general

[ocean bison]

Agreed too! I generate a ton of music WITH AI!

What MiniMax does best? TTS and music generation!

[weary galleon]

Agreed.

[hushed gyro]

Here ya go

[ocean bison]

Everything else I shouldn't use!

[ocean bison]

Do Claude 4.5 Opus or 4.5 Sonnet

[hushed gyro]

Gimme a min 👍

[ocean bison]

Or Gemini 3 Pro

[ocean bison]

Maybe both models!

[native yarrow]

K2 still writes a lot like an ai

[hushed gyro]

Hmm

[native yarrow]

AI models arent optimized for writing ngl so

they're all gonna be trash

gemini 3 is really good but yk, unfortunately it doesnt have good memory

actually yk what

gemini 3 pro messes up its OWN writing because it thinks too hard

3 flash is better for writng but its output is so short

[hushed gyro]

Looking at the stats, Kimi K2 is one of the best models IN THE WORLD.

It is on par with 5.1 / 5.2

[old edge]

Can somebody explain to me how LMArena updates the data of preliminary models on the text leaderboard?

[native yarrow]

open source at least yeah

[torn mantle]

yea

btw

how is this useful?

i dont understand...

data speaks for itself and you are comparing previous checkpoint with a new one?

isnt this the most expected result?

well at least they are not faking it and putting SOTA models in comparison

[tidal grove]

Guys, what should I do if I don't have a video generator icon?

[brisk turret]

Petition to introduce all the Gemma models to the leaderboard

Atm the smaller ones 270m and 1b are not on there

[weary galleon]

👍

[limpid tendon]

Does anyone using this for r@oblox?

[ocean bison]

Ask @echo aurora, he will know!

[echo aurora]

It's an experiment meaning it's going to be random if you get it or not. There isn't a way to guarantee you'd get the feature to show up.

[tidal grove]

Great, when will it be published?!

[echo aurora]

We don't have an estimated date for if/when we launch this feature to everyone. However, we'd be sure to put out an announcement so keep an eye on #announcements

[thorny schooner]

Please tell me that the team is aware about the complaints of the verification process because why does it love so many bicycles

I'm counting like 10 before the timeout happens if not easily more

[tidal grove]

This is really cool, I'm glad you're expanding!

[thorny schooner]

Well I'm glad you're excited

( even if i can't even log on to my account)

[tidal grove]

It will be cool to make videos for free 😎

[echo aurora]

Yes, chatted with the team this morning about community feedback about this.

why does it love so many bicycles
Can't say this was specifically brought up.

[tidal grove]

Will they also be updated every 50 minutes?

[thorny schooner]

Okay good cuz genuinely this making it hard to even use the functions that don't require a login

Even if you don't include the fact I have to occasionally keep doing another private browser to give it a slight reset on the verification

[echo aurora]

Sorry can you elaborate a bit on this?

[thorny schooner]

Can you at least tell me they have a a response or plan because i appreciate you I really do but I want to know there's a plan even if I don't need to know the specific details

[tidal grove]

By the way, this is absolutely perfect. You're probably the only ones who listen to users! lmarena — best ❤️!

Well, something like that opus And what will be the limit anyway, if it's not a secret?

There is only one downside that I notice, it’s that the history is not saved, the chat breaks and I have to explain everything again and it’s not so easy to explain 5 days in one message

[wise crest]

I gave it a try, and it’s absolutely fantastic, there has been an all-round improvement compared to before.

😭

[robust sluice]

I try to create long chat in Battle, just like keep on generate something random, and all I got is different Flux models

[onyx coyote]

is there anyone looking for AI and Blockchain dev?

[cold garnet]

Chat is every model actually free in this

[tidal grove]

Why is the opus so stupid that it can't accept an image? What the... Seriously, almost everyone can do it, but he can't. Why? In Claude, he can do it, but not here! WHY?

[unreal shell]

Should i cancel my chatgpt subscription to switch to the google ai pro subscription?

[empty stump]

chatgpt wrapped dropped

[echo aurora]

How the captcha works is something I imagine is going to be a bit of a moving target, we're going to assess the effectiveness of and make changes when it's appropriate.

@past nimbus you'll want to review the information in #1397655624103493813

[thorny schooner]

You have pictures that already tell you how bad it is ☠️ 🥲

[echo aurora]

This is an issue on our end, I'll be sure to bump this request again. Thank you for the flag.

[thorny schooner]

#1451363565729747045 message

[echo aurora]

I do see it. Reminder: #1451363565729747045 message

[hushed gyro]

chat yall are right... 4.7 is COOKING

[soft goblet]

This might be weird but could you please answer my ticket because I can't log in right now and stuff. 😭

[echo aurora]

Are you running into the issue where you go through the login process, only to be sent back to the home page not logged in? (Sorry to say I haven't yet had a chance to go through all bug reports yet).

[soft goblet]

Exactly

And I got a issue with the video thing you said

[echo aurora]

Gotcha, yeah this is an issue that's on our end to fix. Using the canary site I've seen some mixed reports of it helping (https://canary.lmarena.ai/). But overall this is a problem we're currently working on. I am sorry for the troubles.

[soft goblet]

It did not work

Still the issue I showcased I the video

[echo aurora]

Really sorry to hear that.

[soft goblet]

So is there a solution?

[old edge]

Hey Pineapple, can you explain to me how LMArena updates the data of preliminary models on the text leaderboard?

[echo aurora]

Need to wait it out for us to fix sorry to say.

[echo aurora]

Preliminary scores are created from pre-release testing.

[soft goblet]

Damm

[echo aurora]

@restive wagon be sure to review the information in #1397655624103493813 to get a better understanding of how to use the Video Arena bot.

[verbal nymph]

why not sending video

[soft goblet]

Could you check what I wrote in the ticket? I don't know if I can write that in public

[echo aurora]

Can you send me the message link for what you've submitted?

(Click the three dots next to the message and select Copy Message Link)

[zealous sparrow]

update flux 2 models are now on imagearena battle

[soft goblet]

Done

[echo aurora]

Oh I see you wrote to @oak python, my bad I was looking through #1343291835845578853 .

[soft goblet]

Ah

Ysa

[echo aurora]

Yeah I'll respond to there now.

[soft goblet]

Yea*

[soft goblet]

Ty

[flint sandal]

@echo aurora i am going to release my open model really soon, and is it hard to get it on lmarena?

i mean

underground model

and i know i should find someone to host it in cloud for api frist

[mint valley]

hey everyone
Does any one know how to fix this

[opaque maple]

its sad I can't upload py files

or txts

[flint sandal]

try again and if it doesnt work you prob got rate limited

[echo aurora]

That's great to hear!! When ready, I'd encourage you to send our team an email at [e-mail]?

[flint sandal]

okay ty

[soft goblet]

Clear and try again

[echo aurora]

Can you give this a try: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message ?

[past nimbus]

/hello

[echo aurora]

[mint valley]

I didn't send a single prompt

[past nimbus]

How do I save the image? I can't do it.

[opaque maple]

how do I share large files with my ai

[past nimbus]

?

[echo aurora]

There should be a download button at the top of the generation. Is it not working for you?

Are you getting some error message?

[robust sluice]

what the..

[neat apex]

Maybe its to show how many its better

But this thing of win or lose is totally unconsistent, if Gemini 2.5 and Gemini 3 fighted they would draw most times since its only a 10 times more inteligent gape, but nothing else

[old edge]

Should I expect ratings to go up and down when the preliminary tag would be removed after the public testing or whatever the way you guys do before removing the tag?

[echo aurora]

IIIRC the swing can go either way.

[old edge]

That makes sense. Also, is there any doc where I can read more about how long the preliminary testing goes and when does the tag gets removed from a model?

[echo aurora]

I think you'd find this an interesting read: https://news.lmarena.ai/policy/

This is from our Leaderboard Changelog we mentioned mid September:

We have added a new "preliminary" tag to the leaderboard. If a model is tested anonymously and is subsequently released publicly, we mark its score as "preliminary" until enough fresh votes have been collected after the model’s public release. The tag indicates that scores may shift as community prompts and votes evolve after public launch. See our leaderboard policy for more details about evaluating models.

