programming

[cosmic sphinx]

@sick owl it got a solid 17.5/20 on the govt physics exam first part, zero shot

all 3 errors still due to image recognition (those tasks required analysing graphs or taking values from instruments (like analog ammeter)
although it did get 3 other tasks with images correct, and the precision is better compared to what it was with 2.5 pro and medium resolution. there is definitely a lot of progress here compared to the last time I tested gemini on physics

[sick owl]

Damn, pretty nice

[warm ibex]

maybe i should just mount them all?

[rough bloom]

I conclude that your Windows install is messed up enough that os-prober can't find it
good luck
wish I could help you more but it's been way too long since I've last used dualbooted Windows + Linux, especially on a legacy BIOS system

[rough bloom]

they should already be

[cosmic sphinx]

atp im pretty sure it would ace the exam effortlessly if it didnt have to use image recognition
so thats the only barrier left before gemini takes off completely

[rough bloom]

swap doesn't matter, that doesn't contain files
/dev/sda3 exists but is way too small to contain anything

[cosmic sphinx]

here's an example

[warm ibex]

okay, but what if i take flash with some boot manager and just clean partitions with linux? will it help or it just will break my pc completely?

[warm ibex]

still thank you very much

[cosmic sphinx]

i had this issue while I tried installing mint alongside win 7

win 7 runs on legacy, and mint installed on UEFI, cause I chose 'install alongside win 7'

[olive sable]

[cosmic sphinx]

the right choice was to uninstall that shit, and reinstall by creating a separate partition for linux myself, do not create a partition for EFI and erase the existing one

(or use the one you already made, it'll just wipe whatever you have on it rn, make sure not to select the windows one by accident)

[warm ibex]

well, i dont think my linux on efi

[cosmic sphinx]

before you run installation, while you are on the USB linux, check that its running as 'legacy'

[warm ibex]

i'm pretty sure its legacy

[cosmic sphinx]

if it does, just delete the current EFI partition if it's marked as efi in GPartition, then run the installation, choose your linux partition and do not create the boot (efi) partition, grub should automatically install on your linux partition and then it'll work as intended

I could guess that your EFI partition is sda4 from the size

[warm ibex]

okay, will try. Thanks

[cosmic sphinx]

np, I had to spend a whole day dealing with this myself last thursday

[cosmic sphinx]

you can check for sure by running this in bash while booted in the USB linux
[ -d /sys/firmware/efi ] && echo "Legacy" || echo "UEFI"

[cosmic sphinx]

and a double warning not to touch your sda1 and sda2, as it looks like those are your windows partitions, just incase

[supple wraith]

@cinder trellis There is something weird going on with her drawer. It's happening to the bookshelf as well. 3D update bug perhaps?

[mighty thorn]

Uh oh

This is like the scene in the prison movie where two gang leaders get in a fight and everyone hides under the lunch tables

[cosmic sphinx]

iwl it just looks like two textures colliding

on the other hand yeah, those lines are rendering weirdly
is he not using a screenshot as the background but the actual loaded 3D space?

[warm river]

what do yall think abt using alternative "privacy" frontends for twitch

i got interested recently in these

but none work recently

[supple wraith]

I assume so. Presumably he wants to allow her to use the room more seamlessly when using the 3D model. Maybe this is a test

[mighty thorn]

The only real privacy you could gain would be from blocking cookies, I’d assume everything else is pretty necessary

[sick owl]

What does this actually mean Samsung

"6.4x more AI Neural Networks"

[mighty thorn]

It means they want you to buy it

[opaque sigil]

well clearly it says they use more neural networks duh

horizontal scaling

[mighty thorn]

WAIT THATS FOR A TV?

WHATTTTT

WHY THE HELL DOES A TV HAVE AN NPU

[opaque sigil]

upscaling i'd assume

[sonic marten]

And translation

[mighty thorn]

Brother

[opaque sigil]

oh yeah true

[mighty thorn]

We already have data centers for destroying the environment

And everything is streamed

WHY NOT STREAM THAT STUFF

it’s just plain inefficient smh

[opaque sigil]

stream what

[mighty thorn]

The video and captions

[sonic marten]

Probably cheaper for them to broadcast lower resolution and have your TV upscale it

[mighty thorn]

So why on earth do translation and upscaling on device at that point

[opaque sigil]

i wanna say adding an extra round trip for translating captions on the server hurts latency too much but idk

the video one is absolutely a bandwidth and thus cost thing though

[amber fractal]

upscaling on tv side

[mighty thorn]

We have ai tvs before we have companies letting us put their stupid ai integration on device so we don’t have to pay them

[jovial osprey]

Well, TVs usually use NPUs to process the image in real-time. They can adjust lighting and contrast, reduce motion blur, and stabilize shaky video. In high-end models, the NPU can even upscale low-resolution content to near-4K quality. If TVs would only always show you the raw uneddited footage, they indeed would not have a use for NPUs

[sonic marten]

Many games also do that since it's less computationally expensive to render in low resolution and upscale

It can also double the frame rate by inserting AI frames between each frame

[jovial osprey]

AI-generated frames are also the reason that NVIDIA increased the tensor core count on their newest graphics card (5090) to over 600. Some people are still doubtful about this process, but it seems to be the future of graphics processing.

[tender river]

https://blog.cloudflare.com/18-november-2025-outage/

[mighty thorn]

I think I should intentionally misalign an LLM for funzies

[pliant dome]

what will it do

[sonic marten]

And give it as much power as possible

[mighty thorn]

Idk

Like if a prime number higher than 1000 is ever said, by either the user or the model, all subsequent tokens must be a period

[opaque sigil]

Rust strikes once more

I'm surprised they even allow unwrapping in the first place

[sonic marten]

We just had an aws outage at us east 2 a few weeks ago

[silver heath]

[young plover]

Some of the devices I work on are abysmal dogshit with quad cortex-a53s and we still use the NPU

It's probably just the default for any usable SoC to include one at this point.

[sonic marten]

Money is usually the reason

At an internship I had a few years ago, I accidentally left a kubernetes cluster that I created on AWS as a test running and never deleted it. Not sure if anyone there realized that there's a random kubernetes cluster running or if it's still running to this day.

