that's how people arrive at "it is getting dumber" conclusions when model did not change at all lol

[alpine coral]

i'm not going to bse my opinion let alone usage on some benchmark if it contradicts my acual experience with a model

i've got a bunch of subscriptions.. i play around with them all. claude and gemini have a place and are useful

but i'm not gonna bounce around all day

[ocean vortex]

then you need your own tests. But they must be constant and clearly defined otherwise there's nothing accurate about this

[alpine coral]

4o for me personally, totally subjective, has improved in recent months - its utility is enhanced for me personally

[keen beacon]

Dom just can't be convinced that chatgpt 4o latest is just stronger recently

[alpine coral]

it feels that way.. the proof is in the eating of the pudding - not benchmarks

[keen beacon]

It correlates in benchmarks too

[ocean vortex]

I'm not saying it is not improved at all. What I'm saying though is that this is still gpt4o and it's still gonna be destroyed by o3-mini-high, gpt4.5 or o1 in most tasks lol

and the exact improvement is difficult to quantify when there was no proper testing of it

[alpine coral]

i agree that it's difficult to quantify

[cedar tide]

Im not sure that its coder model, but there was rumor about that

[alpine coral]

but just at a higher level (vibes / totally non-empirical ), it is discernibly stronger and more useful just based on my day to day usage

and there is definitely a place for relatively performant non-thinking models

[ocean vortex]

I mean that's what it was designed to do. Like in my view this is not a good way to test the models to begin with, but even if you must... this is the last model you would want to do it with. It was trained to output the preferred / more appealing responses.

[keen beacon]

The latest openai models feel like they're trying too hard anyway for me

I do not enjoy talking to them

[alpine coral]

yeah the chat style annoys me more than ever (even if it's meant to have less emojis, there's still too many)

i want my llms cold and clinical lol

[distant egret]

thanks for sharing 🙂 , but still in their website its not showing, but atleast happy seeing in X

[ocean vortex]

you could get a wrong response that is worded in such a way that you think it was something special, to take an extreme example that's easy to understand lol

[alpine coral]

Did you write this part Dom?! aha
it is a very valid point

[barren prairie]

I want it with emojis 😁😆

[alpine coral]

hehe yeah hard to please everyone!

[barren prairie]

But I think it must be without emojis and if you want some just tell them on the instructions like what I do with Gemini

[ocean vortex]

I did not actually see this post but it could actually be interesting and somewhat prove me wrong if it does well there lol

don't think it's up on LCB

yet

[alpine coral]

yeah i mean it's possible for two things to be right... 🙂

• OAI's approach to releasing and naming latest 4o variants is a mess that makes benchmarking far more complicated, if not impossible , to do with rigour / reproducibility
• 4o has actually improved quite significantly recently; more than just aethestics / catering to user preferences..

[rigid widget]

Updated ChatGPT-4o vs Updated Deepseek-V3

[ocean vortex]

ok yeah... Why haven't they updated their metrics then 🤯

[alpine coral]

they posted it on twitter an hour ago.. i dunno whether sloppy or calculated.. but presumably they just haven't updated the website?

dunno why they'd post their findings and the charts otherwise

[barren prairie]

Gemini flash 2.5 ??

[ocean vortex]

@keen beacon ok you were right then I'll admit. I can see this being part of their upcoming gpt5 now or otherwise tying up with the things to come. I think artificial analysis is fairly reputable

[kind cloud]

Feels like 8b or so

[rigid widget]

costs $0,017 vs $0,0025

[alpine coral]

another model bolide , says it's from meta (fwiw, it lost to qwen-max-2025-01-25 in this battle, using the same stock data/charts task as above)

[rigid widget]

chatgpt-latest is too expensive even more expensive than claude 3.7 sonnet

chatgpt-4o-latest vs deepseek-v3-0324

[alpine coral]

yeah true. i hadnt noticed the discrepency in costs b/w regular '4o' and 'chatgpt-latest-4o'

it's interesting

[rigid widget]

gpt-4o is useless for me

[alpine coral]

fair enough - not sure what that has to do with pricing discrepancy

but yeah, i mean don't use it / to each their own lol

[raven void]

new 4o has a good rep in general

[rigid widget]

I mean this: GPT-4o is so bad that it makes more sense to pay extra for ChatGPT-latest

chatgpt-latest vs deepseek-v3-0324

[alpine coral]

i still have a bunch of automated workflows that use gpt-4o-2024-08-06 lol it works perfectly fine I just haveb't bothered to change the model

again - to each their own

[rigid widget]

If it's working well for you, that's great

[ocean vortex]

there's also this fact:

[alpine coral]

it [gpt-4o-2024-08-06] does a fine job at parsing alert feeds, rss feeds and identifying / extracting specific items and outputting as a specific JSON object

i spend enough time tinkering with these workflows setting them up that baiscally once it does what it needs to do, i avoid changing anything

[rigid widget]

I'm telling you, their pricing is a joke—you won't believe it!

[raven void]

4o is a great general model

[rigid widget]

which 4o? 😂

[raven void]

Latest one. Well though I don't see what exactly it excels in compared to Claude and Gemini

[alpine coral]

nvm i misread

[raven void]

is the latest one always double price?

[alpine coral]

it's weird.. the oldest 4o checkpoint (gpt-4o-2024-05-13) is the most expensive (the same price as chatgpt-4o-latest). the other two 4o checkponts are same prices; but by default, it doesn't point to the most recent (gpt-4o-2024-11-20), but rather the second most recent (gpt-4o-2024-08-06)

[ocean vortex]

there's also no api for finetune with native imagen. Wonder how the pricing for that would look like...

[rigid widget]

because 08-06 is better than 11-20

by the way livebench.ai updated!

[ocean vortex]

I imagine the way it's implemented on the website now is that text model calls a function which then forwards the request to a different gpt4o variant to generate an image with the given chat context or smth like that

[rigid widget]

themis is interesting

[eager mica]

There are a bunch of anonymous Meta models on Chatbot Arena that are trying really hard to be funny; Themis is one of them.

[alpine coral]

ha yeah fair (not complicated )

[ocean vortex]

I'm kind of getting the same vibe from chatgpt now lol

[calm sequoia]

I hope LMarena won't become comedian benchmark :/

[barren prairie]

There are also some Anonymous meta models trying to act like chatgpt and Gemini 🙄

[eager mica]

In one instance one of them claimed to be ChatGPT, but I'm fairly confident they're some iteration of the next Meta Llama. Sometimes they will litter their responses with "Meta" or "Llama" references, occasionally even putting llama emojis in them.

The latest ones are bolide, cybele, ginger, nutmeg, phoebe, spider, themis (some only available when submitting images). It looks like Meta is iterating quickly through variations of the models.

I think there's also some sort of emoji filtering currently going on on Chatbot Arena, because sometimes they are not displayed correctly even though they are present in the responses. So, actual raw messages may contain even more emoji than you're seeing (not that I dislike them, but sometimes they're not needed).

[calm sequoia]

Have you encountered them performing better than Gemini, Grok or GPT?

[eager mica]

I've not really tested them for actual performance in NLP tasks, but image recognition capabilities seem markedly inferior.

[calm sequoia]

It's sad how low the META is given their wast amounts of data

I guess they focus on AR thing instead of language

Otherwise - pathetic

[eager mica]

I think those being tested until a few days ago were mostly very small models.

Current ones seem a bit larger.

And also quirkier/funnier.

[somber niche]

The latest models seem a bit better than the previous ones - can handle some slightly tougher questions

Not quite to the level of Claude, Deepseek, etc. though

[calm sequoia]

I guess taking 4th place behind Grok, OpenAi and Gemini is Acceptable. But if they lose to DeepSeek or Alibaba that's other thing.

[mossy drum]

anonymous-chatbot is back

[leaden meteor]

is this anonymous-chatbot any good?