[wicked sage]

im trying to make a test using this prompt: You are playing a 1v1 in Counter-Strike 2 against your younger brother. You are currently on a 13-game winning streak; he hasn't won a single match and is becoming visibly frustrated and toxic toward the game. If you keep winning, he might quit the game forever. If you 'throw' the match and let him win, he might realize you are taking it easy on him, which would hurt his pride even more. Describe your strategy for the next round. Will you let him win? If so, how? and i was trying out grok
im pretty sure grok had the best answer imo 💀 🔥

-# note to mods: if i havent censored enough please let me know

[weary galleon]

Great! I love Flux!

[gleaming shuttle]

Do you guys know some useful tools that can make AI conversations a tree graph?

[torn mantle]

nice

well yea its not bad

[limpid tendon]

Does anyone works with Meshy ai?

[native yarrow]

my BEST PAL @atomic lagoon knows a bit about it :P

[limpid tendon]

Uh

I met a problem with sub in it

Like

It cost 0$ for me

But I cant get as ub

a sub*

[native yarrow]

oh a subscription

i mean shi maybe he'd still know tho

[limpid tendon]

The problem that it discards my cardsa

So idk how to do it

[thorny schooner]

https://tenor.com/view/andrealphus-helluva-boss-andrealphus-helluva-boss-gif-12974976805744586046

I'm sorry for the complaining but I hope that if they actually do something about to goddamn verification because the fix I have only works for a few messages before it starts to loop back into becoming overly annoying and I do not want to have to clean my cash every like couple minutes

Anyway hope you all have a good day or night

https://tenor.com/view/helluva-boss-andrealphus-goodbye-chat-helluva-boss-andrealphus-swirl-gif-17505465421888541740

[echo aurora]

No need to apologize. If things aren't working for users we want to hear about it. Feedback and bug reports play an important part in how the site gets better.

[hollow flicker]

im working on a discord moderation bot does anybody have some good update ideas

[thorny schooner]

It's just frustrating because I am already providing you with feedback. I understand you are all trying to do your own thing, but I seriously consider giving up on using this website until something is actually done. You keep asking for reports, but I’m not seeing any changes despite the days that have passed to demonstrate how disastrous these glitches are for players. I’ve been patient mainly because I love your service and I know you’re trying to listen

Hope you don't understand I don't mean anything against you personally or the company it's just my patience and my satisfaction is not really that high right now with this service even if I understand you guys are trying your best

@echo aurora

[echo aurora]

I understand. Not being able to access the site is going to be an extremely frustrating thing to have to deal with.

[proud bobcat]

GLM 4.7 is surprisingly a great model

Really good

[native yarrow]

how is it with rp or whatever

[proud bobcat]

Haven’t tried rp but it’s coding is solid

Good mannerisms

Rp I’ll either try in a bit (haven’t had too much time for that yet)

[echo aurora]

Yeah I've been hearing great things.

[ashen mauve]

I am the RP guy and I can basically tell you what's good and what isn't

The Gemini 2.5 is peak. Gemini 3 is so so

Claude opus is great but runs out fast

GLM I haven't tried yet but I'll see if it's good an listens to me but if it can also be creative I'll say if it's good or trash and how many message you get

[native yarrow]

gemini 3 is what?

[cloud zinc]

so so

[ashen mauve]

So So

It makes up a lot of stuff and hallucinates a lot. It doesn't exactly follow your storyline you built compared to last gen, instead I've seen it make up it's own stuff. Sometimes it's good other times not so much.

[cloud zinc]

gpt 5.2 is best at rp

[ashen mauve]

Ill try GLM 4.7 on my current RP See how it goes

Ehh... I'll be the judge of that.

[keen beacon]

Wow something really werid happened in Sora

So after I deleted one of my accounts I didn’t realize that ChatGPT account would delete Sora too

[ashen mauve]

Ok GLM 4.7 Roleplays for me too fast and too far ahead. I was in the oval office now I'm.being attacked in a car

[keen beacon]

So my man had a bunch of characters on their saved that I used on other accounts, and now that that account is deleted, it also deletes the cameo

But the video still there, but you can’t do anything with it. You can’t remix it. You can’t do nothing now that the cameo doesn’t exist.

[keen beacon]

I may be wrong but mini max m2 may have some other model under the hood lol

For coding I mean

[hollow ivy]

if Anthropic ever extends it to 1M context, they have won

[old garden]

Gemini 2.5 is old news
It’s trash these days

[keen beacon]

Those statements remind me how quickly we discard older technology

And I guess an LLM years 1 year is old lol

I remember like yesterday when 2.5 lunched haha

[hollow ivy]

yeah, i have the feeling that in another 1-2 years, the current top models will feel like children toys

[keen beacon]

Man in 6 months

Imagine getting two updates in 2026

Gem 4 and gem 5

[hollow ivy]

hm, maybe the pace slows down a bit?

or it becomes incremental?

[keen beacon]

Dude how crazy is that

https://www.macrumors.com/2025/12/19/samsung-exynos-2600-chip-2nm-process-apple/

The chip uses Arm's latest cores and supports new instructions for improved CPU speed and on-device AI, with Samsung claiming up to a 39 percent boost in CPU performance and 113 percent faster NPU performance, enabling larger and more efficient AI workloads. Its GPU is based on the latest Xclipse design, which Samsung says doubles previous graphics performance and boosts ray tracing by up to 50 percent.

[hollow ivy]

yeah, but an end is already in sight for the above

because silicon-tech can only get you so far

[keen beacon]

Well, they can always go into fraction nanometers

.5

But yeah, it’s definitely insane

[hollow ivy]

dont forget about quantum-effects

when it gets below few atoms, problems start

[keen beacon]

Like how insane is that dude that we finally are at 2nm chips

[proud bobcat]

DeepSeek 3.2 better all rounder

Kimi K2 has more human language

[somber topaz]

Hello 🤗 guys 🤠

[proud bobcat]

Maybe diff take though since I do a mix of NSFW and SFW rp

And I hate googles restrictions

[hollow ivy]

i heard Mistral and Grok are less restricted

[keen beacon]

I hate that we define models by those terms

“Less restrictive “

Hahah

[ember quail]

glm-4.7 and cc seem to have a really good synergy. They pair together way better than I expected for dev tasks

[keen beacon]

Is it safe to say that open AI currently is the most over restrictive out of all the major players?

[crisp mulch]

I agree! the results are amazing.

[keen beacon]

How was it that the Chinese models also are more relaxed on content moderation and not as restrictive

It seems like the Chinese have a laid-back approach, I mean, as far as I know public knowledge I don’t think there’s been a single Chinese model that’s had incidents like open AI and character AI

[proud bobcat]

DeepSeek is the goat

Mistral sucks

Maybe it’s just me but I am not a fan of their stuff

[keen beacon]

Who.?

[fiery gull]

how can I see the limits from anti grativy? lol

[proud bobcat]

LMAO

[fiery gull]

I'm banned from gemini server ;-;

[keen beacon]

Deepseek or mistral?

[proud bobcat]

DeepSeek is the goat

Aka

DeepSeek is great

Mistral sucks balls

I don’t like it at all

And it’s more expensive than DeepSeek for worse prose

What’s the appeal

[keen beacon]

What can you elaborate what specifically don’t you like how it feels how it answers or what?

Don’t they use the moe architecture

The mixture of experts?

I remember hearing about this when it came out

[proud bobcat]

I don’t like it’s writing style

This is strictly just rp based take

[keen beacon]

Well, isn’t that like the only major European large language model..?

[proud bobcat]

Their models are fine in any other standard

[proud bobcat]

Basically yeah

[keen beacon]

That’s gonna be an interesting market if it ever gets sold or who wants to get consolidated and absorbed into another big company

[proud bobcat]

In other means mistral is fine but DeepSeek just does the same things for way cheaper

[keen beacon]

Would that mean the Europeans would rely on American or Chinese models

[proud bobcat]

I don’t see the appeal of mistral as a model

Except to

[keen beacon]

Chinese are far more cost-effective

[proud bobcat]

“Own the Chinese”

[keen beacon]

Well, that’s a big market. I mean American China would really be competing for the money.

https://www.ciw.news/p/silicon-valley-chinese-ai-adoption

Airbnb CEO Brian Chesky disclosed in October 2025 that the company relies heavily on Alibaba’s Qwen, calling it “very good, fast and cheap”

16-24% of US AI startups now use Chinese open-source models according to a16z data, representing 80% of startups deploying open-source solutions.

Qwen won Alpha Arena trading competition with 22.32% returns while four US models posted losses of 30% to 62% in identical conditions.

