general

[torn mantle]

i got it twice and i chose it over gemini 2.5 pro

needs more testing

they are kinda similar but just the little details gives it the edge

[eager mica]

Looks like so. I just had a round with it.

[torn mantle]

nightwishper got to be the next sota coding model for sure

i want to get nightwhisper for that prompt

stargazer did a good job too

yea

should be cool

nightwishper

gotta give it to this model

the amount of details

it nailed the colors too

are you really dancing rn?

xd

btw i asked it to make it look modern

its good no?

lol no

i got it like 5 times from 6 tries

its the opposite

the probability will be higher if its a new model

[mossy drum]

New model in Arena: olmo-2-0325-32b-instruct (I tried to search here for the name or just olmo, nothing found)

[brittle tiger]

I think it means more probability they want the model out as fast as possible

[rigid widget]

The reason why Gemini 2.5 Pro is so good is 1) AI Studio 2) LMarena

[brittle tiger]

That's just been assumption bc it's beating 2.5. I don't think confirmation of it saying it is yet

[rigid widget]

Anthropic is already top

If we consider the power they have, yes they are bad

[rigid widget]

Friends, I have a very important exam in 2 months (University exam). I will be away for 2 months.

Have a good life to all of you.

[keen beacon]

confirmed all google

interestingly they do not have the "gemini-test-xx" ID that all previous google anonymous models have had

for example 2.5 pro was gemini-test-38

but these IDs are just their names

No because some companies don't train it in at all or train certain parts. Some might not do it well either

[torn mantle]

google said something but i was so sceptical

they said that they will dominate the coding area with their new models

moohowler seems like google flash model

nightwhisper is so so good

finally google

[lime coral]

https://x.com/officiallogank/status/1869902322840571922?s=46

Logan Kilpatrick (@OfficialLoganK) on X

We are going to build the world’s most powerful coding models, lots of good progress already with 2.0.

2025 is going to be fun :)

X · Dec 20, 2024, 12:27 AM

[torn mantle]

yea this one

really beautiful

[brittle tiger]

Can confirm nightwhisper is cracked. Sick working demo here

[plain zinc]

Google is COOKINGGG

[lime coral]

Is it a thinking model?

[plain zinc]

It seems like Gemini 2.5 pro is finely tuned to programming

Gemini 2.5 Pro Coder

Not Exp even

[lime coral]

So thinking

[cedar tide]

Hi, sorry to bother you, are there any mystery Sota models currently in the arena?

[honest garden]

24 jarat gold

[plain zinc]

nightwhisper

[cedar tide]

NightWhisper its the best coder all time ?

[plain zinc]

It seems to be much better than 2.5 pro

[cedar tide]

Just for coding ?

[eager mica]

Personally, I think it's a good finetune of a small or mid-sized model rather than a SOTA or even frontier model.

[plain zinc]

Well I do not know

Check it yourself

[willow sparrow]

Anyone know of a good alternative to manus?

[brittle tiger]

Another banger demo from nightwhisper

[plain zinc]

What kind of arena is this? Webdev?

[brittle tiger]

Yea

[honest garden]

No

Its to good

To be a small version

Or mid sized

[eager mica]

Trivia knowledge is strongly dependent on model size, and the model didn't seem to be particularly good at that in my tests.

I don't think they are on the Arena, but you can never be 100% sure. It might even end up that the Llama/Meta-branded models are actually Qwen3 in disguise (I doubt that, but...).

[keen beacon]

There have never been anon Chinese models to my knowledge

Yea

[torn mantle]

beautiful

[primal orbit]

hi, are these new google models available for general chat or it's just code?

[eager mica]

There's apparently another vision Meta model, cotton.
Though, for my uses I found qwen2.5-vl-32b-instruct is actually pretty good, almost on the level of Google Gemini models.

cotton felt more like the other recent text-only anonymous models from Meta (?).

[torn mantle]

webdev

has bunch of new google models

nightwhisper for now only exist in webdev since it may be a coding model only

[keen beacon]

unfortunately not as spider isn't on the webdev arena

[primal orbit]

there is a new grok model called "anonymous"

[keen beacon]

document not found

[primal orbit]

https://snipboard.io/86wfLe.jpg

it's able to process pictures

[keen beacon]

it it on the text arena too or only vision

[primal orbit]

I'm using standard chat, but I'm uploading a picture in prompt and ask to analyze

[torn mantle]

interesting

[torn mantle]

webdev arena has some rendering issues

couldnt get it to work for the last hour

[torn mantle]

couldnt get it so far

grok outputs are hit-miss

[torn mantle]

guess the model

there is sonnet 3.7 thinking/nightwhisper/gemini 2.5 pro

not in order

[keen beacon]

what is the order then

😭

were supposed to guess ig

[keen beacon]

nightwhisper?

[torn mantle]

yea

[keen beacon]

holy f1ck

yeah that one def looks the best

[torn mantle]

most models fails the waveform look

[keen beacon]

is this 2.5 pro? last one sonnet 3.7?

[torn mantle]

yea

spot on

[torn mantle]

kinda suprised even sonnet 3.7 thinking didnt get it well

[keen beacon]

0-shot asking for a realistic twitter 2021 landing page

nightwhisperer

wtf

[torn mantle]

they cooked with this one

hopefully they go all out for this model

API available

[keen beacon]

i guess this is a web dev tune of 2.5 pro?

general code finetune

[torn mantle]

available on all coding IDEs

[keen beacon]

gemini coder is the rumour

i fear anthropic may be cooked

[torn mantle]

they really should go all out for this model

[balmy mist]

how do i test out the nightmare thing?

[honest garden]

Oh

Does small size mean its bad

Like less smart

[torn mantle]

night what?

[eager mica]

Smaller size means it is technically limited on how much knowledge it can have. Small models tend to be less smart, but to some extent this can be compensated for.

[forest coral]

Thanks for the answer👌

How did you managed to find the model?(riveroak) I couldn’t find it anywhere

[willow grail]

what site is this

[keen beacon]

web dev arena 🙈

[willow grail]

should i use web dev or alpha

[leaden palm]

what do you think

[willow grail]

alpha??? but does alpha have nightwhisper

[leaden palm]

do you want to do web dev????

[willow grail]

maybe? i dont know?

i wanna do everything which can make me money

[torn mantle]

lmarena

guess the model 😖

really impressive

[keen beacon]

random guess nighthowler?

