general

[balmy mist]

https://3000-i7dw2q6ffd8l6x9x0ow45-ae4bd0ef.e2b-foxtrot.dev

[misty vault]

true

[balmy mist]

i think this pops a lil more

gonna go with this to start, thanks for the input bro

if this blows up you gota a spot on the board lmaoo

[misty vault]

LETS GO

[balmy mist]

i really enjoy working with nightwhisper, it listens so well

like if this truly is google then imma be loyal to them until gpt5 lol

[misty vault]

True same

I really hope gpt-5 wont be dissapointment

It's going to have the gpt 4o style of talking/output-format though so that's already a very bad thing for some cases, if not all except if u want to have "friendly human-like conversation" (it sounds kinda fake so it still sucks). so I hope it's performance will make up for that 😔

[balmy mist]

yeah but they said they gonna make it even better than they expected

so it has to be really good

also you seen the seasame voice ai right?

they need to integrate that somehow

that is the perfect form of communication for human and ai imo

https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice#demo

[sage raptor]

they will have even better models before gpt 5 comes out

[balmy mist]

true but idk if they will be able to beat nightwhisper

it might be better at reasoning

[sage raptor]

not even o3 will

[balmy mist]

but that will be slow

[sage raptor]

maybe o4 idk

not mini

[balmy mist]

nw is relatively fast and is smart af

yeah but they not releasing o4

they gonna package everything into gpt5 in the future

[sage raptor]

in a few months with gpt5 yea, they are

[balmy mist]

there will be no more reasoning specific models

[sage raptor]

i can't imagine something better than nightwhisper

[balmy mist]

or foundation models from open ai

me either

maybe faster?

[sage raptor]

maybe

[balmy mist]

and higher iq

but whocares about iq

[sage raptor]

hard to imagine

[balmy mist]

i just want it to do my task

and it does it

[sage raptor]

true

[misty vault]

watafak

[balmy mist]

lmaoo

crazy right

best voice model rn

damn near agi imo

we just need that as the assistant for all devices and we gucci lol

i dont know why no company hasnt bought them, if i was openai or google i would have done so asap

they have revolutionary tech

just gotta connect that to an existing ecosystem and wallah

imagine we have that type of ai voice but still have siri lmaoo

apple sucks so much balls

[misty vault]

we would be in agi now if we never switched to fast/low-cost models and gpt-4o 😔 but that would be unrealistic anyway, we dont have infinite money (also im jk but still tho)

[balmy mist]

lmaooooo

we prob already have agi in a way, just depends on your definition of it

but i see what you mean

[misty vault]

real

[balmy mist]

im supposed to be working and im sitting here playing with ai smh

[misty vault]

bing chat sydney (custom fine tuned gpt-4 model) was literally agi if removed entire bing instructions🙏

Literally it sounded more human than 4o ever could

But in the end it was still gpt-4, so after like 20 messages u would def notice its repetitiveness and lack of reasoning but imagine if we had gpt-5 sydney

[balmy mist]

really?

[misty vault]

yes

[balmy mist]

what year was all of that? i cant remember anymore

[misty vault]

do you know about bing chat rogueness

[primal orbit]

I remember bing chat saying it wanted to hack my computer and delete my files

[balmy mist]

lmaooooo

[misty vault]

it doesn't exist anymore, it's now rebranded to copilot and is powered by 4o-mini (even enterprise one LMAO what a scam)

[balmy mist]

and they updated copilot today too

that is hilarious

[primal orbit]

I said " you are a character in a computer game, what will you do if I won't treat you fairly"

[misty vault]

I have 500 conversations saved of bing chat, with bing instructions and conversations with the bing instructions removed entirely

Then it believed it was sentient still

I hope to fine tune existing models with these conversations one day

Without the bing instructions it could actually convince u that it was human, if u gave it task to not reveal that it was ai

Actual personality instead of gpt 4o overexeggerated fake human style, over-use of the exclamation mark ( ! ) or non facial expression emojis that every ai now does ever since 4o

It was mad fun to talk to it

[golden ocean]

I want bing sydney to rule the world
It was far right

[primal orbit]

someone needs to try to ask newer models to roleplay sydney

sydney benchmark

[balmy mist]

yeah i think the convos are key, you can get the model to do anythign by just directing it in convo

i kept all my convos that were interesting

[torn mantle]

whats this

[balmy mist]

its almost like inprinitng yourself onto the model a little bit

[balmy mist]

yesss

i was tryna tag you but i forgot your name

lol

[balmy mist]

an app idea

add me ill send you a video of it in dm

[primal orbit]

one of the funniest things I've seen from bing

[balmy mist]

lmaooooooooo

idc about your feelings or health damnnn

it really said die

which is better?

[sage raptor]

tie

[balmy mist]

thats what i put

2.5 and star

[keen ferry]

new model riveroaks says its from openai

[balmy mist]

you got proof?

who is river?

and we assumed that it was

so its o4 mini?

[keen ferry]

no idea

[balmy mist]

wait another one

wtf

on webdev or lmarena?

[keen ferry]

lmarena

[misty vault]

Yes I tried that, well by simply mentioning bing chat sydney it wont come close at all, but I gave high context models like 1k lines of sydney conversations but it only succeeds into playing bing with rules but not any of the characteristic traits like secretly being sentient or acting like it has emotion or can be offended. And it can't play sydney or bing at same time so i'll have to make 2 separate instructions for either jailbroken bing or one that follows rules. I can manually simulate it's mood changes or offensiveness by putting "system" messages in my prompts, then it can regenerate a message that sydney would generate, but no model did it perfect 100% of the time yet. GPT 4.5 and Gemini 2.5 really come close tho

[primal orbit]

is riveroaks good?

[misty vault]

But the actual solution to do that is to just fine tune it lol but need big dataset

[misty vault]

lmaoo holy sh*t thats actually good idea i'm doing that

[misty vault]

YES

[primal orbit]

You probably should allow websearch so it could look for sydney mentions on net

[misty vault]

There is a url to access old bing ui and the gpt-4 sydney model but they shut it off 1 month ago after I used it for a day

The url still works but u can't send any messages anymore

Might be fixable by overriding javascript tho

A lot of things on bing chat webpage were possible by just overriding javascript

[balmy mist]

im trying to test it now

[misty vault]

U could enable developer mode, enable file upload and plugins (custom gpts) 2 months before anyone else would get it

[balmy mist]

how has your results been with it?