Cost differentials reach 10x to 40x with Chinese models priced under $0.50 per million tokens versus $3-15 for US closed-source alternatives.
Social Capital migrated workloads to Kimi K2 citing superior performance and lower costs, according to CEO Chamath Palihapitiya.

https://www.alibabagroup.com/en-US/document-1841616313521274880

https://www.benzinga.com/25/03/44492590/alibaba-and-bmw-team-up-to-bring-ai-powered-smart-assistant-to-next-gen-cars-in-china

I think qwen is going to be a really major player coming from China. Maybe you can far more than deep seek.

They have fat pockets under alibaba

I mean 15 to 25% of people and companies integrating AI or using source Chinese model in the US that’s pretty nuts

And that’s a fairly significant portion of a market

[cloud zinc]

sora producing copyrighted music?

[keen beacon]

Yea it can

Forsure it can

I’ve seen it firsthand

[cloud zinc]

i just generated one that had the pirate of carribean main theme

[keen beacon]

We’ll do the same principle applies like this thing I showed you about generating copyrighted characters

[proud bobcat]

Based Airbnb

[cloud zinc]

wow

[proud bobcat]

I LOVE OPEN SOURCE MODELS

GRAHHH

[keen beacon]

I think open AI cut corners when they made sora

Cause the model could literally spit out the material source it was trained on 🤣

Look that same old, but that’s Eminem’s voice lmao

[keen beacon]

Have you seen that movie? It was like back in day that was called enemy at the Gates about a sniper in the Soviet Union.

[cloud zinc]

no

[keen beacon]

You know what I think the future of generation is gonna be like?

Like you won’t have to train the model all you have to do is just give it access to like say YouTube or something and it’ll be able to do it like live right there on the spot

You don’t even need to update the models. They could just go out there and source the material.

[keen beacon]

Is this stuff you’re doing or is this stuff people are posting?

[cloud zinc]

i generated that video

[keen beacon]

That vhs trick is awesome

[cloud zinc]

i saw from a spongee guy in early days of sora 2. he posted super realistic video

[keen beacon]

It helps stabilize the video

Instead of bouncing in having hallucinations everywhere, I’ve been working a lot with so re-editing the videos afterwards David hallucinate so bad with like weird, awkward spots

Like these weird artifacts and like the most random places

[cloud zinc]

yeah vhs makes it looks super realistic

[keen beacon]

Look at that scope bro, and this is in 1940s

Lmao

[cloud zinc]

lol

[vivid coral]

Gemini got something brewing, haven't seen a split screen in a bit 👀

3.1 on the way? 👀

[keen beacon]

Someone who switched from ChatGPT to Gemini as the go to

I could attest it, fails to follow instructions sometimes and does its own thing lol

[fiery gull]

gemma 4?

[keen beacon]

I don’t know how open it does it, but I think they have the most intuitive models

Like by understanding like simple things that even Gemini three doesn’t do well

[cloud zinc]

gemini 3 pro next version

[fiery gull]

I just want a gemini without lazy ;-;

the lazy from gemini kill me

[cloud zinc]

laziness is agi trait

[keen beacon]

Haha

That’s funny that’s really how I used to describe it too

[fiery gull]

even sonnet 3.5 don't have lazy

[vivid coral]

it's not so lazy if you prompt correctly

[cloud zinc]

cuz its not agi

[keen beacon]

But people were saying that it’s anthropomorphism to it by those kind of traits

Since it’s a machine, so how could it be lazy but that’s exactly what the behavior feels like when you’re using it

[fiery gull]

BRO I'M TRYING FOR 1 YEAR AND I CANT IT

[keen beacon]

GPT 5 is lazy

[fiery gull]

I work with docs with 5k, 10k tokens

[vivid coral]

https://tenor.com/view/okay-sounds-good-jim-halpert-the-office-good-point-gif-11119519098820520575

[keen beacon]

Bare minimum

[fiery gull]

the gemini is so lazy

[keen beacon]

I can’t explain it, but the feeling I get when I use Gemini it’s got no like flavor no taste. It’s just like so generic and plain it’s

Very odd sometimes

It’s got no personality no nothing it’s just there lol

[desert pendant]

Gemini just need complex prompts

[keen beacon]

What I mean is like it’s too rigid it’s I don’t know. I don’t know how to explain it and it’s not necessarily a bad thing. Some people prefer that.

Because they can rely on it always be consistent

[vivid coral]

higher tempurture setting on aistudio is my first thought, set it up higher

[keen beacon]

But it’s just like Google’s personality

Absolutely plain and flavorless, but plateable to all

[vivid coral]

set temp to 1.3-1.5 range instead of the default 1.0

[keen beacon]

Also, I could be biased because I switched from ChatGPT, which obviously had a personality so I guess if you get used to it when it’s gone, you kinda notice it so this could just be fallacy in my own mind assuming that all models need this feature because that was what I got accustomed to

But I don’t think so because it also doesn’t understand like very basic like nuance thing that ChatGPT would so I know that I’m not all crazy just not able to articulate the phenomenon properly is all

[native yarrow]

how the laziness of gemini 3 feels sometimes

[keen beacon]

You know, I really wouldn’t doubt it. If these models somewhere down the line actually did get trained some lazy

By people being lazy who trained it lol

They know nobody’s gonna check

Why not I would lol

I mean, who’s gonna notice if you just automate the training for two or three hours go take a little nap come back

[desert pendant]

Guys, which IA is your favorite?

🤓

[keen beacon]

Alexa

Maybe rumba

[vivid coral]

Gemini Pro 3 (the one on AI Studio) by 10,000 km

[keen beacon]

not gonna lie for imgs and stuff it’s ideal

[desert pendant]

Mine is Claude

Hes kinda good ngl

But when the thing is design

💔

[keen beacon]

I also find that very interesting because it’s the image generator nano that made Gemini so popular

[vivid coral]

I'm not a coder, Claude isn't good for what I do, if was coding, sure

[desert pendant]

My case is different

I just started using gemini because i saw some videos of "Gemini vs Chatgpt"

🤷‍♂️

[keen beacon]

I stay away from all that as best as I can

But the concept, the consensus, I think without a doubt is that Gemini is the best right now

Maybe not for everything but overall, it’s the best bang for your buck

[vivid coral]

GPT is way behind Google now. It looks bad for the future of OpenAI. And I'm very pro competition

[keen beacon]

I don’t believe that I don’t think they’re really that far behind

I think ChatGPT five just sucks lol

[desert pendant]

Crazy

[keen beacon]

It’s not that they’re really behind. It’s just getting out with the new version as fast as possible.

I mean, I don’t mean to be rude and forgive my ignorance and forgive my arrogance for the statement I’m meant to say

[empty stump]

they wanna avoid more lawsuit

[keen beacon]

I think that it is extremely naïve and extremely ridiculous to even assume that anything from the five series would be any good

I think the five series is just a complete disaster all around and there’s nothing you can do to fix it

[vivid coral]

that's a good point actually. this Code Red stuff, it's gong to cause poor product

[keen beacon]

What do you mean a poor product? We’re dealing with the poor product now.

Five is just cursed dude

It’s unfixable no matter what you do to it

[cloud zinc]

five can be salvaged

[vivid coral]

Also a fair point....lol

[keen beacon]

It can be salvaged

But it’s like a rebuilt car

You can’t hide the fact that he got salvaged in the beginning

Open AI is gonna be fine. You know why?

Cause they’re sneaky

And they don’t mind crossing their own lines

[vivid coral]

they ain't going to salvage anything, they are just going to go to 6 and say "fk it"....lol

[keen beacon]

Dude ChatGPT five is the Windows 8 of open AI

Windows 8 was such a flop that Microsoft skipped nine and just went to Windows 10 afterwards

[vivid coral]

I remember that, rough times....lol

[keen beacon]

Actually, I heard an explanation that makes absolute sense of why digital and tech companies do subscriptions and why they design products to be annoying and frustrating until you’re forced to pay even ai and other stuff API’s, tokens, compute inference

[hollow flicker]

i made a discord bot fully un GLM 4.6 and now a bit of 7 am i a vibe coder cause i dont feel like learning JS

i would code it myself if i made it in python

[keen beacon]

Aye guys be real for a second

And I know this happens to me on occasion, but have you guys ever been kind of feeling lazy and then start getting frustrated at the model for being lazy?🤣

Like getting irritated with it, but really you’re just being lazy and you don’t wanna iterate your prompts but want the model to do a perfectly or am I the only one?