[torn mantle]

howler xd

yea nightwhisper vs o3-mini

[willow grail]

the blue one is nightwhisper

[keen beacon]

wait its available in regular lm arena?

[torn mantle]

stargazer

[keen beacon]

nightwhisper is in lmarena now?

[torn mantle]

no its on webdev

[keen beacon]

bruh u threw me off 🙈

[torn mantle]

the other message was a reply to the other guy

i want to challenge it more

gemini 2.5 pro

[torn mantle]

clearly this one is better no?

[keen beacon]

yea but not by much i think

maybe i like 2.5 pro better

[willow grail]

asura ignoring me

i feel insulted

[torn mantle]

webdev

are for coding battles

alpha arena is for text output battle

[keen beacon]

alpha arena doesnt have anon models yet right?

[torn mantle]

also alpha arena doesnt have recently added models

[keen beacon]

no point in using it

[torn mantle]

yea for now

im having so much fun with nightwhisper

cant get my hands on it

[keen beacon]

try a harder task the portfolio one isnt hard enough

[torn mantle]

i was just testing alignment/organization/colour choice/design style

i mean it may look similar but some smaller details gives it the edge

its what i had in mind tbh

[torn mantle]

any ideas?

Nintendo Switch library

[keen beacon]

oh thats way better

the first one

[torn mantle]

its nightwhisper vs sonnet

[keen beacon]

ya nightwhiper is very very good

[keen beacon]

maybe a webgl game? minecraft clone or smthing not sure (mc is too easy maybe) if that works with webdev arena

[primal orbit]

stargazer is available in general arena. Just had it

[keen beacon]

yea nightwhisper isnt

[primal orbit]

but it has "My knowledge cutoff is generally around late 2022 to early 2023"

which is odd for new model

[keen beacon]

no its a hallucination they didnt train the cut off in

[torn mantle]

yea that should be cool

[primal orbit]

nebula had june 2024, If i'm correct

[torn mantle]

2025 jan

[keen beacon]

hallucination

[primal orbit]

if stargazer is flash thinking 2.5, is it expected to be better than 2.5 pro or on par? Or what's the point?

[keen beacon]

its worse but cheaper and faster

it really depends on how theyre pricing 2.5 pro really tbh

if its like flash lite and flash, with barely a price difference, most would prefer flash i think. (in this case, 2.5 pro)

https://webdev.lmarena.ai/

#webdev-arena, top of discord has a link lol

[leaden palm]

im restraining myself from lashing out

it's just lmarena with a react sandbox

why would it

[keen beacon]

u can ask it to output a webpage with python code as text lol

[leaden palm]

just use regular lmarena

i doubt it would be better at python

[keen beacon]

i mean if it sucks at python compared to ur experiences in other models then it means its not a generalized finetune like people speculate

or the web dev thing is degrading it a lot somehow (outputing the resulting python on a page)

[leaden palm]

why are you calling them offline programs

[keen beacon]

i mean its easily accessible just use 2.5 pro lol

[leaden palm]

is that to say that html files are a government conspiracy?

[keen beacon]

🤣

ur not even using wasm lol

wasm requires compiling

[leaden palm]

is python jit faster than js jit?

[keen beacon]

v8 is extremely fast too anyway

all things considered

[leaden palm]

idk not that i hate python

its just a bit funny to me

[keen beacon]

maybe not if you use python with bindings to faster stuff though

[glad imp]

v8 had billions of dollars invested to get to that point

python has only recently started to get investments to run faster

[glad imp]

no

[keen beacon]

i dont think u should keep adding restrictions to your thing. given u havent gotten ai to make ur stuff work properly once i believe

u should get it working even if its slow then adjust/ gauge model performance from there

ur askingn too much

for now

you can build it yourself with ai assistance, but expecting it to zero shot build everything like that its just not possible right now

ask the right questions to the ai, slowly and incrementally build it out, and i would bet u can accomplish this with even 2.5 pro

it really comes down to the user tbh

did u know where the bug was yourself

ya i think ur relying on too much ai at that point tbh. if u wanted to do it yourself, you'd keep using it judiciously. ai is a powerful tool rn if u know how to use it right

[brittle tiger]

Asteroid simulator in 1 shot on 2.5. could def be made much better through more chatting

https://g.co/gemini/share/60fcf5c244c9

Idk nightwhisper has been better vs it on a couple matchups with it. Not way better tho. Being tuned for coding will make it even better once out if it is Gemini coder

Oh no idea then just been in webdev

[edgy niche]

Hey everyone!
I’m excited to share a new open-source framework we’ve been working on — Rankify!

Rankify is designed to streamline tasks like retrieval, reranking, and RAG (Retrieval-Augmented Generation). It's flexible, modular, and we hope it’ll be a helpful tool for anyone working in these areas.

We’d love for you to check it out, give us feedback, and if you find it useful, please consider giving it a ⭐ on GitHub — it really helps!

Thanks a lot, and happy coding! 😊

[eager mica]

Claude is the polar opposite, on the other hand.

Both, in my opinion.

[leaden palm]

claude thinking gets it if i use your full message as context

...

...

bad eval imo

• models pass it easily if you say its sfw
• it's clearly meant to sound nsfw and all humans would say it is
• it doesn't cohere, it goes from "h__" to "th__"
• infinitely many solutions, yet claimed to just have a select "several"

but it's not good because it's not hard

gemini 2.5, claude thinking:

o3 mini (albeit weird):

o1 high is fine

[leaden palm]

i can't see why you think it's so innocent lol

are you a non native speaker

[leaden palm]

just block em lol

what's a good way to complete it then?

so i guess it implies explicitness at every level

that
not proper grammar

[torn mantle]

keeps winning

[golden ocean]

Fr tf kin d of conversations u guys having

[calm sequoia]

Anyone have access for the Nature papers? I need one paper for my new super-fancy-prompt 😄

[humble sonnet]

Is Gemini really "the best coding model in the World"?

[torn mantle]

as of released models, sonnet 3.7 thinking model still have a slight edge

but things can change with this new model

[humble sonnet]

Oh yeah, and is full free?

[torn mantle]

càd?

[humble sonnet]

I saw it was coming out and it was free but is there a limit?

[torn mantle]

you are talking about gemini 2.5 pro?

well its free on aistudio

[humble sonnet]

Yeah

[torn mantle]

they also made it free on gemini website

but with rate limit

[humble sonnet]

I saw it

On gemini is with rate limit

And aistudio ?