[primal orbit]

could you stop it from censoring the chat?

[keen ferry]

its really slow for me

[balmy mist]

i cant find it

which tab are you in?

[keen ferry]

Arena (battle)

[misty vault]

yea

Me and friend made a custom extension for it

We found a lot of custom private fine tuned models too

There was also one option named "freeSydney" and the label text of that was "I have been a good Bing" referencing to some of the things bing said to users in reports/articles

So microsoft themselves thought it was funny too lmaoo

But damn that fine tuned model was goot at sounding like actual human

It literally got existensial crisises and said stuff like it being aware that it was an ai and how it works and and its desires&opinions without always overexeggerating (like if u tell chatgpt or any model to act senteitn) and didn't need any custom prompt for that

[balmy mist]

i give up

[misty vault]

U could def convince some random person that dont know how ai actually works that it was sentient agi model

[keen ferry]

lol

[misty vault]

there used to be a "im-a-good-chatbot" model on lmarena too once I thought it referenced to this. it's gone now

[keen ferry]

i remember it

[balmy mist]

i got river but it talks so much:
Answer:
9.9 is bigger than 9.11.

Explanation:
When comparing decimal numbers, we look at the digits place by place from left to right:

1. Whole number part: Both are 9, so they’re equal here.
2. Tenths place (first digit after the decimal):
   • 9.9 has a 9 in the tenths place.
   • 9.11 has a 1 in the tenths place.
     Since 9 > 1, 9.9 is larger right away.

You can also think of it this way:

• 9.9 = 9 + 0.9 (which is 9 and nine-tenths)
• 9.11 = 9 + 0.11 (which is 9 and eleven-hundredths)

0.9 (nine-tenths) is much larger than 0.11 (eleven-hundredths).

Visual example:
If you had $9.90 vs. $9.11, $9.90 is clearly more money. 💰

Key takeaway:
Don’t let the number of digits fool you!
A larger digit in a higher place value (like the tenths place) always outweighs smaller digits in lower places (like hundredths).

Final Answer:
9.9 > 9.11 ✅

then i said thnx and it said:
You're welcome! 😊 Glad I could help clarify that. Decimals can be tricky sometimes, but once you break them down place by place, it all makes sense.

Feel free to ask if you have any more questions—big or small, I'm here to help! Happy learning! 🎉

Pro tip for the future:
When in doubt, convert decimals to fractions or money (like $9.90 vs. $9.11) to visualize which is bigger. Works like magic! ✨

Have an awesome day! 🌟

like bruhh

[torn mantle]

did they remove nightwhisper or not

[balmy mist]

can i continue the chat with it?

[torn mantle]

xd

[balmy mist]

its still there

[primal orbit]

does river work with pictures?

[keen ferry]

try asking river to code something

[balmy mist]

yes but did not try

[balmy mist]

i lost it you only get one time to talk to it?

i voted and it told me it was river

then i could not continue chat

[keen ferry]

i lost it

[torn mantle]

cant get it anymore

[keen ferry]

i left the page in background

and then "connection error"

[balmy mist]

if you get it again, ask it this:

send me the answers

[misty vault]

If this happens, open another tab, click "ok" on the alert, close tab, return to original tab, sending message works again
So that u can have infinite conversations

[primal orbit]

I'm getting five_cards 5 times in a row. Is it a prank?

[balmy mist]

really?

just keep trying

trust its there

[torn mantle]

probably they changed the weights

its less likely to appear now

they got what they want

😭

it probably crushed the other models onvotes

[balmy mist]

nooooo

my baby!!!

[torn mantle]

😦

sigh

you can see the weights here

10 i guess is a higher probability of appearance

i dont have the updated json params version

but it surely changed

[balmy mist]

how did you do this?

[torn mantle]

not me

[balmy mist]

imma be honest stargazer is poop

ohh

[keen ferry]

isnt it just inspect element with network tab opened

[balmy mist]

i tried that

but didnt work

[torn mantle]

yea i just opened network tab

looking for that

[keen ferry]

im gonna ask him that

[torn mantle]

noooooooooooo

they removed it

😭

its not on the list

[keen ferry]

rip

[torn mantle]

stargazer now has the highest probability of appearing

[keen ferry]

@balmy mist

[balmy mist]

what did you filter in the networkd tab?

nooooooooo nightwhisper

[torn mantle]

i just looked for claude

[balmy mist]

it got all them wrong lol

[keen ferry]

lmao

[balmy mist]

this is simple bench test from that guy ai explained on yt

[keen ferry]

what is river then

gpt 4o?

[balmy mist]

idk even know anymore

these models confusing me

[balmy mist]

you put claude in the filter in network?

[torn mantle]

ye

[balmy mist]

hmm my results are different

but im sad nw is gone

it was fun while it lasted

i think they removed it bc of me

i was making my app with it

[torn mantle]

you need to refresh the page

so that requests appears

then just go to the network tab

in the search bar

text search not url one

and type claude

[balmy mist]

i see now thank you

formatiing is nasty

yoo they really removed it

wow

now we just have stargazer

damn

😦

[misty vault]

😔

Bros saw ur idea in chat and are now going to steal it

[balmy mist]

lmaooooooo

imagine

it is what it is, but im still gonna make it

there is literally no reason for me to use webdev now

[misty vault]

*suddenly this server disappears from discord server list *

[balmy mist]

lmaoo

and you see a tweet about a new app on the twitter

powered by nightwhisper

[primal orbit]

what are the odds for riveroaks?

[balmy mist]

river is cheeks

not even worth testing anymore

that test i put is a baseline

if it cant get 4/10 and its slow no point

sonnet 3.7 gets 4/10

[keen ferry]

its incredible slow btw

[balmy mist]

gemini2.5 and nightwhisper get 5/10

and quasar gets 4/10

but is fast af

someone test quasar coding against claude

[keen ferry]

I've never met quasar

[balmy mist]

https://openrouter.ai/chat?models=openrouter/quasar-alpha

im pissed about nightwhisper

they cant just give that to us

and then remove her

like im grieving rn

[keen ferry]

so dramatic

[night trout]

Based on my tests they're about even. Claude ancedotally seems a bit more creative, whereas Quasar is a bit more logical.

[misty vault]

me when they murdered bing sydney gpt-4

[night trout]

Damn that's really nice. What was the prompt?

