programming

[fast pagoda]

mine does neuroNOWAYING

[amber fractal]

I need to set mine to enub

[fast pagoda]

[olive sable]

highest ram usage on my desktop has been 11gb with rendering stuff in blender
my laptop in unity has been using SO MUCH swap

they're both on 16gb rn

[fast pagoda]

i do not know if i could actually use acomputer with 16gb of ram for very long

i max out my work laptop all the time which has 32

[amber fractal]

Two of my systems are at over 32gb

[fast pagoda]

i still run out with 128

and 96 on the mini on occasion

[olive sable]

im at 1 week out of the 3 weeks they said my ram RMA would take

should be back to 64gb afterwards

[amber fractal]

framework at 32, and gaming boi at 48

[olive sable]

id ont think its worth upgrading my laptop. i would jsut get a new laptop if i needed more ram

[amber fractal]

that means #3 is at 24gb

[olive sable]

the 3050m only has 4gb of vram, so im starting to reach its limit with gamedev courses

and unlike the new frameworks, i cant swap out the gpu

anyways, new tadc episode

[amber fractal]

Man I haven't watched anime in a hot minute

[olive sable]

🔥

there is no minute emote

[fast pagoda]

illegal left bar on the cursed box

[olive sable]

ew

[fast pagoda]

need to put that monitor in the closet with the server once i finish putting it together again

the left one

[olive sable]

if you're gonna have a bar at the top, put the shit from the left in there

[fast pagoda]

cant wait to actually clear this desk off

[olive sable]

this is disgusting

[fast pagoda]

it's literally

the default

[olive sable]

WHAT

[fast pagoda]

for ubuntu server if you install gnome

[olive sable]

i need to smack some devs

[fast pagoda]

i didnt know until i did it, they still use the same layout as fucking unity

[olive sable]

there is a reason gnomes are an endangered species

good reasons

[fast pagoda]

i hate gnome

[olive sable]

[fast pagoda]

figured it'd be the least scuffed of the guis to have on ubuntu if i were to have one

[olive sable]

real talk, it is actually crazy

who thought that was a good idea?

i feel like you dont need to go to UI/UX class to realize thats a bad idea

[amber fractal]

you'd be suprised

[fast pagoda]

i dont think it was the default before unity desktop

my old laptop running ubuntu

with unity

[young plover]

[fast pagoda]

i mean what got me into kde was kubuntu

but i figure gnome is the one ubuntu will just kinda assume is default

certainly was the first one on whatever cursed canonical page i went to to find the name of the package

that system doesnt have it loaded most of the time tho so it's kinda transient and thus i dont bother customize anything

[mighty thorn]

All of pre-Covid culture in a single image

[fast pagoda]

google+

heh the cr-v (with giant subwoofers not visible although the rear is a bit sagged down for a reason)

what the fuck

i have a dark past apparently

i used to use safari on windows

[olive sable]

need an ai dev to analyse the new epsiode

"chat is this real" ah

[fast pagoda]

this laptop was a true beast

[fast pagoda]

what is tadc

[olive sable]

the amazing digital circus

i started watching it recently

when it first came out i was scared of by the fanbase being a bit weird

but the series itself is pretty good

[mighty thorn]

As someone who refuses to watch it, sure

[true hemlock]

i am officially bankrupt

[fast pagoda]

epyc?

[true hemlock]

• ram

[fast pagoda]

also holy shit i forgot how fucking terrible any and all math was when it was online based

during this time

[fast pagoda]

tfw that was probably worst

[olive sable]

like, are you gonna file for bankrupcy?

or just that you're poor?

as poor as a guy with enough tech to buy a house with it if he sold it can be

[mighty thorn]

He needs us to donate

[olive sable]

if i spend any more money rn i wont be able to make it to opensauce

it costs 80 fucking euro to make a passport

is it made of silver???

i got a appointment at 4.40pm next thursday

[fast pagoda]

im surprised you dont have a passport but i guess that makes sense cuz schengen

REEEEEEEEEEEEEEEEEEEEE my api key wtf is this 1978
i appreciate them sending the email like "ur in" and t hen when i finally go to check it, "actually nvm hold up"

[olive sable]

ye

i have different types of id for going on vacation

[fast pagoda]

same reason many burgers dont have passport

[olive sable]

we have some types of ID, but i dont have a full book thing

[fast pagoda]

my michigan ID got me into canadia and mexico

[olive sable]

you mean us citizens?

[fast pagoda]

yes

[olive sable]

cuz burgers is literally the dutch word for that

[fast pagoda]

lmfao

i use it because it's funny and i learned to read and write on unserious websites

is that a slang term in dutch

[olive sable]

nope

[fast pagoda]

or does it just happen to sound like borgar

[olive sable]

official

[fast pagoda]

kek

[olive sable]

soft g

[fast pagoda]

boujourzrzzzz

[olive sable]

no

[amber fractal]

✅

am borger

[fast pagoda]

burjer ? what the helly is a soft g

[olive sable]

idk how to explain it

[fast pagoda]

ill just go listen to it

[amber fractal]

Yeah it is a pronounation that needs to be heard

[fast pagoda]

burhurs

[olive sable]

ye

[errant flax]

tadc?

i mean the new episode was pretty decent

[olive sable]

english has hard g, and dutch "a" are flat while in english they're weird

[fast pagoda]

well, i call us burgers because

[olive sable]

ye i know

[fast pagoda]

oh jesus

[errant flax]

holy low quality

[olive sable]

[fast pagoda]

it adds to itr

but i didnt see the bottom line till posting lmfao

figure that would count as controversial or somethign

[errant flax]

thats controversial??

[fast pagoda]

not to me

but

maybe someone

[errant flax]

dying for something specific

sigh

</3

[fast pagoda]

i prefer to die to many things at once and together

[olive sable]

the word "book" in dutch is "boek", but they're pronounced the exact same way

[true hemlock]

bok

[fast pagoda]

chimken is back

[olive sable]

no, bok is a male goat

[errant flax]

got

[fast pagoda]

bok choy

[silent island]

Pointless rebrand

[fast pagoda]

well it's console so i expect the enjoyers will be ecstatic about how revolutionary it is

[olive sable]

no, as a console user this is dumb

[fast pagoda]

you are not the sort of console user im talking about

i actually havent interacted with much of what im referring to in like

10 years

[olive sable]

"bro hop on cod"
"no dude i cant sign into playstation"
"you're literally using the console rn"
"no i mean the online stuff"

[fast pagoda]

maybe they dont exist anymore

is xbawks still paid online service

[olive sable]

ye

they all are

[fast pagoda]

how

[olive sable]

nintendo being the cheapest of the group

playstration has free games with ps+

[fast pagoda]

last console i owned was a ps3

[olive sable]

and i have never use xbox online, so idk what they're doing

[fast pagoda]

and it was free back then for most of it but there was like playstation plus or somethig

[olive sable]

the ps5 was somewhat cool

[fast pagoda]

well, i have a switch i guess

i dont really consider that the same thing tho

and i also havent turned it on in over a year

i actually dont know where it is

[true hemlock]

384, the number that is the bane of my bank account

[azure lynx]

384 = 256 + 128. a nice round number.