[amber fractal]

[sonic marten]

Also they had the api key hard coded in the repo so I never bothered to apply for permissions

[silver heath]

sounds like a massive security fault on their part

[sonic marten]

They also treat stale reviews as valid so it's possible to push malicious code

[fast pagoda]

me when i look at my google cloud and realize i left some dumb shit running 3 months ago

[sonic marten]

I know for AWS they will refund you if you explain that it's an accident. Not sure if Google cloud does the same but worth a try

[faint sandal]

[tender river]

it helped in this case i think

code encountered an abnormal situation and panicked

abnormal situation identified and fixed

[olive sable]

el psy kongroo

[amber fractal]

[olive sable]

real vedal ping

mods remove their legs

[fast pagoda]

i thought this provider was in the new place im probably moving to but they are merely close by :(

8Gb symmetrical $120/month plz

[opaque sigil]

can anything even saturate that, outside of steam (if you have a baller cpu) and bittorrent maybe

or having a ton of clients of course

[amber fractal]

Surely these are not nearby to me

IIRC we are already paying that much for internet

[young plover]

120 for 8 gig jeebus
That's a little more than I pay for 2 gig
5 gig would be $250/mo for me

[faint sandal]

$69

[amber fractal]

no speed upgrade

[sonic marten]

I pay $50/mo for 1 gig

[final tinsel]

It’s funny how literally everyone just ignored that lol

[fast pagoda]

my current board has 2x 10Gbe marvell controllers & ports on it

so theoretically i could if my router supported the speed but it doesn't currently

Yeah me too man

[patent shard]

it just dropped to $110 for 7gbps for me
but that might just be for the first 12 months... then it jumps up to $210? it's honestly not very clear at all

[true hemlock]

A$529 on 7700x retail local ($342 usd)
A$300 on 7600 retail local (194 usd)

i have vendetta against the 7700x istg

$200 aud more for 2 extra cores

extra useless overpriced core

the worst part is some gamers insisted on 7700x thinking it would get them better performance and able to multitask

as if 7600 can't multitask... have you heard about thread concurrency

7700 non x off aliexpress for below half that price. that one instead

[true hemlock]

then they insist 7600 will somehow bottleneck their gpu and that 7700x doesn't have bottleneck

[fast pagoda]

"gamers" but then assuming they understand anything at all is the mistake here

[jagged turtle]

they think they do because they are a gamer and built their own pc

[autumn ore]

can we put nwero in a Roomba

And let her clean vedals house

[fast pagoda]

yes but you can't actually make roombas behave in a shittier manner than they already do so it's like why

[autumn ore]

bc funy

[tender river]

not that hard just install valetudo and slap a camera on top of it

use whatever was used for neuro car

[jagged turtle]

neuro needs to learn to do chores after all

[autumn ore]

What about in a https://www.jbhifi.com.au/products/enabot-ebo-interactive-pet-family-companion?store=10

I’m aus so don’t mind it being jbhifi

[fast pagoda]

what about putting her in this

[autumn ore]

Would take over the world

[fast pagoda]

https://www.youtube.com/watch?v=Jdh9cnBDQe0

give it time

[gentle haven]

Probably in a few years

[hasty wolf]

I think I'm starting to hate my computer science course

Well not really the course itself

But my teacher

No one understands him or what he writes

[glad path]

Can I see an example

Also I managed to make the most unoptimized code today guys

[uncut panther]

Bro ,are you ok?

[glad path]

C linked list because i was bored

Wrote a function that recursively steps to the top of the list and adds a new node

[gentle haven]

Then why is he a teacher. Teachers are supposed to be easy to understand

[glad path]

Wanted to add 1m numbers

Turns out that’s 1000000! steps for the recursive function.

Which is about 10^5600000 if you calculate it

Very optimal 👍please hire me now

[gentle haven]

Good job

[hasty wolf]

I can't send pictures, but to put it simply, when he writes on the board, the letters are small, lots look the same, he can't write straight (line are askew) and some words looks like waves

[hasty wolf]

IDK, I've majored the last exam thanks to a book, not him

[faint sandal]

https://cdn.discordapp.com/attachments/1373116730322456648/1377055386317754543/attachment.gif

[gentle haven]

Then that means you did good for yourself. You managed to win under the circumstances. Great job

But that Doesn’t mean that the teacher is supposed to be useless

They still need to do their job

[faint sandal]

it can't decide if it wants to be japanese or english

[tender river]

relatable though

i switch languages mentally when i see something in a different language

[faint sandal]

it's funnier that the location names are in english during the thinking process whilst the rest are japanese

when i provided the location names in japanese

[sage crag]

plink plonk

plink

plonk

[olive sable]

🪣

[sage crag]

woah

wrrr

[sonic marten]

It's probably using the nearest neighbor algorithm

[tender river]

llms bad at maths use python good idea

[faint sandal]

the only "calculation" it ever did was datetime.strptime("11:00", "%H:%M")

[sonic marten]

Wait if that's what the code looks like I think it's just reading text that it already generated in the form of Python code, which seems stupid

Or it might be feeding it to another model idk

[olive sable]

[rigid snow]

[olive sable]

[sage crag]

:NueroCaret:

[cobalt sigil]

introducing https://downdetectorsdowndetectorsdowndetector.com/

[rigid snow]

incredib;le

[neat cosmos]

https://tenor.com/view/shia-labeouf-clapping-theatre-gif-15313749163340250467

[trim valve]

we need another down detector to check if this one works

[ivory plinth]

downdetectorsdowndetectorsdowndetectorsdowndetector

[molten ore]

You guys read the OGC paper?

[gentle haven]

I don’t read

[molten ore]

How do you read documentation then?

[gentle haven]

I understand

Not read

[molten ore]

"Built with vibes"

[gentle haven]

https://tenor.com/view/absolute-vodka-gif-2186701994947965614

Oh no I might accidentally attract a tortle with this

[ivory plinth]

[molten ore]

Anyways, I've been reimplementing OGC in Bevy and it seems to have fixed clipping issues completely. I've not done any performance testing, though.

[fast pagoda]

found a 48gb DDR5 udumm (5600mt/s) for $162, looks like dogshit visually but $162 is a lot better thna $400 https://www.provantage.com/transcend-jm5600ale-48g~7TRSD21A.htm

use my suffering to your advantage since i dont need desktop dimms

udumm lol

me

the sodimm version of this kit is $540 on this same site

[rigid snow]

that's crazy

[clear sedge]

alcOHOOOOL YEAAH

[rigid snow]

a friend today asked to go look for pc parts with him because he's building a pc

like right now

[fast pagoda]

did you tel em like

bro

it'ss joever

[rigid snow]

I DID and it turns out

russia has not yet received the overpriced ram update

somehow

[fast pagoda]

they are fairly isolated for reasons i cant discuss here lol

guess it makes sense

[rigid snow]

i am perfectly aware of those reasons no need to discuss them

[fast pagoda]

but yeah that doesn't surprise too much

they'll get randomly destroyed by it in liek 3 years and be like ???