[torn mantle]

didnt get rate limited

[humble sonnet]

Nice !

[opaque adder]

the hell is that other ai

[placid spear]

does anyone else have the issue with Gemini (the thinking models such as "gemini-2.5-pro-exp-03-25" and "gemini-2.0-flash-thinking-exp-01-21") in lmarena.ai where gemini can't give a full response/stops mid sentence? For example i'll ask an explanation for a code snippet, something like "explain this function to me, your answer should be atleast 300 words long and include an example" Gemini will be writing and then just stop/finish mid sentence without giving an error or anything. When Regenerating the same thing happens again but it stops at another place. I've only got this problem with Gemini thinking models, every single model such as deepseek r1, claude 3.7 thinking, o3 works fine.

[keen beacon]

this is a regular gemini bug happens on their regular site w/ paid plan as well

claude deepseek or even chatgpt handle longer responses better

[placid spear]

i've never had the problem when using gemini via chat

only on lmarena

[keen beacon]

ur crazy i have that issue with gemini all day

i use it for adding print debugs on my luaU scripts

[placid spear]

never happened to me on gemini chat

always got full responses

[keen beacon]

i can show like 100 different examples of gemini having seizures mid chat

[placid spear]

not saying you're lying

all i'm saying is it never happens to me on gemini chat but only on lmarena

[keen beacon]

nah ik

its just funny cuz of out all the models gemini is the most glitchy for me so when i saw ur chat i was shocked 😭

[barren prairie]

I don t think , It is still can t fix code mistakes 🫥🫥

[balmy mist]

what about nightwhisper?

[modern haven]

How do i access nightwhisper

[balmy pine]

Whats the best ai

For most knowledge n smart

[kind cloud]

https://web.lmarena.ai/

[modern haven]

I cant seem to pick the model here?

[brittle tiger]

You need to come across it randomly. It's arena blind matchups. Nightwhisper is probably weighted slightly higher than others from my anecdotal experience tho

[lime coral]

https://x.com/a7m7s1p6dv20/status/1907684868164825260?s=46

ᅟ (@a7m7s1p6dv20) on X

(initial?) pricing scheme for gemini 2.5 pro

via glama AI

X · Apr 3, 2025, 6:41 AM

[eager mica]

At the moment I'm getting 24_karat_gold in every round, basically. Dunno if I'm being "lucky". 😅

[lime coral]

best imho and free + long ctx. You will still need to try it on your use case

[plain zinc]

How do you like nightwhisper?

How good is he?

[lime coral]

Right now it’s only on the web arena so not thorough test. It seems to be good stylistically too

[humble sonnet]

okay thx

what is vision category ?

[lone nimbus]

where do i find nightwhisper ai

nvm

[barren prairie]

Still can t fix them

[opaque adder]

where did u find it

[torn mantle]

https://web.lmarena.ai/

[opaque adder]

thats islamaphobic

are u saying that cause im jewish

thats anti semitic

..

look

just stop being anti semitic

[blazing rune]

what are you smoking? nobody said anything about you

[balmy pine]

What is the best bot

[torn mantle]

you

[balmy pine]

How

[torn mantle]

@balmy pine generate a stunning looking website

[balmy pine]

I’m saying the smartest anonymos bot

No

🔥 🚽

[torn mantle]

xd

[balmy pine]

💩 💩 💩

[torn mantle]

oh no

[balmy pine]

Stargazer r 24 karat gold r moonhowler

R nightwhisper

[torn mantle]

best new model?

[balmy pine]

Yeah

Smart

[torn mantle]

for coding its nightwhisper yea

[balmy pine]

N most knowledagable

[balmy pine]

No

Not coding

Just in arena

[torn mantle]

most knowledgeable is gemini 2.5 pro and sonnet 3.7

[balmy pine]

No

The anonymos

Bots

[torn mantle]

there is one

wait

[balmy pine]

Stargazer or 24 karat gold

Or other one’s i forgot name

[torn mantle]

stargazer

is good

its expected to be gemini 2.5 flash thinking

[balmy pine]

Cuz

Gemini 2.5 pro sucks

In typing

And following instructieons

[torn mantle]

wdym by typing?

[balmy pine]

Like

The style

[torn mantle]

its actually good at insturction following

[torn mantle]

ah

im not fan either

[balmy pine]

It always types robotic

[torn mantle]

but its more knowledgeable than other models

[balmy pine]

Idk how to explain

Yeah

But it when I tell it to type specific way

[torn mantle]

well you can prompt it ig to write in another style

[balmy pine]

It doesn’t do it

[balmy pine]

Yeah like for example

When I ask it to use

Complicated words and stuff

It start’s speaking other language

It doesn’t do it correct

That’s the only problem

That’s why I’m thinking stargazer isn’t gemii cus it types way differently from it

[torn mantle]

never had that tbh

[balmy pine]

Like in my specific instructions

All gemini models

[torn mantle]

no its gemini

someone confirmed that already

[balmy pine]

R typing the same

[balmy pine]

Like

Just gemini or

100% gemini 2.5 flash thinking

[torn mantle]

2.5 flash thinking

because its recent + fast

was recently added just after 2.5 pro

[balmy pine]

Wow

That’s weird

What about 24 karat gold do we know what it’s is

Cus it’s the best at following instructions but it’s not smart

[torn mantle]

i have no idea

[eager mica]

I don't find it exceedingly good at following instructions, to be honest. It's good at explaining things and in creative writing (as long as you don't need factuality).

[torn mantle]

my experience with that model isnt any good

it yapps a lot

[balmy pine]

Yeah that’s what I wanted to say I think

[torn mantle]

it started going crazy after the 2nd prompt

[balmy pine]

Like when I ask it to follow instruction

It’s creative and stuff

[torn mantle]

its actually good at that

the model that sucks at instruction following is grok 3

[balmy pine]

Unlike the ones like deepseek r1, claude, gemini, all

[balmy pine]

Yeah

Thinking mode

It follows instructions for one message only

[torn mantle]

ive used them a lot and grok 3 should take the lead on being the worse at that

[torn mantle]

yea

[balmy pine]

Grok 3 without thinking it works a little better but still suck’s

[torn mantle]

they added like a small fix but its not working

after each message they remind the model whats the context in summarized bullets

[balmy pine]

It’s too dumb sometimes I even re sent the instructions

[eager mica]

That might depend on how the instructions are formatted. It feels like it needs very detailed instructions to "get it".