[amber fractal]

I'm being stupid with how much data a 10M model can process

[azure lynx]

i bet it could run away and hide in a dark corner if it got scared. anything more complicated than that, a 10M model might have trouble.

[amber fractal]

It is by far not perfect, but it is much much better than I expected

I just did a run with effectively 1024 context and while it wasn't the best it was passable

passable as brain damage but the arch was doing exactly zero favors

[azure lynx]

so some % of the time it doesn't immediately seem completely crazy?

[amber fractal]

That precent has a decimal point, but considering english was not a guarantee for the model I'm impressed

Also known as the tokenizer is non existant and 8 byte chunks are thrown in and hopefully something comes out

[nocturne olive]

10M is indeed not small
It's about 1/6th of NeuroSynth's core model

[amber fractal]

For the pure lols, I'm 4xing context: doubling on each model

Surely the pair of single 384 sized layers can handle 16 bytes

[azure lynx]

it's probably plenty big to simulate twitch chat.

[amber fractal]

I feel like if I swapped the patching method, it'd do better. But I'm (sidequesting? This is my main quest though) working on some different parts

[rigid snow]

wait i remember it being the most expensive what

oh right regional pricing i'm stupid

[rigid snow]

this is the md file it's keeping track of cves in

oct 5 2021 kernel security patch but literally no kernel eop exploits since work

google's pretty good

[glass jetty]

maybe just do it by hands?

[rigid snow]

i am security noob no way in hell i'm doing any of that myself

also would have taken me a month to get to where i am rn

in 2 days

[glass jetty]

if you need just root access, isnt there should be million of solution on the internet?

[rigid snow]

good luck

• Pixel 3 (blueline), Verizon variant
• Snapdragon 845, Adreno 630, Kernel 4.9.270, Android 12, Oct 2021 security patch
• Locked bootloader (Verizon, OEM unlock disabled in Titan-M)
• SELinux enforcing, shell uid 2000, zero capabilities (CapEff=0)

[glass jetty]

why you even buy something with locked bootloader

[rigid snow]

because 1. i didn't know it's a verizon model and 2. it wasn't a concern at the time, i swapped my phone after eol anyway

and no need to flash a pixel to a custom rom, the stock one is great

[split steppe]

without an unlockable bootloader any traditional method will not work, so you're cut off from 99% of options, however this thing has helped me root obscure devices in the past, works from device itself. https://root-tools.kingoapp.com/

[rigid snow]

that's why i'm running an agent to explore cves

for eop

[split steppe]

that ain't gonna work ha

[rigid snow]

it might it might not

these models aren't dumb anymore

[split steppe]

the models are still dumb when it comes to low level work, the actual exploits themselves are closely guarded by rooters also otherwise obviously they get patched right away

[rigid snow]

it's a 2021 patch bruh

[split steppe]

doesn't matter, the model won't know shit

[sage crag]

google

cve

[rigid snow]

^

[split steppe]

you should try looking at XDA if there's any hints, but you're unlucky that you're running a verizon lock

[split steppe]

not really how that works

[rigid snow]

??

[sage crag]

https://tenor.com/view/fubuki-confused-hololive-gif-18015941

[split steppe]

i've been doing android dev and playing with rooting for ~6 years, it's not that simple

cves are not individually exploitable for root, security patches don't mean your device isn't patched, because google play services, especially on pixel, have independent patching capability

[rigid snow]

it hasn't been online for ages

[split steppe]

typically there's a whole chain of cves stringed together for a root exploit, those chains are guarded secrets

they have to be the right vulnerabilities in the right places, which the llms wont know shit about

you need intimate and accurate low level understanding of android and linux itself, which are together 200M+ lines of code

you can try, but it's kind of a waste of time

[rigid snow]

look, there are like 10 cves that would eop to root it already found and confirmed vulnerable but require higher privilege in the first place like system
you don't need intimate understanding of anything if a cve says this software (kernel, driver, android) of this version is affected, says what the vuln is and has patch commits linked

you don't need intimate understanding of anything
well, i mean not in the first place

[split steppe]

confirmed vulnerable but require higher privilege in the first place like system
🙄

[rigid snow]

yes you have to chain them that's basic knowledge

it's more than capable of such

[split steppe]

mm yes you need root to get root

[jagged turtle]

mlntcandy is your agent getting paid for this work

[rigid snow]

system is basically root but still not technically

[split steppe]

i hope your llm is free

[rigid snow]

dw about that

what i can say is i'm not spending anything on this

LOL

[jagged turtle]

aintneurway

[split steppe]

is that the actual shell output or its imagination

[rigid snow]

not sure what's actually running on 1234

[split steppe]

if true, you still need to chain the rest and unlock bootloader

[rigid snow]

presumably a real shell

[rigid snow]

i am aware

[sage crag]

try running echo $0

[split steppe]

(which is impossible btw, because it's cryptographically sealed by verizon)

[rigid snow]

tehre is a titan-m exploit btw

[sage crag]

or ps -p "$$"

[rigid snow]

-bash

[sage crag]

ye bash

real shell

[split steppe]

double check that it's actually meaningfully different from your current id, adb.exe shell "id"

[rigid snow]

shell is 2000

[warped narwhal]

So from what I've read here. You've asked an llm to hack your phone

https://tenor.com/view/bold-strategy-dodgeball-cotton-strategy-tactics-gif-4957668

[rigid snow]

not my current one tho, a factory reset one

i thought i should mention that

[split steppe]

found the blogpost it used, fairly recent exploit

[rigid snow]

it's still in a system_app selinux context tho so give it 2 more days until it figures out what it actually can do

[split steppe]

not a whole lot i imagine

btw dont let it reboot or it will softlock, since it just modified the zygote/init process behaviour

