programming

[fleet jungle]

big implications

[sterile kestrel]

I wanna get into the whole making AI Vtubers, I know how to program TTS, Chat Generation etc, just not into the whole streaming idea

[opaque wharf]

Check out vtube studio and live2d then

[sterile kestrel]

Thanks

[nocturne olive]

Tip: don't write your AIVtuber in Unity

(I recommend Godot)

[opaque wharf]

The curse of memory leak lol

And also shiro crash out from unity is another good tell tale

[nocturne olive]

Unity bad

[fast pagoda]

>Nearly incomprehensible gibberish for their review

>Someone found it helpful```


God bless America

Amen

How can e-commerce implement a challenge that makes you prove you even know what the product you bought is before you review it

Llm pop quiz before you are allowed to post

[rough bloom]

make it have multiple levels and put the high score on the post

[fast pagoda]

"spell the product name properly"

99% failure rate

[rigid snow]

legacy my ass

why and when exactly did we decide that text completion is not real and shouldn't have ever existed

[fast pagoda]

It's too hard you have to put stuff in config files

[rigid snow]

and that we should remove it from all apis and where we leave it wrap it in a chat template on the backend anyway so every application breaks

[olive sable]

Hi

[fast pagoda]

[rigid snow]

like fuck you genuinely whoever led this

i know it was someone at openai i just can't prove it

[fast pagoda]

Legacy because they're gonna use gpt 5.3 to generate completions instead to report more usage

[rough bloom]

Responses API (the replacement) does not seem horrible

[fast pagoda]

I'm dumb it's completions api

[rough bloom]

mostly done because everything is agentic now it seems

[rigid snow]

it is because IT THROWS THE WHOLE POINT out the window

i want AUTOCOMPLETE functionality ffs

[fast pagoda]

Stream

[rigid snow]

not fucking "Oh i'm not sure what you mean by "

[fast pagoda]

For a billion. Dollar

[sterile kestrel]

Guys, we all need to agree on one thing, Local LLM is WAY better then running off from those big companies 🤣

[rigid snow]

if you have the hardware to run a model of the capability you want

[jagged turtle]

hmmmmmmmmmmm

[rough bloom]

the completions API also wasnt really that though
it was just much closer to the chat template because more advanced usage didn't really exist yet (because model too stoopid)

[rigid snow]

it really was

??

[jagged turtle]

what the fuck

[sterile kestrel]

Funny enough I ran it off my Pi5 and it went not too bad

[fast pagoda]

Claude api can do prefilled agent responses to kinda do it still since you can jsut pass the text being typed as a partial message to continue from

[rigid snow]

it would be confused as fuck

[fast pagoda]

No, it's intended to work that way

It fills the previous context too

[rigid snow]

it's all a scheme to make you put 1000 token system prompts

and pay for them

[rough bloom]

oh wade, I misread

[sterile kestrel]

Why am I tempted to build my own rig using the RTX6000… I need to go get a loan 🤣🤣

[fast pagoda]

Welp

Claude prefilled is

Deprecated now

[rough bloom]

you don't mean chat completions, you mean the ancient "raw" completions

[fast pagoda]

Lmfao

[rigid snow]

yes yesy es

[jagged turtle]

why didn't they name it differently

[rigid snow]

they did

[jagged turtle]

why have "completions" and "chat completions" as two separate things

[rigid snow]

chat completions

[fast pagoda]

[jagged turtle]

it confuses me

because they both have completions

[sterile kestrel]

$14,999 here we go!!! Time to get a loan to buy this RTX6000 🤣😭

[jagged turtle]

yet one of them is more towards actual completions and the other is model yapping

[rough bloom]

I'm surprised that newer models even supported that, I thought it was deprecated ages ago because of how useless it is

[rigid snow]

the thing is there are applications for it

still

[fast pagoda]

I saw one probably scam listed for like 4k but I'm so mad I was tempted as much as I was to just drop 4k, it really fucks with your perspective after a while

[rigid snow]

not possible anymore

well, possible on openai

[rough bloom]

ye, but surely you wouldn't use an instruction-tuned model for those

[rigid snow]

it doesn't matter

[sterile kestrel]

I’d just buy it at $14,999

[fast pagoda]

That's more than I've seen it for by like a lot even for non scams

Unless that's dollar-y doos

[sterile kestrel]

I’m talking Aussie money haha

[fast pagoda]

Welp that checks out lol

[rigid snow]

this isn't possible anymore

[fast pagoda]

Price looks insane? Probably au

You guys have it rough

[fast pagoda]

Theres fill in the middle

Some models support that

[rigid snow]

yeah for tab complete

[sterile kestrel]

What’s a decent GPU that’s not gonna break the bank 🤣🤣

5090?

[fast pagoda]

/v1/fim/completions Mistral has this endpoint

[rigid snow]

though funnily enough actual github copilot doesn't use that since forever

[nocturne olive]

Used 3090?

[rigid snow]

they just prompt 5 nano or whatever to fill it in

[fast pagoda]

I didn't want to pay double retail for a 5090 but wanted 32gb of vram so I got the ai pro r9700 at $1300

Used 3090 is good to stay in Nvidia

Haven't had issues with rocm/hip really tho

[nocturne olive]

It depends if you need CUDA

[fast pagoda]

Yeah naturally

[nocturne olive]

I specifically have an expicit CUDA dependency

[sterile kestrel]

I have a 3060 rn haha but I need something if I wanna build a complete AI workstation to run everything

[fast pagoda]

Most want cuda real bad

[nocturne olive]

You could go multi-3090

Each 3090 can be as cheap as 500 euros

[fast pagoda]

I haven't had any inference related stuff not work with rocm so far surprisingly

I doubt AU is gonna have cheap 3090s

Well, I guess idk

[sterile kestrel]

I can always go on Ebay

[fast pagoda]

But 500 is fantasy price in the US

[nocturne olive]

Unfortunately NeuroSynth has a hard-dependency on CUDA for Linux GPU inference

(because nothing else is implemented)

Option is CPU

And that takes 10 minutes to render a full project

[fast pagoda]

Might be able to use vulkan easily

[sterile kestrel]

I could go for threadripper days

[nocturne olive]

Does OnnxRuntime have Vulkan EPs?

[sterile kestrel]

Actually stuff it, buy myself a tower server, get two Xeons and 1 5090, that’ll do just fine

[rough bloom]

onnxruntime supports basically everything

[nocturne olive]

Well doesn't seem like it with a lack of OpenCL

[sterile kestrel]

And don’t forget the worse part of all, 256GB of ram

[fast pagoda]

On amd it's MIGraphX

https://github.com/ROCm/AMDMIGraphX

This is the onnx provider for amd cards

[nocturne olive]

Well I don't have an AMD system to test on

[fast pagoda]

And my card uses opencl just fine

[sick owl]

Mogged holy

[nocturne olive]

OnnxRuntime is the one that doesn't OpenCL

[fast pagoda]

Oh

[sterile kestrel]

Wait, do you think it’ll be better running off a PC or a Tower Server… I always get confused on what to do with that haha

[nocturne olive]

But either way, I don't have systems to test anything other than CPU and CUDA

[fast pagoda]

Rocm thru migraph has it instead

Send me a build I'll test it lmao

[nocturne olive]

Em

I'll consider it later once we get to near NeuroSynth release

[fast pagoda]

Need more inference workloads to see how it clocks, it goes silly high in clock when doing compute loads

[sterile kestrel]

Temps are surprisingly good

[fast pagoda]

It's loud as shit doing this even at 82%

But yeah the blower has surprised me

[sterile kestrel]