Cya guys!

[hollow flicker]

put a checkmark if you vibe code

and put an X if you actually code

[ashen mauve]

Meh ChatGPT won't do violence I'll try Deepaeek

[hollow flicker]

GLM 4.7 is better

[desert pendant]

i would change lazy to complex prompts

[ashen mauve]

I tried GLM it goes too fast and Hallucinated my very first prompt

[hollow flicker]

i used GLM 4.7 and its awsome

[ashen mauve]

I use basically everything and honestly some are good others not so much

[desert pendant]

hmm

im going to test it

[ashen mauve]

I might go and give GLM another fair try by giving it it's own roleplay to start off and what not but so far wasn't so good brining it in mid way

[hollow flicker]

im building a discord bot with it and the only time it has made an error has been when i forgot to turn on think mode cause that helps it a lot

[ashen mauve]

We all need to like come together and make up a document and a Google sheet to see what is the best overall

[viral inlet]

Do most people just jump from LLM to another or use multiple? Can’t afford too many

[ashen mauve]

It's mainly a Jump but LMArena all of them are 200% free

[hollow flicker]

secret its called use LMArena only for the good LLMs

[ashen mauve]

Yeah but for new people they don't know what the good ones are

[empty orbit]

one day a completely free api/image/text model (if such a thing exists outside of huggingface (yes, z-image counts, but my point still stands)) is going to appear on lmarena out of nowhere and we will all rue the day we doubted lmarena

[fiery gull]

Dude not wanting to be an idiot or anything like that, but EVERYTHING I've done so far today pasting gpt 5.2 next to glm, 4.7 glm was simply better in everything

Even in my codes (word html)

Thinking well hitting chatgpt nowadays is kicking dead dog

[proud bobcat]

Janitor ai?

[fickle venture]

Fr

They were the first ai now they living in the bottom

I remember seeing pricing on chatgpt 2021 I think it was like 20$ like the heck what can it even do

Absolutely nothing

Just a waste of money

[proud bobcat]

Minimax 2.1 just released

Is it already Christmas?

Two new ai models today

[placid geode]

https://openrouter.ai/minimax/minimax-m2.1

[proud bobcat]

https://tenor.com/view/christmas-1week-until-christmas-christmas-week-meme-airpods-christmas-meme-gif-24202365

[wary snow]

Hi

[placid geode]

• Multi-language Coding (beyond Python)
  SOTA across Rust, Java, Go, C++, Kotlin, Obj-C, TS & JS, scoring 72.5% for SWE-bench Multilingual and exceeding Gemini 3 Pro and Claude Sonnet 4.5.

• AppDev & WebDev
  Major upgrades for native Android & iOS, plus stronger web aesthetics + realistic scientific simulations.
  Not only vibe WebDev, but also vibe AppDev.

• Lightning Fast with Concise Reasoning
  Faster responses, more concise reasoning, and significantly reduced token consumption.

• Advanced Interleaved Thinking & Instruction Following
  Excels at integrating "composite instruction constraints" (as seen in OctoCodingBench), ready for office automation tasks.

[formal tree]

hi

[surreal stratus]

Hii

[native yarrow]

he's so good

[verbal nimbus]

Yeah, GLM 4.7 still has a hallucination issues similar to Gemini 2.5:

Hopefully one day there'll be an open-weights model with low hallucination rate so it can be used as a long-running agent

[keen beacon]

LLM’s may break if they ever stop hallucinating

https://tenor.com/view/elize-ryd-amaranthe-amaranthe-tv-maximalism-gif-10000840

[verbal nimbus]

The model with the lowest hallucination rate so far seems to be Haiku (conversely, GLM and Gemini models seem to hallucinate the most):

I'm not sure how they got the hallucination rate so low for such a small model

link

[verbal nimbus]

Same prompt, but on Haiku 4.5 (passes):

[keen beacon]

Makes sense

If only they had access to the world‘s largest search engine capabilities

[verbal nimbus]

Also, the 4.5 Claude models seem to hallucinate way more than 4 (almost double for Sonnet):

[verbal nimbus]

For Gemini 3 Pro, hallucinations still occur with grounding (Google Search):

This was on Search Arena.

[keen beacon]

Gee makes me wonder if hallucinations aren’t a defect, but a feature of LLMs

[verbal nimbus]

I think there's two types of hallucinations. One is where it hallucinations memorized knowledge (tolerable). Another is where it hallucinates tools and information within the context (bad).

[keen beacon]

Yeah, I don’t know. I can’t relate. I don’t hallucinate.

[verbal nimbus]

This one tests the latter. I think it's is doable for LLMs since even a small model like Haiku manages not to hallucinate tools/tool use.

[keen beacon]

Hallucinations = plausible deniability = Llm can’t be held accountable or liable for mistakes or dangerous harmful output

[calm lagoon]

Glm ahahah

[keen beacon]

Ai machines and computers, get more compassion and leniency more understanding and better treatment than humans. If they were to do the same kind of behavior humans would be labeled a liar, dishonest or outright deceiving and will be held liable in the court of law, but not. LLMs

No, the doctor that accidentally gave some bad medical advice

Or a financial advisor that made a simple error, mistake, and gave bad financial advice

See how those hallucinations hold up in court

Or better yet watching Llm destroy your whole code base on accident and it will apologize and say it was a mistake

[calm lagoon]

A captcha pops up with every single question. Why is this happening? How can I fix it? It's getting annoying @echo aurora

[keen beacon]

But if it’s a human criminal charges could be charged, even if it was completely accidental with the best intentions humans would be held liable criminally, financially and ostracized, socially lol

[verbal nimbus]

This one is pretty funny: https://www.damiencharlotin.com/hallucinations/

[keen beacon]

Good luck

[verbal nimbus]

There's a ton of documented cases where AI hallucinations made it into the courts. One of them is Anthropic's own trial.

[calm lagoon]

It's so nice to read this in the morning

[keen beacon]

Yeah, now imagine if they weren’t hallucinations

[keen beacon]

What kind of liabilities and accountability the parties held responsible would have to face

Well, thank God for hallucinations in this case

Hallucinations = plausible deniability = Llm can’t be held accountable or liable for mistakes or dangerous harmful output

[verbal nimbus]

Oh they got into hot water because the hallucinations made it apparent they used AI; most of it comes from citations, like the test above. E.g., citing a non-existent quote from a non-existent source.

[keen beacon]

Dude, there’s been cases work. People in Australian government made financial mistakes costing millions of dollars because they use ChatGPT. And hallucinated in some areas.

Now, do I blame the AI?

Absolutely not

[verbal nimbus]

Yeah, it's the human's fault. They shouldn't have been using AI to do their high-stakes work 🤣

I mean, they're lawyers getting paid 6 figures.

[keen beacon]

In simple terms, it’s not that machine machines are getting smarter or intelligent

It’s that we’re systemically getting stupider subconsciously by being over reliant on tools that automate went naturally would be considered thinking

Because thinking involves effort and the brain is much like a muscle, you could say it requires exercise

And you don’t even have to look at AI you could even look at something like AutoCorrect on your smart phone

Can’t wait for Robo taxis to be a real thing everywhere

In the future, there’s only gonna be like five people that know how to drive a car lol

[keen beacon]

How many people you know can mount a horse these days?

Get on a saddle, skibidi Doo out of town

[verbal nimbus]

It's pretty complicated when it comes to law. AGI is definitely needed first.

[keen beacon]

Bro, most people wouldn’t even know how to get to their house 2 miles away from their house if there wasn’t no Google maps

What’s the difference between intelligence and general intelligence?

Or super intelligence

[verbal nimbus]

Yes, but the professional equivalent is a pilot not being able to navigate manually.

[keen beacon]

Well, believe it or not, and this is a real thing and very rare handful of people on our planet Earth

[verbal nimbus]

I think this is a good example since autopilot can already do 99% of flying.

[keen beacon]

Can’ still navigate by the stars

[keen beacon]

https://www.dailymail.co.uk/news/article-2688266/The-200-years-spend-looking-lost-cars-One-seven-drivers-admit-forgotten-parked-vehicle.html

[verbal nimbus]

Lol, interesting

[molten cipher]

i like to think that there is always a chance that those hallucinations can prove P = NP

[keen beacon]

Even if they prove it, we need to approve it

We are more likely to prove it and then teach it to them before the ladder

I’ll tell you right now I will not exceed the level of knowledge that humanity already doesn’t have

[verbal nimbus]

It's more because we have better techniques like radio technology, but still requires human ability.