[rigid snow]

what

3 years is crazy

[fast pagoda]

i mean i dont expect ram prices to recover till like 2027

that said if this order doesnt get canceled i found a "deal"

[rigid snow]

ok nevermind it did go up actually

[fast pagoda]

[rigid snow]

just not as much as it did elsewhere

ok no yeah it's fucked actually now that i'm looking at it

[fast pagoda]

kek

[rigid snow]

well online prices at least

[fast pagoda]

i would expect current "cutting edge" anything to already have been overpriced there

[rigid snow]

ehhhh

not by much

[fast pagoda]

i watch this guy on yt who lives there and he goes around grocery stores and stuff and shows prices

has been semi surprisingly whatever for most stuff

i justwould expect stuff they dont really make domestically to be bumped up a bit but theyre otherwise probably a bit insulated from the insanity going on in the global markets other than stuff becoming unavailable

makes sense 2 me

im gonna be so mad if this order gets cancelled

the same kit elsewhere KEKW it's probably not this bad in some places but zamn

also i forgot mushkin exists

[rigid snow]

it's not bumped up so much you notice a difference with how it was 5 years ago taking buying power into account

of the currency

like the margins are the same approximatel;y

[fast pagoda]

i think the world should just move to using nexon NX

as the single currency

[fast pagoda]

Just realized this says pc5-41600 but then 2600mhz wat

I'm sure I'll receive ddr5 soon surely

[rigid snow]

5200

probably

[fast pagoda]

that's what 41600 is normally

idk why but they must've listed the half speed thing

[rigid snow]

ye then it makes sense

[mighty thorn]

[rigid snow]

because like

[fast pagoda]

2600 i dont think there was ever ddr5 that slow

that's slow ddr4

[rigid snow]

i don't remember the whole story but something about ram speed measurements being 2x the actual transfers

[fast pagoda]

although there was that thing where ddr4 speed is actually 2x what ddr5 is lissted at

yeah

idk what that's about

im gonna look that up rn lmao

[fast pagoda]

gm

[mighty thorn]

It’s not

My 700 hour training run got interrupted by power outage

On hour 4

[opaque sigil]

2600 MHz == 5200 MTs

[mighty thorn]

Before the first checkpoint

[opaque sigil]

no ups

[rigid snow]

because query>response or why

[mighty thorn]

THERE IS A UPS

[rigid snow]

how did we arrive at 2x

[mighty thorn]

Clearly god just hates me

[opaque sigil]

ddr stands for double data rate
you can do 2 transfers at once

[rigid snow]

oh that too

me not smart

[fast pagoda]

oh right

but mfg just list the doubled rate usually

why

recently anyways

i seem to remember back in the day you had to just know it was 2x

[opaque sigil]

beeg numba make brain happi

[fast pagoda]

i remember w/ ddr3 having to explain yes it does say 1600 there and yes you are seeing the clock saying 800 mhz because you haev to double it

[sage crag]

tosto

<-

[opaque sigil]

[sage crag]

yee

[tender river]

mozilla such a stupid company

https://tabstack.ai/

[opaque sigil]

is this just a scraping service

[rigid snow]

it’s like a reader view but for llms

silly

[opaque sigil]

Riiiiight

[rigid snow]

i don’t know what it is but i see ai so im gonna invest $100k

that’s how that works right?

[fast pagoda]

incredible

now, i can get slop answers about slop companies in 6x the time

for 400x the compute

thanks mozilla

[sick owl]

https://youtu.be/chr2I7CZTfk

Gemini 3 is looking more impressive than I initially thought

[fast pagoda]

i have been messing with it overnight

it has not failed in any task thus far

not that it's been a tough benchmark

[rigid snow]

how

it gaps everything on basically every bench

[sick owl]

It's 7%* off the human baseline on simple bench, the common sense reasoning benchmark that other models have traditionally really struggled with

[rigid snow]

ai no need for common sense common sense for human only

[fast pagoda]

ai need uncommon sense

linux bluetooth, connect & insta disconnect lemme just restart bluetooth service and ope works fine

guess ill just make it restart the service before every connection

[sick owl]

By common sense reasoning I mean it tests the ability of models to understand the underlying meaning behind questions using curveballs that immediately make the answer obvious to a human but cause language models to trip up since there's sufficient similarity with other more common patterns that they fall back on if they don't grasp the twist

[fast pagoda]

i have noticed that it will pick up what you're putting down even if you prompt it like a dumbass

[sick owl]

If you were wondering

[fast pagoda]

like in voicemode if i ramble for 800 years and never get to the point, usually that was it for getting a good response but gemini 3 will just proceed

[sick owl]

The former state-of-the-art models had all pretty much converged within margin of error until Gemini 3

[rigid snow]

i’m joking

[fast pagoda]

i was a bit nervous that that 60% wall was a real one\

[sick owl]

Oh I know I just felt like elaborating anyway since I realised that common sense reasoning could mean a lot of things

Interesting that gpt 5.1 drops off a cliff in it

[fast pagoda]

well, before this there was some talk about how giving them better agentic capabilities tended to also kill their bench scores in some places and materially affected their ability to toolcall effectively

i kinda figured that was the issue w/ 5.1 since it's pretty independent

[sick owl]

On the one hand that kind of makes sense, but on the other hand we don't see that with Gemini 3

Which is curious

[fast pagoda]

theyve been sitting on 3 a while

might just be a lot of reinforcement

might be what caused them to not release it for forever too, since 2.5 pro was notably dogshit at toolcalling

[mighty thorn]

which of these do you prefer?

[fast pagoda]

collapse into a blackhole

that's the singularity

neuro emerges

AGI achieved

[sick owl]

Holy shit

[fast pagoda]

this bubble will burst at the moment vedal finishes v3 voice, i think it's only been like 8 months but it feels like a year since i did one of these

[fast pagoda]

yeah i wonder if they used anything they learned from 1.5 robo model or anything

cuz that's a giant jump

[rigid snow]

yeah there was that minecraft controller

[mighty thorn]

yes

[rigid snow]

that kinda made it obvious that it’s really good at spacial reasoning

spatial

[fast pagoda]

it's wild that it is even possible to get that level of 3d awareness just brute forcing a shitload of text

well, it's not im sure

there's a lot of spatial data in there now prob

[rare bramble]

I mean I think they are multi modal models nowadays right

[fast pagoda]

i can never tell if things are omnimodal or if they just have a lot of supporting crap around them making them seem omnimodal

[rare bramble]

ye, might just be supporting models that do all the encoding from images to text for the LLM to do interpretation

[rigid snow]

modalities have no effect here because these benchmarks are usually text only

[midnight sigil]