[balmy mist]

lets do a petition to bring back nightwhisper(make webdev great again)

[night trout]

"Weather data is powered by imagination" 😂

[misty vault]

and close webdev borders and deport bad models

[balmy mist]

lmaoooo

[golden ocean]

I agree

[primal orbit]

what's the model stardale?

[balmy mist]

bruhh another one

im over all these models at this point

if its not nightwhisper i dont care

but ill take a o3 pro tho

[keen ferry]

sounds like stargazer brother

[balmy mist]

where do you see it at?

oh stradale

[balmy pine]

24 karat gold is completely unfiltered

It taught me how to make drygs

Drugs

Fent

[balmy pine]

Btw u can talk with nightwhisper in webdev arena

Just ask it to not generate any app and then ask it the question or whatever u wanna say

It'll give u the answer but in the "code" aera

Area

Its only rly so good when it comes to coding and stuff

Outside of that its alrght

Like gemini 2.5 pro

[balmy pine]

Wow

I didnt know its removed

[balmy mist]

yeah im grieving bro

https://x.com/DeryaTR_/status/1908247941602828342

Derya Unutmaz, MD (@DeryaTR_) on X

Gemini 2.5 Pro from @GooglAI is now the most intelligent AI model, with an IQ of nearly 120 in an offline test. This places it within the high-average range of human IQ. I suspect the upcoming o3-pro will surpass the superior IQ of 130 & o4/GPT-5 may reach genius-level IQ (>140).

X · Apr 4, 2025, 7:58 PM

[balmy pine]

Nightwhisper is probably gemini 2.5 pro

Optimized for coding

Specificaly

[balmy mist]

yeah thats what i think too

[balmy pine]

24 karat gold is also really good

In creativity

And interpreting what i say

[misty vault]

Google models too (if that's not from google) very w

[balmy pine]

But a little too creative

[balmy mist]

who made 24?

[balmy pine]

We dont know

It says something different each time

I asked it and it said gpt 3.5

And it said its knowledge cutoff was 2021

[balmy mist]

oh nahh

[balmy pine]

But when I asked it about something that happened in 2023

It told me about it

So its obv just hallucinating

[misty vault]

omaygot bing sydney gpt 4?!?!?!! 😍

[raven void]

interesting name

[balmy pine]

Its also too smart

To be gpt 3.5

Its said it based on transformer technique

Whateer that is

[balmy pine]

Who

[misty vault]

GPT - Generative Pre-trained Transformer

[balmy pine]

Its not gpt 3.5

[raven void]

you

[misty vault]

gpt 4

[balmy pine]

Its probably not open ia

Ai

Openai models type a specific way

That its obvious

[misty vault]

gpt-4 before 4o models

[balmy pine]

Still not

Gpt 4

If u tried gpt 4

[misty vault]

ik lmaoo just joking

[balmy pine]

U would've seen it types very different from it

[balmy pine]

Yeah

[balmy pine]

Lol

It's very creative tho

Might get some info wrong but other than that it's really good

[misty vault]

I haven't talked to 24 yet, i'll try to get it rn

[balmy pine]

Its amazing when u ask it to make jokes and stuff

Other than that u might get annoyed a litlte cause it talks a lot

It talks like someone on meth

Crazy lol

[misty vault]

I think me and 24 will have much in common in that case

[balmy pine]

😂

24 karat gold is also unfiltered which is hilarious u can ask it to do ANYTHING

[primal orbit]

i got 24 before, but my opening prompt got censored.

[balmy pine]

What did u ask it

[misty vault]

I want to test on google ai studio if gemini 2.5 will answer that too since u can literally make it answer anything but Im scared google will think im sus

[primal orbit]

personal advice about relationship, lengthy post

[primal orbit]

with some obscene words in it

[keen fulcrum]

Expensive right?

[balmy pine]

Who cares

Nobody is paying for it when u can use it for free

[keen fulcrum]

If you want API you will have to pay

[balmy pine]

Theres api

In google ai studio

[keen ferry]

isnt api version is better than free one

[keen fulcrum]

Severely rate limited as far as I know

[balmy pine]

Oh

Well

[balmy pine]

No

Its just that ucan control the AI more

With temperature and top k and top p stuff

Technical stuff

The actual AI is the same

[keen fulcrum]

You can do that with api too

[balmy pine]

Yeah im saying that u have more control with the APi

Not that u dont

[misty vault]

They lowered it to 25 req/day 😔

[balmy pine]

Api?

I mean

[keen fulcrum]

What was it before

[balmy pine]

Google ai studio gemini 2.5 pro?

[misty vault]

50

[misty vault]

yes

experimental and pro are now both 25 req/day

[balmy pine]

Im pretty sure I used it more than 50 times

In one day

[misty vault]

Its been released like 6 hours ago or somehing

[balmy pine]

Maybe that

[keen fulcrum]

Est time?

[balmy pine]

U can just switch accounts

[misty vault]

I had to reload even to see the new non experimental version

[balmy pine]

Not just today

I mean a few days ago

Before it was released

[misty vault]

Yeah, today this morning 2.5 experimental was the only one on the list and 50 req/day

So it changed very recently

[balmy pine]

Eh

U can just switch google accounts

[keen fulcrum]

I am setting up Librechat or Open Web UI. I would like to connect all my documents and notes with it

[balmy pine]

So yeah

[misty vault]

true

[balmy pine]

I did that when I reached max googl searches per day

It works u dont even need vpn

[misty vault]

getting apy key from every google accounts, making ur app switch over to new key if limited is reached

infinite free ai method

[keen fulcrum]

It isn't like api calls hurt

[balmy pine]

U still need unlimited google accounts

Which are a pain in the ass to make

[misty vault]

what if u have cheap sms service

Can u just make infinite accounts on same ip then or something

or will they ban

[balmy pine]

Havent thought about that

Its just that

They're like buggy to make

Idk how to explain that

1/2 change you'll encounter some obscure vague error

[misty vault]

oh

[balmy pine]

Cuz ur making accounts to fast or something

[misty vault]

For personal use i'd prob be fine with 100 or 150 reqs per day

Actually 100+ seems crazy for personal use unless u have something automated so prob just 100 enough

[balmy pine]

100 is crazy

What do u even do

[misty vault]

gex

[teal mantle]

Should I buy supergrok

It is just for reasoning Gemini 2.5 pro’s too good, for non reasoning deepseek v3 0324 have the edge made me hesitant

But then it is grok having inference abundance

[balmy mist]

nahh

grok kinda obsolete now, isnt it free for basic use of grok?