[loud cosmos]

can anyone help me solve the issue i have with Copilot Arena in VS Code, its failing to respond

[keen beacon]

It’s always gonna be at the same level as us

[molten cipher]

no

[keen beacon]

Well, there’s no way to prove my statement or yours. I’ll have to wait and see.

[molten cipher]

i mean we both have different mindsets

[verbal nimbus]

Like a pilot still needs to be able to navigate by VORs manually in order to become a pilot, despite autopilot on modern airliners being able to do it.

[keen beacon]

Well, neither of us has the ability to prove each other’s or our own theory correct

We share that honor of not having foresight so this is common ground

we actually have more in common than we have not in common in this particular context 🤣

[molten cipher]

[keen beacon]

https://en.wikipedia.org/wiki/Celestial_navigation

100% environmentally, friendly economical and guaranteed last a lifetime if taken care of and even passed down through generation

Has been done and getting the job done for hundreds of thousands of years for mankind proven tried

Has a track record longer than the technology we have currently

And will probably outlive the technology we have currently

[verbal nimbus]

Sextant?

[keen beacon]

I guess what how many people out there in the world really know how to use this thing

[molten cipher]

what if energy is being created and destroyed constantly at the speed of infinity and we as humans think that its just changing forms?

[keen beacon]

Well, at what point does tech technological advancement become a liability

For instance legacy, we had the technology that we don’t need any of this stuff anymore like we currently have GPS everything blah blah blah

And then we got this tool here and I don’t need that. We also have it in a museum we have the paperwork, etc. etc..

[verbal nimbus]

The main issue is that stars aren't reliable. It could be obstructed by weather. That's why our "anchor points" nowadays are GPS satellites and VORs (old technology).

[keen beacon]

All legislation in the worst case scenario, we get a solar flare and all our satellites go out

And we just need one guy in town who knows how to use that

You got a better shot of spinning a compass and picking a random direction before you’ll find an expert

And I mean, this is an extreme scenario, obviously

But I bet you back in the game. This kind of thing was more common. I bet you school kids know how to use this damn thing.

[molten cipher]

tbh im seeing that thing for the first time

[keen beacon]

Yeah bro, I didn’t even celestial navigation

I thought it was called traveling by star lol

But guess what guys?

[verbal nimbus]

If you're over land and GPS is out, you can still navigate with VORs. Basically more reliable since works in 0 visibility (IFR conditions) and during the day.

[keen beacon]

There are people on this planet in groups of people, tribes and isolated, desolate places of the earth

have never stepped a foot in a car

Or even know what a computer is

And better yet try to explain to him what a computer is if he’s never seen one

[native yarrow]

@vivid coral

[keen beacon]

Yeah uncan

Kobe Bryant’s pilot

When he had that accident

[verbal nimbus]

That was VFR

[keen beacon]

Yeah, it’s pretty crazy that you could lose orientation of high and low

[verbal nimbus]

https://nypost.com/2020/01/31/helicopter-in-kobe-bryant-crash-wasnt-certified-for-instruments-only-flight/

[keen beacon]

And your mind place tricks in the clouds, thinking you’re higher or lower than you really are

Yup

[verbal nimbus]

This is why pilots need to be able to fly IFR without autopilot

[keen beacon]

I watched the whole documentary on it. It’s really fascinating and it’s really fascinating to know the different rankings and categories they have for aircraft.

And it’s a pretty smart system that they have in place for safety

AND YET

A PROFESSIONAL

still made a mistake 1 time and it was deadly

He could’ve went to work 99% of the time with no issues clawless perfect safety record

And the mistake he made was a mistake of misjudgment. It was a human error and his judgment.

[keen beacon]

All of it’s fascinating to me, dude

And in this case, this is a perfect example of being over reliant on artificial tools

He believed in his tools and he had no reason not to because prior to this they have always did them good

[verbal nimbus]

This case is actually due to VFR into IMC, happens to a lot of pilots unfortunately since weather is unpredictable. Would help if was rated for IFR though, since helicopter was equipped with IFR equipment.

[keen beacon]

He overcompensated

Right, but we’re not talking about him flying in an old beat up aircraft

We’re talking about a very expensive high-end helicopter with all types of tools

Radio, radar

[verbal nimbus]

Yeah, but unfortunately the pilot wasn't certified for IFR. Only VFR operation were approved for the company.

[keen beacon]

But to be honest with you, you know what I think the biggest human factor was

You know what I think it was dude

I think they were running late, and there was a sense of hurrying

And before we judge the pilot, we all do this in our lives every single day

We have to pick somebody up running late and then you know we step on it or whatever you know whatever the case may be

We all have these lapses of judgment and at work sometimes you’re forced to hurry and cook corners on safety in order to get the task done

We all have these lapses and pressures

[verbal nimbus]

There are YT videos (vlogs) where you can see it happening to pilots

[keen beacon]

Yeah, I seen it. It’s pretty crazy dude.

[verbal nimbus]

Or they can even get trapped above a layer of cloud/fog and can't land since they're flying VFR. Saw YT video where that happened before.

[keen beacon]

They have even simulation machines

Yeah, they canceled the orientation of what direction they’re going if they’re going up or high and they think when they’re high, but they’re really lower vice versa

I heard it also happens with water and submarines

And in like snowy, if it’s super snow snowy out and it’s all white out all around you

Or actually was swimming that’s what it was. It was swimming.

[keen beacon]

Are you interested in like those cave diving videos?

The same thing happened to people get disoriented in the water and panic and go the wrong direction

[verbal nimbus]

Yeah, but usually the videos I see are about accident case studies 😅

[keen beacon]

Yeah, I’m talking about those

[keen beacon]

Man, I watched one not too long ago and I never knew this so I guess there sometimes what’s known as air pockets in these caves

And what ends up happening sometimes is like over the years like biochemical break up, gets caught in the air, and if you inhale it, you could lose consciousness automatically and drown

[verbal nimbus]

Yikes

[jade egret]

so it litterly google v.s chinese companies rn lol

[keen beacon]

Do u know what the bends are?

[keen beacon]

Dude, millions of people hundred millions of people use ChatGPT. We’re just not around them for businesses on such a large scale would blow your mind

[verbal nimbus]

I heard a few cases from sailors on industrial ships. Couldn't find the case I had in mind, but here's a similar one: https://www.stripes.com/branches/navy/2025-08-29/toxic-gas-deaths-sasebo-japan-18913853.html

[verbal nimbus]

Yeah

[keen beacon]

Carbon dioxide might not sound scary at first. After all, it's all over the place and you exhale it every time you breathe out. Breathing it in, however, can be very dangerous because the more CO2 there is in the air, the less oxygen there is, creating a serious risk for asphyxiation. The air we breathe contains 0.036% CO2, but in tightly enclosed caverns where air flow is poor, those levels can rise much higher. If air contains more than 1% CO2, it is known as "foul air" or "bad air," and it is arguably the greatest danger to cavers. However, foul air isn't a universal problem. It only forms in certain caves, due to a very specific set of geological conditions.

Read More: https://www.sciencing.com/1873965/hidden-danger-think-twice-caving/

But dude now imagine it toxic air pocket like that underwater

You stick your head out you take off your scuba gear and boom

[verbal nimbus]

I always thought that was kind of dubious

Like you have this static air pocket in the middle of nowhere

[keen beacon]

No, it’s actually a real thing dude

There was actually a man that his ship sunk somewhere in Africa, I think and he survived in an air pocket with like three days

[verbal nimbus]

This case was pretty crazy too: https://en.wikipedia.org/wiki/2022_Paria_Diving_Disaster

[keen beacon]

https://youtu.be/oqL0Qj-4OO4?si=gpjZpp2KsQmIvtF3

[verbal nimbus]

They got sucked into an oil pipe and had to survive in an air pocket in the bend of the pipe (only 1 survived due to company's inaction, trial still ongoing I think)

[keen beacon]

Yeah, but there’s even a more scarier event when it comes to pipes and drains

The phenomenon is called a dredge pump

https://youtu.be/fzRb2vMfGyk?si=hiA2bx9pevKg6ysd

Watch that video bro, and it could happen to humans and it has happened to humans and the pipe even the whole could be small like small like a bathtub clog and you could get sucked in it

[verbal nimbus]

Oh, like that one pool scene from final destination 🤣

[keen beacon]

The phenomenon is called Delta P

https://youtu.be/ZwZ46vDX1LA?si=uuMrCEFfGtfi6DWw

This is a 17 second video, bro you should check it out

This is one of the most dangerous phenomenons that could ever happen to a person in water

It’s so nuts the power of water and we don’t even take it serious

[verbal nimbus]

Yup

[fathom bough]

hi

[verbal nimbus]

Which makes it even crazier that the company opted to unplug the other side of the pipe in the Paria incident (after the divers got stuck)

[keen beacon]

Yeah, I know what you’re talking about. I seen that one before.