Oh trust me, I know how servers are like haha

[opaque wharf]

You may want to chat with the one and only deal finder here. Duck

[sterile kestrel]

I can always use Ebay 🤣🤣

[opaque sigil]

Have they made this available to rdna cards yet

[sage crag]

pssk

[fast pagoda]

I have it I'm pretty sure because I've been using onnx

[sage crag]

https://tenor.com/view/cosmic-princess-kaguya-yachiyo-cosmic-princess-kaguya-yachiyo-jumping-fast-gif-8114031472651899845

[fast pagoda]

I need to get to my desk again to check though

[sterile kestrel]

I should be in bed but noooo here I am talking about AI 🤣

[fast pagoda]

To be sure I'm not just running it on CPU since I barely run anything in onnx

[opaque sigil]

I distinctly remember some of their graph stuff being cdna only

Maybe it was something else idk

[sterile kestrel]

You know what, imma just run everything off my Pi5 🤣🤣🤣 it’ll surviveeeeee I promise 🤣

[sage crag]

https://cdn.discordapp.com/attachments/1067638175478071307/1458837136777150464/doc_2026-01-08_07-48-27.gif

[opaque sigil]

ah wait, miopen used to be cdna only that was it

[olive sable]

Thats a new one

[fast pagoda]

Llms work well on mini PCs to an extent, I've not had a horrible time with reasonably sized ones

[sterile kestrel]

I used Ollama on it, never an issue, maybe slow but not bad using 1-2.7b on it

I wanna train an AI to be smart but it is only on 1b params… it’s gonna be so difficult but doable I reckon

[opaque wharf]

Mac and their unified memory: It's a free real estate

[fast pagoda]

Yeah I mean the Mac mini farm is real

[opaque wharf]

You know nvidia is charging insane price if mac mini farm makes more sense than a traditional gpu cluster

[opaque sigil]

inb4 we'll get to a point where you may as well buy a gh200 instead

[fast pagoda]

PS3 folding@home and DoD compute clusters: am I a joke to you

[quick condor]

It isn't loud. I had an issue getting pwm control working on my PC setting up dual boot so my fans were at 100% for an hour

[sterile kestrel]

Alright appreciate chatting with ya all!! @fast pagoda I added u if u ever wanna talk more!! Night night everyone!! ❤️

[fast pagoda]

Loud compared to what a normal PC would sound like, it was like 75 db near the computer which is would make your average grandma go ???*

[fast pagoda]

Goodnight

[quick condor]

11 pfc1212DEs at full speed

[fast pagoda]

But yeah I mean it's not loud compared to server rooms

It's a whole hell of a lot louder than my non-blower cards have been is all I'm getting at with that one

I would like a rackmount with those small diameter huge airflow fans they use all in a line in the case
And just turn them up to max

Rip ears

[opaque wharf]

You may want to start investing into industrial noise cancelling ear protection

[fast pagoda]

I almost bought a single rack the other day for absolutely no fucking reason it would've been so unnecessary

Just the rack itself

[opaque wharf]

You can always DIY a rack

[quick condor]

The pfc1212DEs aren't 40mm (what you are talking about) they are 120s

[fast pagoda]

Yeah but I want 40mms not 120s

[quick condor]

This is 8/11

[fast pagoda]

There's a lot of like "network cart" or "network rack" type half racks

Lmfao that's the high pitched whine I was imagining with the 40s at least

[opaque wharf]

Hmmm, how long until take off?

[opaque wharf]

Exactly. The rack is using a standard L steel so you only need to construct the frame

[sterile kestrel]

Hold onnnn one thing before I leave, Which is a better coding GPT? ChatGPT 5.2 Codex orrrrr Opus 2.5 from Claude

[opaque wharf]

I myself have tried gemini 3 from google

And only that

[quick condor]

7k rpm

[opaque wharf]

My result is, it is quite acceptable if you give it a popular framework/library

[quick condor]

Opus imo

[sterile kestrel]

I feel like Opus is exceeding expectations nowadays

[fast pagoda]

4.6 is crazy good at what I've asked it to do so far but that wasn't coding (it was research), 5.3 fixed an api bug that I was mad at easily

But that's all I've tried

[opaque sigil]

I like codex, it doesn't yap as much

[sterile kestrel]

I love Codex in VSC plugin. God damn it helps

Only thing I’d say is it hallucinates a lot

[quick condor]

I use Devstral2, codex and claude equally

[fast pagoda]

I used this Claude usage on robot vacuum shopping research lmfao

Not as much as has using Claude code for dataset stuff but this makes me laugh to see considering what it was

[sterile kestrel]

Never used Devstral2

Is it LLM or?

[fast pagoda]

It's Mistral coding model

Local

Well also api but

[sterile kestrel]

Maybe I should look into it

[fast pagoda]

It's like what 32b I can't remember

[quick condor]

24

[fast pagoda]

24?

Oh ueah

[sterile kestrel]

I’m surprised nobody has made a dolphin version of it 🤣🤣 aka uncensored unless they already did

[fast pagoda]

There's a dolphin Mistral 24b

[quick condor]

Mistral doesn't heavily censor in the first place

[fast pagoda]

But that ^ as well

[sterile kestrel]

At least it’s not so censored like ChatGPT

[fast pagoda]

I have honestly almost never run into/up against any guardrails except for early on a few times of them refusing to mess w/ song lyrics

[sterile kestrel]

Cause I do malware analysis, like build my own and test and then work how to stop it etc and it hates when I try and get it to help me 🤣

[fast pagoda]

Oh that will do it

[sterile kestrel]

Like I understand, it stops bad actors

But it makes it harder for the good ones like myself, 0day etc

[fast pagoda]

Pliny managed to get 4.6 opus completely mind broken the other day

[sterile kestrel]

Wouldn’t be surprised

Most “newly” trained AIs gets jailbroken or destroyed outta its mind within 24 hrs

[fast pagoda]

There's really no reason to believe they'll ever truly stop that from happening as long as llms are llms

[sterile kestrel]

Exactly, it’s called finetuning 😉

[sick owl]

Opus is a yapper but its so much more more reliable than codex in my experience

[fast pagoda]

i tend to go codex if i am totally dead in the water and dont want to spend more time tryying to fix something

i find it's reliable for that specifically

opus can yap itself out of being useful for that

[sick owl]

An optimal mix would probably be Opus writing code and Codex testing it tbh

Opus is definitely stronger at programming but 5.3 Codex is insanely good at agentic tasks

[fast pagoda]

i think i upscaled this too much, it is almost too large for me to upload w/ nitro

it's 4x what that res says lule

[opaque sigil]

[alpine cloud]

462mb png? wtf

[fast pagoda]

wow discord didnt resize it

getting my money's worth

[maiden geyser]

this doesn't make me feel like i'm downloading it through dialup enough, try harder

[fast pagoda]

i had a 700+ mb version but it is too large to upload

500 is the cap

pretty sure huggingface is so mad

[dull egret]

https://www.reddit.com/r/AyyMD/s/MooUXKNd04

BRO, CEO'S OF AMD AND NVIDIA ARE COUSINS?! WHAT THE HELL?!

[fleet jungle]

yea

[maiden geyser]

thanks, estonian intelligence

[fleet jungle]

I wonder if this inhibits competition

[dull egret]

Huh?

[fickle rain]

I don’t think they meaningfully interact

AMD’s perfectly capable of shooting themselves in the foot without NVIDIA help

[azure lynx]

i wouldn't recognize any of my first cousins once removed. I have difficulty enough with some ordinary 1st cousins.
but 'tism.