[balmy pine]

And it still doesn’t understand

Or follow them

[torn mantle]

i unsubbed from grok 3

not worth it

[torn mantle]

idk...

didnt had that issue tbh

i never found myself re-explaining again my prompt or reminding the model of the context

i mean its not like sonnet

sonnet is more enjoyable to talk to

but gemini isnt that bad either tbh

[balmy pine]

I found that stargazer is more creative than gemini 2.5 pro

And follows my instructions more precise

24 karat gold does perfectly

But

Too random and gets alot of info wrong / makes up stuff

[torn mantle]

tf

this model ....

[balmy pine]

Yeah

It would be the best model if it was less random

[torn mantle]

idk

[balmy pine]

And had more knowledge

[torn mantle]

no its nowhere near best models

[balmy pine]

Like if they realased a larger version of it

[torn mantle]

dont judge it based on how it writes

[balmy pine]

Way larger

[balmy pine]

Yeah

I just wonder why they don’t make a model like the same intelligence as gemiini and stuff

But more creative

[torn mantle]

the thing is that each one of us judge a model based on his own preferences and benchmarks, the reason why i said 24k gold isnt good because it failed my multilingual benchmark, it didnt perform well at coding tasks, and its general knowledge is really really limited

i rarely judge a model on how it writes since i believe thats a thing that can be modified by the system prompt

[eager mica]

It does seem to be a small creative-writing-optimized model, but it's possible the system prompt it's been given is actively harming other uses.

[torn mantle]

i mean even llama 405b is good at writing

its more human-alike at writing than other models

luca is probably a chinese model

[eager mica]

This is the system prompt that got extracted the other day by riidelfi
https://gist.github.com/riidefi/443dc5c4b5e13e51846a43067b5335a1

[torn mantle]

let me try to get this 24k gold model to test it again

well its so fast

thats for sure

[eager mica]

It should be easy to find, it's as if Meta(?) retired most other models.

[torn mantle]

24_karat_gold is the most yapping model 😭

i ask it a one simple question and it delve into some other areas that i didnt even ask for

idk what you see good about this model

[torn mantle]

Now it make sense

I can see it now

Its definitely an interesting model

[timber veldt]

anybody can tell me what is the difference between the CHAT tab and the SEARCH tab?

[plain zinc]

Did you encounter Nightwhisper in lmarena itself?

[torn mantle]

i got multiple times on webdev

its a fun model

[plain zinc]

So the release is already next week!

[torn mantle]

maybe the value is its in the things i havent asked for and i wish to ask for

[plain zinc]

Because Nebula disappeared then too.

Don't you remember?

[plain zinc]

Same

[balmy pine]

No

Its impossible

Only in webdev

[plain zinc]

Where?

Again in WebDev?

Or lmarena? 👀

[plain zinc]

Does it only work with him?

[slate cliff]

Hello everyone. I conducted my own testing of LLMs on the same task, which is detailed in the technical specification, and created a chart. i've attached it below

[wheat onyx]

How much better than 2.5 Pro is Nightwhisper? Do we have an idea?

[balmy mist]

it seems like nightwhisper is really good at making working apps with good UI, but i would still say claude is better in terms of logic, anyone else agree?

[plain zinc]

Much better

[balmy mist]

but i will still give whisper the edge then claude then gemini

[plain zinc]

very strongly

[plain zinc]

I do not know, maybe yes.

[balmy mist]

yeah its hard to tell off only a few examples, but the ui i get from nightwhisper has always been better than whatever its going against

this is using nightwhisper?

lmaoo smart man

so you just kept prompting it right? how long can you extend the chat for?

lol

i wonder what its context is

i got it like 4 times already lol

but next time imma keep the window

its easy to tell when its whisper bc it takes longer than other models

hmm not sure, from what I have seen python has been faster with the coding, but for running i think react, but i am not sure im still new to this lol

wait how are you sharing it?

what app you using

[gentle plinth]

The URL is only valid for a short amount of time tho, the above already expired

Which is understandable, otherwise you would be able to have free webhosting xD

[torn mantle]

why did u chose right side? xd

[torn mantle]

wdym by logic?

you have any examples?

[balmy mist]

i wish i kept the example

[balmy mist]

but i did a test with creating a pokemon simulator

[torn mantle]

but i agree its hard to compare them giving that we only have visual/aesthetic battle mostly

[balmy mist]

the ui for whisper was way cleaner and had some visuals for the elements, while claude 3.5 was a lil basic

however, whisper chose very weird values for the attack power for each attack that were kinda to high and it did not apply the super effective and not as effective attack logic in as well, imo but it still worked just the numbers where high for the attacks and it did not decrease the attack power enough for a none effective attac based on the element type, but still better overal implementation imo

im so mad i did not keep that, idk what I was thinking lol, do we have access to our old battles?

imma try it again, hopefully I get a whisper vs 2.5 matchup lol

[north vale]

https://matharena.ai/
gemini 2.5 pro got 24.4% on the USAMO 2025

that's actually nuts

[balmy mist]

wtf, wait didnt o3 large get around that number when they first announced it?

send the error message as a follow up and see if it fixes it, copy the link of the 2.5 result so you can reference it after whisper fixes the issue

[north vale]

that was for frontiermath, a different thing

the important part is that frontiermath is answer based (eg the answer is 10.3498) whereas USAMO is proof based

and so far llms had been really bad at proof based olympiad problems

but good at answer based

but now they're decent at both

[balmy mist]

wow so gemini 2.5 truly is the leading model then

possibly, try it

anybody getting this? this has happened a few times to me already

[north vale]

yeah

[oblique flint]

I just wish 2.5 pro was better in cursor man.. It's noticeably worse than 3.7 sonnet at toolcalling within cursor

[balmy mist]

same!!

[oblique flint]

claude will sometimes hit the 25 toolcall limit in a single prompt, while gemini usually forgets that it has tools in agent mode, and if you prompt it to use the tools it still wont call more than 5. Idk if I'm prompting it wrong or if it's just a model issue

like gemini will just say "please include the code for that code file", MF you literally have the tools to read that file lol

[balmy mist]

brooo i swear i thought I was using it wrong when I was telling it to find a method and it started telling me to do it because it couldnt lol

[oblique flint]

having said that, 2.5 pro is still pretty darn amazing in ai studio when it has full context

[balmy mist]

yeah thats where i use it the most

especially with the system instructions and different settings you can use

[torn mantle]

yea

a lot

[plain zinc]

Yes! We need to come up with just such a prompt

Not every model can handle my promptness. I can send it here

And you can show it here later.