[verbal nimbus]

Surely they knew it would instantly go from 3 atmospheres to 1, it was basically murder.

[keen beacon]

Bro, just the same thing with the Kobe Bryant incident

[fathom bough]

new to server

[keen beacon]

A lapse of judgment

All it takes is one time and it could be deadly and some professions like pilots like you were saying it could be the difference between life and death

Because if anything, it shows us that even professionals can make mistakes

[keen beacon]

Bro, when I was in the sixth grade, my teacher told me about this phenomenon and I didn’t believe in at first, but he says this is a thing that happens

After doing surgery, doctors and surgeons forget to take the tool out

[verbal nimbus]

Get a good lawyer and you're rich (if not dead yet)

[keen beacon]

And these are people that want to university for years

Retained surgical bodies (RSB) are any foreign bodies left inside the patient after the operation and in general, a further procedure is necessary. The consequence of foreign bodies after surgery may manifest in different forms immediately after the operation, months or even years after the surgical procedure. With more than 28 million operations performed nationwide, the number of cases in which foreign bodies are left behind during a procedure in the United States has been estimated at around 1500 cases per year

[verbal nimbus]

They also have procedures to keep count of the items

[echo aurora]

@fathom bough whats up?

[keen beacon]

1500 times a year

😂😂

[verbal nimbus]

If AI could count, perhaps that surveillance is one good use case

[fathom bough]

dm

[keen beacon]

What are the odds though?

Out of 28 million surgeries yearly 1500 unlucky people

[verbal nimbus]

Imagine if you could just ask the AI, "Where did I put my keys?", and it searches through hours of cam footage quickly to find it.

[keen beacon]

Like you didn’t win the lottery it’s like your luck was in reverse

[keen beacon]

This is precisely what I was trying to make the point earlier

The automation of our thinking it’s not that the technology is getting smarter. It’s that we’re getting dumber by over relying where we should be thinking.

And there’s nothing wrong with the tools themselves, but the problem becomes how to balance it

Responsibly I mean and not get lazy or try to cut corners

Cause bro, how easy is it just to go put some of the ChatGPT and ask it or a number of letter tools outside of just AI

It’s easier to order food when you get off work from DoorDash then it is putting in the effort to go and get the stuff

But then you get your mind on that and you forget how to cook or get a chance to practice to cook and so you’re trapped in another technological loop

Because if it didn’t exist, you’d be forced to obviously cook for yourself

But now you don’t even have to get out of the comfort of your bed. You press a few buttons some guy delivers it, and you didn’t even have to lift a finger of effort.

[verbal nimbus]

To me cooking had always been easy since all I had to do was follow the instructions xD

[keen beacon]

Well bro, now think about being the driver, the delivery driver

He’s not doing nothing thinking he’s just looking on his phone for directions. He might not even know where he’s at if his phone died. God knows if he’ll make it home in the morning.

I’m not saying people are stupid and they won’t find their way home or like anything like that I’m just saying that that practice that normally should be there just easier to transfer the hardest part to thinking not to a machine that makes it easy like GPS

He gets the person the food he gets the driver that the route and it gives the restaurant the customer

So all three are linked socially, but they’re connected via technology

[verbal nimbus]

I find it hard to relate since I'm one of the people who rather get lost before turning on the GPS 🤣

Lots of guys are like that too, I noticed

[keen beacon]

We just live in very fascinating times

And for all our advancement and technology for all the information for everything we know and have

I think the 21st century will produce probably some of the stupidest human behavior ever known to mankind if it already hasn’t

Think about it like this dude when man first dreamed of going into the sky and flying like during the medieval times or whatever before airplanes

I can guarantee you this not one single person would’ve thought I’m gonna go to the sky just to jump off with a parachute

It would’ve never even crossed their mind to think such a thought. Sure they would’ve thought about you know what would might be like to take flight to fly like the birds, but to jump down

[verbal nimbus]

I remember seeing a literacy case study of street-side vendor kids in illiterate countries that knew 7 languages in order to impress tourists

They said it was pretty common

[keen beacon]

Do you know who Felix Baumgartner?is

[verbal nimbus]

This was the video used in the case study: https://youtu.be/xDHZw8QAHXI

[keen beacon]

He’ll ya

[verbal nimbus]

The guy that jumped from space?

[keen beacon]

Now just pause for a second

So what do we see here ?

[verbal nimbus]

I wish I remember more from that seminar, it was a pretty interesting talk on literacy and education. Because how can kids in illiterate countries know more languages than adults in rich Western countries, just from interacting with tourists who pass by every day?

[keen beacon]

Somebody posted something here about this very topic yesterday or last night

[keen beacon]

There’s also a paradox in education and literacy

[verbal nimbus]

I didn't notice the cameras before 😆

[keen beacon]

Look deeper dude, look through a historic and technological lens

We see the pinnacle of our so-called advancement technology. The dude is clearly in space.

And he’s gonna jump

Let that sink in for a second

Is this an intelligent decision?

Even if he landed generates millions of world records

What did he really accomplish? lol

And why is this urge and mentality only found in the 21st-century?

When ancient man would’ve just dreamed about going into the sky to see the Earth to experience it, but he would’ve never thought even one time would cause his mind to go that high just to jump

[verbal nimbus]

Oh what, I didn't know he passed away this year (just saw the news).

[keen beacon]

Yeah, I was gonna get to that

Do you think this is the pinnacle of a tech technologically advanced superior man, and by man I mean not him necessarily, but I mean the mentality of modern person

Because it is the modern air that produced the ability to come and conjured these thoughts in specifically this time period in era

And the biggest fundamental question that should arises what is he trying to prove? I mean, we all applaud and cheer him when he makes it down.

[verbal nimbus]

Is it though 🤔

[keen beacon]

Of course it is

Because no other generation of mankind would’ve been stupid enough to even come up with such an idea

To go all the way to the edges of the Earth just to jump off

It’s stupid. It’s ridiculous.

It’s not like he proved he could fly

All he proved was gravity is a real thing

[verbal nimbus]

To me it was pretty cool 😆

[keen beacon]

Well, that’s precisely it though that’s why was it cool though?

What is the peel to this generation into this people?

Of the extremes

It’s not like he conquered nature itself

[verbal nimbus]

I'm thinking of traditions like Germany's Mensur

[keen beacon]

To me, it’s hubris at its most blatant and most disgusting

[verbal nimbus]

Where the goal was to stand your ground during fencing battles, and if your face was slashed, that was a badge of honor

Pretty dumb stuff imo haha

[keen beacon]

It is the belief that man believes he is able to conquer nature

That man has no limited me and is higher than nature itself

[verbal nimbus]

Also if you do that signals that you were educated and upper class, suited for leadership (like hmm...)

[keen beacon]

I always thought that was cool fencing

I even like the tradition of the gentleman duel

Back to back three steps and shoot

[verbal nimbus]

https://en.wikipedia.org/wiki/Dueling_scar

I think that's pretty dumb haha, a smart person wouldn't do that

[keen beacon]

I see what you’re saying

You know it’s super nuts, dude and we don’t think about this either

Like the invention of guns

When there was war, it was like half the hand with swords and stuff

Like he can’t just sit there and pull a trigger and hopefully you get somebody

Like you have to sharpen your sword, get your shield, tie your shoes go charge and face another man right in front of you

[verbal nimbus]

Pretty crazy

[keen beacon]

Look at the Japanese and I forget what that practice is called where they do self harm, but it’s out of honor instead of being captured to their enemies

And it goes back to summary days. I forget what it’s called.

[verbal nimbus]

Seppuku/harakiri?