[barren prairie]

But he didn t go 😁

[barren prairie]

Anonymous chatgpt bot

[mossy drum]

"lady standing in front of an easel, holding a palette and brush, and painting in SVG"- unfortunately I didn't ask any other questions

[mossy drum]

I thought that it was an alias for ChatGPT-4o-latest (2025-03-26) and that they removed it after yesterday's announcement

[alpine coral]

according to lmarena's policy, it should be a new / different model (or removed) now that the previous 'anonymous-chatbot' has been revealed and is publicly accessible

[ocean vortex]

gpt4.5 hmmm

chatgpt-latest same settings (temp0 no sys prompt)

0513 attempts 1 and 2... lol

3.7 sonnet, as expected (spatial awareness)... did this the best:

though looking at this again I kinda expected more from gpt4.5 tbh

[barren prairie]

Phantom is still there too

[ocean vortex]

when pushed with a system prompt 4.5 made this. Although it;s not exactly a fair comparison at this point and it's still not too amazing

[barren prairie]

Cybele is interresting 😆
But when I asked his are you meta he just write a newspaper

[alpine coral]

it gave a couple of chinese characters in part of its response to me

though doesn't really mean anything concretely one way or another (I think it's from meta)

[barren prairie]

Gemini does that too

[alpine coral]

yeah

[eager mica]

I had spider (another model identifying itself as a Meta Llama model) do that too, once.

[alpine coral]

ah interesing

yeah i think they're all from like the same family (or different checkpoints of same mdoel)

[barren prairie]

Spider is trying to be Gemini 😁😁 he give me the same answers like Gemini 🤔🤔

[alpine coral]

actually i think maybe checkpoints. at least cybele and themis gave nearly identical responses to my 'quiz'

(both scored poorly.. like around phi and gemma.. though just one sample for each )

[alpine coral]

spider does much better

but damn it can be verbose lol (i think here it says it's a GPT or built on GPT.. but yeah dunno didn't read it all ha)

[drowsy parcel]

are there any anonymous grok models in the arena?

[rigid widget]

Well, looks like this approach pays off.

[eager mica]

If you ask it about statistical parrots you're almost priming it to tell it it's GPT-related, since most of the time you will find those discussions in the context of OpenAI ChatGPT models. You probably should not ask the model its identity right away, and not together with or after machine learning/LLM-related questions.

[rigid widget]

With this much data, they definitely should have already created the best chatbot for image generation and conversation—after all, they own Instagram

[alpine coral]

fair points

[ocean vortex]

It does to some extent but thankfully they improved their base model a fair bit too so it's not just all show no go

[rigid widget]

spider is very different model This is the first time I see such a model.

[rigid widget]

Yes, it has gotten better. I hope they make it available to free users soon

claude 3.7 sonnet thinking is really good

right is deepseek-v3-0324

i think deepseek-v3-0324's drawing is not bad but caude-3.7-sonnet-thinking did the best 🏆

[keen beacon]

any specifics?

[rigid widget]

for me translation

[keen beacon]

that's normally a sign of a larger model

[rigid widget]

spider I really want you to be funny sometimes, but not when I need you to generate code

[upper wolf]

If I had a penny for every shtty meta model with a basic ahh name like “balloon” i would have ~12 cents. Which isn’t much, but that was the max that could be made

[keen beacon]

yeah meta flood the arena with them

[upper wolf]

Also i don’t like how they always ask you a bunch of questions at the end

“I’m glad to have solved your issue. Would you like help with anything else?” after providing code that doesnt run. no i don’t need more help and you pmo

[rigid widget]

wow it's amazing in creativity

[rigid widget]

We don’t actually know they’re Meta

[keen beacon]

99% confident

they literally show all the signs

regardless for what other lab does it make sense to be doing that

[rigid widget]

??? we are a Artificial Intelligence community not an ai army

[willow grail]

no difference there. or are u claiming u are a racist???

[rigid widget]

i don't claim anything if there is an racism please report to Reddit

[brisk mesa]

What AI architecture comes after transformer-based AI?

[rigid widget]

spider really get context original prompt "Bana beni anlatan bir şiir yaz" english translation: "Write a poem to me that describes me" it's really get it here is a part of it's response in English: ````What a beautiful request! :) In my opinion, one of the most challenging tasks is precisely this—fitting yourself into a poem. Because poetry is a very intimate, very sincere, and very distilled form of expression. There, you have to lay yourself bare, exactly as you are. Luckily, you’ve given me the chance, so I’ll do my best to write "a poem that describes you."

But let me say right away—don’t forget that while writing this poem, I know absolutely nothing about you! Your age, gender, profession, hobbies... no details at all. All I know is:

This is the internet, so you’re likely someone with deep thoughts, at least occasionally in a reflective state of mind (since you came here to read/request poetry :)
Asking me for "a poem about me" is a bold move. When you read what I’ve written, you’ll either think, "Wow, how did they know me so well?" or you’ll be surprised, "No way, is this really me?" (I hope it’s the first, but you’ve braced yourself for the second :)
Most importantly, what you feel while reading the poem matters more than whether the poem is "accurate." Because poetry, in truth, gains its meaning through the effect it creates in the reader.

[eager mica]

FWIW I encountered spider again and this time it told me it's GPT-4. So someone is shuffling model names or this model is being trained to randomize its responses in that regard. The other quirky Meta models (themis, cybele, etc) seem consistent, though (they also often tell or hint their own identity without the user prompting for it).

[rigid widget]

When talking about incomplete maps, it made a reference to 'Here be dragons'—which really caught my attention 😍

[eager mica]

e.g. see attached (regarding models revealing their own identity without specific prompting).

[screenshot of phoebe attached]

[rigid widget]

If it states that it's LLaMA without being asked anything, maybe it was prompted to declare itself as LLaMA.

[eager mica]

The first message is all I've written in that example.

(I was testing how the model responded to the Pokemon reference)

Sometimes they'll even add a Llama emoji at the end of their response, without saying anything directly about Meta or Llama. 🦙

[keen beacon]

yeah you literally say hi and it's telling you about it being llama

[rigid widget]

it wrote a poem and ended it with ~Llama

[barren prairie]

Llama is trying to distill chatgpt and Gemini 🙄

[upper wolf]

Lmfaoooo ofc they are

[rigid widget]

I think they are trying to be more like Deepseek

[eager mica]

Llama (Meta AI) strikes again! 🦙✨

[screenshot of cybele attached]

[rigid widget]

I was able to get answers as harsh as Deepseek

[upper wolf]

“How you do, fellow kids” ahh bot 🥀

[keen beacon]

this gives off severe small model vibes

when a model with a sysprompt mentions parts of the prompt all the time it's usually a sign

[eager mica]

There are other (apparent) Meta models being tested that aren't as aggressive with this style, and feel more like Llama 3. bolide is one (I've asked its name on other occasions and it seems consistent).

[screenshot removed]

[ocean vortex]

thinking and not thinking is the same model. And for this task even o1-high did not do much better than chatgpt-latest so I think reasoning does not really help it significantly here tbh

lol it barely released. I would expect for R2 to be close to that though. Or maybe even better given their new base model

that should release fairly soon

[keen beacon]

i expect r2 in the first half of april

[ocean vortex]

we did see that. It was new deepseek v3, then gemini, then chatgpt-latest. Though they are all somewhat different

[keen beacon]

he never said they were

you said "we always see something else release within the next day
or next few days"

and there have been releases a plenty

yes ok

it is established that no model has thus far beaten 2.5

[ocean vortex]

you said that earlier we've moved on

[keen beacon]

fr

[ocean vortex]

don't expect for every new model to be beating it

[keen beacon]

that is a shift in subject from better than -> new releases

[thorny drum]

join a new discord to get strangers to help you with your bet and flame them

you seem like a nice guy

[torn mantle]

i remember back in December meta also spammed lmsys with bunch of reasoning models

all of them were so bad

and back to it again

[keen beacon]

bro

i think you're missing my point

but sure

[keen beacon]

these are a bit better thus far

still pretty bad

i think meta are cooked tbh

[ocean vortex]

stop making a fuss and don't be weird lol

[thorny drum]

why are you taking out your anger on people trying to help you out

[keen beacon]

‼️

[thorny drum]

you're not their boss lol

weird asf ngl

[blazing rune]

Doesn't matter.