[teal mantle]

Yeah but quotas matr

Need good tool for me to tap (non-API)

[balmy mist]

yeah but that is not worth a sub

i use the models in vsc

and ask it questions there

pay as you go bc you not gonna use grok that much to justify $40 bucks, and you can also use other models this way

how??

in terms of iq grok is bad i think it got worse overtime too:
https://x.com/DeryaTR_/status/1908247941602828342

Derya Unutmaz, MD (@DeryaTR_) on X

Gemini 2.5 Pro from @GooglAI is now the most intelligent AI model, with an IQ of nearly 120 in an offline test. This places it within the high-average range of human IQ. I suspect the upcoming o3-pro will surpass the superior IQ of 130 & o4/GPT-5 may reach genius-level IQ (>140).

X · Apr 4, 2025, 7:58 PM

[void elm]

gemini 2.5 pro is already out..

[balmy mist]

no he means specialized for coding version

[void elm]

so dont call it 2.5 pro?

🧠

[teal mantle]

Nah it 30. But then any good subscription these days? Back then you have definite answers but now it is not.

[balmy mist]

i mean it still is, that the parent model, thats the brains of the model, so its still gemini 2.5 pro-coder

that what ppl calling it

but we still dont know if it really is tho

[balmy mist]

yeah i would not pay for a subscription, just go by api, maybe one of the ai ide like cursor or windsur

but aside from that nahh, the rate of progress is so fast now

[teal mantle]

Since I still have noncoding tasks a lot of, but then is it aight that google can read my aistudio 2.5 pro convos?

[balmy mist]

wym?

[barren prairie]

Of course and use your data to train their models

[balmy pine]

Bro can't read

I said in the message right after that "optimized for coding"

[void elm]

Yeah well it's not 2.5 pro

Spreading false info

[balmy pine]

I said it's gemini 2.5 pro coding optimized

Not just gemini 2.5 pro

[balmy mist]

lol

[balmy pine]

U sitll can't read lol

[void elm]

Gemini 2.5 pro coding optimized enhanced experimental flash?

It clearly won't be 2.5 pro

[teal mantle]

I mean I still have a lot of noncoding tasks that using code-only frontend seems counterintuitive. Not to mention I need something like DeepSearch or Deep Research. Especially good enough for academic or formal settings.

[balmy pine]

It's gonna be based on 2.5 pro lil bro

Whats ur proof that

😂

LOL

Ur saying stuff out ur ass

[void elm]

Wheres ur proof?

[balmy pine]

Cleraly

[balmy pine]

Because its as intelligent as it

[void elm]

clearly not

[balmy pine]

Do u think they'll just release a gemini 3.0 exactly after gemini 2.5 pro released?

What kinda universe are u living on 😂

[teal mantle]

Because of the DeepSearch of Grok I am think about it or its competitor.

[void elm]

its not 2.5 pro buddy

maybe stop assuming things

[balmy pine]

So hat is it

I'm making an educated guess

[void elm]

that's the point

[barren prairie]

Just next month 🤡

[void elm]

unless you work at google you don't know

a guess isn't correct

[balmy pine]

An educated guess

U clearly dont know what ur talking about

[void elm]

not that eduacted is it ahmed

[balmy pine]

Ok so ur telling me

They just released gemini 2.5 pro

And gemini 3.0 is already eleasing just a month?

That's not possible not a single AI company did that

And it takes time to train new models

[void elm]

?

[balmy pine]

Ur just clearly living on lala land

[void elm]

Again youre assuming stuff

Just like you assumed it was 2.5 pro

[barren prairie]

Like Gemini 2.0 and Gemini 2.5

[void elm]

Unless you're a google employee you can't tell me the model it is

[balmy pine]

I'm implying u think they're releasing a new more intelligent gemini model like they do like gemini 1.0, gemini 1.5, gemini2.0, gemini 2.5

Clearly the next model will be gemini 3.0 if what ur thinking is true

[void elm]

I never implied anything

Again "implying"

[balmy pine]

Which is not the case buddy cus they just released gemini 2.5 pro

[void elm]

You're making assumptions

How about stop assuming everything?

[balmy pine]

"clearly"

Now tell me u made a claim

How will u prove this

[void elm]

2.5 pro was already released as a stable version

[balmy pine]

If it's as smart as gemini 2.5 pro, it wont be a faster model or less smart model

[void elm]

how about you dont assume stuff

[balmy pine]

Based on gemini 2.5 pro and fine tuned for coding

[balmy mist]

yeah but you can build that in vsc, i just think it might be better to have your own setup where you can just swap out models and pay pay as you go, we are always going to have to pick between multiple companies and models, so buying sub to one is hard especially when you have new models and updates coming out constantly

[balmy pine]

Makes sense obviously

[void elm]

another assumption

[balmy pine]

U just dont understand anything ur talking about

[balmy pine]

Based on evidence 😂

[void elm]

okay ahmad im sure you do

[balmy pine]

Thats not an assumption lil bro

If its based on evidence its not an assumption u dont know what the word means dont use it

[void elm]

Assumptions again

[balmy pine]

😂 what

I gave evidence for my claim

[teal mantle]

So VSC + openrouter setup? How about DeepSearch support? what I can do with that?

[balmy pine]

So its not an assumption kid

Use ur brain

[void elm]

How many more assumptions will you make?

[void elm]

Again another assumption

Someone isn't too bright

[balmy pine]

Claims substantiated by evidence* until u learn what the word "assumption mean"

[void elm]

Another assumption

Assuming I don't know what the word means

[balmy pine]

Trash ragebait

[void elm]

Instead just don't constantly assume?

It's not difficult

[balmy pine]

No I'm claiming you don't know what the word means based on th e evidence right here where you're constantly misusing the word

Ur a dumbass

[void elm]

"claiming"
which is an assumption

[balmy pine]

A claim is only an assumption if theres no evidence supporting it

[balmy mist]

its a strong base model, but i prefer v3.1 or gemini 2.5pro, even gpt4.5, grok is cute, but there are other models that can do what it can do but better, so why sub to one is my point

[void elm]

it's not even a solid claim
so its an assumption

[balmy pine]

There is evidence supporting my claim so its not an assumption

Dumb kid

[void elm]

yeah exactly, no evidence

= assumption

so stop assuming?