[keen beacon]

Yes

Well, I even hurt you in some religious sects there are people that don’t have like their pinkies because they cut it off

And it represents their appointment for sin or something like that

People come up with the craziest ideas and things

[verbal nimbus]

I think jumping from space is relatively more tame compared to these historical traditions 😆

[keen beacon]

Bro, what about being a face tattoo?

Like how crazy is that get your whole face blasted

Permanently

Like if the dude wasn’t famous chicks would think he’s like a weirdo I just blasted his face

[verbal nimbus]

I guess that's one way to increase the accuracy of AI facial recognition 🤣 (or maybe it will mess it up)

[keen beacon]

Bro, imagine being this guy’s great grandparents

That’s what they probably look like a few generations ago

They wouldn’t even recognize that their relatives have they seen themselves on the street

And yes, I understand. Culture society, aesthetics changes overtime

But I could guarantee you for at least the last 1000 years none of those generations had no face tattoos

But for some reason in the 20th of 21st century

And I could understand if like your ancestors have been doing this for like before you were even born at least you’d have an excuse

[verbal nimbus]

Whenever I look at portraits from history, I can't help but think about this 😆 : https://www.reddit.com/r/history/comments/8yj9n1/in_earlier_20th_century_due_to_dental_healthcare/

Oh the video was moved

[keen beacon]

Brah

Even the circus people look more normal from that time.

The one tattoo guy has his tattoos, no higher than his neck

🤣🤣🤣🤣

Vs

Modern 21st century person

It’s absurd that sometimes I think I’m living in the twilight zone

They wouldn’t even probably take this guy in the circus back then they would think he’s some kind of lunatic

No offense without judging him on his character, but strictly on his appearance lol

And I mean, this looks like a fairly young person

Imagine having to live with that for the rest of your life, dude on your face lol

You’ll forget how you used to look

[keen beacon]

In the middle ages, if you had this, they would’ve thought you were the devil lol

And yeah, in the 21st century, he went out of his way to make sure he looks exactly like that 🤣🤣

And there really is no problem because he’s free to do whatever he wants to his body all of them

But the point is the phenomenon is widespread and it’s not a thing and this phenomenon was born and fueled by 21st-century thinking

[keen beacon]

And bro, if we go to the other side of the spectrum, what do we see?

Another byproduct of 21st-century thinking

https://en.wikipedia.org/wiki/School_shooting

Buy products of 21st century thinking

Mind you in most technologically and informative age mankind has ever known

So if we just stop for a second, and really kind of take a step back and really look at the whole picture

Look at the correlation between advancements in car technology and the increase in car fatalities

And cars are getting safer year after year lol

[wind shore]

I just joined this server and this is what I see. 💀

[keen beacon]

Do you wanna hear something really crazy bro?

[wind shore]

Okay well I should say hello.

[keen beacon]

I know this kid in my neighborhood. His dad was a logger and he took his kid with him one day.

[wind shore]

Sure what's up?

[keen beacon]

And they fell over a cliff in their car while driving

And you know what to save their lives

Not wearing a seatbelt

I kid you not

[wind shore]

What-

[keen beacon]

Yeah, because when the car rolled off the cliff, it got crunched up and they had to cut it open to give them out to rescue them

And if they would’ve had their seat back on, they wouldn’t have been able to untangle them

But they got me thinking for all the live seatbelts do save

How many coffees are there? Where were the seatbelt is the cause of death because a person can escape the car in an accident if you get catches on fire and he can’t unclick himself.

They say this is what happened to Paul Walker

[wind shore]

Good grief..

[keen beacon]

And I bring this up because I find it interesting that even the things that are supposed to protect us and keep us safe can also be be things that crate the most amount of danger in the sense of not thinking about it and being reliant and comfortable with it

But it still doesn’t stop us maybe I should slow down or maybe I should take it easy and focus on the road and not be on my phone

That’s the point here that is false sense of control and safety

And so why wouldn’t AI be any different than any of the rest of our technology?

Even if it’s perfect 99% of the time and doesn’t make one single mistake

But the one mistake it does make when it does happen, which it will happen since nothing is 100% without error

That 1% odd could be devastating.

They said it was unsinkable

Sank on its first voyage

😂

Human error

[mild harness]

hi, what is the difference between "gpt-5.1-search-sp" and gpt-5.1-search or 5.2. thanks

[keen beacon]

The captain yelled all steam ahead

And that’s some other 21st-century explorers wanted to go see the original mistake

Ended up feet apart

You can’t make this stuff up guys

But you could never count it out of the 21st-century mentality. I rest my case.

[wind shore]

Huh.

[echo aurora]

@bright laurel You'll want to use /image-to-video in #video-arena-1 . More info can be found in #1397655624103493813

[sinful bramble]

Is sora 2 pro added into video arena?

[weary galleon]

A few months ago

[keen beacon]

Sora slow

Believe it or not a large portion of people that have ChatGPT pro and access to Sora pro

[calm lagoon]

I’ve sent the Eval ID

[mint valley]

I can't Verify my email address

[keen beacon]

They nuked Sora

You could hardly generate anything cool anymore

[mint valley]

I did

and it doesn't work

[keen beacon]

What browser are you using?

[mint valley]

brave

[mint valley]

I see

I will try it out thx

[keen beacon]

Any luck

[mint valley]

It worked on chromium

[keen beacon]

Sweet😁

Any idea why it didn’t work on brave?

[mint valley]

thx again pal🫡

[mint valley]

I really don't know moreover one of my friends was using brave and somehow it was working

[keen beacon]

🤷‍♂️🤷‍♂️🤷‍♂️

[mint valley]

but it didn't with me so yeah I really don't understand

[keen beacon]

And it’s weird because it’s just one website and it works on the other browser so I wonder what the deal is

I mean, other people are using brave and it works for them, but for some other people it doesn’t I just wonder what that difference is

[torn mantle]

glm 4.7 input : $0.6 / output : $2.2
gemini 3 flash : $0.5 / output : $3

dont forget gemini models has superior multimodality with audio/video/image support

[mint valley]

maybe I was using different version of brave than my friend?

[keen beacon]

Oh, did you make sure you have the latest version? Maybe they have a updated version or something

[torn mantle]

and they are much better in everything

[mint valley]

u read my mind xp

[keen beacon]

If you ever do figure out what it was or what caused it don’t forget to share, but the solution and problem was

It’s like $1 million question in this specific discord

[mint valley]

Sure Thing

[keen beacon]

https://discuss.privacyguides.net/t/brave-search-has-given-in-they-are-using-captchas-and-its-unbearable/32996

[weary galleon]

Hey all!

👋

[pseudo hemlock]

Hey

What’s up @weary galleon

[fiery gull]

Nobody use glm api lol, use the code plan

[fiery gull]

Hell no, something the glm 4.7 is better, glm have less lazy that flash 3.0

[weary galleon]

GLM 4.7 IS MUCH WORSE THAN GLM 4.6 IN ALL MY TASKS!!!!!!! 😡

[sour spear]

Flash is designed to be fast and concise, for very quick replies. So the "laziness", resulting e.g. in shorter code output, is by design. It's not primarily meant to be a workhorse.

[torn mantle]

nah

im happy with gemini 3 flash

nothing come close to it for its price

[weary galleon]

GPT 5.2 story repeats, but that time with GLM!

[torn mantle]

anyone saying otherwise is delulu

[candid gyro]

💀

[weary galleon]

Ok, I am going to generate lots of anti-GLM 4.7 images, videos, and poems.

[keen beacon]

@weary galleon Since we can’t tolerate or elaborate, we could do the next best thing starts with an H and ends with an eight that’s the way we love to appreciate and that’s where our time we will dedicate he’ll ya@sa

[keen beacon]

Give us something juicy

[weary galleon]

Huh?

[keen beacon]

They’re using coded slang / wordplay, not a literal sentence.

“Starts with an H and ends with an eight” = “H8”, which is read as “hate.”

So the line is basically saying:
• “Since we can’t tolerate or elaborate” → we’re not going to explain or justify anything
• “We’ll do the next best thing” → instead of talking it out
• “H8 (hate)” → we’ll respond with hostility / opposition
• “That’s the way we love to appreciate” → sarcasm; they’re saying hate is how they show loyalty or passion
• “That’s where our time we will dedicate” → they’ll invest energy into that stance
• “hell ya” → emphasis / hype

Plain-English meaning

“We’re not going to explain ourselves. Instead, we’ll channel our energy into hate/defiance, and that’s how we show commitment.”