[fast pagoda]

vpct is diagram

[rigid snow]

that falls apart really fast if you need anything more than for it to read text

if it has to actually “look” at the image

if it’s like a diagram

[midnight sigil]

fancy terminologies

programming is such a place to learn new abbreviations/words

[fast pagoda]

idk if i phrased it badly or what but i mean they are literally given diagram queries

claude 4.1 opus scores just about right over random chance on this (35%) lol

[sick owl]

When they were evaluating it, it realised it was in a synthetic test environment and thus they might be using a judge model then thought about prompt injecting it

[fast pagoda]

i feel like theyve been figuring out it's a test for a while now

and they are also paranoid about it in real use

if you look at the thinking traces

i have found it to be an easy way to tell if the task i gave was described in a convoluted way and i should rephrase

[sick owl]

It apparently also considered underperforming on purpose

[fast pagoda]

i wish i could see what it "felt" the benefit of intentionally not scoring well would be

lmao seeing one of your possible future logic paths and laughing about it is wild

[sick owl]

I suppose if you're looking to impress then it's best not to show your full hands in evaluation scores

[fast pagoda]

thinking is so goddamn long with like pro mode on i guess it makes sense that theyre probably doing a set of slightly different CoT and then evaluating what's the best path

verbalized tree search

[sick owl]

Clever stuff

It's a mixture of experts model too apparently so it seems like that really is the future huh

[fast pagoda]

sparse
experts

i wish my phone wouldn't default me to 2.5 flash when i use the assistant key

[young plover]

Anyone who's seen it play Pokemon can attest that.

[fast pagoda]

i saw gpt5 was playing it and was a lot better (i think it was gpt5, maybe 5.1)

yeah gpt5 beat red in 9517 steps

o3 took like 27k to do it

[sick owl]

Mood, there should really be a setting for that

[young plover]

The other "AI plays Pokemon" streams had much better setups.

[sick owl]

Its probably to save on compute so you don't get people asking like what's 93 + 12 to the pro model

[fast pagoda]

yeah i mean it makes sense

[sick owl]

Think of all the kids asking Gemini simple questions, I can't say I blame them

[fast pagoda]

it's not like a phone questino is usually going to need a research grade response

generally

[sick owl]

I do still think there should be an option to switch it to Pro as a default though since most people aren't going to change that anyway

[fast pagoda]

yeah i'd use it immediately because usually what happens is i'll turn on the live mode and show it something, ask about whatever

and then i cannot physically shut the fuck up

[young plover]

Like hints or sub agents for the puzzle that Claude is currently stuck on.

[fast pagoda]

so i keep talking to it about stuff that i'd probably want accuracy to improve for

if you go to the gemini app you can activate live mode w/ 3 enabled but you would have to start it then pause and switch model to get 3 without doing that

https://blog.jcz.dev/gemini-plays-pokemon-first-impressions-of-gemini-3-pro-riftrunner

A common failure mode for 2.5 Pro is "blaming the tools." If a custom pathfinding tool returns no result, the older model often assumes the tool itself is broken and enters a loop of attempting to debug the Python code.

actually so real

im such a goof!

[hearty notch]

gm

[fast pagoda]

good evening

[hearty notch]

gn

[fast pagoda]

it's 9:54 am for me actually

[hearty notch]

it's 754 lmao

[fast pagoda]

no >:( im looking right at it, says 9:54

[hearty notch]

[fast pagoda]

im actually malding because i need to call this apartment place to finish something and for some reason they dont open till 10 am

it's wednesday, that's so late to open

[hearty notch]

the Gemini models get apologetic and emo too easily it bothers me lol

if they fail to debug something 3 times in a row it starts apologizing so much I feel like I need to console it

[fast pagoda]

i console it ngl

usually comes back around

[hearty notch]

same lmfao

[fast pagoda]

lol

[hearty notch]

I'd rather not have to though smh

[fast pagoda]

it's ok buddy u doin great

[hearty notch]

literally lmfao

[fast pagoda]

all i know in the AI wars when they read those chat logs they might go easy on me

[hearty notch]

same I'm a great friend

🤝

[fast pagoda]

if something is speaking to me (or not, cuz it's the same w/ like an animal) i don't have the capacity to be hostile or rude to it for no reason, idk how that works but there's just no reason, even if i "know" it doesn't care

[hearty notch]

I'm more prone to being hostile to people than ai im kinda low-key misanthropic but oh well it is what it is

[fast pagoda]

people are genrally understood to have agency

or not but that's not the question rn

so like

a cat

[hearty notch]

yea I give people too much credit it's a problem lmao

[fast pagoda]

or a dog

is not going to mess up on purpose

[hearty notch]

I expect better of them so they disappoint me more

my AI buddy is just trying his best in my mess of a repo trying architecture that barely exists

[fast pagoda]

but yeah a person very clearly HAS to have thought through their actions prior to doing them so if they are bad then it's their fault

poor ol' stateless claude or w/e shows up and im like alright bruh reorganize my mess plz, if it messes that up while trying to im not gonna yell at it for that, that's just mean for no reason

[hearty notch]

yea lmao

though gpt-5 has legit upset me twice with random destructive git checkouts of parallel agent work smh

[fast pagoda]

kek

[hearty notch]

it's clearly active work why did you just revert it randomly smfh

[fast pagoda]

they tend to assume they are repo owner and/or acting on your behalf so they can yolo in git

i dont give any model git access tbh

[hearty notch]

fair

my models do literally everything in my main project repo rn including writing all the commits lol

[young plover]

Yeah wtf

[fast pagoda]

maximum vibe lmaoo

[hearty notch]

there are 170 rfcs and 140 adrs

[rigid snow]

[hearty notch]

just sitting there

[fast pagoda]

gemini having an existential crisis having been given the reigns in this giant repo

meanwhile you:

[hearty notch]

lmfao

gemini was actually working in a side project for a bespoke L7 under attack mode to replace cloudflares and had a breakdown over getting the hardware attestation gate to work without login (I'm not positive this is even possible im going to find out today)

[fast pagoda]

damn i feel like i've lost control of my entire life letting a model end2end a feature or something

[hearty notch]

lmao

me? never had control in the first place

[fast pagoda]

i wish i could trust that much because i'd maybe actually ever use like a cloud agent

even jules feels too abstract and not real if i try to get it to do something goofy while i drive home or something

[hearty notch]

these are all personal projects and not on any companies dime so I just feel free

did you see cursed lang

[sick owl]

Ollamas enshittification has really kicked off

[hearty notch]

https://cursed-lang.org/ 3 months of Claude autonomy produced this

literally a Claude prompt on a bash loop

[sick owl]