[ebon basin]

The newest claude model for agentic coding , ChatGPT for reviewing , Google Gimini & Deepseek for algorithms improvement

[ebon basin]

It is all about connection , most company CEO know each other

[fast pagoda]

when you run ruff but accidentally catch a stray .venv in there

only 17000 files to format

[fast pagoda]

vscode on linux moment

Closed all the files but I'm pretty sure they somehow just remained in vram

[maiden geyser]

https://discord.com/press-releases/discord-launches-teen-by-default-settings-globally

[azure karma]

Hello Guys, can you recommend some datasets for AI pre-training? The target is a regular conversation. But everything I found so far seems kinda bad or weird.

[maiden geyser]

it's so over

[fast pagoda]

Make your own with your own discord data

Like request it from discord they will send you your messages

[simple sky]

man, I can understand lambda calculus, but I couldn't tell you why

[fast pagoda]

Maybe you're just wicked smaht

[simple sky]

I need to review beta reduction

[maiden geyser]

remember: scraping is bad and you should never ever ever EVER infringe on people's copyrights, even the biggest ai companies would never do such an outrageous activity nor get away with it

[simple mural]

Horrible. And they just got hacked not too long ago and had data for 70k accounts stolen

[azure karma]

I am not scraping data from strangers. I was searching on regular websites for Datasets. But, they are kinda bad. I will try it with the Discord data as recommended.

[azure lynx]

definitely safer to scrape data from strangers than your friends. they'll be less likely to notice.

and what's regular conversation?

[azure karma]

Normal dialogs. One is asking something and someone else is answering

[azure lynx]

i was about to answer for "natural unconstrained informal speech" but that's different

and normal conversations aren't "please do this" "certtainly, do you need anything else?"

Air Traffic Control speech data matches what you are asking but not what you mean.

there is 20 hours of it in the corpora though.

maybe you can make up for quality with quantiy ;]

[azure karma]

Yeah, thank you :)

[opaque wharf]

Time to get back to mumble!

[rough bloom]

Matrix

[quick condor]

LibriSpeech is what I'm working with right now you really want varied - good, OK and bad training data

That's 960

[fast pagoda]

it works fine for a style lora in my experience however if your messages are not good (like many of them might be just some url or something) you probably need to do a good bit of cleaning but you don't need that many messages to make a good style

in the times ive done a lora without really cleaning the data it's suppperrrr random and they love to start spewing youtube links (that dont exist)

i have also given larger models (thru batch api) a bunch of examples of how i respond to things from the messages and then had them generate a feasible response pair for it to do its best to match the intended style and then used those as synthetic pairs to do both

not cleaning the data much has worked before (the one i've liked the most as far as result was a fucking base pretrained model without instruct where i yolo'd the entire set of my msgs at it with a small subset of fake pairs which were also made to be a sort of shitty instruct data at the same time)

it makes the result quite spontaneous and funny if nutty

[maiden geyser]

get back to mumble
that's an interesting implication your got here

[fast pagoda]

xfire is back

we should just go back to that

[opaque wharf]

What implication btw?

Or stoat is fine too

[maiden geyser]

you can't get back to a place where you have never been before

[opaque wharf]

Ah I see lol

[tender river]

you dont want "specific" data for pretraining, you want quantity over quality, the goal is to get the model "close enough", finetuning is where clean data is important

basically, your goal for pretraining is to teach it english, not teach it dialogue

[silent cloak]

Is that a screen ontop?

[olive sable]

im waiting for my packages
the first electronics arrived in a week, and the 2nd batch all seems to be getting delayed by border control

[silent cloak]

Stop ordering nukes

[olive sable]

a capacitor is only explosive if you feed it too much power

[silent cloak]

https://tenor.com/view/walking-capacitor-capacitor-发呆小屋-sword-electrolytic-capacitor-gif-7916112749430420474

[fast pagoda]

me when i completely missed that he said pre-training

[opaque wharf]

Naah, capacitor can also explode if you "accidentally" reverse the polarity

[olive sable]

[opaque wharf]

So yes, double check your connection. Especially if there is a polarized capacitor down the line

[olive sable]

i ordered a pack of 10, should be fine

nah

i ordered 20 apparently

[sage crag]

https://tenor.com/view/cosmic-princess-kaguya-yachiyo-cosmic-princess-kaguya-yachiyo-jumping-fast-gif-8114031472651899845

[olive sable]

hi konii

[shrewd cloak]

also just keep this in mind, the data is highly biased that also depends on server, chats (for example, general chat/art chat)

[sage crag]

topipoprrr

[olive sable]

[shrewd cloak]

so it would be pretty bad idea to train it on few people and only specific servers

[sage crag]

[silent cloak]

Trained on NN

[sage crag]

phrr

[shrewd cloak]

if u have a particular use case in mind, you should probably focus on chats/server focused on that topic

[tender river]

https://tenor.com/view/heart-locket-cosmic-princess-kaguya-gif-2615852714989385299

[sage crag]

https://cdn.discordapp.com/attachments/1421861060507602944/1425936142754123927/neuroCatModeOfOfnnfOOf.webp

[silent cloak]

Train chat to speak properly using datasets

[shrewd cloak]

fine tuning, we dont train not in this economy

[sage crag]

https://cdn.discordapp.com/attachments/1126496895246807123/1433049661270261871/ShouldICelebrate.gif

fine tune based on

konii dataset

[tender river]

[silent cloak]

Im gonna fine tune you

[shrewd cloak]

[sage crag]

https://cdn.discordapp.com/attachments/1061113840852533339/1456314197993459712/SPOILER_NeuroNewYearCute5.gif

im already overfit

[shrewd cloak]

to your own data

[silent cloak]

Vedals secret plan

Neuro is used to fine-tune chat

[mighty thorn]

Yeah I myself would never

[warped narwhal]

Oh the choices

[opaque wharf]

Capacitor?

[olive sable]

ye

[olive sable]

5950x

[opaque wharf]

Make one of this
https://www.youtube.com/watch?v=5dYt34JIDGw

[olive sable]

no

[opaque wharf]

[warped narwhal]

I prefer his guitar more

[olive sable]

0.47UF 50V electrolytic capacitor

[opaque wharf]

Nah, his guitar is understandably upsetting when used. But the alarm clock? Put that 2m away encased in a metal box and you're safe with a "fun" novelty

[opaque wharf]

For some reason people always forget nanofarads is a thing for capacitor, and I hate that when searching for component. Because it could be 0.47u or 470n

I think that is also an entry for Cursed Unit 3 video

[olive sable]

idkw aht nanofrada is so

[opaque wharf]

nanofarads is nF. microfarads is uF

millifarads is mF

[olive sable]

it should be u

0.47UF

[opaque wharf]

Yes, and 1u = 1000n

That's just SI prefix

[olive sable]

[opaque wharf]

When designing schematic, it is customary to not use dot because on ye olden age, it could be unreadable. So we usually use the unit to denote the dot. 4.7K resistor for example, is written as 4K7R. 0.1 Ohm is 0R1

[olive sable]

damn

thats crazy

[opaque wharf]

Which is why I hate 0.47uF and prefer 470nF

[olive sable]

its a nonissue in the digital age

i will keep using .

[warped narwhal]

What about mega farads?

[opaque wharf]

https://tenor.com/view/venom-gif-27487650

[olive sable]

[opaque wharf]

Then the M is capital

Also, I don't wanna be anywhere near a megafarad capacitor

[rough bloom]

GitHub is die

[opaque wharf]

MICROSOOOOFT

Damn, that was quick

So old man yelling at the cloud does work

I take that back

[stray dragon]

it just has an energy capacity better expressed in kilowatt-hours, what's the issue?