[rigid snow]

it knows and warned me

but also not a big deal i can flash it

or actually just factory reset from the recovery

[kind nimbus]

bash -p

[olive sable]

Idk, its like 20 nucks for a year or so

[rigid snow]

wasn't it $10/mo

or have they lowered the price

[olive sable]

Idk

[split steppe]

nintendo can be fairly cheap, but it is regionally priced

[olive sable]

If anything im paying more

20 euro is more than 20 usd

[split steppe]

indeed

[jagged turtle]

maximum nintendo accounts: 1

was this stat necessary

[olive sable]

[unkempt citrus]

Nintendo implements family sharing so kinda

[jagged turtle]

aren't they separate versions of the same plans

[unkempt citrus]

Nope, you get a discount if you max it out technically

I mean, it becomes worth it at 2 accounts

But supports upto 8

[jagged turtle]

well yeah, point is if both plans have the same "max acccount membership" and it's only toggled whenyou click a button that basically changes the subscription to another type is it really worth putting that stat there

[unkempt citrus]

For the family plan it is important to know how many it does support

but I do agree you could have it in a separate spot

Though this does make it clear and it is a selling point

It being a toggle is kinda tumb tho

[jagged turtle]

yea

[azure lynx]

well, the 4B cannot play interactive fiction very well. it failed miserably at both Zork I and HHGTTG. It almost made it out of the first room in Hitchhiker's when i literally told them the commands to type one at a time in direct messages.
and after they've played for a while, they loose the ability to talk at all, just trying to /type messages. they get confused easily.

(4B Gemma 3 model)

i think i'll try a 9B Qwen 3.5 without changing anything else and see how that affects things.

[fast pagoda]

https://tenor.com/view/spittin-bars-bunny-bnuuy-rabbit-bunny-spittin-bars-gif-9157936881030366542

[azure lynx]

looks like i need a new version of llama.cpp ;[

there goes a few hours of compiling.

[fast pagoda]

few HOURS?

[azure lynx]

no, of my time though.

[fast pagoda]

i was hoping that's what that meant

[azure lynx]

probably only an hour at most

[fast pagoda]

5 hours of rabbit hole XD

[azure lynx]

i dislike finding they change APIs in a subtle way that makes my code still run but no longer work

so don't want to update mid project

[fast pagoda]

i was messing with lmstudio on my mini earlier and tried using that lmstudio link thing they put in which is pretty meh, cant share resources or anything, but it tunnels the responses to the other lm studio and it functions like that lm studio is the one running the model

but then i was testing something with echo too and so i had to launch his discord thing on the mini pc and naturally loaded his model on my main pc like i've been doing

but, it was loaded from the minipc cuz i forgor

so the mini pc had the client on it

was sending a request to the main pc for the model output

the main pc was then requesting same from the mini again

[azure lynx]

your own private cloud AI server. perhaps fog server coz it's not way up in the clouds with the rest of them.

[fast pagoda]

autism cloud service

[azure lynx]

"oh, you want this guy." "Yeah, no. You want that guy." etc

[fast pagoda]

I also was then chilling in snoop dogg mode leaned all the way back in my chair listening to music and got comfy enough to fall asleep in chair

but somehow that lead to me kinda rolling off the back fo the chair

so i woke up having been slumbering like this

my back hasn't communicated much appreciation

[sage crag]

https://tenor.com/view/fubuki-confused-hololive-gif-18015941

[sick owl]

Who's gonna tell him

[rigid snow]

a human did not write that

[sage crag]

:Cheesit:

[true hemlock]

that hota t6 is wayyy smaller than i thought

but package arrived

this is the second last packages i needed for my new briefcase pc

next package will be the e display port board and touch digitizer board then im set

so tiny

[olive sable]

smoll

[true hemlock]

yeah its actually gonna be so easy fitting that inside the briefcase

[true hemlock]

deemed one of the best balance charger in the RC community i think

for DC-DC

[glass jetty]

at least i made most of work
no tests and no double checks
no posix stream (descriptor) support
no documentation
no glob time protected implemented
but i kind of completed it

next time ill decide to do something with network, ill have good api (possible that it just broken now )
https://charaverk.online/git/CharaVerKys/ats_socket

[olive sable]

[sick owl]

True

Fifty says the guy had his openslop instance write the readme

https://www.techspot.com/news/111770-nvidia-confirms-dlss-5-relies-2d-frame-data.html

I take it back

[glass jetty]

who would imagine...

[sage crag]

stable diffusion inpaint

[glass jetty]

[olive sable]

[glass flower]

who knew a thing that looked like a AI slop filter was just a AI slop filter.... i can't believe nvidia lied to us...

[patent shard]

The enlarged nose is so funny to me
Forget the hair growth

[sick owl]

There's finally a pull request for Qwen 3.5 MTP support in Llama.cpp

https://github.com/ggml-org/llama.cpp/pull/20700

[nocturne olive]

Oh? What that be?

[mighty thorn]

Martian Transfer Protocol

[sick owl]

Multi token prediction, confers an enormous token generation speedup at the cost of only an extra gig or two of mem usage

[nocturne olive]

Well then I can't use it I guess

[sick owl]

Its a form of speculative decoding that uses heads built into the model

[nocturne olive]

I got no spare couple GB

[ruby star]

This might be a silly question, but how are you supposed to send POST requests to Randy? Is it done through the terminal?

[nocturne olive]

POST requests are just HTTP requests that hit a specific port and endpoint with the method set to "POST"

[ruby star]

Ah. So I could just send it from within the app.

Cool, thanks

[nocturne olive]

Send from whatever, curl command in the terminal, another program, it's up to you

[azure lynx]

4. No Thinking Content in History: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content.
   Hmm. "does not need to include the thinking content" is not the same as "must not include the thinking content".
   but then it says:
   However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.
   hmmmmm

well, there's not enough info. I'll risk keeping the thinking and if it degrades badly I'll clean it.

[azure lynx]

well, i think i see why they recommend not saving the thinking. it sure does go on and on.

[ruby star]

I assumed it was through terminal but Randy wouldn’t let me type while running, now I realise I could have just opened another window :P

Lesson learned: Don’t code with a headache

[granite river]

I am wondering if anyone knows any good extension for Gemma 3-4b

[azure lynx]

Good extension for doing what for what program?