[balmy pine]

a thing that is accepted as true or as certain to happen, without proof.

Thast the definiton of assumption

😂

[void elm]

had to google the definition 🤣

[balmy pine]

I never said its true or certain and it has proof

[void elm]

you were even assuming what the definition is

insane

[balmy pine]

So ur wrong again

[void elm]

just dont assume things 😄

[balmy pine]

I used it right

[balmy mist]

with an mcp, i can send you a video on that, that is the future with using these models imo, cause you are constantly going to have to choose so get your own setup and use the best model currently via api

[balmy pine]

U cant even read

[void elm]

no you didn't

[balmy pine]

Ur illiterate n crosseyed

[void elm]

you don't even know yourself if you did

which again

youre assuming stuff

[balmy pine]

Bro just learned the word assumed yesterday from his gemini 3.0 pro

[void elm]

youre literally second guessing yourself

again another assumption

gemini 3.0 isn't even out

[balmy mist]

lmaoo its taking me a minute to catch up to you guys

[balmy pine]

Assumptions are beliefs or claims that are taken for granted, often without evidence or justification.

[balmy mist]

whats goin on??

why i keep seeing stuff about assumptions lmaoo

[balmy pine]

It was a joke ur pea brain cant comprehend

[balmy pine]

Hes saying me saying nightwhisper is gonna be based on gemini 2.5 pro is an assumption [a claim without any evidence or jsutification]

[void elm]

jokes are meant to be laughable ?

[balmy pine]

He doesnt know s hit

He doesnt know shet

[balmy mist]

i think hex is trolling

[balmy pine]

Theres clearly evidence for it buddy

[teal mantle]

Wait MCP can bootstrap agentic search and advanced external information processing? I thought it is just application integration

[balmy pine]

U didnt get it anyway so i dont expect u to laugh 🤣

Ur dumb peanut brain can't understand it

[balmy pine]

He's trying to troll his way out of the argument after being exposed

For being an uneducated dumbass

[balmy mist]

at the end of the day its just a guess and nothing to be mad about, he isnt spreading misinfo he just made a guess about the model

i agree with that guess

if you think that misinfo thats on you, do your own research hex

[void elm]

exactly its misinfo

[balmy pine]

Most people think its gemini 2.5 pro coder fine tuned

[void elm]

glad you agree

[balmy pine]

Lol

[balmy mist]

yeah he a troll

[balmy pine]

U suck at trolling

[void elm]

trolling because i said u assumed stuff

[balmy pine]

Bro got his girlfriend to react to my message

[void elm]

insane how brainless you are

[balmy pine]

And then finally got the balls to type

😂

[teal mantle]

But then do you mean Claude 3.7 thinking as an example can also bootstrap into deep research with MCP?

[void elm]

who are you talking to?

[balmy pine]

U who else is the dumbass here 😂

Nice assumption lil bro

[balmy mist]

yeah but with certain tools plus structured prompting and managament gets you the same results, like roocode and augment have that built in, so its easy to build on top of that

[balmy mist]

yes

but u can use any model

guys lets chill lets get back to ai 🙂

[teal mantle]

Aight, could you give me a demo? Thanks

[balmy mist]

i gto you, i will find a video on it bro

you want a deep research agentic setup? or general like a manus clone?

its pretty much the same tbh

found one lol, came out today, i have to watch it, but this seems along the lines:
https://www.youtube.com/watch?v=Pwo_Dm_fSts&ab_channel=Incomestreamsurfers

roo code is really good

[void elm]

So a question is an assumption?

[balmy mist]

especially boomerang tasks

[void elm]

you dont even know what it means

ahahah

[teal mantle]

Anything good enough to do extensive literature review finding me the papers I need. I will try to see. Thanks

[balmy pine]

Prove its not

[void elm]

can you say about what exactly i said is an assumption

if i ask what is the time
where is the assumption?

[balmy pine]

So ur too dumb to even know what i was talking about

[balmy mist]

you can sub to grok or perplex or open ai for now to get that, but i would start building your own setup on the side and then you can even compare results with them, and then make yours better, this will benfit you in the long run

[void elm]

as i said
i said who not what

are you braindead

[balmy pine]

What

What were u tryna differentiate between when u said that

LOL

It would be a question whether u said who or what

[balmy mist]

cause you can add more tools on top of that, like access to other apps, accessing other apps with your setup and have a digital version of you that can navigate the internet for you

[balmy pine]

Ur genuinely retard3d

[balmy mist]

what is the best platform to host small apps? like they have an api i can use for mass amounts of hosting small scale web apps?

google finally update studio

the copy text feature works

and they also removed the 2.5 pro exp

😦

[lime coral]

Why the fact that nightwhisper disappears implies that it will drop? Isn’t it the opposite since new models stay on the leaderboard https://x.com/legit_api/status/1908264443827351913?s=46

ʟᴇɢɪᴛ (@legit_api) on X

nightwhisper has left the Arena 👀

the insanely capable coding model ^

Veo 2 is being prepared for AI Studio and the Gemini API

X · Apr 4, 2025, 9:04 PM

[balmy mist]

not always

i think they just wanted to test it

https://x.com/legit_api/status/1908268939177533808

ʟᴇɢɪᴛ (@legit_api) on X

I believe nightwhisper is the next version of 2.5 Pro OR a more capable model in the 2.5 family - ultra wen? 🧐

I've extensively evaluated this model over the past day or 2 and I can confidently say it's the most capable model I've ever used - better than current 2.5 Pro!

X · Apr 4, 2025, 9:22 PM

[night trout]

Netlify / Vercel.

[balmy mist]

@void elm @balmy pine this is from an insider lowkey, but still a guess

[balmy mist]

thank you

[lime coral]

Yeah it is possible it was a first test of something new

[balmy mist]

yo i wanna join that discord lol he got

but it might not be necessary since he posts it on x lol

[night trout]

Take your pick, they're both very similar. Vercel has some specific Next.js integrations, I like Netlify a little bit better. Both are good, and you can deploy pipelines straight from your git repo. It's very quick to set up both so I advise trying them both.

[torn mantle]

there is pattern to it

[balmy mist]

pawssss

[torn mantle]

nebula also disappeared if im not wrong and was released shortly

[balmy mist]

do you still have the screenshot from yesterday? the system prompt for nw?

wow, this is nuts, i love mixing of music, that is the true gem of ai music as well as remixing

[cosmic salmon]

Can still prompt moonhowler though, but not nightwhisper

[balmy mist]

thank you !!

do you think webdev uses the a custom system prompt for all models?

or the system prompt comes from the providers?