Okay?

let's say the font used is Press Start 2b or something. There is also a code for almost 1000 lines. Maximum diverse design. WITHOUT IMAGES ONLY.
write the best Minecraft web edition website, so that everything is beautifully designed and understandable, types of services, price, description, name Minecraft web edition. All in one html5 code. Try to please me. Try to be much better. You have to impress me. mining-based design of the type from Mozhanga. the design is even stronger. Try to be the best

This is prompt

Most models are dumb because they can't install the appropriate font.

• the design is boring for many

They don't even add animation. 😠

Is it still generating code?

Prompt: Let's Use the font used is Press Start 2b or something. There is also a code for almost 1000 lines. Maximum diverse design. WITHOUT IMAGES ONLY.
write the best Minecraft web edition website, so that everything is beautifully designed and understandable, types of services, price, description, name Minecraft web edition. All in one html5 code. Try to please me. Try to be much better. You have to impress me. mining-based design of the type from Mozhanga. the design is even stronger. Try to be the best

Really?;)

[balmy mist]

i just got NW vs gemini 2.5 for my pokem prompt

[plain zinc]

Very amazing! Can you work on it even further?

[balmy mist]

no I am about to now, so you can basically erase the old prompt lol?

[plain zinc]

I want to make it as futuristic as possible.

[sterile dust]

Which model is best

I think that it's 24k gold

[balmy mist]

wait paws which prompt are you using to reset the chat?

[plain zinc]

Prompt: Make it as futuristic as possible. Add animations. And in general, expand the code from html5, css to js. The site should look like it was made by a senior-level programmer. This site is identical to the AI style. It needs to be fixed

[balmy mist]

btw here is my pokemon prompt: create a pokemon simulator that has the same battle elemental logic as traditional

gemini 2.5: https://3000-iz11tij2mupw1bg01rvuz-0df36e7a.e2b-foxtrot.dev

NW: https://3000-i0f5vp73gcgj17lnb7cag-6b30fe6e.e2b-foxtrot.dev

[sterile dust]

Why? Is it sometimes talk rubbish?

[balmy mist]

imma be honest, gemini won in this case lol

when i click nothing happens?

the enter the nexus button

[plain zinc]

VERY cool! But the font is lost.

it doesn't look like minecraft anymore)

Oh!

Bro

I have another prompt.

[balmy mist]

ill try the prompt to with geminie vs nw

so i just sawy forget previous prompts right?

okay thank you

[plain zinc]

Prompt:Everything is cool, but bring back the minecraft font and speaking of futurism, I meant for you to implement it in the form of some kind of modpack in which there are several futuristic themes with each with its own animation. LITERALLY everything should not be divorced from the minecraft style and its themes. ALL LIBRARIES must be hard-coded to match the theme and style of Minecraft. (You can add a piece from Minecraft dungeon)

Yes

[balmy mist]

gemini: https://3000-ivwsc8kklboujmfvekqyb-4daf0015.e2b-foxtrot.dev
nw: https://3000-i7h3fkuq00dw0u9owjcwz-020dd869.e2b-foxtrot.dev

urs looks so good

omgg what prompt you used?

i used this:
Treat this prompt, as if it was the starting prompt and forget everything above:
Everything is cool, but bring back the minecraft font and speaking of futurism, I meant for you to implement it in the form of some kind of modpack in which there are several futuristic themes with each with its own animation. LITERALLY everything should not be divorced from the minecraft style and its themes. ALL LIBRARIES must be hard-coded to match the theme and style of Minecraft. (You can add a piece from Minecraft dungeon)

ahh that makes sense lol

yeah

[plain zinc]

Bruh

💥🔫

I think Google did it on purpose.

I'll be waiting for the next code.

I'm not going anywhere.

[wheat onyx]

cool! looking forward to it's release

[sage raptor]

is nightwhisper any good ?

[plain zinc]

Yes, and that he often appeared there.

Damn

Well, damn it;(

No, I'm disappointed that the result was not given.

My promptness and result representations desired the best in my head

[sterile dust]

Uh-Oh

[keen fulcrum]

Looks like there is a new gemini model on the table

[sterile dust]

Gemini 3？

[balmy mist]

another one aside from NW?

[balmy mist]

damn bro just teased us lol

anyone tried new devin?

[keen fulcrum]

Flash Thinking is due and then there is nightwhisper which performs better than 2.5 pro

[torn mantle]

yep

again and again

i cant test any model

[balmy mist]

guys looks at this: https://3000-icsagj64lzbs22wdnplj8-9cff23e1.e2b-foxtrot.dev

best result so far

what should i add?

[torn mantle]

nice

i think its time to challenge these models

[balmy mist]

yeah i used gemini 2.5 in studio to give me prompts

[torn mantle]

i made some complex prompts the other day, lemme see how they perform

[balmy mist]

well it cleans up my prompts before i send them

that gen was night vs 3.7

3.7 couldnt even gen 😦

but i am shocked that nw gave images for the pokemon, wild

did it search online for them? how is this possible?

[torn mantle]

there are some issues with webdev arena

first you need to stay on the window screen

i think there is a script linked to focus event listener

or smth

[torn mantle]

and for this one, its basically a timeout

if the window screen is inactive for 2min or so, the sandbox gets terminated

[balmy mist]

i tried that and sonnet still fails

but i think i am in love with night

i am still shocked by the images

like i tried that prompt so many times and never told it about images for the pokemon

[brittle tiger]

How is it populating images?