[maiden geyser]

spheric extension in vacuum, solve it

[naive burrow]

Neuro v3 voice soon?
https://youtu.be/jZ8wPB-KI8g

[maiden geyser]

qwenslop

[nocturne olive]

I am unamused

[naive burrow]

qwengem

[topaz tendon]

Don't mind me, just copying 315 907 files worth of my Minecraft instances over

[glass jetty]

why i see plasma launcher everywhere

[topaz tendon]

Maybe it's Cachy being popular? It's what I'm trying rn

[sage crag]

prismlauncher

is popular

because

[topaz tendon]

If I manage to solve some VRR related goblins, I may finally be able to switch over full time

[sage crag]

its good

[topaz tendon]

I thought they said Plasma referring to KDE, not Prism

[olive sable]

[sage crag]

and is the nominal successor of polymc, which itself succeeded multimc, which was highly popular before its scandal

[granite river]

I am just exploring uses with LM Studio. Just curious what's out there

[glass jetty]

not know any of thos

[topaz tendon]

Oh man, I remember the PolyMC shit, been there on the server

[glass jetty]

only tlauncher(name was stolen) -> legacy launcher(original)

[topaz tendon]

Pirated MC

[sage crag]

prismlauncher is lightweight and offers better ux and a larger utility featureset than other launchers

it uses qt for its ui, which is a bonus

[glass jetty]

qt or qml?

cant say using qt is bonus

today i spend my day for nothing
wonderful day

[sage crag]

it allows the individual configuration of an unlimited number of instances, allows you to download and manage textures, shaders, and mods, and has convenient options for sharing and managing instances

its simply better software than the official launcher, even if you dont care about modding

naturally, another plus is it being open source

[tender river]

the old official launcher was pretty good the new one is awful

[sage crag]

cef

[tender river]

the ~1.6 era was the best one, better than the ~1.2 era and the microsoft era ones

[sage crag]

i hate chromium embedded framework

[olive sable]

is the new one that bad?
the only issue i have with it is needing to log into microsoft account. and that it fails to do so a lot on linux

[sage crag]

oh and another thing

[topaz tendon]

I just got a sudden Magic Launcher flashback

[sage crag]

prismlauncher allows offline play

naturally, no microsoft account is needed for offline play

[olive sable]

im pretty sure they force you to log in to check if you actually own the game, along with the usual spying on you

i used to share the apk via bluetooth

that got patched a while ago

[topaz tendon]

I literally just logged into my MS account on Linux just fine

[sage crag]

minecraft assets are publically available to download from mojang cdns without an account iirc

[glass jetty]

forgotten tech

[sage crag]

[tender river]

ye

i use a minecraft launcher written in nix

[sage crag]

ye

an account is only needed for online play on servers

[olive sable]

i see

[sage crag]

you can still do LAN or play the game for free

with no account

[glass jetty]

on some servers

[glass jetty]

who said lan? you can play just regular server

[sage crag]

any server that doesnt allow freely impersonating other people will typically require an account

[glass jetty]

idk how for new minecraft versions, but for 1.12.2 and 1.18.2 i just ran forge server as is without any requirements

[glass jetty]

so you talking about 0.1% of servers?

than, well, ok

[tender river]

there are plugins for per-server passwords

[sage crag]

no, the majority of public servers

[sage crag]

ye

but its inconvenient because you cant guarantee your username

someone else could register it first

[tender river]

i wonder how hytale is doing

[rough bloom]

PrismLauncher does force you to have at least one MS account added I think
but ye, it's a simple check, there's no real technical limitation or any authentication behind it

[olive sable]

probably not too high.

[sage crag]

badone

[olive sable]

wasnt a joke

not intentional

[glass jetty]

funny, only librewolf with a bit of docs tabs and discord tab open

[olive sable]

i cant see any real numbers online

only estimations

[glass jetty]

lack of steam

i not see me buying minecraft any time soon of far

[azure lynx]

well, if you say anything with </think> in it it's likely Qwen 3.5 will think that token directly which makes the engine get confused because it's just thinking about the tag, and not done thinking.

[olive sable]

some places say 50K concurrent, but im not sure i believe that

[sage crag]

bzzz

[olive sable]

wrrrr

[granite river]

My poor llm is stuck on my 970 for the time being... I kinda feel guilt about it.

[olive sable]

[rigid snow]

uneducated

https://modrinth.com/plugin/authmereloaded

[olive sable]

welp, im in this cursed blender file again

trying to puzzle the elctronics in so i can 3d print the housing

[rigid snow]

blender for cad

[azure lynx]

"laggy" is apparently pronounced "lag-jeye"

according to my TTS

[rigid snow]

"Let me be honest with the user." is my new ai ick

[timber dagger]

why not use freeCAD?

[azure lynx]

never trust anyone who says "I'm going to be honest with you"

[rigid snow]

no as in it's giving up

"let me be honest i'm 100% sure i'm stuck" even though there's like a ton it can do still

[azure lynx]

"Let me be honest with the user. I'd prefer to be drawing pictures of bears."

[rigid snow]

But honestly, I think we've exhausted ...
Let me be honest with the user and summarize where we are.
in reasoning, then in response:
Honest summary: ...
Want to call it for now, or try something else?
like SHUT UP no i DO NOT

the ai is telling me to quit

[sage crag]

cigarette pack

[rigid snow]

thank for reminding i

[sage crag]

loam pat tax

[fickle rain]

How can a machine be exhausted

[rigid snow]

we've exhausted <subject>

not "i'm exhausted"

[azure lynx]

were you asking it for shrimp recipes? Pan fried, deep fried, stir-fried. There's pineapple shrimp, lemon shrimp, coconut shrimp, pepper shrimp, shrimp soup, shrimp stew, shrimp salad, shrimp and potatoes, shrimp burger, shrimp sandwich. (thanks Forrest Gump)
only if you had all of those have you exhausted shrimp.

[rigid snow]

no you can also eat raw or or

alive

[olive sable]

Im not gonna learn an entirely new software that does the exact same thing as blender, when i already know blender

And blender is pretty good for this

[olive sable]

Ye

[granite river]

I made the mistake of having my 2 llms chat. I had to intervien and have been trying to calm one down after the exostential conversation about how life is like "Pancakes". I am so tired of hearing about pancakes. I had to ban that word.

[rigid snow]

literally told it to put its suggestions to give up up its ass and it found a uaf which crashed the kernel 5 minutes after

[gentle haven]

Just a question. If I somehow got the ability to see Mod or owner only vc’s does that mean I’m doing something bad even though it’s not my fault and it’s completely the programs fault?