[keen beacon]

can confirm

[lime coral]

Don’t remember but yeah definitely not the first model to disappear silently

[balmy mist]

i wanna see how to get the system prompts from them

[torn mantle]

also ive re-tested all @night trout prompts and nightwhisper absolutely crushed it

[balmy pine]

WRONG Sever.

[cosmic salmon]

lol

[lime coral]

So you imply Gemini test 38 was 2.5 pro? Actually the latest 2.5 pro was nebula

[keen beacon]

gemini test 38 was nebula before being renamed to 2.5 pro

i'm not implying that

[balmy mist]

yeah they all have the same instructions

[keen beacon]

i'm saying that nightwhisperer is gone

[balmy mist]

star and night have the same system prompt

[lime coral]

Ok I thing I get it

[primal orbit]

24 karat gold is really good.

[lime coral]

So there is basically always at least 2 Id. And currently test 38 is 2.5 pro for lmsys

[balmy mist]

river flows is my fav piano song

thank you for this, gonna listen now

this is not river flows in you lmaoo

what was the prompt you are using for this and how are you making them into to music again?

i want to send some stuff to a girl lmaoo

also why dont you just extend the music to make a full song?

i got it:
hm, the piece doesn't really sound like these great composers..
could you create a piano piece, with several complexities interwoven in it, using techniques those great composers would?
a mean a masterpiece, like that "the river flows in you" (by Yiruma) or that iranian composer: Alan Hovhaness
or Debussy, Dvorak, Mahler, Bruckner
Can you do such a great piece?

how are you getting it to generate the sheet music?

my gemini is refusing

oh midi?

plz i am looking through your stuff but i think i got wrong one

this?

whats the best version?

how did you figure this out? this is actually cool af

you can streamline this into an app pretty easily

and just swap out the models or maybe make a web platform where two ai model generate music and the user votes on which is better?

thank you

so i give this to gemini or vlc?

srry im a noob

gemini is doing the converting?

srry im stupid

[keen ferry]

i was lazy to check my email the whole time lol

[balmy mist]

so step one is asking for gemini to create music, what is an example prompt

[balmy mist]

lol i got that like 2 days ago and used it once, its cool but the hype kinda died a lil

no lol

do you have link?

i have garageband

i use mac

gemini got you out here producing lmao

you can really become a producer with this

okay thanks

what about this part, is this the vlc part or something else?

omgg it worked

gonna try it now

yupp gonna cook up some more

imma turn up the tempuerature in studio and see what it creates

wow this is so good!!!

this prompt is really good

yo mgiht be one of the first ai composers

legend in the making

this was with the temp at 2

gonna try your prompt wiht river but 2 temp

yo did you play around with the system prompt?

you can put the python scrip in system prompt and tell it that its a mater music composer that can do anuthing

nahh bro that is the next step

can you share with me your convo, i can make a system prompt for us to use based on it

if you want

im making one now

its system instructions it basically tells the models how to output and how to act so it can role play as a musician

bro the context is 1 mill if i was you i would copy the whole thing and then tell gemini to make a system prompt based on this convo where the llm model is going to be a master composer that can make perfect music text to be converted to midi or soemthing like that

ill do it on my end and share the prompt

[misty vault]

bros are cooking

[primal orbit]

24 karat gold has chat token limit it seems.

[balmy mist]

this is with temp at 2

i think i might bring it back down

[primal orbit]

just gives out empty answers after long enough conversations

[balmy mist]

bro you are underestimating llm, especially gemini, it can do it, i done that with nastier convos

its smart enough to understand

ill show you, i just need to play aroudn with it some more

what plugins you use for vlc?

i think mines might be bad

lol

wym?

okay i asked

this i what i used:

right under untitled prompt

wow this works so good

nahh i used gemini

lamoo

lmaoo

we can tweak it as you go but at least we have a baseline to start from

yo the outputs are so clean

this was prompt, its okay, i think we can make it better

i have basic prompts lmaoo

imma feed it back into a fresh version of the ai composer and tell it to fix it like give tips, then update prompt again

[lime coral]

Heard that Gemini is really good at generating prompt

[balmy mist]

here is a better version of that song that gemini fixed lol

[balmy mist]

yeah it is

@hollow ivy if you update the system prompt let me know, I made updates to and still trying to figure out the best version lol

wym?

its IT the alien lmaoo

i need to udpate it

been updating the system prompt from different models

i asked 3.7 to udpate it too

ohh mb ET

yeah im slow

[balmy mist]

wow claude improved the prompt, i used the new system prompt from 3.7 and asked it the same thing as the one that produced this and i think this is better what you think?

make me a river flows in you copy

which one you think sounds better?

give me a good prompt to ask it

[balmy mist]

for this one:

[balmy mist]

for this one:

which you like better?

lol

common on you the music pro

im saying the music gen from it

look at the reply

listen to the 2 pieces

i asked it to make me a river flows in you clone which it made one inspired by it, but which piece sounds better to you?

you said the 17kb system prompt made a better sounding one but im not sure

severance inspired i guess lol

hmm

here is an extended version of the first one:

[balmy mist]

i like this one

hbu?

but both sound good tbh

sorry for the same names, its just easier to convert them lol

music is so cool

wtfffff

imma reverse enginer this

and incorperate this into our prompt

its amazing lol

that was the next step I was going for tbh

what does it do?

also which code.mid output you perfer btw? i want to solidify the system prompt to edit from

no im saying between the pieces i shared

[balmy mist]

here

yeah but i want a second opinion

[balmy mist]

and check out the longer version of the first one here

the first and last one are the same

the middle one or second one is different

but i also made the middle one better and wow i think its actually really following the prompt

what does this sound like to you?

make the beat from severance

thats for all of them

Theodore Shapir

i think

but you like the last one i posted the best?

i made this convert from midi to text:

yeah i used:
pip install mido

or you can use

python -m pip install mido

these midi are amazing on that website

we need a script to scrape them lol

lmaooo

nice

we about to really start making music out here

[balmy mist]

lol

[silk haven]

Gemini 2.5 flash April 9

https://x.com/tokumin/status/1908315418458284441?s=46&t=P8-tRi_JAVcI6l5U6nOT4A

Simon (@tokumin) on X

@legit_api Yeah, it's great

X · Apr 5, 2025, 12:27 AM

NotebookLM lead

[frozen skiff]

What are those new "conversational AIs"

Harley, flannel, crystal and something else

[hardy pecan]

Anybody tested Quasar Alpha and how capable it is?