[rigid snow]

good thing you ordered 20

[unkempt citrus]

Supercapacitors when

[rigid snow]

not like they explode if you overvolt them or anything no

not like that would happen to you definitely not

[olive sable]

i got ones with way higher voltage than im gonna use

jsut cuz

the motor thing is 12V and the cap is 50V

margins and stuff

i was originally gonna get 35V ones, but the 50V were the same price

[opaque wharf]

Technically speaking, you don't even need the capacitor since all you use is already a module with the required capacitor on the module board

[olive sable]

i did some research and supposedly a cap at the BTS7960 input should help with noise int he power

i need the power for the oher devices to be pretty clean

i also ordered a diode, resister, and a fuse

[opaque wharf]

Oh, you didn't use the L298?

[olive sable]

supposedly the BTS7960 is a lot better in terms of not killing the other devices

i do have a L298, just wont use it

[opaque wharf]

[fickle rain]

Sorry very sorry the Copilot Agent made an update

[quick condor]

> Peer access from NVIDIA GeForce RTX 3090 Ti (GPU0) -> NVIDIA GeForce RTX 3090 Ti (GPU1) : Yes
> Peer access from NVIDIA GeForce RTX 3090 Ti (GPU0) -> NVIDIA RTX A4000 (GPU2) : No
> Peer access from NVIDIA GeForce RTX 3090 Ti (GPU1) -> NVIDIA GeForce RTX 3090 Ti (GPU0) : Yes
> Peer access from NVIDIA GeForce RTX 3090 Ti (GPU1) -> NVIDIA RTX A4000 (GPU2) : No
> Peer access from NVIDIA RTX A4000 (GPU2) -> NVIDIA GeForce RTX 3090 Ti (GPU0) : No
> Peer access from NVIDIA RTX A4000 (GPU2) -> NVIDIA GeForce RTX 3090 Ti (GPU1) : No

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 12.8, CUDA Runtime Version = 12.8, NumDevs = 3
Result = PASS```
lets go

[leaden crest]

is the use case high amperage

[olive sable]

nah

3-5A at most if it stalls

the idea is it wont stall tho

its a GA25-370

[leaden crest]

uhhh yeah I think thats above l298n capacity 💀 bro got pushed over to bts7960 for 1A

[olive sable]

the L298 does 2A per channel concurrent, GA25-370 should only do 1.5A at most if it isnt stalling

BTS7960 can supposedly do 43A

its not really about the amperage that im switching.
its about if the L298 dies it will take everything else with it

[azure lynx]

A thinker is one who thinks. A seer is one who sees. What is a peer?

don't need to say.

[olive sable]

pes?

nah wait, the er is part of the base verb

so its peers

[leaden crest]

cool

[quick condor]

89t/s sustained gen (1200 token output) on a 24b reasoning model

I'm stupid happy rn

[simple sky]

I take everything back, programming languages is difficult.

WTF is a closure

[tacit plover]

Is there any algorithm to convert a decimal number into a n-base number with non-repeating digits

[mighty thorn]

https://x.com/googlecloudtech/status/2020271355623620801

am I crazy for thinking a new agentic prompting system doesn’t qualify as an innovation, certainly not enough so to warrant a name change from LLM?

Google Cloud Tech (@GoogleCloudTech)

Recursive Language Models (RLMs) let agents manage 10M+ tokens by delegating tasks recursively.

This Google Cloud Community Article explains why ADK was the perfect choice for re-implementing the original RLM codebase in a more enterprise-ready format →https://t.co/p3MsNtLVJL

X · Feb 7, 2026, 11:00 PM

[olive sable]

i think you jsut devide by the n, and the outcome will be devided by n for the next digit.
and each time the remainder is the new digit

idk how to explain it

[rough bloom]

function with some data/variables glued to it (stored as part of the closure)
so you can access variables from the scope where you created the closure even if you pass it around a bunch

idk me not programming language theory expert

[mighty thorn]

In this paper we propose FLMs, an innovation in LLM inference for entertainment. Our groundbreaking framework automatically appends “and be funny.” to the end of all system prompts

[olive sable]

823 to Base 8
823/8 = 102 with remainder 7
102/8 = 12 with remainder 6
12/8 = 1 with remainder 4
1/8 = 0 with remainder 1
1467 in base 8

[azure lynx]

if it was 822 it would have a repeated digit though

[olive sable]

idk what you mean with non-repeating digits

if you dont want repeating digits, then you cant convert 822 into base 8

there isnt an algorithm to solve that

[azure lynx]

any representation of the number in an arbitrary base would not be the "actual" number, but it is possible to map any number to any base-n (greater than 1) with non-repeating digit sequences of arbitrary length but i don't see what problem it's solving other than the abstract one given.

[leaden crest]

?? Isn’t it just take away powers of n until you have 0

[azure lynx]

the "non-repeating digits" is the tricky part.

[leaden crest]

wdym

[azure lynx]

822 in base 8 is 1466 which repeats the 6.

[leaden crest]

oh that

mb thought you meant $\overline$

[olive sable]

if the n is fixed then it depends entirely on the input number and its relation to that base.
so what exactly is the question?
an algorithm to find the n that would make it so the input number doesn't have repeating digits int he new base?

if n is fixed it will jsut not work with certain inputs

[leaden crest]

is it any repeated digits or just consecutive digits cant be repeating

[azure lynx]

there are ways to always force it not to repeat, but that is not the same as taking the base-n representation. it can still be calculated, but is more work

[simple sky]

I feel like I'm being gaslit into thinking I'm stupid

[olive sable]

wdym with "force it not to repeat"?
it either repeats in base n or it doesnt

[simple sky]

maybe I am stupid

[olive sable]

the local variable in the scope stays alive so you can access it from outside iirc???

[azure lynx]

one simple way would be to convert the number to binary, then for the first digit if it is 0, put 0, otherwise 1. if the 2nd digit is 0, put 2, otherwise put 3. if the 3rd digit is 0, put 4, otherwise put 5, and continue with putting the kth digit as ((2k+bitvalue))%n). that would work for all n greater than 2.

[rough bloom]

dw closures sound difficult but are actually really simple in practice

[simple sky]

I don't even know what this means

[azure lynx]

there can't be repeated digits with that encoding

[olive sable]

do you know what a scope is?

[simple sky]

give me a sec

yes!

my brain lagged for a bit

[azure lynx]

i think they don't want "literally the number base N" they want "a unique mapping from the number using only the digits base N such that none of the sequential digits repeat"

if they want "literally the number base N" then that's not possible for arbitrary N

[tacit plover]

Essentially I wanted some sort of system that would let me automatically assign a number consisting of between 3-6 digits in base 6 to a decimal number like a hash (don't know if that's the proper term)

[olive sable]

so the base is fixed?

to 6 i guess

[azure lynx]

what do you mean by "not repeating" (which you didn't say this time)

coz that's the hard part

[opaque wharf]

Not repeating a number for different input?

Or not repeating a digit?

[olive sable]

do you know how the lifetime of a variable is dependent on the scope?

[opaque wharf]

A digit and number are two different things

[rough bloom]

here have Rust example

fn example() -> impl Fn() {
    let a = 1;
    let f = move || {
        println!("{}", a);
    };
    f
}

fn main() {
    example()();
}

in this case the value of a gets moved into the closure
so it can still be used when later calling the closure from main even though a isn't in scope anymore

[tacit plover]

Yes

No wait I'm stupid

Base 7 sorry

[olive sable]

ok sure, base 7.
now how does the non-repeating come into this?