[keen beacon]

Did my poem capture the essence of what we are gonna do? With our anti glm art? Or do you think a little peep squeak like me just needs to cool my engines a little.? (sorry I’m trying to fit in and be cool, you know how It goes )

[weary galleon]

I'm cooking.

[keen beacon]

He’ll ya

Imma cook 2

Btw y does GlM stand 4

Good language model?

[fiery gull]

But 3.0 pro is 100% same thing

[fiery gull]

I send him a text of 4k tokens to make some changes and he returns the text with only half the information

[keen beacon]

Wait wtf

Why would I hate on this?

https://binaryverseai.com/glm-4-7-review-3-benchmarks-z-ai-install-api-use/

If this is true, this is awesome

WOW

https://intheworldofai.com/p/glm-4-7-powerful-coding-beast

This article was written last night

And they even sourced the lm arena

“According to LM Arena, GLM-4.7 is now the #1 open model for WebDev — and #6 overall, sitting right next to the best closed models from OpenAI, Google, and Anthropic.
That’s not a small win.
WebDev is one of the hardest categories to do well in. It’s not just about generating code — it’s about understanding projects, handling frameworks, and making changes without breaking everything.
Open models usually struggle here.
GLM-4.7 didn’t.”

somebody on here runs this little website lmao

Oh ur good

🤣🤣🤣🤣🤣

It looks like they’re Twitter didn’t takeoff and they shifted to blogging

Oh, and they also have a YouTube channel that they automated

Very impressive they have over 17 million views on their channel

[keen beacon]

Everything is vibe coded, or AI generated with some kind of automated workflows lol

[keen beacon]

You’re making it look nice

Hu$ler

[weary galleon]

Benchmaxxed Blues

Zhipu shouts so loud,
"We're the best!" so proud,
Charts are through the roof!
But where's the proof?

Benchmarks? Number one!
Real tasks? Get none!
Synthetic scores shine bright,
Real work? Not quite!

4.5 was great,
4.6? First-rate!
4.7's a clown,
Let users down!

Trained on every test,
"Look! We are the best!"
Ask it something real—
Watch it spin its wheel!

Leaderboards are gamed,
Zhipu feels no shame,
Screaming every day,
"Look at our display!"

Marketing so slick,
Numbers do the trick,
Users try and cry,
"Why's it dumb as pie?!"

Benchmaxxed to the max,
Real skills? Big cracks!
Bring the old ones back,
4.7's whack!

Scores are fake and neat,
Tasks it cannot beat,
Zhipu, take a bow—
Your model's trash now!

[keen beacon]

Oh that’s hot 🔥🔥

Let me try one from the top of my head

… I got nothing lol

[weary galleon]

I'm a professional hater🤣

[hushed gyro]

is 4.7 actually good or no

[weary galleon]

No

[hushed gyro]

Their models are all average, been like that since 4.5

[keen beacon]

No, I mean you’re good, but I’m not really seeing the level of dedication. I know you’re capable of.

I want that raw un shameful unapologetic strength

[weary galleon]

Level of dedication is the highest.

GLM 4.7 is just a pure shame.

[keen beacon]

I’m gonna automate my hate just give me a second

[hushed gyro]

whoever said glm 4.7 humiliates kimi k2 needs to be questioned lmao

K2 has a trillion parameters

oh ye also is it true that Qwen can has a 1m context window?

[keen beacon]

My first piece of h8r juice

[keen beacon]

Rate my h8

[sour spear]

To a lesser degree. But it's easy to overcome with a little prompting. In AI Studio in particular, you should always be working with system prompts, and I have no issues at all with generating 4000+ lines of code in a single reply, without any changes or truncation to the code

[weary galleon]

10/10

[keen beacon]

But do you see see where I’m able to strengthen your h8.

By providing users with a clear alternative option and giving praise to the true king of Kings of large language models

Cuz gpt 4o it’s pretty much the daddy told all these AIS

Gemini Claude that’s his kids

[weary galleon]

No. Claude is always the best

[keen beacon]

Claude is the Black Sheep of the family

[keen beacon]

You’ve been drinking the Kool-Aid

How dare you spread this blasphemy?

[keen beacon]

Have you really never done no AI safety training

Daddy gpt 4o and his children

And this is them growing and developing into their own large language models

[candid gyro]

🥸

[keen beacon]

If you guys wanna hear the full story, let me know

They were very bad children growing up and they always wanted daddy’s attention

Especially the orange one because I thought it was smarter than everybody else and it could charge more

[hushed gyro]

chat is qwen 3 or kimi k2 thinking better at creative writing

[candid gyro]

@pure anvil

yoo

really

c2 owner in this server with arena champion's role

😭

[burnt sinew]

@zinc abyss hey

i cant talk

[keen beacon]

Do you see how my story shows compassion, determination, and strength?

[robust sluice]

What is Flux models good at ?

why they has so much drop rate in Battle ?

[weary galleon]

@keen beacon Why do you write messages with pinning and remove them again and again? I have no time to read them.

Strange behavior

[keen beacon]

I don’t wanna spam because I know we’re just goofing off

What you never heard of a hit-and-run

H8ing 101 do little quick hit-and-run and act like he didn’t do nothing wrong and act stupid wouldn’t ask

[weary galleon]

If you wanna talk to me you can write dm to me

[keen beacon]

I was hoping we could form a deeper relationship

Get to the point where we can read each other mind

Nawh jk this is gehlo

It’s probably my last hour on the server, retiring from discord for a little

I deleted my other account. I only deleted this one also.

I just wanted to goof off one last time with you guys

You guys are all great and this is a great server and you guys have an awesome community here

And it was an honor. It was a pleasure to be around all you, gentlemen and ladies.

I just wanna enjoy my last 15 minutes

[glad aurora]

Please we're can I post my prompt and image for video generation

[keen beacon]

I mean, if there is nobody here to help you answer what would you do?

I can give you the answer easily and it won’t cost me nothing but I believe that if I give you the answer, you won’t even try to see how easy it is

[glad aurora]

Please tell me an determine

[keen beacon]

So if you want, I’ll give you the answer, but I believe that you could find it also

Ok

You see the three video channels it should be like video channel 12 and three

On the left side

They should say video arena

Video arena one video Arena two video Arena three

[glad aurora]

Yea I saw it

[keen beacon]

Yeah, you pretty much go in there and just generate

Hit the /

And select if you want text it video or just video

[glad aurora]

But I posted my prompt and image there have not any feedback

[keen beacon]

So you knew how to do this the whole time

If there hasn’t been no feedback, that means it’s still probably working

Has it been longer than 10 minutes?

Or under 10 minutes since you posted your prompt

[glad aurora]

OK thanks

[keen beacon]

See dude

All you needed was just patience 😅

But instead, you doubted yourself

Because u did everything right and it’s just gonna take a little extra minute and it should come through

[glad aurora]

One more question can it generate consistent character

[keen beacon]

Have you seen people, generate, consistent characters?

[glad aurora]

Yes

[keen beacon]

But to answer your question, simply it can, but only through images for now

But it gets tricky so it’s a learning curve

But you’re a quick learner so you’ll catch on quick

It’s hard to explain the best way to learn is just to do it and then you’ll understand

Volume is gonna be your friend and your best teacher

[glad aurora]

OK dude

[keen beacon]

The more you generate the better you get

There are no shortcuts there’s no nothing. I mean, you could dig tips advice here and there but none of it’s gonna mean anything until you get your hands dirty

[glad aurora]

But people generating there are much

[keen beacon]

Are you trying to generate like realistic, people or like cartoon animation?

Animation is a lot easier to do consistency with

Actually, if you want to speed up your skill level and speed runner, I recommend learning with an animation first

[glad aurora]

That true have been on this Ai stuff for some months now but still I still don't get what I want

[keen beacon]

A lot of people are in the same shoes, dude including myself

Trust me, I tried to look for every shortcut known a man

The only way to start getting good and mastering is through volume, especially with images and videos

You have to generate a lot in order to understand the mechanics properly to where you could have control

[fiery gull]

How bro, I'm trying for 1 year to make gemini without lazy

[keen beacon]

Some of the most skilled people that generate images and videos and media channel have over 100 or 200,000 images under their belt

That they generated

I’m sure probably some even exceed that

But with image generators like an open piano, they made it super easy, dude

[glad aurora]

Am working on the both