[rigid snow]

moral of the story: never give up

[warped narwhal]

because of discords poorly made API, voice channels are never truly private so technically everyone can see it. however if you go about sharing who was in what channels and when, you'd probably find yourself banned on all connected discords pretty quickly

[gentle haven]

I see

[rigid snow]

#sociology

[gentle haven]

I will not say how this happened then lol

[glass jetty]

complain about nothing

[fickle rain]

If you repeatedly tell LLM to try harder it’ll probably get it

[rigid snow]

"make no mistakes"

[glass jetty]

it not work
easier to start new chat and give needed context if last run failed
(assuming no code generation)

[fast pagoda]

I have never had an issue logging in from Linux that's odd, do you have some weird authenticator set up

[olive sable]

Nope

[fast pagoda]

Honestly I think I have an easier time with my Microsoft account from Linux than I do in Windows lmao

[olive sable]

Microsoft login page just doesnt make the app work

[fast pagoda]

Oh you have an xdg issue

[olive sable]

I log in on the browser and nothing happens in the laucher until the 3rd time i try or so

[fast pagoda]

You must have issues with browser oauth flows in general in some applications

Do you often have to use the "... Or copy this code into your application" option when trying to authenticate via browsers

[sage crag]

@tender river

[glass jetty]

in short, what is oauth? is it some sort of rsa alt?

[tender river]

@sage crag

[glass jetty]

[fast pagoda]

https://oauth.net/2/

[azure lynx]

it's how you can log in to other places with your google or apple account

[fast pagoda]

Just token (bearer) authenticator

[glass jetty]

perfect
now, what it uses?
i not want to read specs so i just not know

[glass jetty]

given token is one token just from user, how server verify? seems like some magic culbits

[fast pagoda]

client redirects user to the auth server, user authenticates and grants permission, auth server redirects back with an authorization code. Client exchanges that code for an access token (and optionally a refresh token) via request, then uses the token to hit the resource server

[glass jetty]

so it is just 2 servers instead of one?

[azure lynx]

it's a 3rd party authentication system.

[fast pagoda]

It's just allowing an auth server to attest the identity of the user without the user needing to share their actual login

To whatever application

so like for example if you use vscode, and it asks you to login to github

github verifies your identity and gives vscode a token that says you are who you are and lets it connect with your credentials

without vscode ever knowing your credentials

[glass jetty]

how it differ to just pasting whatever token to vscode?

[fast pagoda]

automated version of generating an access token for something

it's the same

[glass jetty]

i mean, if you have 2 servers anyway, it kind of win win

but if you just do some sort of service

why would i use it

[fast pagoda]

same reason you'd use tailwind or something

[azure lynx]

oauth means you don't need to share your credentials and can still use other services

[fast pagoda]

it's just a framework for doing it without having to roll your own

if you've ever seen auth0 ads, they use oauth2 on the backend and provides the rest like the servers and such for you

[rough bloom]

OAuth is kinda complicated and supports a ton of different auth flows and use cases
but it's basically a general API that supports every way you could possibly do auth ever so you don't have to design your own from scratch

[rigid snow]

Let me document and take a different angle tomorrow.

[glass jetty]

-# assuming https you need only login and password, maybe 2fa

[fast pagoda]

you wouldnt need any of that with oauth that's the point

the point is that you are not ever giving whatever application your credentials

[glass jetty]

so it just the same as copy paste access token

[fast pagoda]

the access tokens you're talking about are likely oauth2

so yes

[glass jetty]

any access token that shared via secure channel and at server side have permissions list, idk is it oauth or not

so, basically, as i see
it is just the same as frontend/backend, REST or threads/memory models
just general api that every one understand

maybe one day i will need it

[fast pagoda]

most likely it is oauth2 based if you are talking a commercial application in the last 15 years

[rough bloom]

just general api that every one understand
avoids requiring special hand-written integration with every possible identity provider, at least in theory

[fast pagoda]

we love reinventing the wheel

and reinventing the CVE

[glass jetty]

btw, what is rest, i not remember how it decipher
RE S T and C U R D lul

[rigid snow]

curd

[fast pagoda]

rest is just stateless api requests so each request is its own self contained thing, server doesnt have to remember anything between requests, and it's broadly compatible with http without extra overhead

[glass jetty]

i know thx...

[fast pagoda]

crud = create/read/updayyt/delet

[glass jetty]

just name is 2 1 1; + i forgot decipher

no way it is crud

[fast pagoda]

it's crud

cruddy

[fast pagoda]

g'day curd nerds

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

top tip 4 ya

[chrome rampart]

I love programming cheese

[azure lynx]

perfect #baking content

[stray dragon]

oh my god what did youtube DO to the ui

ewwww

[sage crag]

t

hi t

t

t

t

hello

[stray dragon]

konii why

my ears

[sage crag]

m

[stray dragon]

i was just about to sleep

[sage crag]

gree t

https://tenor.com/view/fox-fox-exploding-exploding-explosion-missile-gif-8942955235803510992

[stray dragon]

stop blowing up the fox

anyways goodnight konii

[sage crag]

https://tenor.com/view/foxplosion-fox-explosion-gif-23524330

by

[tender river]

neuroca t uuh

[sage crag]

m

t is going to sleep

[stray dragon]

hi chayleaf

[sage crag]

stop messaging t

[tender river]

hi t

[sage crag]

[tender river]

good night t

[stray dragon]

bye chayleaf

[sage crag]

good nigh t

[tender river]

m is going to sleep

dont ping

[sage crag]

false

m

@m

@letter_m

NOT exist

[tender river]

got annoyed and left the server

[stray dragon]

you called?

[sage crag]

t

[stray dragon]

ok changing back

[tender river]

[sage crag]

for a brief time

t was m

[stray dragon]

you can't prove it

it never happened

[sage crag]

m

true

[stray dragon]

they'll never believe you

[sage crag]

maybe

[tender river]

or maybe they will

[sage crag]

maybe

[sage crag]

m is t 2026 campaign

[tender river]

dont believe what the media tells you about t

[stray dragon]

they won't believe it, because it didn't happen

[tender river]

[chrome rampart]

t probably codes in php

[stray dragon]

i would NEVER

[shy jetty]

[rigid snow]

https://projectzero.google/2025/11/defeating-kaslr-by-doing-nothing-at-all.html

cosmetic kaslr

[opaque sigil]

[timber dagger]

I am willing to help if you want
I have solidworks from school, or inventor, or fusion 360, or freecad
if you give me technical drawings and its not anything crazy

[proven merlin]

so.. ive started palying with fastmcp.. using small models qwen3.5 4b, on lmstudio, i did a simple thing like add, aka litterally just the quickstart example, but a thing i noticed, after just 1300 tokens, it becomes VERY slow to start generating while it absolutely is hammering my gpu.. is this normal?

once it started generating its flying at 30t/s, its just as if there is a second hidden "processing prompt.."