[tacit plover]

If you want to look at it from a more practical angle, make a non-repeating permutation of 6 elements off of a decimal number

[opaque wharf]

So you want a way to convert base 7 numbers that has 3-6 digit to other base, without the number appearing twice (numbers not digit)?

[tacit plover]

Yes

[azure lynx]

you mean you want a permutation of the digits?

[tacit plover]

Because I want to connect 6 dots

[azure lynx]

often when asking a question, saying what you are trying to do is more helpful than asking a specific question.

[tacit plover]

Sorry

[azure lynx]

that's a totally different problem

easy though. just make a list of numbers and shuffle them

[olive sable]

no worries

[azure lynx]

and use a hash of the input as the seed to the shuffle

[opaque wharf]

Literally just do base conversion from 7 to any n. You won't have any repeat. If you want to have less digit, go for a higher base

[azure lynx]

that will give digits outside the base and doesn't prevent repeats.

[opaque wharf]

Oh

[opaque wharf]

The initial question was about base conversion. I answer it for that specifically

[azure lynx]

so did I. but we've moved on. np.

[simple sky]

Yes!

[rough bloom]

because the surrounding variables basically become part of the closure you can increment an integer variable like this too

fn example() -> impl FnMut() {
    let mut a = 1;
    let f = move || {
        println!("{}", a);
        a += 1;
    };
    f
}

fn main() {
    let mut f = example();
    f();
    f();
    f();
}

if you were to call example() again then it would give you a newly initialized closure, which would start counting up from 1 again

[olive sable]

closure basically jsut lets the variable data survive outside the function

from my understanding

[simple sky]

So a closure is just defining a function in a neater package?

[rough bloom]

package of function (or rather, function pointer) + some variables from scope where closure is created (data)

[opaque sigil]

think of it this way, every function internally gets a map of variable -> value
closures automatically extend this map with specific local variables

[simple sky]

I think… I’ll understand this after some rest and relaxation

[silent cloak]

i love getting closure allocations in C#

[simple sky]

I’m about done with my homework for the day, my brain hurts

[azure lynx]

a closure is like what is inside a function you just called

"is like" does a lot of lifting there

[tacit plover]

Wait how

Oh just a random call

[simple sky]

Thankfully the rest of the assignment is just independent representation which is easy

[tacit plover]

Uh I think I got it

[opaque sigil]

also important to note that people like to use the words closure and anonymous function interchangeably

[azure lynx]

you could just literally count instead of using a random number.

[rough bloom]

closure = lambda = anonymous function (usually)

[opaque sigil]

they're technically not the same if you wanna be pedantic but not like it matters

[simple sky]

Okay THIS makes sense

FUCK

[opaque sigil]

you can treat them as equivalent

[simple sky]

https://tenor.com/view/ishowspeed-reverse-smile-ishowspeed-confused-gif-17362243368421867313

[opaque wharf]

Wait until he gets to currying

[tacit plover]

But for 6 dots to connect in arrangements of 3-6 dots that comes at 1920 different numbers

[simple sky]

I’ll think about that later

[sterile kestrel]

Damn Toast, still in here 🤣🤣

[opaque wharf]

I love currying

[opaque sigil]

was i supposed to leave

[sterile kestrel]

No 🤣 just had a big sleep and then coming back and seeing you still in here

[opaque sigil]

[simple sky]

I’m not even a comp sci major, I’m an information sciences major specializing in computer science 😭

I’m not going to be programming for my day job, I’m going to be a system analyst, I’ll never use this ever again

Well, except maybe functional programming for personal projects

[rough bloom]

you're here
it's already too late for you

[simple sky]

I should’ve taken app development

[opaque sigil]

you sure

[azure lynx]

what language are you using? if it's python

>>> from itertools import permutations
>>> for i in permutations(range(3),2):
...     print(i)
...
(0, 1)
(0, 2)
(1, 0)
(1, 2)
(2, 0)
(2, 1)

[simple sky]

Racket, it’s a dialect of Scheme

Am I allowed to post files here? When I’m done with my homework I’ll post what I’m working on

[rough bloom]

two parallel conversations

[tacit plover]

I hadn't got to programming yet, I was thinking to search if someone had already done it

[rough bloom]

as long as it doesn't fall under the promotion rules and you just wanna yap about it I don't see why not

[opaque sigil]

does anyone know cheap places where one can rent b200s

[tacit plover]

You can probably halve this number since for the purpose of arrangements of connecting dots a 123 and 321 are the same

[rough bloom]

probably still Runpod or vast.ai
-# idk I'm behind the times I've used neither in the last year or so

[azure lynx]

yay! finding graph isomorphisms!

[opaque sigil]

runpod lets you pay per second huh

[simple sky]

At the end of the assignment my professor has the sentence

“I hope you find the closure you deserve”

[rough bloom]

they don't have any in the cheapo tier

[real sierra]

hey

i have an odd question which you can ignore if you're not comfortable answering but

[opaque sigil]

$2.5/h doesn't seem too bad

[real sierra]

are you canadian too by chance

[opaque sigil]

i think i'll ask my supervisor first though if he can hook me up somehow

[opaque wharf]

Now we have 4 conversation in parallel

[rough bloom]

have you found the one on vast.ai too?

[opaque sigil]

ye

[rough bloom]

ye that one should be good

[real sierra]

[olive sable]

Hi shiro

[real sierra]

hiiiiii

[rough bloom]

[opaque sigil]

where's my r200 with 22TB/s

[real sierra]

programmers

[opaque sigil]

[real sierra]

i hope you're all doing really well..!

[rough bloom]

shir

[olive sable]

My package is still at border control, its been almost a week now.
But they did have the audacity to send me a picture

[real sierra]

wait your username is your real name

i thought you just made it up

[olive sable]

[opaque sigil]

as long as they don't rip open the package and destroy whatever is inside

[azure lynx]

"We are treating the hostage well and they are comfortable and safe."

[rough bloom]

they send you pictures?
what a luxury, usually it just disappears into the void without the tracking ever updating

[olive sable]

I do see why i had to pay import tax on it. They put the price of the item on the package sticker

[silent cloak]

https://tenor.com/view/no-nope-no-way-noo-absolutely-not-gif-20244171

[opaque sigil]

that's how that works yes

[real sierra]

also that return address isn't sensitive at all is it

[tacit plover]

Hibo help

Shibo

[silent cloak]

Shibo

[real sierra]

hi

hello

[silent cloak]

Shibo help this person

[olive sable]

They ussualy dont, but for this they do i guess.
Cuz of the border control

[real sierra]

can i help you

[olive sable]

Idk, it probably isnt but i wont take any chances

[olive sable]

Ussualy they dont from my experience.

[silent cloak]

sorry no capacitors for you

[amber fractal]

hmmmmmmm soup

[silent cloak]

ur gonna have to take out a ton from other stuff and chain them together

[rough bloom]

ye, for domestic shipments of course they don't, except maybe on delivery
I mean for border control standards

[olive sable]

Thats not the caps, thats the o pi 5 ultra

[silent cloak]

like from a microwave (dont do this a microwave has a trillion different ways to fatality you)

[olive sable]

Im not gonna take electronics from a microwave.
I have the patience to wait a week

[real sierra]

-# sam i mean the one at the top-right of the label

[silent cloak]

yeah its scary discharging them

[rough bloom]

tax fraud my beloved
actually I guess it can't be tax fraud since it's not really a tax. tariff fraud? import fraud? some kind of fraud
-# all of the above mabe

[tacit plover]

How to make a hash function to associate a decimal number to a non-repeating permutation of elements in groups of 3-6

[silent cloak]

its crazy how much those capacitors can hold MONTHS after unplugging

[sage crag]

phrr

[real sierra]

phrrrrrrrrrrr

[azure lynx]

lies! it's like 4 ways. none of them pleasant but they probably won't even give you time to realize you made a mistake.