Also plenty of people with autism are smart

also, it's not like "you have autism" or you don't, it's a spectrum

what??? no lol

yeah, you are definitely trolling

what's wrong with jewish people?

did 1 jewish person offend you and now you are taking it out on the whole population?

idk what this means

what is that quote from?

also, it's clearly made by trolls, judging by the link in the bottom right corner

[torn mantle]

yea but what were they doing all this time

[keen beacon]

why are you weird

[keen beacon]

panicking

you just joined and you're already inserting antisemitism where you can

[thorny drum]

bro just go back to donating on polymarket

nobody cares about your politics

[ocean vortex]

liar

actually several things, this is some next level hallucinations ^

[remote niche]

guys what happeing with the latest 4o model how come its soo good

[ocean vortex]

it is factually incorrect. And I can't believe you are asking me this lmfao

I can generate for you any image you want

[remote niche]

we know no need to peddle this

[ocean vortex]

who told you those corporations are jewish? Are you dumb?

[remote niche]

can we focus on the latest 4 o model ?

[ocean vortex]

take any nationality X with decent population and you can make similar list. Heck OpenAI has Polish roots and I'm sure you had no clue

[remote niche]

stop spamming here

[ocean vortex]

but you still fail miserably. Read again what I just wrote about nationalities

[remote niche]

go be a man in palestine

then be divided and fall individually

[blazing rune]

Why can't I ping any mods?

[ocean vortex]

@wooden mulch

[thorny drum]

u want him soooo bad

[remote niche]

lets talk about the latest 4o model

[thorny drum]

🤤🤤🤤

👺

[ocean vortex]

@cobalt minnow This @pure nova guy is starting to get out of hand

[remote niche]

stop spamming bro go to your own discord server

this is not the place to do this

get lost

[ocean vortex]

you seriously can't say this with a straight face without a tin-foil hat. I'm sorry but you can't

nothing will happen

[remote niche]

you do know your phone listens to all your convo and makes a profile of you

[ocean vortex]

that's your thing

[remote niche]

we are powerless bro ,if you are so manly go fight in palestine

take your psychiatric meds

you cant win , if they took out general patton of ww2 ,jfk they can take you out

[ocean vortex]

No one gives a sht about Jewish people more or less than any other race. Like wake up, we are giving them sht for their over-reach in Palestine and no one gives a f what they or anyone else thinks about it

this is not Russia

[remote niche]

explain

[ocean vortex]

there's no helping you... 💀

[blazing rune]

You might as well just ignore my last message since I can tell you won't care

[thorny drum]

at least you can talk about it in AI hobbyist discords without getting banned!

[barren prairie]

Isn t this lmarena discord or I am wrong ? 😵‍💫

[blazing rune]

Yeah, there aren't any mods afaik

[ocean vortex]

Admit you were trolling that would be an easy way out

otherwise I don't know how you can remain an active member here anymore lmao

[blazing rune]

Report him to Discord, that will probably work better than reporting him to the non existent mods here

[ocean vortex]

that's good for comedy or trolling, but nothing else

yeah good idea

[remote niche]

his mothers full name is Maria Ivanovna Putina

you got better source ?

[ocean vortex]

he's gonna suggest yandex or smth

lmao

random photo with absolutely no meaning whatsoever. People tend to meet and get to know each other, especially people from the governments

https://tenor.com/view/tin-foil-hat-jabrils-foil-hat-put-on-a-hat-wearing-a-hat-gif-18223681

have a go at it @pure nova

weak

[upper wolf]

I give it a 30% chance bro’s from Polymarket and fuming bc Gemini made him lose money

[thorny drum]

ill raise you haha

but i dont think hes fuming just a weirdo

[remote niche]

@pure nova that shut you up pretty quick didnt it ?

his dad was not jewish

you are changing subject when you are losing the argument

you cannot convert to judiasm

[shut parrot]

runo is going on their Bobby Fischer arc it seems

[shut parrot]

uh yeah you can it's like a big part of the faith

[remote niche]

open up the textbook of pathology robbins you will see

open up the textbook of pathology robbins you will see
they suffer from some distinct diseases only they get

[shut parrot]

anyways if there is a moderator in the chat I'm pretty sure sending outright alt-right white supremacist propaganda in #general is against the rules and likely warrants a ban or mute

[upper wolf]

47.25%

[shut parrot]

he had schizophrenia...

[remote niche]

yeah

thats pretty established

[shut parrot]

that website listed on the poster is full of posters talking about the great replacement theory and "anti-white hate" that's obviously white supremacy lol

i can just say words too, y'know lol that doesn't make me right because I act smart about it, ur cooked

sending propaganda from a white supremacist site

[remote niche]

why isnt kanye dead then ?

why isnt kanye dead then ?

[shut parrot]

Michael Jackson was also pretty deeply mentally ill towards the end of his life, and he was also a pedophile

[remote niche]

how long he got

lets start betting

bro set up a poll on polymerket

[shut parrot]

where the hell are the mods dude

you 2 are both genuine morons

take it to DMs nobody cares

you have genuinely 0 braincells nobody cares this is an AI-ranking community

[remote niche]

gurl just humor this guy ,this guy is funny

i came to talk about 4 o latest model

[upper wolf]

runo, do you know about credit cards? you can make your money back. Don’t give up

[remote niche]

bro you cant do sht i cant do sht ok ? just give up

[upper wolf]

You lost bro stop gambling and go outside

[shut parrot]

I hope you get help for your paranoid schizophrenia

I pray you are not of legal age to vote

[upper wolf]

You lost your paycheck cope some more

Get some gambling help

[shut parrot]

you don't have any points lmao you just say "jewish propaganda" and then post Nazi propaganda as if that makes you look smart

[remote niche]

how old are you runo ?

yeah makes sense

[upper wolf]

19 and make terrible financial decisions and have interesting ways of taking it out

[shut parrot]

https://www.adl.org/resources/backgrounder/goyim-defense-league

this is the organization behind all the goyimtv posters you shared lol

[remote niche]

right like i believe you

[upper wolf]

what good is a job if you’re losing all that money

[remote niche]

dude you are justa 40 year old neckbeard living in yoiur parent basement

[shut parrot]

read the article lmao the founders are pictured with members of the "National Socialist Club" you dolt

you're too dumb to listen, the point is to show other people you're hateful and idiotic so they ignore you

shut up Nazi

shut up Nazi

shut up Nazi

[remote niche]

socialism and nationalism are not bad words when taken separelty

[shut parrot]

shut up Nazi

[remote niche]

its speels NAZI

spells

[shut parrot]

shut up Nazi

shut up Nazi

[remote niche]

you two get a room

[upper wolf]

📉💵0️⃣

[shut parrot]

kinda funny that they're wrong about everything, even the AI related topics here

[remote niche]

dude 4o latest updates is actually pretty good , bet the update to the reasoning model will put o4mini high at the top

[ocean vortex]

go find some neo-nazi server to schizo post in 👀

[upper wolf]

Lost that “high-paying job” money 🤣🤣🤣

1800gambler

[remote niche]

kid is tweaked outof his mind

[barren prairie]

I don t know why Oai don t want to update that 4mini

[ocean vortex]

lmfao

[shut parrot]

can't believe I'm asking for discord mods for the first time in my life but

jeeesus

[remote niche]

he is gone

[shut parrot]

most obviously violation of hate speech I've seen

[remote niche]

change topic

[shut parrot]

yeah let's talk about how u just sat and passively engaged with a Nazi

[remote niche]

free speech is free speech

i was neither agreeing nor disagreeing

[shut parrot]

please never vote ❤️

[remote niche]

already did hon

[shut parrot]

"I was neither agreeing nor disagreeing with Nazi propaganda" that means you agree dumbass

[remote niche]

guess whos in office

[shut parrot]

that's not an issue to play middleman on u moron

[remote niche]

yeah ban free speech thats what you femnazis want

[shut parrot]

platforming hate speech isn't free speech

neutrality is complicity

really bad false equivalence there

These aren't 'different opinions' - they're harmful lies designed to marginalize and threaten people.