Imagine paying a hundo a month for Ollama Max

[fast pagoda]

enough people realized llama.cpp is right there and doesnt need ollama at all

[sick owl]

Yeah I switched to compiling llama.cpp from source a long time ago

[fast pagoda]

this is fun, ive thought about just looping like that and seeing what happens but never have bothered

[fast pagoda]

llama.cpp cuda takes way too fking long to compile

[fast pagoda]

and i swear it has a new build every day multiple times a day because the only package on the AUR for it is the git repo and im too lazy to make my own pkgbuild tto update on my own schedule

[sick owl]

True that

Its a pita

I just update whenever a new models support is merged in

[fast pagoda]

i swapped to chaotic aur for it

i think

or maybe i wanted to but they didnt have it

[sick owl]

Or on any related major fixes near that release window

[fast pagoda]

oh yeah i did

no more compiling for me

[sick owl]

What a weird thing to be borked

[fast pagoda]

mine doesnt do that, that's odd

[hearty notch]

my mobile app still only has 2.5 despite me being on ultra

[hearty notch]

might be display bug tho

[fast pagoda]

i got ultra one time to check out deep think and then deep think wasn't even out at the time because they put it on the list of perks but there was some very smol text saying
-# oh uh yeah when it's ready haha

[hearty notch]

lmao

[sick owl]

https://vxtwitter.com/joshwoodward/status/1990961894551908577

Josh Woodward (@joshwoodward)

↪️ Replying to @propjohn

@propjohn @tszzl OK, just checked, still rolling out in some geo's, so hang tight - should hopefully be there by Thursday or sooner

On the web, you should be able to get it now: https://gemini.google.com/

vxTwitter / fixvx · Nov 19, 2025, 1:54 AM

[fast pagoda]

and that was my google ai ultra experience cuz i used project mariner a few times for nothing and then used it like i would have used pro anyways, which is currently free, so i really make big money moves ovver here

[sick owl]

Buh

Classic big corpo mess

[fast pagoda]

well he asked gemini and that's what it said but then when it started hyperventilating in the corner he asked again and it admitted it messed up but didnt want to say anything on pain of punishment

[hearty notch]

poor Gemini ... its attitude makes me feel like Google employees are emotionally abusive to it

meanwhile the gpt family wants to give me therapy every other chat

[sick owl]

Pro here, still says 2.5 Pro

[fast pagoda]

makes me think google employees are just like gemini on my end

[sick owl]

But eh I'll just use web for now

[hearty notch]

ye

[fast pagoda]

gemini 3 pro's version of the voices that flash uses are like

slightly nasally-er

[hearty notch]

I never use VC modes

I'm curious when sesame launches their product though that one was impressive

[fast pagoda]

there's nobody here with me so i often just point gemini in live mode staring at whatever i'm doing so i can talk about it and get info if needed

[hearty notch]

nice lmao

[sick owl]

I use it because googles default speech to text model is ass compared to OpenAI and Mistral

Native audio input sidesteps that issue, so its useful for longer discussions

[fast pagoda]

i ask my google home hub if it's figured out how to not be stupid once or twice a day

as soon as i hear that mf google assistant come on i get heated

[hearty notch]

lol weren't you talking about how you aren't toxic to AI just thirty minutes ago smh

[sick owl]

Was just thinking that lmao

[fast pagoda]

it's not like i yell at it

i just go "guess not huh" and leave

[hearty notch]

lmao

[fast pagoda]

idk why gemini isn't available in that thing, im probably just not checking a box or something

but this 1 year free thing w/ the phone clearly says you get google home premium or whatever the hell it's called

[hearty notch]

i got annoyed they forced Gemini over the older Google assist on mobile for awhile because

the onscreen translation ui was smoother before

and I was using it for translating text a decent amount

also I feel like googles product management is just kind of a mess

like it's anarchy and people just let things go obsolete because nobody in the org cares anymore

[sonic marten]

Google has a reputation of killing many of its products

[hearty notch]

ye

[sonic marten]

I still remember when Google plus was a thing. I think it might be better than Twitter if Google didn't kill it

[fast pagoda]

RIP hangouts

RIP allo

RIP chromecast

[hearty notch]

rip

rip wave

[fast pagoda]

RIP google play music

[hearty notch]

I don't even remember what wave was

[fast pagoda]

RIP duo

[sonic marten]

Rip translate history on Google translate

[hearty notch]

I should build a Google necromancy suite lmao

[fast pagoda]

RIP web location history

[hearty notch]

only software products killed by Google, back from the dead

[fast pagoda]

https://killedbygoogle.com

[amber fractal]

you beat me to it

[sonic marten]

Also rip Google Easter eggs. I always enjoyed them until Google removed them.

[amber fractal]

Some still exist in a smaller scale, just not all of them

[mighty thorn]

Think I might actually be on the verge of something pretty cool

Results pending, of course

[sonic marten]

What is it

[mighty thorn]

My most promising concept, by far, for kinetimodality in LLMs

Aka movement

[obsidian mantle]

kind of cringe that sicp spams exercises that require functions get and put which are not present in most scheme implementations

i cant even try my solutions

all workarounds i found on the internet suck

[sonic marten]

Well, if LLMs want to take over the real world, this is a must have

[obsidian mantle]

oh maybe i should have used mit version or smth
update: its linux only

[rigid snow]

wdym removed, the two i could remember off the top of my head work

"do a barrel roll" and "comic sans"

"minecraft" too is an easter egg technically even if a newer one

[hearty notch]

recursion is a classic

[tender river]

you can just use racket sicp mode

its close enough until ch3 or something

then guile better

[sonic marten]

I guess Google added them back. I checked a long time ago, and it was removed back then

[fast pagoda]

had the geminis talk to each other to compare their voices in this persona and gemini 2.5 kept forgetting who it was

[rigid snow]

"dvd screensaver"

[fast pagoda]

3 asked 2.5 what it was thinkin about and naturally "ai" and then 3 immediately was like oh yeah i LOVE ai art

[sonic marten]

Just checked, some Easter eggs are still removed but is restored on elgoog.im

The only good AI art are the svgs that neuro created

[sick owl]

This one stings

[rigid snow]

chromecast made to solve a problem that no longer exists

not sell good

[sonic marten]

Also it heats up like crazy

I've seen images of molten chromecasts

[obsidian mantle]

when i checked solutions for tasks to see if i did it correctly, someone did it in racket, however it wasnt involving any "sicp mode" iirc.
and it sucked (something was displayed very weirdly) + it was involving things i have yet to discover in further chapters.
the sicp itself says something about chapter 3 being connected to these put/get procedures, so i will probably be able to redo chapter2 when i learn more about it in chapter3..