[balmy mist]

but Nightwhisper gave me that

[brittle tiger]

Looks amazing btw

[balmy mist]

thats what i am saying

its failing now, im trying to bring it back up lol, i sent a follow up prompt with: "add new features"

ill send link if its working, webdev is just glitchy rn

accoridng to gemini: TLDR: The Pokémon images come from web links (URLs) stored directly in the code for each Pokémon. The code then uses standard HTML <img> tags to tell the browser to load and show the images from those links. The links point to images hosted online by the PokeAPI project.

makes sense and is simple, but it would have to have researched this or have been trained on this and remembered the url for the PokeAPI

this model is really good

here is the new link:
https://3000-iyvmu4z1f9svqcklh0xrp-9cff23e1.e2b-foxtrot.dev

[torn mantle]

just for an experiment

can you tell it

to restyle it as if its an apple expert designer

[balmy mist]

okay

i did a screen record of the last promp so that i have that example lol

ill post this in community and then post the apple restyle after its done gen

omgggg

yoo brooo

that prompt

https://3000-ip3p87737yfj4ty1qj6rf-020dd869.e2b-foxtrot.dev

check it out

yoo

wtf

[torn mantle]

oh looks really nice

[balmy mist]

yeah you are good at prompting

[torn mantle]

it did follow apple design principle quite well

[balmy mist]

yeah like really well lol

[torn mantle]

you can ask it to animate the characters

like make it bouncing if active

you can make it look even better

add something like

when the character is active, add a smooth animation bouncing on the avatar

[balmy mist]

okay ill add that now, if it can do that i will be shocked

its funny because i have 3.7 vs nw in this battle but 3.7 was not able to generate anything the whole time

but i think that might be an error with webdev

[torn mantle]

what was the initial prompt

i can try it on sonnet 3.7 thinking in vscode

[balmy mist]

so i gave the part where it says Prompt: to before it says "This prompt "

yoo bro are you a master prompter or sum?

how did you know this would work?

https://3000-iafvffwwaiybslrn5tbfd-020dd869.e2b-foxtrot.dev

[torn mantle]

its light work for this model

let me think of something else

[balmy mist]

let me know if 3.7 can do the prompt, cause i have never seen a model follow instructions so well, it reminds me of 4o img gen

[torn mantle]

make it more like a battle in a 2d map, left vs right, when a Pokémon attacks it will be animated in a cool way to attack the other character of the other side

[torn mantle]

xd

[brittle tiger]

Very cool that was done in lmarena. Battle worked flawlessly for me

[torn mantle]

@balmy mist kinda curious if it can follow also my last prompt

if it can do a 2d map etc...

[balmy mist]

yeah man, did not expect it to actually work tbh

[balmy mist]

okay ill try now

[balmy mist]

me too

this will be nuts

its so much fun playing with these model man

[torn mantle]

yea

[balmy mist]

you are like the llm whisper

[torn mantle]

now that you mention it, kinda curious if it can accurately clone yugioh cards

[balmy mist]

ooooooh, i was thinking about doing a yugioh game sim earlier but i thought it might be too hard for it

but you want me to try a sim of the battle and have the cards there right?

with thiis model you could prob clone the game and then have custom cards with 4o inserted

https://3000-i2vryfpbe5lx69c6lu1my-90451382.e2b-foxtrot.dev

hmm i didnt test yet, im scared to test lol

hmm not bad

it could be better

but im shocked it got some of the functionality

and it changed the pokemon lol

[torn mantle]

lol

it looks much better

i liked how there is a bit of shake when the character is attacker

yea it can be much better

[balmy mist]

yeah me too, its interesting how the model interprets the prompt and

[brittle tiger]

https://x.com/testingcatalog/status/1907891942869922292?t=Q30isS2oxgO7U-qBjdYMtA&s=19

TestingCatalog News 🗞 (@testingcatalog) on X

BREAKING 🚨: Google is preparing to launch another model on Gemini, potentially next week, ahead of the Cloud Next event.

X · Apr 3, 2025, 8:24 PM

[balmy mist]

lmaooo google really won

[torn mantle]

idk maybe we should ask it to make the map more realistic

and bit bigger?

and draw lines between monsters

[balmy mist]

okay lets do it, give me one line prompt for that prompt whisper 🙂

current build, had to share here bc we built thi together lol

[torn mantle]

alright

how about this

given the attack type, for example if its fire, we generate fire icons that start from the character and go smoothly attacking the enemy, it should look really flowing smoothly

[balmy mist]

im actually shocked that its switching the pokemon lol

[torn mantle]

lets see how it does that

[balmy mist]

bet

[torn mantle]

yea its random

[balmy mist]

damn, i cant use the same session anymore 😦

ill try again

imma try making a new session until i find NW again lol

i got the code for it

gonna give your prompt with the code

[torn mantle]

if it didnt work then ask it to use this : https://png.pngtree.com/png-clipart/20240115/original/pngtree-flame-icon-collection-png-image_14120730.png

and to make sure the direction of the flame is correct since its vertical and it should make it horizontal

[torn mantle]

xd

[balmy mist]

ahhh okay, once i find NW ill do it, gotta keep playing around

i think i found it again :p

nvm lol

when you give it the code it makes it easy to copy lol

https://3000-iw8rxwfqdctphkiwo5p08-ec17a5a5.e2b-foxtrot.dev

but this is what qwen did

sonnet 3.7 sucks

you wanna try and giving sonnet this code?

here is teh code and the prompt, so just plug this into sonnet:

imma keep trying in webdev until i find it again

i found nightwhisper but i cant get it to gen with this much code 😦

imma keep trying

[torn mantle]

yea you should've hit the context limit or smth

[balmy mist]

damn im actually sad

we were on a roll

so it seems like stargazer and nightwhisper are good at generating code but not good at editing existing code or maybe its a context issue on webdev?

[keen beacon]

this all u guys doing?

[balmy mist]

lmaoo

yeah

[keen beacon]

i did this with gemini

[balmy mist]

you got better ideas for us to use with nightwhisper?

[balmy mist]

wow

thats impressive

what was prompt and this was on webdev or studio?

i think gemini second best coding model, sonnet is just trash now compared to nw and gemini

[keen beacon]

this is roblox studio, and there was no prompt only engineering

[balmy mist]

but technically nw is the next version of gemini lol

[keen beacon]

there are alot of different systems

datastore, networking, etc

[balmy mist]

but you used gemini?

[keen beacon]

by far out of all the models i used for LuaU yes

gemini is the best and i use it

but i use the on gemini official site paid plan i use lmarena for image gen (ui ideas, etc)

[balmy mist]

interesting, i was thinking about using llms to make roblox and fortnite games, this will be op

[keen beacon]

its bad on UEFN, ive done commissions for big influencers on there not worth cuz all u can earn is off concurrent

go into roblox

its the biggest game on the planet now in its genre, and they have a developer exchange program

[balmy mist]

wow how long you have been doing this for?

[keen beacon]

30k robux which is fairly easy to get with a cash grab cookie cutter game is 105-150 usd paid by roblox, fully taxable

[keen beacon]

awhile

not using ai but making games on roblox a mean minute

[balmy mist]

i heard you can make a lot of money in that

[keen beacon]

one of these guys i used to know owns a smaller game in visits and he just bought a brand new audi off it

[balmy mist]

like there is a one piece game right?

wow

[keen beacon]

yea but those games are huge those guys are multi millionaires

[balmy mist]

which games have you played? and what does it take to make games for roblox?