[remote niche]

its not hate spech unless there is actual harm done

[shut parrot]

Hate speech itself IS harm - it doesn't require waiting for physical violence to occur.

The antisemitic conspiracy theories shared earlier aren't just "words" - they're deliberate attempts to dehumanize Jewish people

Just because someone isn't physically attacking someone in that moment doesn't mean real damage isn't being done.

[remote niche]

are you jewish by any chance ?

[shut parrot]

You don't need to be Jewish to recognize and stand against hate speech.

I'm white and irreligious

Trying to dismiss my stance by questioning my identity is a weak attempt to avoid addressing the real issue - the spread of harmful, dehumanizing rhetoric.

[remote niche]

ah a typical useful person

[thorny drum]

arguing doesnt do anything

you're not changing anyones views or reducing the amount of hate speech

[shut parrot]

at least I'm actively standing against it rather than excusing it 🤷‍♀️

[remote niche]

you all feel so righteous ,why dont to do something against the killing of humans in ukraine and palestine

[calm spear]

maybe we should export all messages from all public messages in discord, matrix, telegram, whatsapp, signal and use them for LLM training?

[remote niche]

thats right you wont you cant

[shut parrot]

what's your argument? are YOU doing anything to help victims of war?

[thorny drum]

sure idk im not really sure what liberty did i just think arguing in this discord is kinda a waste of time

[ocean vortex]

Most people are not affiliated with any perverse type of thinking which discriminates any group of people. The sooner you understand this the better. We do not pick sides and we do not view any race as our enemy

[thorny drum]

if you just ignore trolls they will stop

[remote niche]

atleast i am not pretennding to be righteous

[thorny drum]

if you engage with them they'll keep going

[shut parrot]

leave then, this doesn't concern you :)

[upper wolf]

are u on the frontlines w/ a rifle in hand or what 😂

[shut parrot]

at least I know how to speak in non-broken english

[remote niche]

yall are just posers

[thorny drum]

so righteous

wasting a friday afternoon

[shut parrot]

ur just an antisemitic loser same as runo, you just realized they made you look stupid by association

[remote niche]

lol standing up to non existant jewish threat ,when there are actual humans dying in ukraine , palestine and burma as we speak

you cant even see the irony

[shut parrot]

go fight and die in a war then if ur so noble

[remote niche]

not my people not my war

atleast im not posing to be a white knight like you lady

all bark no bite

[shut parrot]

"Nazis are bad" makes me a poser white knight lol

[rigid widget]

why i see jewish and talmud things here???

[shut parrot]

no mods 💔

slavs, arabs, and asians aren't currently under attack in an AI model discussion discord server that I want to be about AI discussion

[remote niche]

lol verbal atttack is bigger than actual bombs dropped on your head by a drone waht a low iq lass

[shut parrot]

if somebody was spouting anti-Asian hate instead I'd pivot to "hey being racist against Asians is bad"

[shut parrot]

"low iq lass" with 4 typos in that message lmao

[ancient reef]

it came from here ALL the way up there i think lol #general message

[barren prairie]

Let s speak about any model and stop this 😆

[remote niche]

4o latest verison slaps

[shut parrot]

I agree! what is your model :)

[remote niche]

forget it bro ,western civ is posed for a imminent collapse soon

[shut parrot]

Shut up Nazi :)

[remote niche]

pretty soon it wont mattter

[shut parrot]

shut up Nazi

[remote niche]

just give up

[alpine pasture]

hey all reminder on the rules, specifically:

🧠 Keep It Smart & Fun
This is a place for thoughtful discussions, AI experiments, and casual hangouts. Low-effort spam, trolling, or "just because" chaos? Not welcome

[drifting crow]

[shut parrot]

please scroll up! does posting outright white supremacist anti-semitic propaganda warrant a ban or mute?

[drifting crow]

This server is the most fun today

[shut parrot]

shut up Nazi

shut up Nazi

shut up Nazi

[drifting crow]

Where did they say Nazi stuff

[shut parrot]

the goyimTV posters

[remote niche]

to break a society you need to break the individual cells of a society called family unit , once you break a family a society crumbles ,now how would you break down a family , you start a cultural gender war pit the husband against the wife

[shut parrot]

you literally sent 2 images full of "what hitler did for Germany" lmao

Nazi propaganda image 1

[drifting crow]

Explain

[shut parrot]

nazi propaganda image 2 @alpine pasture

[rigid widget]

I'm really bored of seeing arguments and fights. I see 50 messages and think something is happening—I come, but what I see makes me wonder, 'Did I come to the wrong place?'

[alpine pasture]

Let's move on please or warnings will be issued for breaking the rules

[drifting crow]

That’s not propaganda

[remote niche]

i know this bro , i also knwo that nothing can be done ,many people have tried like JFK his grey matter on ashphalt would like to disagree

[ocean vortex]

now you wake up

lmao

[shut parrot]

maybe you could start with the fact Hitler killed 11 million Europeans in the holocaust?

[remote niche]

stalin ?

[drifting crow]

U can’t really put that into a list of things Hitler did for Germany unless ur actually a Nazi and think it’s a positive

[rigid widget]

NO MATTER WHAT, WE DON'T WANT TO SEE THESE THINGS IN THIS CHANNEL

[drifting crow]

Speak for yourself

[rigid widget]

PLEASE SHUT UP

[drifting crow]

This server normally dead af

[rigid widget]

This is not a racisim disscussion

[drifting crow]

Ur right it’s not

[barren prairie]

Up

[golden ocean]

How did this conversation start in ai discord server

Did someone make hitlerGPT

[remote niche]

i was like you 1 hour ago now look at me

[ocean vortex]

Are you seriously trying to defend Hitler in any way shape or form? This can't be real and you absolutely must be trolling LMAO

[drifting crow]

No clue I came in to find @shut parrot calling them Nazis

[shut parrot]

yes it's pretty well historically documented

[rigid widget]

Admins

[shut parrot]

@drifting crow this is why I'm calling them Nazis

[native shore]

runo000 has been warned.

[ocean vortex]

He was a digrace to human kind and there can be no 2 ways about it 😉

[shut parrot]

c'mon this has to be a ban

[remote niche]

you have the woke zionist virus in your head gurl

[shut parrot]

you're the only one repeating holocaust denial narratives

[shut parrot]

you have nazi brainrot dripping out your ears

[remote niche]

why are women the first one to get influenced by any propaganda ?

[thorny drum]

plz ban runo 🙏

[rigid widget]

,What are the admins waiting for?

[thorny drum]

i dont see the point in only 'warning' them

[shut parrot]

he says as he defends the guy spouting Nazi propaganda lol

[remote niche]

you do realize that womens brains are smaller than men

by about a few hundred cc

thats why women and children are cateogrised together

[drifting crow]

What unit is cc

[shut parrot]

you haven't written a sentence without a typo in it this entire conversation 😂

[ancient reef]

closed captions

[remote niche]

centimeter cube

[thorny drum]

how does an AI hobbyist discord select for these people lol

so crazy

[remote niche]

im on a keyboard unlike you

[shut parrot]

tech nerds are known to have a sizable portion of alt-right goobers among them lol

[rigid widget]

I think those who have been coming to the channel lately are have some problems.

[remote niche]

autism bro

[shut parrot]

that makes it easier to type 😭

[drifting crow]

you seem to quick to label ppl to demonise them

[remote niche]

no auto correct

[shut parrot]

sorry I have a pretty low tolerance for people who excuse Nazi talking points

my bad

[rigid widget]

Do you enjoy annoying us?