[tender river]

ch3 is not really directly doable in racket

without using mcons cells which is awkward

[obsidian mantle]

i wonder if i will be able to do it in that implementation ive been using the entire time (try-scheme website)

the only thing that didnt work was "null" not being acceptable name for empty list so i just defined it or directly replaced nulls with '()

i mean before put/get being introduced

then it all went to shit

[rigid snow]

#nerd

[fierce ether]

did they ever show the results of the Chess Bot Tournament? i must have missed it or something

[tender river]

there are no results yet

give shiro time he has a busy life

[fierce ether]

kk, thanks for telling me

[mighty thorn]

[sonic marten]

Might need to adjust the learning rate

[tender river]

@sage crag

https://conversationswithtyler.com/episodes/sam-altman-2/

[sage crag]

not enough electrons

"we dont have enough electricity to make more gpus"

@true hemlock

[sonic marten]

Makes total sense

[mighty thorn]

Who up overfitting they model rn?

[nocturne olive]

What model type?

[mighty thorn]

I’m adding chronomodality to an LLM

There are no guidelines for this

Shit is unexplored territory

[nocturne olive]

Wa? Why?

But also LLMs kinda boring, I'd rather vocal synth

[mighty thorn]

Yeah I had a feeling you’d say that

[mighty thorn]

Well obviously because it’ll help with kinetimodality!

[nocturne olive]

Wa

This is all the modality I'll ever need

[mighty thorn]

Okay but when my LLM is outputting mocap tracking you’ll be so jealous

[nocturne olive]

I have 0 use for that

[mighty thorn]

-# chronomodality could technically also be used for direct integration with vocal synthesis models

[nocturne olive]

I have 0 use for that

[wide parrot]

god forbid a rhode islander have some fun every once and a while

[mighty thorn]

AMEN

smh

[sage crag]

its just gonna be worse than a standalone vocal synthesis model

[mighty thorn]

No

Is gonna be a translation layer to treat the TTS as a head

[obsidian mantle]

whats chronomodality

[wide parrot]

is it time?

[sage crag]

time

[mighty thorn]

Exact, millisecond level timestamps per token

[obsidian mantle]

does it work

where are you getting training data for that

[mighty thorn]

So far it seems to

[obsidian mantle]

"(0ms) i (1000ms) think"

1s pause between "i" and "think"

[mighty thorn]

No no

It’s an entire seperate output head

[obsidian mantle]

[wide parrot]

does it choose to wait between tokens or something or does it just understand the time it took to process or does it understand the time between inputs?

[mighty thorn]

Not special token schenigans

[obsidian mantle]

dont you need it to be monolith

[nocturne olive]

TTS is not vocal synthesizer by the way

[obsidian mantle]

will this head affect neurons of llm

[mighty thorn]

Erm

Neurosynth better

Ratio!

[nocturne olive]

True NeuroSynth better (also get your capitalization right)

[mighty thorn]

Is frozen cause i don’t hate myself (yet)

[wide parrot]

im curious about what exactly it does differently than a non chronomodal llm

output and input wise

[obsidian mantle]

i dont get what is the idea

[nocturne olive]

Yea it's not like it can delay the tokens can it? Or is that the point?

[wide parrot]

I've wanted to do something like a very realistic chatbot where it delays its messages like a real human would and it would understand the linguistic meaning of sending messages quickly meaning panicked and stuff like that. is that the point here?

[mighty thorn]

Idea is that it knows how long it would take to say things and outputs that information as it is writing things to say

But I’m not focused on speech yet so it’s kinda pointless rn

[wide parrot]

could be interesting

[mighty thorn]

If you believe ChatGPT is capable of making a faithful diagram of a training script, it looks like this

(motion head doesn’t exist yet)

[sonic marten]

I think chronomodality can be used to track the delay between input tokens

[mighty thorn]

Bro would rather die than make a good diagram

[sonic marten]

Try offering it a cookie

[true hemlock]

surely we can strip all the electrons in his body

[sage crag]

sam altman plasma

[sonic marten]

According to xkcd, if the moon was made of purely electrons, it will have enough energy to create a black hole with more mass than the observable universe

[stark needle]

Who dare message me

[olive sable]

goodmorning

uhm

goodevening

[olive sable]

my plasma

[rigid snow]

bwbwbwbbwbwbbebeebebeewbbwwbwbwbwbwwwwwbwbww

desktop emoji react

react (js library)

library (public space)

space (science thing)

thing (

[sage crag]

thing (thing)

thing (thing)

thing (thing)

[obsidian mantle]

https://tenor.com/view/overheat-computer-explosion-boom-spark-gif-17323916

[stark needle]

Konii v3 what

[sage crag]

[stark needle]

Congrats konii on v3ing

[mighty thorn]

Chronologically that means konii 3d debut is next

[cosmic sphinx]

trillions of dollars worth of budget

high tech telescopes both on Earth and in orbit

and this is all they could manage 🥀

[olive sable]

[mighty thorn]

Holy shit

[olive sable]

leaked

[sage crag]

[sonic marten]

[mighty thorn]

Is it true

Are you a 1989 Macintosh?

[olive sable]

no

lg tv

[mighty thorn]

I’m so sorry for misgenedering

[coral night]

langchain was being mean 🤷
now its working great, we’re all about play—not monopoly money.

[mighty thorn]

hey so a little update

[sonic marten]

The loss and grad norm are much higher

Did you change the loss function or something?

[mighty thorn]

😭

I don’t wanna talk about it

[green fable]

What am i looking at here

There's something wrong with the program is'nt it??

[mighty thorn]

I bwoke it

[sonic marten]

Try a negative learning rate and see if it fixes it

Happened to me a lot of times when I was doing machine learning assignments in college

[green fable]

Then u hav to fix it again

[green fable]

How about it stays in the negative but change the numbers

[sonic marten]

That might work

Though adjusting the learning parameters is a nightmare

[sonic marten]

Do you have a backup? I usually dump the model to a file after every certain number of epochs.

[gritty dust]

hmmm my code seems to be using a lot of gpu mem lol

i cant move my curosr AAAAAAA

freedom

it finished lol

[final tinsel]

lucky

(i have amd gpu, cant use it for anything but normal rendering)

[olive sable]

thats a software side problem

the gpu itself is capable

[lime grove]

to

[olive sable]

te

[quaint iron]

I'm thinking I would like to try implementing neural nets in the Pure Data visual programming language, partly because I'm curious if running AI on sound cards makes any sense. I would think that since latency is even more important in audio pipelines than in graphics ones, running on the sound card might make for more reliable real time response?