[sage crag]

https://cdn.discordapp.com/emojis/1154436961277313126.webp?size=48&animated=true&name=evilSlam&lossless=true

shimo

[real sierra]

uhhhh let me think

[olive sable]

Didnt even notice that

[silent cloak]

exactly

same with CRTs

[sage crag]

shimo, for programming dress, do you prefer purple or green

https://cdn.discordapp.com/emojis/1421140197369839687.webp?size=48&name=enub~1&lossless=true

[real sierra]

PROGRAMMING DRESS?

[silent cloak]

old vaccum TVs arent too fun if ur not careful discharging them

[sage crag]

ye

topical dress

[silent cloak]

pick one shimo

[real sierra]

i like the green here

[silent cloak]

shimo shibo

[sage crag]

[tacit plover]

To help illustrate my issue

[rough bloom]

weird looking clock

[silent cloak]

dementia clocks

[sage crag]

to illustrate my issue (constellation diagram)

[olive sable]

so, we all know im colourblind, but that's supposed to be purple and green?

[silent cloak]

Shiba Inu how has ur game progress been?

[sage crag]

yes camuel

[amber fractal]

your trying to use base 10 digits as a case idenftifiers

[sage crag]

it green and purple

[tacit plover]

My hash function will include the current orbit of mercury

[tacit plover]

YES

Well, in this case the indentifier is a base 7 number

[real sierra]

well the easiest way i can think of is to write down the permutation indices as an ordered sequence {a_1, a_2, ..., a_n} and then sum over (a_i)(n^(i-1)) for all 0 < i < n

[sage crag]

ye

that most trivial way

[olive sable]

beige and blue

[sage crag]

treat each index as a unique symbol in base N

[real sierra]

konii gets it

as expected

[tacit plover]

But that's over 1k cases

[amber fractal]

I just got done trashing a similar binary system in a project

[sage crag]

ne, form of arithmetic compression

simple

unique

hash

phrr

[tacit plover]

Here's the identifier for 6 dots

[real sierra]

if you have a max limit on the decimal output, just modulo it after

your hash will no longer be reversible (or collision-free) but i assume that's not required

since you're calling it a hash

[olive sable]

shiro green dress art when?

[tacit plover]

I will look up what collision-free means

[sage crag]

https://cdn.discordapp.com/emojis/1154436961277313126.webp?size=48&animated=true&name=evilSlam&lossless=true

[sage crag]

collision-free means that no two inputs have the same hash

[simple sky]

It's finished

[azure lynx]

can you tell me WHY you want to do this? perhaps there is an even simpler way to get what you actually want.
because i thought the itertools.permutation() function would give you all the values you said you wanted.
but obviously you want to do something else.

what are you actually trying to do?

[opaque sigil]

cant you just sort the permutations and then have an incrementing index

[olive sable]

what the hell is a .rkt?

[simple sky]

racket

[real sierra]

racket

[olive sable]

what the hell is a racket?

[silent cloak]

rootkit

obviously

[real sierra]

a language

functional programming

[tacit plover]

These are supposed to be "runes"

[simple sky]

https://racket-lang.org/

[rough bloom]

Lisp

[simple sky]

I'm not promoting this

fuck this langauge

[olive sable]

thats crazy

[sage crag]

lisshp

[olive sable]

anyways

[tacit plover]

That you collect through exploration

[olive sable]

back to shiro dress

[silent cloak]

i only ever slightly touched runes from svelte

beyond that na

[sage crag]

i prefer purple

[real sierra]

wait that dress is for me

[sage crag]

apparently

[olive sable]

honestly same

[opaque sigil]

well your best bet is probably to try and frame this as one of the bazillion graph problems there are

[real sierra]

i thought it was for konii

[opaque sigil]

and then do something with it

[sage crag]

i thought so too

[simple sky]

green

[real sierra]

the green one gives me fauna vibes

[simple sky]

for matrix

[sage crag]

2 green 2 purple

[fiery anchor]

is it just me or is that language awful

[real sierra]

usually i would pick purple over green for colour but in this special case

[tacit plover]

And to make sure that the player actually explores to find these runes, the symbol that describes a rune is randomised at the start of the game

[real sierra]

the green just feels happier

[opaque sigil]

green

[real sierra]

[olive sable]

i mean, you asked for shiro's preference

[silent cloak]

Vedal color

[sage crag]

3 green

[azure lynx]

then generate them once using a python script and something for visualizing them and choose the ones you want to use then keep only those and move on.

[simple sky]

alright, I've got a long journey home

[opaque sigil]

familiarity issue

[opaque wharf]

I am going to even it out then. Because I think the purple one is better

[sage crag]

https://cdn.discordapp.com/emojis/1421697636322377759.webp?size=48&name=FOCUS~1&lossless=true

[amber fractal]

I'll pick purple as well

[sage crag]

strong competition

https://cdn.discordapp.com/emojis/1424098008940675126.webp?size=48&animated=true&name=pogs~1&lossless=true

[olive sable]

im partial to the purple blue one cuz colourblind

[opaque wharf]

Maybe because I see too much green irl

[sage crag]

3 green 4 purple

[azure lynx]

i like the black and blue one.

[olive sable]

idk

[sage crag]

https://cdn.discordapp.com/attachments/1067638175478071307/1458837136777150464/doc_2026-01-08_07-48-27.gif

[tacit plover]

This makes sense but it's boring

[olive sable]

i prefer more bright clothes

[real sierra]

based

[tacit plover]

Thanks for the idea though, will default to it

[sage crag]

phrr

[olive sable]

lets ignore the fact that most of my wardrobe is dark sweaters and jeans

[fiery anchor]

pick both in neon.

[sage crag]

sam you should get this dress

[olive sable]

huh

konii is asking me to cross-dress

[sage crag]

what problem

[olive sable]

idk if i want to cross-dress

[azure lynx]

what are the stats? +3 to programming when alone, -3 when others are around?

[olive sable]

ill consider it

[sage crag]

i dont believe i said "do you want to"

[olive sable]

how expensive is the dress?

[sage crag]

i dont rember

[opaque wharf]

So, for EU citizen. I am not well versed it the lingo of your law. Does EU law applies to a service provider that provides service to EU citizen but is hosted outside of the EU?

[sage crag]

probably

[olive sable]

what?

i think so

[real sierra]

do they mean for gdpr and stuff?

almost certainly

[opaque wharf]

How would you guys enforce that tho?

[sage crag]

fines

[opaque wharf]

Strongly worded letter?

[rough bloom]

usually
protections like the GDPR do at least

[real sierra]

im not european dont ask me

[olive sable]

from my understandign if they dont adhere to eu law they're not allowed to operate in the eu eventho they're centered outside the eu

[sage crag]

euro bean

[leaden crest]

we’re neuropean

[rough bloom]

fine them
if they refuse to pay the fine, ban them

[opaque wharf]

I see

Thanks a lot

[opaque sigil]

[olive sable]

why do you want me to cross-dress tho?
im not like quack where i leak pictures "by accident"

[sage crag]

https://tenor.com/view/cosmic-princess-kaguya-yachiyo-cosmic-princess-kaguya-yachiyo-jumping-fast-gif-8114031472651899845

[opaque wharf]

What did I miss?