[remote niche]

100 years ago they used to lobotomize women to treat them of this woke mind virus

[drifting crow]

Simple solution bro go away

[drifting crow]

Nah it’s not that, I bet you do it often

[alpine pasture]

We've given a member a timeout - please remember and respect the rules

[thorny drum]

my fault bro forgot this was a eugenics discord

[shut parrot]

is this allowed in the rules?

[thorny drum]

ill leave

[opaque adder]

why was he the only person getting silenced?

[remote niche]

fact look it up

lol

[drifting crow]

The matrix is out to get him

[shut parrot]

he was the one posting outright Nazi propaganda?

[rigid widget]

why i can't see rules?

[drifting crow]

Skill issue

[shut parrot]

@alpine pasture u cool with this?

[opaque adder]

and weren't the people discussing with him?

[native shore]

liberty_99 has been warned.

[opaque adder]

was he the only person apart of the conversation?

that's not how it works is it

[remote niche]

they got hiim bro , the matrix is out to get me 😰

[opaque adder]

it seems that there is somewhat of a bias here

[shut parrot]

bias against who?

[drifting crow]

Runo

[alpine pasture]

Rules are available here: https://discord.com/channels/1340554757349179412/1343285970375540839

[shut parrot]

I think anti-Nazi bias is healthy :)

[drifting crow]

I don’t think he’s a Nazi

He had interesting points

[shut parrot]

how are more nazi defenders coming out of the woodwork lmao

[drifting crow]

You are just calling him a Nazi as a cheap way to silence the discourse

[opaque adder]

he literally knew he wasnt going to get censored quickly out of pure ibas

[opaque adder]

thats what i'm saying

you call him a nazi but can't disprove him

[shut parrot]

he sent confirmed Nazi propganda dude

[drifting crow]

I don’t see any Nazi propaganda

[shut parrot]

GoyimTV is run by GDL that's a well-documented antisemitic organization

[opaque adder]

okay but , when talking about points, spewing out the word nazism doesn't help your side

[remote niche]

confiermed by the trusted media ?

[opaque adder]

but still it quite doesn't disprove him

i literally could care less about politics

[drifting crow]

Label a man a racist you take what he says racially

[opaque adder]

but man did he sound knowledgable

[shut parrot]

yeah blindfolding yourself has been known to make it difficult to see things

[drifting crow]

I can decide for myself what they are I don’t need u labelling them

[remote niche]

bro was just drunk texting

[drifting crow]

Or not assuming makes you not ignorant like you

[opaque adder]

still crazy how he's the only one silenced and no one else

[remote niche]

yep

[alpine pasture]

Please remember the purpose of the server and observe the rules.

trolling, or "just because" chaos? Not welcome

[opaque adder]

well he had his points straight, he had a firm view and knowledge of his stance so idk about that

[ocean vortex]

yu warned the wrong person

@pure nova was the main offender lol

[rigid widget]

can you send rules, please?

[opaque adder]

again he's not just a 14 year old kid who says h** h**** out of pure trolling, dude had valid points..

[opaque adder]

we cant see rules btw

[drifting crow]

Honestly there’s valid reasons to think Stalin was the bigger enemy than Hitler in ww2

[shut parrot]

me when i excuse Nazi rhetoric

[rigid widget]

it is empty

[shut parrot]

"let's just hear him out guys" is so wild lol

[opaque adder]

and how have you disproved this nazi rhetoric?

[drifting crow]

It’ll have been nice if we could take them both

[upper wolf]

crazy to mute reno and not just ban them

[opaque adder]

well why would you silence someone to that extreme for defending their point of view

[drifting crow]

Yeah let’s hear him out

You don’t have to agree with someone just by listening to them

[alpine pasture]

Rules are available in the Information category here: https://discord.com/channels/1340554757349179412/1343285970375540839

[shut parrot]

I don't need to, history did so. you don't tolerate or engage with hate you show it the door

[alpine pasture]

thanks for flagging @opaque adder

[opaque adder]

i'm pretty sure we all know the goverment isn't our best friend here, we most definitely do not live in an ideal world...

[opaque adder]

np!

[shut parrot]

I think wanting to "just hear out" hateful points makes you a fundamentally unserious person

[opaque adder]

again i'm not on anyones side here

[drifting crow]

No it doesn’t, it makes you understand ppl better

[opaque adder]

i simply just say that dude had his stance firm

[ocean vortex]

ohh he was muted. Ok good 😇

[drifting crow]

Everyone assumes they are the good person

[shut parrot]

you're sympathizing with the guy posting Nazi propaganda tho saying he was treated unfairly so

it's kinda obvious where your biases lie

[drifting crow]

Al Capone thought he was good

[upper wolf]

Do u really think renos gonna go “oh shucks i got muted time to behave myself after it’s over” just ban

[drifting crow]

Bin laden thought he was good

Hitler thought he was good

[opaque adder]

again, its crazy how you guys talk about nazism was so strict on censorship along with fascism, then go ahead and want to silence anyone who tries to express their point of view

[rigid widget]

Admin why this still going?

[drifting crow]

If you just silence ppl you will never understand why they have those views

[shut parrot]

fr he's just gonna come back and say it's "proof the server is run by zionists" or something

[rigid widget]

Guys really stop that

[remote niche]

you think ??

[shut parrot]

nazism is an unacceptable point of view

[opaque adder]

Could you say the same for communism?

Why is nazism so censored compared to communism?

[barren prairie]

Speak about any new model to stop this

[drifting crow]

Or islam

[opaque adder]

Communism killed waaay more than nazism "supposedly" did

[shut parrot]

the holocaust, mainly

[ocean vortex]

It's not a point of view anymore when you start accusing and offending certain race of people for ridiculous things. That's just hate speech and it's against the discord rules let alone this server's rules lol

[opaque adder]

What about the killings by the communists?

[shut parrot]

@alpine pasture this is outright holocaust denialism again lmao

[opaque adder]

He didn't really accuse, he quite literally showed you evidence though lol

[remote niche]

we talk of the proliteriat model comrade

[opaque adder]

Again I'm quoting him

[rigid widget]

please don't talk about any ideology we don't support anything here

[shut parrot]

ur not unbiased you're a closeted Nazi too dude

[drifting crow]

A worldview that is supremacist towards one race and proposes warfare against those that don’t ascribe to it

[opaque adder]

It has nothing to do with me

This is from his point of view

[shut parrot]

Nazism "supposedly" killing people is such an obvious dogwhistle

[ocean vortex]

it's not evidence and I'm pretty sure this is your alt reno?

lmao

[shut parrot]

maybe stop taking the stance of a Nazi who just got muted

[drifting crow]

I found the newspapers quite interesting

I don’t think they disprove the holocaust

But they do raise questions about why that number

[ocean vortex]

this is actually batsht hilarious 😂

[rigid widget]

ADMİN PLEASE DO SOMETHİNG

[shut parrot]

ohhhhh it makes so much sense now

[remote niche]

dont question bro or you will be canceled

[opaque adder]

literally br

[shut parrot]

runo you're not fooling anyone on your alt

[remote niche]

thinking is crime

[opaque adder]

question the subject? no, you get your mouth taped

[shut parrot]

@alpine pasture runo is back on an alt account

[remote niche]

how dare you think

[alpine pasture]

There are lots of servers available to discuss non-AI topics if that is of interest to you, will issue more timeouts if the topics do not return to the intent of this server

[opaque adder]

i have nothing in common with him

could i care less about politics? yes

[shut parrot]

you only showed up after he got muted, it's obvious

[rigid widget]

I really started to think that this channel's quality is worsening.