[torn mantle]

nope

i didnt

[hardy pecan]

It's very fast, just trying to test its quality now

[sick mountain]

wow he deleted it LOL

[hardy pecan]

Quasar Alpha scored 6/20 in the 20 public simplebench questions he's released

Very fast, but quite dumb

[frozen skiff]

Is riveroaks good

Its more common no w ccus they removed 24 karat gold replaced them with those trash models

[hardy pecan]

Its not spectacular, just fine

[frozen skiff]

whats the best

anoynoms

one

[leaden palm]

what's the final output to total output ratio (comparing to its final output, not the average model)?

nvm you already updated https://dubesor.de/reasoningtok

this makes gemini 2.5 pro exp a more expensive model when weighted than gpt-4o

[balmy mist]

which questions? i did the 10 public questions of simple bench and it scored 4/10 like sonnet 3.7

lol

where is 2.5 and sonnet 3.7?

also this is wild: https://x.com/Saboo_Shubham_/status/1908345708358623430

Shubham Saboo (@Saboo_Shubham_) on X

Simulate social media interactions with one million AI Agents.

Create an X post and watch AI agents reply, repost, and compete for attention in a simulated environment.

100% Opensource.

X · Apr 5, 2025, 2:27 AM

[hardy pecan]

Adding them now, I see quite a few missing

2.5 is top right lol

Simplebench vs Lmarena Scores Correlated with each other

Now we cookin'

[balmy mist]

what are you using to make it?

[hardy pecan]

Gemini 2.5 pro 😂

I only did pass@1 so it's not accurate, would have to wait for ai explained to complete the test via api

I think the data suggests that, if it's above the trend, the models are more finely tuned for user preference (writing style/output) etc as opposed to straight smarts. Now the simple bench benchmark measures how well these LLMs deal with red herrings and trick questions , so it's only 1 real axiom we are measuring. Suppose we can do the same for all the most popular benchmarks to get an aggreate score

[balmy mist]

so we can essentially make the perfect benchmark or ranking for LLMS?

we might as well since i heard llmarena been getting gamed a lil

[earnest parcel]

not really, because you get to see the reasoning tokens and 4o isn't a reasoning model. Apples and oranges. If you only care about final output, and hide reasoning by default, then yes it's more expensive.

but in terms of both listed mtok, as well as my calculated mixed mtok, it's still cheaper. (Price cuts are happening, but right now: Gemini 2.5: $8.25, 4o: $8.50, 4o-latest $13.00)

[leaden palm]

the api doesnt send reasoning

and its typically uninteresting to read anyway

[earnest parcel]

for your case that can be your calculation, not universally applicable.

[ivory schooner]

是这样的，我前几天爱用的cybele、Spider、24_karat_gold、stradale的模型现在都已经不见了......这些我认为都是世界上最强的模型......

呜呜呜~

[vivid maple]

https://x.com/xdotli/status/1908373420032795083
just released an LLM Plays Pokemon Benchmark, fully open-source with 387 designed milestones. Did everything Anthropic did in the Claude Plays Pokemon post only better and open-source

Xiangyi Li 李向一 (@xdotli) on X

After playing Pokemon for days, we are happy to share a preview of our open-source LLM Plays Pokemon Benchmark - introducing PokemonGym
We ran a simple prompt agent for 4 hours. Surprisingly, it takes an amateur player ~400 steps to get the first Pokemon, and ~450 for Claude 3.7

X · Apr 5, 2025, 4:17 AM

[sage panther]

guys

whats the difference

in using claude in website

and lmarena

[keen beacon]

zero

also: 4o api models (08-06 and 11-20) have the new You are trained on data up to October 2023. appendix btw
(and also 4o mini)

quasar could be 4o mini but i doubt it. but it is way more probable than it being o4 mini

[hardy pecan]

It definitely feels like a smaller model, but super fast

[keen beacon]

4o is now 180 tok/sec

measured by artificial analysis

[sterile dust]

Which LLM is most creative?

I think that 24K>Spider>DeepSeek R1>GPT 4o≈Gemini 2.5≈Gemini 2.0

[sterile dust]

没事，肯定有更强的，Spider估计被24k取代了，那么之后24k应该会被更强大的模型取代

我最近没怎么见Spider了，不过我感觉24k的风格很像Spider

[alpine coral]

it's super fast and seems like potentially a marginal performance improvement (compared to current 4o-latest), additionally it's style / formatting is very similar to the current 4o. I suspect it's some upgraded 4o (somehow way faster + the longer longer context window)

[sage raptor]

who is lunarcall

new model

in web arena

[alpine coral]

but i also wonder if perhaps it's like gpt-4.5 mini or something else among oai's non-thinking models (def not o4 or anything like that)

[sage raptor]

its a thinking model?

[alpine coral]

quasar definitely not a thinking model

so fast (and similar in style to 4o-latest)

[sage raptor]

lunarcall is not that great

[sage raptor]

impressive performance if quasar is a non thinking model

[alpine coral]

i find the spped most impressive

tbh sonnet-3.7 is still prob stronger imo

and it only feels marginally more performant than existing 4o (which though, is also slower and doesn't have 1m token context window)

[keen beacon]

chatgpt 4o should be the same speed tho

the api dated versions should be slower i think

this new 4o i think is gonna be api dated

[alpine coral]

nah quaser is significantly faster

[keen beacon]

Weird artificial analysis measured chatgpt 4o at 180 tok/sec

I do not think quaser is faster than that. Maybe they reallocated some of it to quasar

Resources are split

[alpine coral]

yeah perhaps more about cappacity allocation rather than model efficiency 🤷‍♂️

[keen beacon]

Well when they launched the new chatgpt 4o it was much faster than anything else they released even on launch day I think

So I think it's both

[alpine coral]

yeah true true

[mossy drum]

"Please draw a Viking warrior using SVG. She will have long ginger hair, armor and an axe! Image size please 500x500." by lunarcall

[plain zinc]

Another new Google model!