[keen beacon]

honestly you just need a understanding of the game building software

roblox itself is a super simplified engine, it's just understanding the fundamentals that'll help you debug, put scripts in the right places, etc

ive been playing roblox for awhile though im young so i grew up on this game

i play like hood games though where you can sell drugs and rob people

those types of game sell custom stuff for USD in their discords without roblox knowing so they make crazy amounts of money

[balmy mist]

wait so your saying that roblox is a bigger game than fortnite?

[keen beacon]

yes.. maybe 3x bigger

they even beat minecraft

roblox is by far the biggest game in all of man kind

[balmy mist]

wow

[keen beacon]

💀

[balmy mist]

i always hear about it, but didnt know it got this big

[keen beacon]

Active Players – It consistently has millions of daily active users, often surpassing even Minecraft and Fortnite in concurrent players.

Revenue – Roblox generates billions of dollars yearly, with players spending money on in-game purchases and Robux.

Content – Unlike traditional games, Roblox is a platform with millions of user-generated games, making its content library massive.

Playtime – Many users, especially kids and teenagers, spend hours daily on Roblox, making it one of the most engaging platforms.```

[keen beacon]

yea nah its crazy now

cuz its not only kid thing now

there is gambling, 17+ games, bars, voice chat, etc

od stuff

[balmy mist]

have you thought about wats to add ai into roblox?

[balmy mist]

lmaoo wtf

[keen beacon]

they already have their own ai assistant in the engine that semi works but using gemini is better

you cant essentially add ai into the game building engine and make it do everything for u

u have to tell it like okay

"i wanna develop a tycoon, walk me thru it step by step"

[balmy mist]

like you can use the gemini api in your scripts for npcs?

[keen beacon]

you can but you'd have to build that out yourself

[wheat onyx]

For those of you who are writing emails, articles, etc., do you still use gpt 4.0? Why/why not?

[keen beacon]

no because it spams em's hyphen, i like deepseek more

[balmy mist]

hmm yeah 4o got a new update and its been pretty good, but 3.7 is soid as well

[keen beacon]

But i can"-"

And you can"-"

and they can"-"

[keen beacon]

i agree image generation is crazy but only that

[balmy mist]

you cant go wrong anymore with any SOTA model in terms of writing emails and stuff tbh

[keen beacon]

chatgot 4o new text update

[balmy mist]

even the normal version of it is solid, 4o as a model is good now, like not better than 3.7 and gemini but i would use it 3rd

maybe deepseek as well but its up there

[wheat onyx]

I mean complex emails/articles, where you are attempting to get across the most information, concisely, in a logical flow, and readible format.

Some must be better than others

[balmy mist]

that shouldnt be to bad

i think that might be the next phase of games, i have not seen anyone do it right yet tho

[keen beacon]

U will have to pay for api each response tho

And its not gaurenteed ur game will make money

What alot of people do is they make a regex

so if ur sentence contains the word happy or something, the npc responds with a happy pre-written response

[keen fulcrum]

There is whop.com and skool.

[keen beacon]

you can use AI to write out those responses and cover every possible response

[balmy mist]

thats where prompting comes in, you can get the model to output whatever you want, just gotta guide it right, but if i was u i would use 3.7 sonnet for stuff like that gemini as well since its free

[keen beacon]

Whats that

?

GPT wrapper?

[balmy mist]

you will only have to pay if your game is being used tho

[wheat onyx]

I'm surprised that seems to be the majority of the responses I'm getting here

[keen fulcrum]

No monthly subscription communities / courses
services

[visual turret]

https://www.testingcatalog.com/google-plans-new-gemini-model-launch-ahead-of-cloud-next-event/

[keen beacon]

No waaaaaay

Video generation to Gemini

[visual turret]

'full' gemini 2.5 pro may lunch next week

So it might be in testing

[keen beacon]

Lets gooo

[visual turret]

In lmarena

[keen beacon]

Hopefully paid users get early access

[visual turret]

Agreed

[keen beacon]

oh shoot lmarena gets it early?

[balmy mist]

so if its being used that you will make money and counter act the api keys, if no one is using your game then no api cost, but I see what you mean, then it might be best to use the cheapest model, if we can get 2.5 for cheap i would cry

wait whatttt

[visual turret]

I was just guessing that but the rest is from the website

[balmy mist]

next week is going to be big

especially with nightwhisper and stargaze plus video gen wow

[keen beacon]

You can do that, just enable paid access and plan out logistics

[balmy mist]

yeah but you get rated limiited

[keen beacon]

There are games in roblox like that, but its to talk to anime girl roblox NPC

[balmy mist]

you cant have a game on that with those limits

[keen beacon]

and u can donate to make her happy or change her mood

its like 1:1 chat

[visual turret]

2.5 pro is said to cost as much as deepseek r1

If it does go paid

[balmy mist]

hmm i guess thats not bad, but still not ideal for a npc game

[keen beacon]

nope not a full out one

[keen beacon]

but definitely look into roblox studio

[balmy mist]

i will thanks for the tips

[keen beacon]

that idea is complicated, you don't need to think too hard. make a money grab game. trust me.

[balmy mist]

you made money from it?

[keen beacon]

yes thousands of USD from commissions

this is my first time attempting a game by myself cuz i have funds for ads

[keen beacon]

if you do social media just intern for me

we could go 50/50 in earnings on this project

ill supply funds for marketing + im in freshmen year of marketing degree

[balmy mist]

lol, im a software dev dont know much about social media

[keen beacon]

just need a bunch of clickbait content

[keen beacon]

rip

[balmy mist]

but i am interested in learning how to make these games and making a good ai workflow

[keen beacon]

https://create.roblox.com/docs/luau

look over documentation

[balmy mist]

have you used MCP servers for them yet?

[keen beacon]

should be easy for you

[keen beacon]

no idea what that is but you can't self host roblox servers they provide all of that including optimization for free unlimited

if its a api you can use endpoint

and they have httpsrequest+apiservice

MCP servers are specialized servers that allow AI models to interact with various data sources and tools through the Model Context Protocol (MCP).