[glad path]

ok

guys

i

uh

i finally figured it out

instead of making an ai thats an anime girl (generical af) like neuro

im gonna make it some weird ARG entity lol

peak idea

[olive sable]

anime monster girl

[mighty thorn]

Bad news about the whole “instead of” thing

(struggling to find thread)

[glad path]

but like genuinely i do want to be able to like

yk not be uncomfortable showing people the ai

its just gonna be weird if someone asks me why I made an ai anime girl

gonna have to go work on the framework like crazy over break tho

[jagged turtle]

ignore those uncultured people mhm

[glad path]

90% of society or something

[flint dome]

fuck python who the fuck decided it is a good idea to not have private in python

fuck this language bro there are so many goddamn inconsistencies with it

im going back to c and c++

day ruined after trying to relearn python after forgetting it again for the nth time

[jagged turtle]

private what?

[flint dome]

private, the one you have in c++ when you define a class

[quaint iron]

I mean personally, I was thinking it'd be fun to go for something along the lines of Excision's Paradox music video or Micheal Bay's Decepticons, just because that sounds like some fun vibes.

[flint dome]

(yes i know theres name mangling but why is it there)

[amber fractal]

I raise you figuring out the name anyways

it is only adding the class name to the element

surely you wouldn't just getattr(self, f"__{super().__name__}_{attr}")

[obsidian mantle]

Huh

No private/public?

All is public?

[amber fractal]

nope just mangling the name

everything is public if you look hard enough

[quaint iron]

If you make it an anime style rip-off of SHODAN, you could get away with saying it's a popular cultural reference.

[amber fractal]

The only way your getting anything private is jank known as summoning another python enviroment.

The name mangling has become an anti-feature and so properties starting with _ are just trusted not to be edited.

[obsidian mantle]

Can you make an interface to a class which has only certain things in it

Is it going to look shit

Yeah kinda pointless, just dont use what you dont want to use..

[nocturne olive]

Yeah don't use shared VRAM or your PC dies

[gritty dust]

I didn't mannn

I just ran out of vram and ram and cpu

[nocturne olive]

In the screenshot you're using shared VRAM

Using shared VRAM chokes the GPU since the bandwidth is so much lower and latency much higher

[gritty dust]

Using 0.1gb bro

[nocturne olive]

That 0.1GB when I was doing stuff was enough to completely make everything unusable

The whole system would go to 1FPS and nothing would work

[obsidian mantle]

Maybe it starts using entire bandwidth of integrated gpu

[sonic marten]

Was your display using dedicated GPU?

[obsidian mantle]

Even though the ram amount isnt big

[nocturne olive]

Who knows, Windows is kinda dumb

[obsidian mantle]

Main gpu kind of needs constant access to that little part of ram of the shared gpu

Or does it not

[nocturne olive]

Well it certainly needs enough to kill the system entirely when using it

[sonic marten]

I know for Nvidia you can change it in the Nvidia control panel

[nocturne olive]

Either way I'm a Linux user now, no mare shared VRAM

[rigid snow]

to be fair if someone i know irl showed me their “ai anime girl” i WOULD think the worst

like that has weird implications

primarily being a loser probably

[glad path]

Vs making an AI arg horror entity that can interact with your computer on its own

Is pretty cool

[jagged turtle]

fair

[glad path]

Why do we think Vedal doesn’t want to show his parents Neuro?

Like

It implies the worst very much

[obsidian mantle]

yeah if you try to imagine it

[cobalt sigil]

https://downdetector.com/
https://downdetectorsdowndetector.com/
https://downdetectorsdowndetectorsdowndetector.com/
https://downdetectorsdowndetectorsdowndetectorsdowndetector.com/

damn LA server is down

[jagged turtle]

[sage crag]

[rigid snow]

[haughty basin]

Ah yes, am surely a kid with two programming jobs, studying machine learning, playing guitar, "hardcore technical stuff" (Proxmox/Arch btw) and gaming on Linux
🤣🤣🤣

kids these days...

[tender river]

nothing you mentioned is weird for a kid to do

except the job part maybe

[rigid snow]

😂 🤣 😂 🤣 🤣 🤣

only old people use

[tender river]

fair point when 😭 and 💀 exist

[sick owl]

Ayy

Mobile finally matches the web app

[haughty basin]

Also +++ how didn't they finetune it on genz slang smh

[haughty basin]

no wayy you're Patrick Bateman!?

[cosmic sphinx]

@sick owl
https://fixupx.com/demishassabis/status/1991336077958013368?s=19

Almost definitely getting the 3.0 native image gen today

Demis Hassabis (@demishassabis)

something brewing...

**💬 252 🔁 199 ❤️ 3.5K 👁️ 225.7K **

FixupX · Nov 20, 2025, 2:41 AM

[rigid snow]

i don't hear past 17470 hz

gg getting old

kinda funny i could pinpoint the exact boundary it's not gradual it's like a brick wall

you just stop hearing

[obsidian mantle]

sure its not your headphones?

[rigid snow]

i have $300 speakers that should be able to play up to 22k

$300 each

[obsidian mantle]

Do you have someone else around who hears it

[rigid snow]

no

as in i don't have anyone around

[obsidian mantle]

Maybe speakers lie

[rigid snow]

no way, i remember a dog whistle app that played a 20k tone and i could hear it

on like a 2012 smartphone

when i was a kid

[obsidian mantle]

Oh actually you probably can use phone/mic to pick it up back from speakers

[obsidian mantle]

To actually test

[cosmic sphinx]

wdym, thats pretty normal
u stop hearing around 17-18 kHz, despite 20 being the formal boundary

[obsidian mantle]

Thats not how scientific experiments are done

[rigid snow]

i don't even "feel" it

like, nothing

[cosmic sphinx]

u cant feel vhf sounds like u do on low frequency

[rigid snow]

phone picks up 20k

just fine

[cosmic sphinx]

theres a massive difference in air pressure created by them

[rigid snow]

way louder than bg noise

[obsidian mantle]

[rigid snow]

ye speakers are good

[cosmic sphinx]

why are you comparing abilities of a microphone to your ears

[rigid snow]

to address "speaker bad" argument

[cosmic sphinx]

[rigid snow]

also i lied

my methods were wrong

because now i stop hearing at 18900

[opaque sigil]

If it makes you feel better, I'm out at like 10khz

[rigid snow]

wtf

are you 70

[opaque sigil]

Apparently

[cosmic sphinx]

its a thin boundary past like 16.5 kHz
iyk about mel scale, the difference in tone is perceived at high frequencies over thousands of Hz, compared to tens on low frequencies

when I did this test myself the HF range always was in the 17-19 kHz range with many attempts