[olive sable]

dont worry about it

[sage crag]

everyone should wear pretty clothes

[opaque wharf]

God damnit

[sage crag]

not matter of gender

https://cdn.discordapp.com/emojis/1154436961277313126.webp?size=48&animated=true&name=evilSlam&lossless=true

[olive sable]

i do agree that i like dresses

[sage crag]

https://cdn.discordapp.com/emojis/1154436961277313126.webp?size=48&animated=true&name=evilSlam&lossless=true

exactly

[olive sable]

but idk if i like to wear it

[sage crag]

go

https://cdn.discordapp.com/emojis/1428165786672824320.webp?size=48&name=MitoBwaa&lossless=true

[fiery anchor]

kilt

[olive sable]

nah

thats like the last type of clothes ill ever wear

[opaque wharf]

I can tell from experience wearing dress is indeed breezy

[real sierra]

i too tall for silly cosplays

so i just stick to my usual boring clothes

[olive sable]

cross-dressign sure, but a kilt???
you're out of your mind

[sage crag]

shiro is 1 shirometer tall

[leaden crest]

smocks

[olive sable]

im 183cm, so 6ft. im also tall

[opaque wharf]

We finally have a new unit beside bime

[mighty thorn]

https://tenor.com/view/they-all-hated-me-todos-me-odiaban-cabizbajo-tite-caminando-gif-271476022722807553

[fiery anchor]

ye, should go to sleep. you're right.

[azure lynx]

what color hair does Shiro have? coz i think dark hair would go better with the purple dress, light hair with the mint one.

[sage crag]

https://cdn.discordapp.com/emojis/1422641181581377757.webp?size=48&animated=true&name=TortureBy1000TinyAttacks&lossless=true

^like this

[olive sable]

does shiro have hair liek that irl?

[sage crag]

everyone looks like their online avatar irl

[olive sable]

fuck

[opaque sigil]

[azure lynx]

does IRL matter for this question? ;] the green one then

[olive sable]

i have become an anime girl irl

[sage crag]

[rough bloom]

[sage crag]

https://cdn.discordapp.com/emojis/1332792300174180392.webp?size=48&animated=true&name=MinyanPet&lossless=true

[mighty thorn]

You are a tv?

[olive sable]

miss《wrrrr♡idol☆princess♡~《《《{▪☆
so idol, princess, and a tv

i wonder what she's the princess of

[mighty thorn]

Tv idols

[sage crag]

[tender river]

[azure lynx]

maybe she is a smart tv.

[olive sable]

a 2hz smart tv would be sad

[mighty thorn]

She isn’t

[opaque wharf]

What am I then?

[sage crag]

[mighty thorn]

In a good way

[olive sable]

you have no pfp, so you're faceless

[mighty thorn]

Trust

[olive sable]

an npc

[opaque wharf]

[mighty thorn]

You are an unknown soldier

[leaden crest]

uh oh looks at my pfp

[olive sable]

streetsign

[leaden crest]

Trainer Tips! Press and hold the off button to delete all your progress for the past few hours!

[olive sable]

if you pay for it ill cross-dress to opensauce

[mighty thorn]

Konii is going????

[olive sable]

i dont think so

[mighty thorn]

Oh

[olive sable]

she's free to join but i doubt she will

[mighty thorn]

Then why does she care

Lmao

[olive sable]

im not gonna crossdress in the same country as my parents are

[mighty thorn]

I meant like

Nevermind

[olive sable]

say it with your full chest kaine

[opaque wharf]

But it would be funny
- Neuro

[mighty thorn]

Why would she pay for it if she isn’t even gonna be there?

Also

[olive sable]

[mighty thorn]

Why would you ever encourage me so openly without knowing ahead of time what I’m gonna say

Do you not know me?

[olive sable]

what do you want me to do? send pictures?

[olive sable]

i was assuming it was not too weird for #programming , but you make me doubt that now

[opaque wharf]

Sent him that one AI generated images I did a long time ago

[olive sable]

him?

[mighty thorn]

I ask again, do you not know me

https://tenor.com/view/silly-dance-sea-lion-gif-18060641275691946665

[olive sable]

[olive sable]

i just meant the clothes btw

[mighty thorn]

I KNOW

now you are being the weird one

https://tenor.com/view/the-horse-of-suspicion-and-distrust-gif-12657911522319693236

[olive sable]

how so?

actually nevermind

this is already outside of the boundaries of "technical talk"

[opaque wharf]

Found it

[olive sable]

too far off-topic even for #baking

[opaque wharf]

Look at him all happy

[olive sable]

that ai generated image looks 30 years old

i should have never sent that image

[mighty thorn]

What

[olive sable]

im haunted by my past mistakes

[mighty thorn]

Baking?

[opaque wharf]

Imma be honest, I've never seen an actual teenager from the west

[mighty thorn]

Me at 1 am every day

[olive sable]

you're not a #programming regular and it shows

[azure lynx]

technically now that we're talking about #programming we're back on topic.

[olive sable]

[mighty thorn]

I love #programming in #python with my #intlellij-idea-ultimate-2025.02

[fiery anchor]

that's it, i'm going to bed. enough #programming for today.

[olive sable]

the crossdressing tangent was too much i guess

[fiery anchor]

nah that was fine

[amber fractal]

[olive sable]

hi iggly

[amber fractal]

[mighty thorn]

hi Intanani !

[olive sable]

what a coincidence that iggly is here with this topic

[amber fractal]

[mighty thorn]

[olive sable]

[olive sable]

the dress appears to be 54.03 euro.
Flying Diary Autumn Love Letter Green & Purple Lolita Long Dress
but its sold out

[amber fractal]

[olive sable]

i bet konii bought them all

[real sierra]

wait are we still talking about crossdressing in here

[mighty thorn]

Obv

This is #programming

[olive sable]

@sage crag

[real sierra]

i wanted to say i think discord's new age estimation AI is interesting from a technical standpoint

[olive sable]

join us at opensauce and come in crossdress

[real sierra]

"on-device processing" makes me wonder how spoof proof that is

[sage crag]

https://media.discordapp.net/stickers/1425866152424505394.png?size=160&name=ElvynSerius&lossless=true

[sage crag]

https://cdn.discordapp.com/emojis/1137861105444724767.webp?size=48&animated=true&name=evilGiggle&lossless=true

wrong

[mighty thorn]

I doubt it’ll be on device

[real sierra]

like what actually gets transmitted to discord?

[mighty thorn]

Tbh

[olive sable]

wdym wronmg?

[real sierra]

i mean it says right here

but i think the suspicious party is where they only say the selfies won't leave a user's device

[opaque wharf]

Maybe a vector

[mighty thorn]

Just like how copilot is “semi on-device”?

[opaque wharf]

Hopefully a vector....

[real sierra]

that still seems concerning to me tho

[mighty thorn]

So true

[real sierra]

like, even if you use SVD to compress the video frames down to a vector or something

[olive sable]

kaine, if everyone is gonna cross-dress we expect you to cross-dress too

[real sierra]

you can still reconstruct an approximate video from that

[opaque wharf]

Video frames?

[amber fractal]

[olive sable]

[azure lynx]

video so you don't just show a picture

[real sierra]

it does say "Video selfies"

their age verification will require a video of yourself i guess

[azure lynx]

similar to youtube for creator's to unlock certain features.

[opaque wharf]

Wait, I thought they were doing age estimation by account behaviour? The other server lied to me

[real sierra]

nobody lied

they're doing both

[sage crag]

check aliexpress

https://cdn.discordapp.com/emojis/1137861105444724767.webp?size=48&animated=true&name=evilGiggle&lossless=true

[olive sable]

okay

give me a sec

[opaque wharf]

Well yeah, the verification is sending a picture right? And the estimation is from behavior? You are talking about the estimation right?