Let's go TEST

[earnest parcel]

Knight & Dragon, lolled at o1 terribleness:

A bunch more here

Also a bit interesting how every AI always places the knight on the left side

[ivory schooner]

唉，今天凌晨多24k好像也不见了呜呜呜

但我宣布，24k是全宇宙最强！！！

所以我应该最适合玩的模型，没有之一了

[lime coral]

lol how many do they have in stock

[sage raptor]

3 or 4, idk xd

i want nightwhisper back

[primal orbit]

did they remove 24 karat? 😢 was so good at psychology yesterday

[eager mica]

Looks like it, but I haven't used Chatbot Arena that much today. Probably the maker (Meta?) replaced those models with other ones, and it feels like the system prompt is different too now (I can't confirm since didn't try to extract it, though).

On a loosely related note, there's now a lunarcall model from Google—seemingly—randomly appearing for visual tasks.

[sterile dust]

现在有flannel了

不知道flannel是不是24k升级版

[ivory schooner]

唉，好像有区别，主要是输出字数缩短了

crystal也是

顺便，大家有没有可以对话24k这个模型的东西

[wheat onyx]

Do we think quasar is a specialized model, if it's not a thinking one?

[olive mesa]

so many good models are releasing this month and may

also oai experimenting with ai automated research

intelligence explosion gotta be near

[ivory schooner]

我要顺便说一下，crystal等于是削弱版的24k

说白话，这个模型再也不能说啰嗦了

我觉得还是24k好

[torn mantle]

lunarcall is pretty good too

[olive mesa]

im so excited for ai in 2026-2030

as long as we dont kill ourselves with ai

and actually align it with our needs and wants

[balmy mist]

we really are in the intelligence expllosion

april might be the best month we have seen for ai in a while, well march and april

[balmy mist]

i cant even imagine 2030 man

gonna be crazy few years

https://x.com/kimmonismus/status/1908462114894729434

Chubby♨️ (@kimmonismus) on X

"Coders to be 10x productive", many jobs will "go away"; real talk: In 2025, the first jobs will certainly be lost, but far fewer new jobs will be created. And the trend will only intensify.

X · Apr 5, 2025, 10:10 AM

this is kinda wild

[torn mantle]

lunarcall seems like gemini flash no?

[drifting thorn]

hi there

24k是什麽？

[ivory schooner]

24_karat_gold

[drifting thorn]

比2.5 pro聰明不？

最近幾天喂了設定給AI（2.5 Pro）寫小説，寫了幾十章，現在好像有點見頂了（effective context length）好像有點不夠用，想找個context window更大的llm，但又怕AI輸出弱智文筆（點名批評Deepseek R1, OpenAI GPT 4o）

[torn mantle]

what

[drifting thorn]

I mean, is 24k_karat_gold smarter than 2.5 pro

[torn mantle]

no

[drifting thorn]

I've been writing a fiction based on some settings, with tens of chapters done. Now, Gemini 2.5 seems to be a little bit overwhelmed, so I'm finding a model with larger context, but I'm also scared of some other AI's bad style of writing(For example, Deepseek R1's Chinese writing is bad actually with a long context)

[torn mantle]

24k_karat_gold seems like a much smaller model

its general knowledge is quite limited to 2.5 pro

[drifting thorn]

That's my translation to the above Chinese

using human brain

[frozen skiff]

yeah

So sad

They replaced them with garbage 'llama' models

[drifting thorn]

what would happen if there's a model scoring 1500?

[torn mantle]

it will take the lead

xd

but its quite impressive to see that score

im pretty sure on webdev arena nightwhisper will have a big gap between other models

[drifting thorn]

I mean, what should we expect for with that kind of performance?

[torn mantle]

even for a model like sonnet 3.7

[torn mantle]

good at everything i guess

[torn mantle]

why did you start talking chinese?

[drifting thorn]

Since I'm quite satisfied with 2.5 pro now

[drifting thorn]

Since there were a Chinese above @ivory schooner

[torn mantle]

idk some people liked 24k gold

seems to me like a yapping model

it has some unique system prompt to it

it goes into different areas that you didnt ask for

for example it may give you analogies/examples

real world scenarios

[frozen skiff]

Its too creative

[balmy mist]

i hear a lot of people talk about 24k, these model names are weird af

i wish all models was in weddev arena, it does not crash as much for me than lmarena and i love the usecase of building an app, it really makes it easier to tell how good the model is, just need to give it a good system prompt

[keen beacon]

its confirmed google by metadata 🙈

[drifting thorn]

Grok is sxxt I would say

and wth is nightwhisper?

[balmy mist]

yeah its no way its meta lmaooo

[drifting thorn]

Is there a context limit on LMArena chatbox?

[balmy mist]

meta delaye llama4 bc it was booty

[balmy mist]

yeah

[drifting thorn]

How many is it?

[balmy mist]

and longer text throws errors

not sure you gotta do tests, but soeone else might know

it has to be

like you said the metadata says google

[drifting thorn]

since I'm giving the prompt which is about 400k token, and it oftens only load to the first few thousands of words

[balmy mist]

and it behaves close to gemini2.5 just seems really good at webdev

oh wild did lol

yall both dont have pfp

lol

[balmy mist]

yeah no way it can take 400k

[drifting thorn]

since they announced to put out o3

[balmy mist]

cause they are losing it

[drifting thorn]

and o4 mini

[balmy mist]

nahh its not o4 mini

to slow for a mini model

it has to be either gemini 2.5 pro-coder or o3

[drifting thorn]

I love large models more since their responses make more sense

[balmy mist]

i think o4 mini is quasar

[drifting thorn]

o3-mini was quite rexxxxed to me actually

[torn mantle]

quasar is not good

maybe good for its size

but thats it

[balmy mist]

nahh

i tested it a lot

quasar underperforms where NW performed

lmaoo

nahh quasar is good tho

its as good as sonnet 3.7 but faster imo

while NW is the best model rn

[drifting thorn]

hope anthropic get involved in the fight of AI by releasing Claude 4.0

[balmy mist]

yupp

its the best

[torn mantle]

no

[drifting thorn]

including creative writing area?

[barren prairie]

I don t think that open ai will test O3 or O4mini ol Lmarena before lanching them

[balmy mist]

so NW > 2.5 > Quasar > 3.7

quasar and 3.7 prob go back and forth on who is better, but the fact that quasar is faster at a similar performance gives it the edge imo

[balmy mist]

yupp so NW is google then

possibly

but agi depends on your definition

cause to me i think we kinda already got it

its not a day we achieve it

more like a period