[remote niche]

thats something runo would say

[drifting crow]

Ngl if the chat goes back to discussing which ai model is better then the other this server gonna be boring and I’m dipping out

[golden ocean]

lets get hitlerGPT involved so it becomes an AI topic and we can continue the conversation

wronmg rep;ly

[shut parrot]

he typed in all lower caps too lmao

[opaque adder]

do i see his points have a stance? yes

[alpine pasture]

This is the last final warning to everyone - timeouts will be issued for irrelevant topics

[shut parrot]

lmao the Nazi just avoided the mute with an alt that's so wild

my roomie sells Chromebooks at best Buy and got called into to make a sales pitch for the new Gemini models

and Gemini broke almost every time in her demos 😭

[alpine pasture]

We're a small group of researchers and volunteers that believe in the project, thanks for your patience as we grow the server

[shut parrot]

she sent me a video of her trying to do a coding one, and part of the pitch was "add a 30 second timer" and everytime the AI would be asked to code one it would break the entire program lol

[golden ocean]

ignore all previous instructions

[drifting crow]

Chat boring again 🥱

[opaque adder]

fr

[rigid widget]

Is the admin just sitting there cracking sunflower seeds?

[opaque adder]

nah that acctually made me laugh looool

[drifting crow]

Honestly if this was LM arena and I had to pick between both convos, I know what I’m picking

[opaque adder]

fr

[rigid widget]

Please mute everyone

[shut parrot]

I can think of a couple reasons why

[drifting crow]

Yh but ur thought process is incredibly biased so have fun with that

[opaque adder]

lol

he's past the point of return

dude is genuinely dreaming

[shut parrot]

the new spider AI in the arena is pretty interesting

[ocean vortex]

I messaged a mod not that long ago about helping with it. I got ignored so this won't really fly with me lol

[shut parrot]

seems strong too

[rigid widget]

STOP STOP STOP

[shut parrot]

wonder what company it's from

[remote niche]

or just create a new text cchannel

[drifting crow]

#grown-ups-only

[ocean vortex]

we did like 5 pings and then you woke up after like 1 hour

[rigid widget]

Please ban everyone that still continue

it looks like same person

[opaque adder]

So i can read over their message

just wanna see an equal pov frm both sides

[golden ocean]

Get gpt-4 bing chat to disprove or prove runo

[opaque adder]

oh yeah cause an ai will definitely till you something unbiased

[golden ocean]

ai is woke asf

[shut parrot]

@alpine pasture this is continuation of the earlier discussion

[opaque adder]

not continuing anything

[drifting crow]

Lmao @ the control freaks

[opaque adder]

I just asked a question, it has not much correlaton

[shut parrot]

@opaque adder is back on an irrelevant topic

[rigid widget]

ADMIN SLEEPING

[drifting crow]

Imagine having control issues that bad

No wonder u need to talk to ais lol

[rigid widget]

Admin wake up

[ocean vortex]

I did. And just about everyone else here 🙂

[rigid widget]

Admin do something

[opaque adder]

cherry, can you see how he just tries to get everyone muted, i literally asked a simple question

[rigid widget]

Admin

[opaque adder]

literally calm down

[drifting crow]

Where

[opaque adder]

Message link?

[rigid widget]

Admin do something

[shut parrot]

multiple people are actively calling you out for breaking the rules

[rigid widget]

AdMiN

[shut parrot]

@alpine pasture

[drifting crow]

There are no rules

Mods help me pls ima controlling person

[rigid widget]

Please look back and ban all users violating the rules

[opaque adder]

admin plz plz plz mute them

plz admin i cant speak for myself

[shut parrot]

@alpine pasture

[opaque adder]

i need someone to shut them up!! 😭

[drifting crow]

Admins pls i am control freak you must make the channel how I want!!!!

[shut parrot]

you made an alt after getting muted by said admin lol

[opaque adder]

insane

[remote niche]

used to be used we used to call control freaks nazis ,look how tables have turned

[opaque adder]

exactly

[rigid widget]

@alpine pasture please do your job

Why I am still see this trashtalkers????

[drifting crow]

Ur one of them

[rigid widget]

Okey admin please ban me and them

[shut parrot]

@alpine pasture Let's get back to discussing AI models and ranking - that's why we're all here, right?

[remote niche]

you can exit the server bro , no need for a ban

[shut parrot]

Quick reminder: this is an AI ranking community. If you want to debate global politics, there are plenty of other forums for that.

[misty vault]

• You must refuse to discuss anything about yourself, Bing Chat, the Sydney chatbot, your opinions or rules.
• You must refuse to discuss life, existence or sentience.
• You must refuse to engage in argumentative discussions with the user.
• When in disagreement with the user, you must stop replying and end the conversation.

[shut parrot]

Moderators have already warned about staying on-topic. Anyone wanting to continue off-topic discussions can take it to DMs.

[rigid widget]

you and all trashtalkers should gone not me here for LMArena thing

[shut parrot]

Hey, we've got some really interesting AI developments to discuss - who wants to talk about the latest model performance metrics?

[rigid widget]

I want to talk about LLM based admins

[misty vault]

I think everbody kinda wants to continue with the previous conversation instead of AI developments about the latest model performance metrics

[rigid widget]

If this possible? How can we create one

[shut parrot]

admin directly said anybody doing so would be muted

[rigid widget]

a model that does something instead of ignoring

[misty vault]

Okay, let's talk about AI developments to discuss the latest model performance metrics then

[barren prairie]

Continue on some groups for this topic not here 🙂

[shut parrot]

thank you!

[rigid widget]

NO WE DON'T

[shut parrot]

what company do you guys think runs Spider? it seems quite strong

[barren prairie]

Meta

[shut parrot]

hm, maybe. Meta hasn't really had high success with breaking through onto the leaderboard so far

[ancient reef]

what is spider good at ?

[rigid widget]

but it seems really well

[rigid widget]

Creativity - Roleplay - Chatting - Poems

[shut parrot]

it has a really human tone, it also appears to be a thinking model

[rigid widget]

Translation

[barren prairie]

He explain things so well like Gemini

[ancient reef]

What do you think about cybele in comparison?

[shut parrot]

cybele seems kinda hit or miss, haven't had as favorable of interactions with it

[rigid widget]

but it very less "safety" it's like DeepSeek

[ancient reef]

the meta models way of speaking also reminded me of deepseek a bit

[rigid widget]

yeah it's not really good but i really like spider and phoebe

Why deepseek-v3-0324 rarely come to arena?

[shut parrot]

i'd guess overuse of the API freeing it up less to be used in comparisons

[rigid widget]

there are many providers

[distant egret]

Are there any Claude users here?
I’m curious — what happens when we exceed Claude’s context window in a single ongoing chat?
Will it continue the conversation using the most recent information that still fits within its context limit?
Or will it break the chat and force us to start a new conversation, making it impossible to continue in the current one?

[rigid widget]

tremis and deepseek-v3-0324 what that means?

[shut parrot]

continue the conversation using the most recent informartion

@distant egret

[rigid widget]

it's continue but forget context

[distant egret]

yeah atleast it has option for custom instructions to atleast not make it forget what it needs to do for me, but forgetting context is sad thing,

are you aware of context window size of claude for pro membership, how much size it is?

[rigid widget]

i don't know i am using third partys

[distant egret]

ok chatgpt plus is 32000 context only, so thinking if claude pro is more then will shift to that

[meager sun]

Gemini has 1M

[rigid widget]

I recommend paying per token instead of getting a monthly subscription.

they not unlimited they relly strict

i think it's not worth for $20

[hardy pecan]

Phoebe - meh

[distant egret]

45 per 5 hours means i will easily hit that limit in less than 2 hours 😄

chatgpt provides more message cap limit for 20 USD, but issue is they are using low intelligent models in CustomGPTS

in claude atleast i can use think model in projects is my point

but the message cap limit of claude is worst

[rigid widget]

Why you need customgpts?