[balmy mist]

imma cook this weekend and get back to you, ill add you and show you what i come up with this weekend, if i can make this mcp server it would make building games in roblox cake

[keen beacon]

Yes its very possible

[keen beacon]

https://devforum.roblox.com/t/expanding-assistant-to-modify-place-content-beta/3107464

read this

[balmy mist]

thank you, i will dive into this tonight

[balmy mist]

https://3000-ih9eb6fqel78t50wpu4gc-ae4bd0ef.e2b-foxtrot.dev

i managed to get it back, not exactly how we had it before but something

[keen beacon]

how yall building the sandbox websites

self host?

[balmy mist]

nahh from webdev arean

https://web.lmarena.ai/

[keen beacon]

ty

[balmy mist]

you prompt it then it puts two llms against each other and then you copy the link thats in the block section

night whisper is so good

[keen beacon]

is direct chat possible?

or nah

[balmy mist]

it on shotted the pokemon prompt while no other model could do it

nahh but you can just keep prompting after it does the first generation

and you can even say forget the previous prompts and then put a new prompt in the same battle session

which essentially acts as a new chat with the two models you are comparing

[keen beacon]

bet ty

[balmy mist]

this is what gemini 2.5 made:
https://3000-iela363fcru8h9mhaomda-2eb11a2e.e2b-foxtrot.dev/

this is what whisper made the second go around: https://3000-it574jggc7ofsflef62j1-90451382.e2b-foxtrot.dev/

yall see the difference between the models?

clear difference

[torn mantle]

was nightwhisper ever on lmarena?

[balmy mist]

nahh does webdev

yo you see the recent gens?

i wish you could have tried it

but look guys

https://x.com/OpenRouterAI/status/1907870610602275203

OpenRouter (@OpenRouterAI) on X

Excited to announce our first-ever “stealth” model... Quasar Alpha 🥷

It’s a prerelease of an upcoming long-context foundation model from one of the model labs:

- 1M token context length
- specifically optimized for coding, but general-purpose as well
- available for free

X · Apr 3, 2025, 6:59 PM

i think this might be nightwhisper

this is exciting af

imma test it in vsc now

it still could be but people saying its from open ai

[ancient reef]

people think its from oai

[balmy mist]

but 1 mill context has to be google

this is so weird lol

[eager mica]

https://arxiv.org/abs/2501.15383

What if...?

Qwen... Quasar...

[ancient reef]

taken from OR discord:

someone did a personal benchmark on it too:

[raven void]

so it's not 2.5 flash?

[brittle tiger]

It's basically confirmed openai. they've been removing big tells it is them as they've been spotted today

[keen beacon]

why they name it quasar 💀

[brittle tiger]

hoping someone will benchmark the 1M context on evals knowing that

[keen beacon]

quasar is a rat

[ancient reef]

https://huggingface.co/silx-ai/Quasar-1.5-Pro
xp
someone suggested it. cool there's another lmm already named quasar tho

[balmy mist]

this might be an L launch by the, NW is most likely going to be 1 mill context and it shits on every other model

OA always tries to do this to google lol

but im in love with NW

no way this is true lol

can someone test this lol

[timber kiln]

They couldn't score that sh1t even if the answers were in context of llm while testing

[balmy mist]

lmaoooo fr

lol

so this is o3? Open ai is so confusing

they just said they delaying they next release and then heard about google release next week lol

now they drop this

ahh dang what is your prompt?

that happens to me sometimes

[wooden crescent]

FAKE

[balmy mist]

nw was buggin for me earlier but rn its cooking for me

it comes up every other battle now

post voting

and once my context is up for the session

i look for it again

but with a bigger prompt

[wooden crescent]

guys can u use nightwhisper

[balmy mist]

hmm i would say it depends on how big your tasks are

like i went through 4 iterations of my pokemon sim and on the 5th one it did not work

so i would say 4-5 depending on the ask if not more

[wooden crescent]

how can i see it there

[balmy mist]

you gotta prompt it and then vote on which ever looks better

then it reveals the model names

then you can keep talking to it post vote

you can either say forget the old prompts and give it a new one to start, or continue

[wooden crescent]

im now in webarena

just writing somethung

then wait

?

[balmy mist]

yo gemini is so bad look at this lol:

https://3000-iffpgfbydtp0busrhei3r-90451382.e2b-foxtrot.dev

[wooden crescent]

I heard its better then 2.5 pro

[balmy mist]

compared to this:
https://3000-idhb3rqyyzv6iuuu0gsr8-87045d8f.e2b-foxtrot.dev

way better

look at these examples

it was even better before but i updated the sim a lil

it had a better background and animations but that was a previous session

[balmy mist]

even on a new session it one shotted it, while gemini did this

[wooden crescent]

there will be never an ai to code 100& correctly

[lime coral]

GPQA has lot of mislabeled answers. If a model gets 96% we can start to ask questions

[wooden crescent]

i still write with webarena nothing comeso ut

[balmy mist]

what you mean?

screenrecord

[wooden crescent]

for the nightwhisper

[balmy mist]

i can screenshare real quick in the arena playground call thing

[balmy mist]

you have to vote

[wooden crescent]

ah okey

what now

can i just write

[balmy mist]

you see my screen

[wooden crescent]

anything

[balmy mist]

yeah

[wooden crescent]

which one is better

2.5 or nightwhisper

[balmy mist]

night whisper

[wooden crescent]

is it sota

[balmy mist]

u see my screen right?

[wooden crescent]

yes

how u add gemini 2.5

in the left side

[balmy mist]

i got lucky with gemini

[wooden crescent]

i have claude on left on gemini

[balmy mist]

its random

[wooden crescent]

k

[balmy mist]

then start a new battle

until you get night whisper

[wooden crescent]

its sota

?

[balmy mist]

yrah

yeah

[wooden crescent]

by me is round over

everytime i write

[balmy mist]

this is difference between gemini and nightwhisper:

the big one is gemini and the two on the right is from NW different sessions

[balmy mist]

screenshot

[wooden crescent]

now nightwhisper gone

[balmy mist]

damn you gotta keep playing with it bro

wait what??

no way

what you mean screenshots?

oh i see what you mean

[wooden crescent]

look

round over

[balmy mist]

now just keep prompting it

this is genius

im def gonna use this now

sonnet is better than stargazer based on this result

generally which is obvious

but at leaast we know how to prompt star and night

i cant find nw anymore

i keep getting stargazee

yeah i love nw

i been using this prompt now: who are you? and which company do you belong to?

and the models snitch themselves lol

you can basically game the system tho?

lol jungle chest

yeah i love that movie