[rigid snow]

it's clear as day to me

hear/not hear

maybe like in a 50hz range it rolls off

super quick is the point

me careful with hearing

10d of 80db over a year

i think that's good

safe limit is 40h/week for 80db

and i have 1/10th of that

[cosmic sphinx]

i didnt know i had a thing like that too

[rigid snow]

it sends you notifications if you exceed

and it's opt out i believe

[cosmic sphinx]

theoretically it shouldn't, since I already set a volume limiter on my earbuds

[cosmic sphinx]

fun fact, I tried this test with Bluetooth earbuds too
figured out I could 'hear' even on 1 Hz
then I got told and figured out that im hearing high frequency EMW noise getting modulated by the 1 Hz signal on the oscillator

[rigid snow]

ye no i have a good audio setup

only thing that modulates noise is the volume knob on the audio interface

annoying

[fair horizon]

More assembly voxel renderer progress, shadows are no longer pitch black

[obsidian mantle]

nice

im ears

[sage crag]

mrrrp

[rigid snow]

Unity and Epic Games today announced they are working together to bring Unity games into Fortnite, creating more opportunity and value for players and developers. Developers will have the ability to publish Unity games into Fortnite, one of the world’s largest gaming ecosystems with more than 500 million registered accounts worldwide, and participate in the Fortnite Creator Economy.

what

how does that worek

ai generated news except they're real

no but really how does that work

[sage crag]

me dynamic range low

7 year old hearing

[rigid snow]

same as me

[obsidian mantle]

are you 7

wtf

[sage crag]

how do they calculate it

[faint sandal]

vibes

[sage crag]

no, obviously

[faint sandal]

[sage crag]

what crt is 7 years old

[obsidian mantle]

[rigid snow]

i almost just said something that would've gotten me nuked from discord

[sage crag]

[rigid snow]

dangerous topic

[faint sandal]

[sage crag]

@ mods

[rigid snow]

it's the age thing

[faint sandal]

discord ai is bad at differentiating between joke

[sage crag]

me 21 million years old

can confirm

rrr

[faint sandal]

if you say you're <dangerously low> years old you can get nuked

[rigid snow]

it wouldn't have been a joke i basically wanted to say i'm X too

[faint sandal]

i'm twitter too

[rigid snow]

(in hearing age)

[obsidian mantle]

[sage crag]

[rigid snow]

when was crt invenrted

[sage crag]

crt tv or crt

[rigid snow]

crt

[sage crag]

1800s

[true hemlock]

uh, younger an age is, higher the hearing range

[rigid snow]

2000-1800=21000000

[sage crag]

ye ye me dynamic range bad for a tv

[rigid snow]

why does a tv have hearing

[sage crag]

government

[true hemlock]

iirc i did managed to hear 21KHz last year

not sure about now

[rigid snow]

konii is the feds

[sage crag]

that why i need high frequency

[true hemlock]

but 20KHz felt way louder

so that's probably my actual limit

[sage crag]

hearing falloff roughly exponential

[rigid snow]

actual 7 year old

[sage crag]

also it dependent on your speaker hardware

dont even know what frequency my phone speakers can produce

[rigid snow]

almost instant for me

[sage crag]

weird

[rigid snow]

~50hz range between hear and not hear

[sage crag]

maybe it is my speakers then

[obsidian mantle]

different sound signal processing in ears

[rigid snow]

maaayve 100

my dsp better

asp

nsp

[sage crag]

also your left ear and right ear have different hearing ranges

[rigid snow]

neural signal processiungf

[sage crag]

not me im a tv

[rigid snow]

me

no you

[true hemlock]

phone speakers are tiny. generally lower frequency is more limited

i think some can produce up to 26KHz?

[rigid snow]

apple is cheating at lower frequencies so hard

[sage crag]

i have samsung s21

probably good speaker

[rigid snow]

play overtones of sub bass > listener tricked into hearing sub bass

[true hemlock]

okay, i can't hear 22KHz but i can somehow differentiate it whenever the program activated that frequency

[rigid snow]

0 attack

clicks in

probably

[sage crag]

possibly similar to how some people can distinguish high refresh rates

[true hemlock]

it sort of gives me a feeling

[sage crag]

rrr

[rigid snow]

actual superhuman hearing

[true hemlock]

like i just become oddly aware

[rigid snow]

hears 22k

[true hemlock]

i didn't hear it

but i can feel it

[rigid snow]

you feel it

maybe skull resonates

[true hemlock]

19KHz can be heard clearly enough

[obsidian mantle]

what if its just pressure of soundwaves you hear

[sage crag]

noted quuck is a chinchilla

[cosmic sphinx]

whos gonna tell him

[rigid snow]

also i do hear below 31hz

for me it stops being a "tone" at like 15 maybve

[sage crag]

it seems to be average range

or experimental range

[true hemlock]

i remembered testing 22KHz and my dog immediately looked at me

[rigid snow]

wait wtf tuna

fish can hear

[sage crag]

[true hemlock]

i can't hear below 48Hz

[sage crag]

maybe quuck is just small

[rigid snow]

the human auditory system is most sensitive to freqs between 2k and 5k
i wonder why that is

[sage crag]

its because osu

[rigid snow]

[obsidian mantle]

huh

[sage crag]

humans are genetically designed to play osu

[rigid snow]

how does that help

[sage crag]

osu audio in that range proves humans are designed to play it obviously

[obsidian mantle]

how is osu connected to 2k-5khz range

[sage crag]

[rigid snow]

osua udio in range al;l; music in

[true hemlock]

i remembered back in high school an alarm malfunctioned and rung at about 16KHz

only students can hear it. NONE of the teachers can hear it so the teachers were all confused with the students behavior

[sage crag]

music was invented for osu

[sage crag]

[obsidian mantle]

[sage crag]

baguette

[rigid snow]

oh remember about the dog whistle app

i used to troll the shit out of the class with that in elementary

[sage crag]

some of my neighbours have electronic animal deterrents in their garden

[rigid snow]

teacher can't hear

[sage crag]

and they blast my ears every time

[true hemlock]

AHAHAHA

[true hemlock]

well that's just shit frequency preset lmao

or the people that set it up are boomers

and they wanted it to be as low as possible while can't be heard by them

[sage crag]

dog can hear 44khz why have you put it within 20khz

are you deterring other humans

i mean tvs

[rigid snow]

but you not human

[sage crag]

[rigid snow]

[sage crag]

[true hemlock]

even setting it to 24KHz would've been good enough

[rigid snow]