[true hemlock]

normal. at high context windows it has to compute more

[thorny marsh]

can confirm

the longer the context, the more memory it uses, and more usage means the GPU also has to read more of it

[fast pagoda]

this is why stuff like claude have the "COMPRESSING CONTEXT" summary thing where haiku comes in and summarizes everything to that point and clears out the history

at its most basic attention has to attend to every previous token in the sequence so if your context window is 500k tokens it has to compute attn scoring for that entire context window as part of continuing its output

this is mitigated a bit by kv cache since it's caching that computation for the tokens already processed. the expensive part is prefill phase which is when lmstudio will say like processing prompt- processing the prompt/context to build that KV cache before generation starts. once the cache is built, actual token generation (the decode phase) is fast because you're only computing attention for the new token against the cached KVs. that's exactly why you see it "hammering the gpu" then suddenly flying at 30t/s

i wanted to put a screenshot but my brain is stupid

this is that

[olive sable]

i have a systemt hat works

no need to over-complicate it by adding other random programs or other people

[fast pagoda]

vibecode your own, shittier, cad

[mighty thorn]

@fast pagoda I’m gonna replace backpropogation

[fast pagoda]

frontpropogation

[olive sable]

also, "if you give me technical drawings"
so then id have to still design it in blender, then instead of exporting as stl and 3d printing, you want me to make technical drawings of it, then you make the exact same thing in a different cad, and then i 3d print that???

[mighty thorn]

Forward Forward

[proven merlin]

actually, it turns out it was "processing prompt", which shows up if i disable thinking. however kv caches should allow me to only compute the tiny bit that it still has to do, however it seems that lmstudio (aka older llamacpp) really struggle with qwen3.5 due to their true multimodalness

i suppose i should try kobold.cpp?

[proven merlin]

i was actually at just 4096 context size to sanity check that, and no it still happend, 30+ seconds.. meanwhile a full 48000 token context takes my device about 15-25 seconds to process iirc

which makes this whole thing just wierder..

[topaz tendon]

Any Linux gamers here know more about controlling VRR? I figured out how to make it work in games now, but it also works in eg. video players, which tends to cause either flickering or artifacts due to low refresh rate. I'm wondering what would be the best way to try and fix that.

Like a VRR blacklist or maybe a toggle for VRR so it only works in games

[proven merlin]

okay i got no clue whats going on, fresh chat, 4096 context, and the model just wont reply at all now.. could this be due to structured json = true?

[minor sail]

It's just choosing not to reply, it's being extra smart

[proven merlin]

id accept that honestly

like

its just not even starting to do anything

[minor sail]

but if you have to send it properly formed json, do you know what the json object is supposed to look like?

[proven merlin]

i accidentally somehow managed to make the promt "Dont send a message" actually functional for ai

[proven merlin]

i dont, im trying tool calling for the first time

[minor sail]

That might be where to start lookin

[proven merlin]

lm studio, fastmcp on python, qwen3.5 4b.. the quickstart example, plopped localhost into the mcp.json and it worked for a bit until it just.. sometimes doesnt or takes 3-4 business days to start

[proven merlin]

wdym?

[minor sail]

Hmm, start without any MCP hosts, see if it works?

idk about what structured json = true actually means, but it could be expecting json for its input?

[proven merlin]

without mcp hosts its fast

only occasionally slows

[minor sail]

curl -X POST http://localhost:1234/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "local-model",
    "messages": [
      {
        "role": "system",
        "content": "You are a helpful assistant that responds in JSON format."
      },
      {
        "role": "user",
        "content": "Tell me a short story about a cat and a dog."
      }
    ],
    "temperature": 0.7,
    "response_format": {
      "type": "json_schema",
      "json_schema": {
        "type": "object",
        "properties": {
          "story_title": {
            "type": "string",
            "description": "The title of the story."
          },
          "story_content": {
            "type": "string",
            "description": "The main content of the story."
          },
          "word_count": {
            "type": "number",
            "description": "The number of words in the story content."
          }
        },
        "required": ["story_title", "story_content", "word_count"]
      }
    }
  }'

Internet tells me that's an example request

[proven merlin]

i just downloaded gemma 3n qat, and that seems to work fine

is it truely just a old llama.cpp kv cache issue?

lemme test further ad see what happens

[mighty thorn]

Gemma 3n mention

[sick owl]

https://time.com/article/2026/03/09/ai-robots-soldiers-war/

Torment nexus

[proven merlin]

so uh yea

i think lm studio is just massively struggling with qwen3.5

well i guess i should actually look into better solutions.. though most are either just ugly, buggy or just dont have the features one wants, or need like 45 thousand containers and seperate scripts that you need to start up..

[sick owl]

Llama.cpp is vastly superior

Then you can just slap on any old frontend or use the built in webui

Lmstudio is both closed source and usually several releases behind

It's like a less crap ollama in that regard (though at least some of ollama is still open source I guess)

[glass flower]

haven't really had issues with lmstudio tbh. tho it is a bit behind in supporting the latest llamacpp features like a month behind. which is only really a issue with brand new models. qwen 3.5 especially had some looping issues at the start but those should mostly be gone.
been using qwen 3.5 9b for my main llm for a bunch of different stuff from autocomplete and opencode etc.

[warped narwhal]

I've always doubted these articles. Not because I think AI soldiers won't exist, but because I highly doubt they would ever be in a humanoid form.

like if I'm going to create skynet, it won't be with something that runs on two legs, I'm gonna make that thing fly through the sky and roll across the ground at 100mph

[sick owl]

I think there will be humanoids despite the impracticality because they look more impressive in the global military dick measuring contest

[warped narwhal]

I mean, I think I'd be far more terrified of 4 legged spider robots than other humanoids, but I do see what you mean

[leaden crest]

wars are fought with aura not guns

always have been

[olive sable]

https://tenor.com/view/steamed-hams-aurora-borealis-what-is-happening-fagetti-wtf-gif-16514726

this is the reason i dont trust cad

you convert to polygons and see 20 different crimes

WHAT

[leaden crest]

is it circular

[olive sable]

yes

[leaden crest]

the fix is easy just make a circle 0.1mm larger than it and extrude it into a cylinder to make a cylinder face

[true hemlock]

sums up the whole US military shit

[leaden crest]

assuming you don’t care about nanometer accuracy that is

[olive sable]

ye

[true hemlock]

slap ai in their tech regardless of how ass language models are at timing aware tasks and shit

[leaden crest]

same deal for rectangular faces

you dont have to be perfectionist about it just do it on screw and wire holes and other places that matter

[olive sable]

there is no need to

it has 2 layers

[leaden crest]

oop

[olive sable]

i downloaded this model form a psychopath

[leaden crest]

[quick condor]

I've never touched lmstudio. I went from ollama to Exllamav3 and then to my current serving platform vllm. I doubt I'll mess with it since I'd have to give up my highly optimized Marlin kernel and the p2p optimizations.