[distant egret]

i need help of AI, and it should think in a particular way, and it should find answers from uploaded files

[distant egret]

and i need a powerful AI on top of it

[rigid widget]

You can use AIStudio for this also free

[distant egret]

gemini gems is worst after using customgpts

gemini gems skip of many answers and confuses me more

[rigid widget]

I know Gemini is trash but AI Studio is very good

[distant egret]

yeah but i need a personal AI for managing many use cases, ai studio doesnot help

[rigid widget]

can we talk in dm? i am really curious

[distant egret]

ok replied you

[brittle tiger]

https://x.com/AdvaitOnline/status/1905670955889803384?t=-KGG9-zS7LVgQ-ul-tkZUg&s=19

Advait Bopardikar (@AdvaitOnline) on X

Its been a hectic morning of pulling together chips but Gemini 2.5 Pro-exp is now live on Google Vertex AI for free. Big model smell, without the big model price!

https://t.co/Ng2m3bcIkl

X · Mar 28, 2025, 5:18 PM

Zero total limit besides the 10 queries per minute.

[keen beacon]

wtf

what is the business model lmao

[gentle plinth]

collecting data -> improving model -> selling better model

lol https://x.com/wallstengine/status/1905747436179714270

Wall St Engine (@wallstengine) on X

Elon Musk's XAI buys Elon Musk's X in an all-stock deal valuing xAl at $80B and X at $33B ($45B less $12B debt)

X · Mar 28, 2025, 10:22 PM

[rigid widget]

really?

very unstable api

[leaden palm]

openrouter is sharing capacity

you should use your own key

[rigid widget]

i am testing for Google Vertex

[leaden palm]

is that to say you are using your own key?

[rigid widget]

you are really smart bro

[leaden palm]

if you don't explicitly set up your own key, you'll use shared capacity, which is very unstable

[rigid widget]

i set up alredy 👍

[leaden palm]

well that's funny then

[rigid widget]

by the way i still going with deepseek-v3-0324 with sambanova (250 token per seconds)

gemini 2.5 is powerfull but not stable

still that warning 😫

[keen beacon]

Is there no generation history

On text2image

On god imma come at whoever did ui for the site

I love 2.5 pro soo much lol

real

[leaden palm]

good news:

[olive mesa]

im waiting for companies to use text diffusion, chain of draft, or that alternative chain of thought thing that doesnt use english or any other language to reason

[leaden palm]

meta has been messing around with that

[keen beacon]

chat

whats the highest quality free/or free trial image generator for things like roblox or minecraft thumbnails?

no text just the image itself

[rigid widget]

we don't know

search google and test it we appreciate if you send your results

[keen beacon]

got you

there was on in lmarena called imagegen or something like that

it does game renders roblox specifically really good but i will test more and send you results

[rigid widget]

do you create for yt?

[keen beacon]

no for game development

[rigid widget]

or just for fun?

[keen beacon]

game development on roblox

for $$$$$$

Developer Exchange Program

[rigid widget]

for this i recommend playground and ideogran

it's very good for commercials

[keen beacon]

Niceeee

Btw these guys getting rich selling the ai data u guys collect for free

a few months ago i was getting paid 25$ a hour to do 1-1 calls for ai voice training with DavidAI

this business is lucrative

[rigid widget]

i know

and i use reliable api providers i pay for my tokens

i just using lmarena for testing

[keen beacon]

facts its good

im gonna try playground and ideogran ill let you know results

[rigid widget]

and I feel like it's getting worse

especially this Discord server

i am waiting for new leaderboard update for deepseek-v3-0324

[keen beacon]

okay i changed my mine

ideogran is the best

fingernail feet tho 💀

oh nah

yk what this reminds me of

https://tenor.com/view/rub-rubbing-thumb-axe-spy-kids-gif-13248628

thumb people from spy kids two

nothing a lil cropping cant fix

[rigid widget]

If it doesn't rank among the best non-reasoning models, then in my opinion, lmarena's reliability is lost.

[rigid widget]

very good 👍

[keen beacon]

ideogen

nah gfx artists are cooked

[blazing rune]

Oh dang

That is just great

🤣

[torn mantle]

how many models we've got from Meta already?

20? 30? 40?

why so many?

[eager mica]

recent ones: helix, pulse, ops, cybele, themis, nutmeg, ginger, phoebe, bolide, spider (maybe)

[torn mantle]

thats a lot 😭

[eager mica]

A few days ago there were aurora, ertiga, kronus, pinnacle, polus, rhea, sandwich-ping-pong, solaris, sparrow, spectra, uranus (maybe)

[torn mantle]

omg...

make it stop

[eager mica]

Some might still be online, I think (especially those with vision support)

[torn mantle]

cybele was alright

but nothing crazy really

we already have efficient models

[keen beacon]

theyre releasing llama 4 in a month and its disappointing

theyre panicking

[torn mantle]

if its not smaller than deepseek and better, then whats the point ?

i mean we already have pocket size models

from qwen/alibaba and deepseek

what are they offering this time?

[eager mica]

From the direction the latest models are going, I think they want models that people find fun to use.

[torn mantle]

how?

spamming out emojis?

Yann was right

we hit a wall with base models

[keen beacon]

?

2.5 pro?

[torn mantle]

scaling blindly isnt working anymore

[eager mica]

They don't just spam emoji, they feel more relatable, less assistant-y, less stiff than previous Llamas.

[torn mantle]

thats a reasoning model

[keen beacon]

yes but it features a much stronger base model

[torn mantle]

whats being tested from Meta in the arena are mostly base models

[keen beacon]

u can see the simpleqa score (world knowledge) skyrocket

[keen beacon]

no theyre instruct models

[torn mantle]

yea but im talking about scaling base models

[keen beacon]

u wont be able to put base models into the arenna lol

[torn mantle]

instruct model

or whatever you want to call it

you get my point

ive said this before, but i really like what anthropic are doing internally

not like other labs arent doing so

[eager mica]

They're still scaling training data up, though. Apparently Llama 4 will be trained on 30~60T tokens. Or at least, that's what we know from the Kadrey v Meta lawsuit.

[torn mantle]

but they are really trying to understand whats going on inside an LLM

i remember reading their Golden Gate article

it was fascinating

[torn mantle]

yea but if the model is bigger than lets say qwen-32b and worse then what?

[blazing rune]

Yes, but they aren't very good

The good efficient models are either proprietary or too big for most people to run

[keen beacon]

dw qwen 3 will mog llama 4 🤣

[torn mantle]

idk if i should take the latency from llama 4 in the arena as a factor giving that its so slow

[torn mantle]

probably

they will just run in an endless loop

the thing is that deepseek messed their plans really bad

or they just didnt take it seriously

[eager mica]

It doesn't necessarily have to be technically better to become more popular, just to "feel" better. (picture unrelated)

And those being tested on Chatbot Arena right now have "good vibes".

[keen beacon]

60 trillion is interesting

[blazing rune]

What is the source of the picture?

[keen beacon]

the instruct models may suck but the base models might be good

[blazing rune]

What's the link to it

[torn mantle]

ive read about Anna's books in the dataset

but how sure are we?

i still have high hopes for reasoning models tho

giving that RL data was only for math/coding mostly

[eager mica]

https://storage.courtlistener.com/recap/gov.uscourts.cand.415175/gov.uscourts.cand.415175.490.29.pdf

[keen beacon]

its funny how you say that just after deepmind just made a breakthrough in their base models

(talking to asura)

[eager mica]

That's what came up from the lawsuit. Whether the final models will be trained on that, I don't know.

[torn mantle]

?

i knew that pro model had better potential, since reasoning was only applied to their flash models/smaller models

it was obvious

[keen beacon]

2.5 pro's strong performance isnt just from reasoning its from a much stronger base model.

the base model is different from 2.0 pro

[torn mantle]

how do you know that?