[real sierra]

no i meant the verification part

that's supposedly done on-device

so im curious about how spoofable it is

[azure lynx]

"sending a pic" still uses estimation.

[opaque wharf]

Ahh, I have no idea then

[real sierra]

yes we're just differentiating between the two age systems used

[opaque wharf]

I mean, some sources says that they are using TF2 game to verify and it readily accepts it

[real sierra]

of course they're both estimates

[opaque wharf]

I imagine inzoi game purchase would suddenly increase

[azure lynx]

"if you first played a game 18 years ago, you are over 18"

[real sierra]

[opaque wharf]

The best spoof is the simplest one

Makes me remember the day of access control system bypassed because they used a relay on the reader

[olive sable]

[sage crag]

ok now try and find a cheaper listing

https://cdn.discordapp.com/emojis/1425862407800230051.webp?size=48&name=TheEvilEye&lossless=true

[azure lynx]

magnet on the side. old trick. real problem was the control logic was on the wrong side of the wall.

[opaque sigil]

they'll first try to match you to an age group based on your behaviour and if it doesn't put you in the 18+ group you get prompted to verify manually

[opaque wharf]

To be fair access control is not a simple problem to solve lol. Especially with building code compliance that needs to be fulfilled vs security

[azure lynx]

i used to have to badge in and badge out from a room i worked in. sometimes i'd follow someone else out and forget to badge out so i couldn't badge in again without getting it reset on the other side of the building.
but then i found i could "badge out" through the wall. then i just had to remember to badge in again before going in.

[olive sable]

i found one that's more expensive

[opaque wharf]

I sure hope discord AI would be wise enough to know that

[opaque sigil]

they do have a lot of data on you

[rough bloom]

how even

[opaque sigil]

can't wait for it to think i made my account when i was 4

[opaque sigil]

btw they're using a different provider now and supposedly fixed a lot of the workarounds so who knows, i'm sure people will find a way though

[opaque wharf]

Well, some are already planning to ask random people on the street

[azure lynx]

probably unreal engine metahumans.

[sage crag]

you are bad at shopping

[azure lynx]

then drive them with Live Link Face.

would be easier than what I'm trying to do now.

[olive sable]

i know

idk what search terms you used to find the first one

i only find one for 50 euro and one for 65 euro

[azure lynx]

but isn't the one they found sold out? that could be because it was cheaper.

and now you've found the current price.

[olive sable]

the one that sold out was 54.03

[opaque wharf]

The same term, sort by cheapest. Or copy the listing name of the higher priced item

[azure lynx]

maybe search in a different language? might help find some results?

[olive sable]

aliexpress search doesnt allow for as many characters as the title of the item, and search with incomplete title gives different results

[opaque wharf]

Check for each key words on the title

[azure lynx]

vintage victorian dress + other low probability words

(jurk means dress, right?)

[olive sable]

ye

eh i give up

[azure lynx]

clothes sizes online are also a risk.
you could spend the money only to find that large meant it was for a large hamster.

[olive sable]

eh

i can sew it to fit if its not too far off

[sage crag]

just use measurements

see

[olive sable]

ill do it if you buy it

[sage crag]

me brokege

this is going in my bookmarks forever

[opaque wharf]

https://youtube.com/shorts/bof8TkZkr1I

[faint sandal]

good now I have clothes for my large hamster

[hollow topaz]

What?

You come in to say strange things sometimes

[lucid shoal]

Has anyone figured out how to run whisper-large-v3 on GPU using a Radeon Windows 11 setup

[silent island]

chat

RIP game optimization frfr

tf u mean 'RTX required' bruhh?
is lowering the res and quality not enough anymore???

[fickle rain]

The game requires raytracing capabilities?

Which you get since fucking 2018 BTW

[silent island]

not that im aware off...
its just a jrpg

[azure lynx]

do they still make drivers for earlier cards?

[fickle rain]

I mean, games have already started requiring some RT capability haven't they

Look at Indiana Jones & Doom: TDA

[silent island]

if im not mistaken, 16 series cards still gets new drivers
rip 9 and 10 series

[fickle rain]

Pascal will be 10 this year it's time to let go

[silent island]

its time to move to consoles frfr

[silent island]

as to why people are still holding on to their 10 and 16 series cards is the same reason why some people are still on PS4 imo

[leaden crest]

$75M 😭 😭 thats a big hype train

[young plover]

Accounts were dark until January 6th and they suddenly got 75M of VC funding? Coincidence?

probably

[unkempt citrus]

I fully expect the pitch to just have been shizuka pointing at the subathon and going "that but we sell it"

[fleet jungle]

my GPU no longer will be receiving updates

maybe for the best

[young plover]

I'm curious what the end-game is. Making back 75M USD from AI streamers is not gonna happen, even in Japan. Making AI characters from existing IPs?

[fleet jungle]

they didn't invest in Kizuna AI? big L

[quick condor]

I just spent 3 hours debugging p2p when it was that my water pump was off

[fleet jungle]

what

what's the connection with peer2peer and a water pump

[azure lynx]

i just spent nearly 2h watching an anime movie and decided i didn't like it enough that i just stopped watching it with about 20 minutes to go. Battle Suit Gundam. (robots makes it on topic)
it got to the part of the movie where the main plot has been resolved and now they have to deal with a brand new and even harder problem after a dramatic twist, except it's obviously just another tacked on part to pad out the movie.
also: not enough actual Gundam fighting. too much character stuff. except the character stuff i think i understood seemed like a "we need a way to humanize this character... let's give them a love interest. then we have to kill the character off about 20 minutes later for plot reasons." (but i also didn't have subtitles on for the first 30 minutes so maybe i missed something?)

[fleet jungle]

yeah improvised story development sucks

[azure lynx]

there wasn't anything wrong with it moment to moment. the plot had a similar feel to an LLM generated one where it sort of makes sense as you're watching it but you can't explain why certain things happen afterwards

[faint sandal]

clearly p2p stands for pipe to pipe

[fleet jungle]

ah he's a plumber

[azure lynx]

if your pipes connect peers you have your urinals connected wrong.

[young plover]

🤢 oh god, he wants to make the grok companion but Japanese
https://x.com/cumulo_autumn/status/2020869813157917163

あき先生(Aki) (@cumulo_autumn)

みなさん、このツイートは覚えていますか？
僕は本気でやると言いました。
明日ちょっと大きな発表があります。どれくらい本気なのかは、その発表を通しても伝えられればと思います。

X · Feb 9, 2026, 2:38 PM

[azure lynx]

tax write off maybe?

"who knew it wasn't going to be a profitable marketplace? i mean other than everybody? oh well, i guess our profits are down by 75M this year."

[rigid snow]

vc is about investing into the founder, their current assets and what they can do in the future

[rigid snow]

why did you not read anything

they are a lab

first and foremost

not a vtuber

well yeah but that’s obviously not what they are gonna make money on

[azure lynx]

if i read everything anyone pasted here, i'd read like 5 things a day which is too many.

[rigid snow]

also 75m is the val not what was invested

[young plover]

I did read it
There's not much substance there

[young plover]

which is why I was speculating about the real product

[rigid snow]

the “real” product may as well not come ever

it’s vc

[young plover]

money furnace go brrrr

[rigid snow]

it’s on the off chance it catches on

[azure lynx]

maybe the investor just really likes vtubers

[rigid snow]

but yeah the stated goal is r&d of ai companions