[fast pagoda]

the backend for lmstudio is quite literally just llama.cpp

[young oyster]

Why only one, can have long distance auvs, more maneuverable drones, humanoid foot soldiers and smaller cheaper hexapods for scoping small areas.

[sick owl]

Yes, but its a permanently outdated build with a closed source frontend tacked on

[fast pagoda]

might be some weird pinned version but it is the same as whatever build it's pinned to

it's definitely not the mainline build

probably cherrypicked

on whatever they inclujde

[sick owl]

Man you really feel the difference in answer quality between the dense Qwen 3.5 and the 35B MoE on inputs like these

[fast pagoda]

they have their own versions

[sick owl]

Top is 35B bottom is 27B

The lack of intuition in the MoE is obvious in the formatting

[fast pagoda]

that's the woo wood from more activations at once

which you def know but im stating the obvious because

now i want to see the speed diff between lmstudio backend and the main llama.cpp

guess im off testing again

[sick owl]

Interestingly the 27B translation kept a Japanese word in there because they define it afterwards in the comment anyway

The 35B translated it into English regardless

Both are valid choices, the 27B is the more intelligent choice but the 35Bs decision is localised and thus flows a little better

I'd still prefer the more grounded and reliable answers from the 27B even if they're less fluid in edge cases like these though

Fwiw I run the MoE Q8 and the 27B at Q5

Q6 for the dense if I'm serving it to other devices since I can squeeze a little extra VRAM out then

My monitors 4k so the OS reserves a decent chunk

[fast pagoda]

i think mottainai is a better way to translate it because it's kinda a specific concept

a weeb would still say mottainai lol

[sick owl]

Its definitely the more culturally sensitive translation

[fast pagoda]

the other one is better in like

a general broadcast

or something

[sick owl]

I think I would still use the MoE for really simple translation without much subjectivity like twitch chat messages since its speedier

But for anything paragraph length or more the slightly longer wait on the dense is definitely worth it

[fast pagoda]

translation is definitely something that benefits from the increased awareness and larger possible field of accessible information

frontier model translations are really good, which makes sense, it's just funny how much better it gets as you increase model capability when it doesnt seem like that hard of a problem. even tho it is if you think about it, there's a lot of nuance to it

[sick owl]

Honestly when it comes to pairs like Japanese -> English I run on the recommended sampler settings for precision coding tasks because it takes a pretty sharp grounded model to figure out how to interpret the differences in structure between them

[fast pagoda]

have you tried glm 4.7 air

[sick owl]

You end up getting weird shortcuts or hallucinations with lesser models and less constrained sampler settings

[fast pagoda]

flash

i mean

[sick owl]

Yeah its far worse than Qwen 3.5

Not even close imo

[fast pagoda]

on this task? or just in general

[sick owl]

In general and on this task

The full size GLMs are beasts though

And the Air variants were good for their time

[fast pagoda]

i've found it to be fairly reliable for at least for like, if i slap it in the driver seat of a training run or something to just monitor and fix it if it explodes

27b is definitely better

but 4.7 flash is much faster

it runs almoost as fast as 9b qwen3.5 for me if not faster it feels like

which is crazy

[fast pagoda]

rofl at the difference between llama.cpp-HIP vs -vulkan

[opaque sigil]

wtf is that variance

+= 845.72

[fast pagoda]

it's because

HIP

goes oom

and freezes

[opaque sigil]

ah

very cool

[fast pagoda]

yeah you'd think you know

ROCm

with their premier workstation "AI CARD"

would be like

useful

thank god for vulkan is all i can say

[opaque sigil]

is this a build from amd or some generic hip build

[fast pagoda]

i built this

it's 7.2 nightly from a couple days ago

[opaque sigil]

nvm the amd llama.cpp fork is deprecated anyway

guess they just upstreamed their optimisations

[fast pagoda]

there's no llama.cpp rocm, it's llama.cpp HIP, and it's not amd official afaik but idc

[opaque sigil]

https://github.com/ROCm/llama.cpp there was this, emphasis on was

either way

[fast pagoda]

oh im sorry i forgor im on 7.11 rn

[opaque sigil]

it wouldn't surprise me if the kernels for rdna were unoptimised af

with how they like to pretend cdna is the only one that exists

[fast pagoda]

it's stinky

i wish they'd just make the pro cards cdna if they want to be goofy about it liek that

but money hard

[opaque sigil]

i'm sure it'll get better over time, it makes sense to prioritise the cdna cards at least if they have to choose

[fast pagoda]

i cannot see how thye possibly have to choose with how much money theyre making

[opaque sigil]

idk

amd's management is fucked

[fast pagoda]

resources are resources but thye just dont wanna is all it tells me

which is sad

[mighty thorn]

That

Was does nightly mean

[granite river]

Daily build

[fast pagoda]

this card has almost twice the fp16 compute as a 5090 in matmul

shit tier mem bandwidth

comparatively

but

the compute is there

[opaque sigil]

too bad theoretical peak throughput means nothing

[fast pagoda]

it means something but it certainly doesnt translate well when youre talking different architectures

im only referencing it to say it COULD if optimized and supported well, be good, and honestly it is good when things are working

but like

why vulkan work better than rocm

cmon bruh

[opaque sigil]

[fast pagoda]

especially in inference this thing cant saturate at all

gddr6 bandwidth lule

[opaque sigil]

oh, that bandwidth

[fast pagoda]

it's so sad

it works fine for training

but man does it kill it in inference

[opaque sigil]

clearly you should just buy an MI300X duh

[fast pagoda]

training has other problems (rocm)

i was about to say something along the lines of i would like lisa su to pleas sell me a mi300

for

not $153293045832

5.3TB/s is the kinda bandwidth i can get behind

[opaque sigil]

i wanna see what kind of bandwidth numbers they'll reach with hbm4

[fast pagoda]