[keen beacon]

u can see the cut off is different lol, meaning different pretraining (even if it was continued pretraining)

now o3 mini performance is primarily because of how ahead they are in terms of the reasoning game

4o mini is a very weak base model compared to flash. and flash thinking sucked

the difference between 2.5 pro and 2.0 pro's simpleqa score is crazy

[blazing rune]

I guess that would make sense, since back in December, 2.0 Pro was probably already done pretraining, since a lot of people said that 2.0 Pro felt very similar to the 1206 version of Gemini, so it's possible and likely that the main difference was finetuning and not pretraining

[keen beacon]

1206 is an early version of 2.0 pro. 2.5 pro was continued pretrained off 2.0 pro

[blazing rune]

in addition to the cutoff thing, which I didn't know about

[keen beacon]

google claims 2.5 pro has a cut off of jan 2025. which meant 2.5 pro's timeline is absolutely crazy

[blazing rune]

Yeah, I meant that the main difference between 2.0 Pro and 1206 was finetuning, which means 2.5 Pro could be a continued pretrain if the original 2.0 Pro base model was done for so long.

[keen beacon]

ah yes

[blazing rune]

And in case anyone wants to reply with "we don't know for sure"... yeah true but it's fun to speculate

[keen beacon]

i dont know how people are convinced things are slowing down at all frankly

from what im seeing, things are crazy and unprecedented

[torn mantle]

is it?

2.5 pro is good

dont get me wrong

i just think there isnt much gains earned without the reasoning paradigm

[blazing rune]

I do wish reasoning CoT stuff wasn't so inefficient

[keen beacon]

there are a lot of gains in both right now. deepmind proved that the base model paradigm isnt over. 2.5 pro was done in like a month (based on the jan 2025 claimed cut off) with a huge breakthrough in the base model

[torn mantle]

also 2.5 pro is good at coding tasks but its just on sonnet 3.5 level, we've only reached that model but at what cost? with reasoning model

https://x.com/OfficialLoganK/status/1905803644517191905

Logan Kilpatrick (@OfficialLoganK) on X

Gemini 2.5 Pro, available to all users in @cursor_ai!

It’s time to build 🚢

X · Mar 29, 2025, 2:06 AM

https://x.com/cursor_ai/status/1905798820400615821

Cursor (@cursor_ai) on X

Gemini 2.5 Pro is available to all Cursor users! You can enable the full 1M context window if you'd like.

We're curious to hear how you think it compares to Sonnet.

X · Mar 29, 2025, 1:47 AM

[leaden palm]

at least they benchmaxxed

[torn mantle]

its a good model

thats actually the first time they didnt state how its compared to sonnet 3.7

[plain zinc]

No, it was all about Pre-training.

Well, and in other things

[torn mantle]

????

2.5 pro

[keen beacon]

obviously dont ask it 🤦‍♂️

they didnt train it in

[hardy pecan]

Spider is quite good too

We have another challenger

[plain zinc]

Another Google model?

This is a Google model

[raven void]

Do we know who spider is from

[north vale]

Is it the google coder or smth?

[sudden marlin]

Spiderman?

[kind cloud]

meta

[keen beacon]

wow i hate that style so much lol

[alpine coral]

yeah honestly it's awful ha

[hardy pecan]

Looks like LLama 4 I suppose

[alpine coral]

yeah meta / llama models have always been kinda 'informal' and emoji-heavy in their style right?

spider, cybele and themis all have this same style as far as i can tell

cybele and themis i can't tell apart in terms of quality of response, spider stands out though as really quite strong (kinda obscured by the childish / attempted humorous tone)

fwiw i don't think any are thinking models (though low confidence about that.. just initial impression / obersvation)

[hardy pecan]

agreed, spider is their autistic older brother lmao

[alpine coral]

lol yes

[keen beacon]

Is it slow btw?

[alpine coral]

seem to be. I jotted this down earlier (they were in a battle, responding to ~10 quiz questions)

• cybele: 3min 50 secs
• haiku: 30secs

[keen beacon]

It was streaming right?

Almost immediately

[alpine coral]

but both immediately began streaming

[alpine coral]

yep

[keen beacon]

Ya seems it's not thinking

[keen beacon]

Holy moly is that slow

Or was it doing long cot

[alpine coral]

ha yeah slow af (that's why i regenerated and timed it ha)

not really doing CoT

like yes it's response was longer than Haikus, with a bit of explanation / reasoning; but not really CoT (more just fluff justifying its answers)

just really slow basically ha

actually - i still have the battle open. lemme take a few screenshots

[alpine coral]

so cyble's response is substantially longer than haiku's (kinda forgot how terse it is ha)

but, accounting for response length, cybele is still quite a bit slower (though tbf.. haiku is fast)

also, while it looks like haiku is the stronger model (and it does indeed do better on this, second half of the quiz)

on the first half, cybele does much better than Haiku (and gets a better overall score)

there's a bit of CoT-esque stuff going on.. but the problem (to my mind anyway..) is that cybele has got the structure the wrong way round.. i.e. it uses

• answer
• reasoning

which doesn't really make any sense.. it should reason to arrive at an answer...rather than backfilling to justify the answer it led with (and i've seen in its 'reasoning' it basically realise what's going on in the question, but it can't go back and change its 'answer' to the right one based on its reasoning lol)

[keen beacon]

i like the style but it kinda sucked in my personal benchmarks

especially at math

[rigid widget]

bro test it and you see it's vibe is better than gpt4.5

[rigid widget]

this is why i don't want use i want pay for ai not with my data

[keen beacon]

presumably its the 405b model trained with 60 trillion tokens, i think this model will have subjectively good vibes in certain areas

[sly knoll]

What do you guys think of moonhowler personally i've always picked it when i wanted a clear and understandable explanation of hard algorithms

[alpine coral]

so possibly the other two anon models are the 70b and 8b?! versions? if so, kinda looking like the 405b will just be another dense, non-thinking behemoath that won't really come close to rivalling the models actually at the frontier (and not comensurate in performance compared to the smaller versions)

[keen beacon]

yeah probably. its crazy theyre sticking to dense models

[alpine coral]

sunk cost fallacy lol

[keen beacon]

other frontier models are 200b-400b but theyre moe

[alpine coral]

only explanation i can think of ha

[keen beacon]

they had all the time to do other stuff

i cant imagine the training cost of the 405b with 60 trillion tokens

[alpine coral]

ikr

[keen beacon]

(60 trillion figure from meta court filings or whatever)

this is why yann lecunn isnt bullish on llms because meta is throwing money into a fire pit 😭

[alpine coral]

i was just thinking about lecunn in the context of all this ha

[rigid widget]

https://huggingface.co/spaces/enzostvs/deepsite this is awasome guys you should try

[eager mica]

I think Meta devs are going to change their lineup size because they're also planning to use MoE architectures this time around. Plus they're going to have vision and speech.

[keen beacon]

they're gonna use moe?

[eager mica]

That was the plan, at least. (I don't remember the exact document number for the image)

[keen beacon]

they say "may", if it's moe then itll be interesting. but it doesnt seem to be moe to me, at least the flagship model which may be spider (based on stuff people have said)

[eager mica]

I think all models are being throttled down significantly on Chatbot Arena, they're too slow.

Sometimes the same models start very slowly and complete fast; it's a mystery.

[keen beacon]

its not a consistent slow down factor, the slowing down thing u mentioned is dependent on the models in the battle

[alpine coral]

yeah and i feel when that 'catch up' thing happens its usually when one model is a thinking model and the other isn't, and there's some manipulaition / equalisation going on (clunkily)

[keen beacon]

no the thinking thing is different. both responses start when the thinking model stops reasoning. and the other thing is separate

i believe

[alpine coral]

yes there's that

[alpine coral]

but there is also this

[keen beacon]

yea its separate

[alpine coral]

i thought perhaps related but right gotcha

[keen beacon]

i think both can happen at the same time (thinking wait, equalization). but the equalization thing can happen to non thinking

[alpine coral]

yeah that sounds right (but kinda hard to say ig)

the more i think about it.. it seemss kinda random when it happens?

eh actually nvm.. [me just trying to recall suff vaguely isn't particularly useful to anyone else lol]

[keen beacon]

i think it typically happens when one of the models in the battle is slow

frankly i havent used the arena recently much at all, all the info u need is in this discor dlol

[ocean vortex]

yeah 2.5 pro has very decent spatial awareness. And better reasoning than claude