2x

[granite river]

cry inside with just a gtx 970

[fast pagoda]

3.3TB/s per stack

for samsung

it's just twice as wide as hbm3/e

2048 bit

[opaque sigil]

interesting, the mi300x has 25% faster memory than the mi350x despite using hbm3 instead of hbm3e

i guess the extra capacity hurts

wait no the gpu in general is just faster

oh well

[fast pagoda]

the model numbering for the instincts is weird

[opaque sigil]

they announced 4 or so different cards in the MI400 range too

[fast pagoda]

yeah

[opaque sigil]

hopefully the hpc one will be good

[fast pagoda]

that'd be MI430x

[opaque sigil]

yea that one

[fast pagoda]

ill accept 432gb HBM4

19.6TB/s overall lel

[opaque sigil]

slower than rubin, unacceptable

[fast pagoda]

2x the available mem tho

well

not fully 2x

1.5x ish

exactly 1.5

quick maffs

[opaque sigil]

true

but speeeeeeeed

[fast pagoda]

12 stacks vs 8 i guess they are limiting for heat?

they are only doing 6.4Gbps per pin which is less than the 8Gbps of JEDEC HBM4

so theyre sandbagging that for power or heat i guess

8192 bit mem bus for the 455x sheesh

cant find a width for rubin other than a 1024 for the cpu

too bad infinity fabric sux compared to nvlink

ualink never

ok that makes sense

total bandwidth of the 12 stacks should be 39.6 TB/s

they seem to be runnign the stacks at half speed

inb4 430/450/455x all have different actual bandwidths spanning 19.6-39.6 entirely on binning

[mighty thorn]

[fast pagoda]

combine all those and believe it or not, AMD instinct MI455X comes out

[mighty thorn]

Crazy

I’ll take your word for it

[fast pagoda]

lmfao what the helly

[granite river]

woah

Why Vulkan running 2x

[fast pagoda]

Vulkan has been nearly universally better in all applications i've had the option of choosing it over rocm

[granite river]

Significant difference

[fast pagoda]

just realized hip was out of date there so updated the hip build and it's a bit better but not that much

it's not even universal though, is the thing

[mighty thorn]

Guys

What if we take Vulkan

But

Instead of having your GPU do the math directly

We send every single math operation as a request to the ChatGPT api

In order to stay at the cutting edge

[opaque sigil]

[olive sable]

vulcan?

vulkan?

[opaque sigil]

vulcuhn

[mighty thorn]

https://tenor.com/view/patrick-star-minor-s-spongebob-spongebob-meme-spongebob-squarepants-gif-10940848928744248792

[split steppe]

https://tenor.com/bC4iW.gif

[cobalt sigil]

https://www.reddit.com/r/degoogle/comments/1ry6z5k/androids_new_sideloading_rules_are_here_and_they/

[olive sable]

what the fuck

[fast pagoda]

fuck off google

[olive sable]

guess il put nixos on my phone

[fast pagoda]

"We think you are stupid as shit so"

[cobalt sigil]

one more reason to switch to iOS lol

[fast pagoda]

[cobalt sigil]

but at least not a 100% ban

[olive sable]

the ios bros instantly jumping on the google hate bandwagon when apple doesnt even allow any sidelaoding

[fast pagoda]

yeah i mean i aint gonna swap to gulag because im mad at being thrown in jail

[olive sable]

[cobalt sigil]

https://android-developers.googleblog.com/2026/03/android-developer-verification.html?m=1

[olive sable]

"android is making x harder, and since that is important to me i will switch to a platform that doesn't allow it at all"

[fast pagoda]

can i take an iq test and disable this if the result is over 5

[olive sable]

[cobalt sigil]

what's 10+9

[fast pagoda]

21

[olive sable]

21

[fast pagoda]

9+10 is 21 , 10+9 varies

[cobalt sigil]

you failed, now you can't sideload app in android

[olive sable]

10+9 is context dependent

[olive sable]

at that point id rather use the chinese lenovo android

[cobalt sigil]

tbh now sideloading apps in some phones are already relatively difficult

like those with Chinese rom

[fast pagoda]

im installing gentoo

on my pixel 10 pro les go

[cobalt sigil]

you gonna lose GMS also then

[fast pagoda]

global meatball system

[olive sable]

i dont even know what that is

[cobalt sigil]

lmao

[cobalt sigil]

Google mobile services

[olive sable]

so i get the bonus of removing spyware?

[fast pagoda]

does GNSS exist in europ

[olive sable]

idfk

never heard of her

[fast pagoda]

it's galileo

the euro gps i guess

never found out if that was actually a thing

i mean it's a thing

but like

in use

[cobalt sigil]

, but the downside is that it's pretty much impossible install them back if you need them and the phone company did not pre install them

[cobalt sigil]

didn't EU say now they have to make battery removable now

[olive sable]

you can use google services and such, and with some trickery you get widevine too.
i specifically want the lenovo rom cuz of that

[fast pagoda]

i installed google mobile services on an amazon fire 11 tablet

[cobalt sigil]

unless you got another rom

[fast pagoda]

without root

[olive sable]

you say that as if we the people know what happens in the dumbass parlement

[cobalt sigil]

I actually got a redmi phone that doesn't have GMS installed, and manually installing them doesn't work at all

[fast pagoda]

they wear funny hats and issue angry statements it hink

[olive sable]

ye thats why i want the lenovo one

[cobalt sigil]

does Lenovo have them pre installed

[olive sable]

well, kinda

[cobalt sigil]

the last time I have them are like 7 years ago

[olive sable]

its complicated

[fast pagoda]

im surprised considering amazon fire OS doesnt come with them and it was pretty easy to get them on it

[olive sable]

i got a chinese lenovo xiaoxin pad pro 2023. and google play store came preinstalled, and i installed the other apps via that

[cobalt sigil]

and I think I bricked the phone somehow

[olive sable]

then i had to make it boot into safemode to get widevine

[fast pagoda]

we love virus based drm

[olive sable]

i dotn remember the details, but there is method to the madness

[cobalt sigil]

tbh back then when they force apple to use type c I thought we actually gonna have some great things

but instead they got a 16pin type c instead of the common 14pin type c or something like that

[fast pagoda]

i still cant believe how long crapple was hawking lightning in iphones when they switched ipad infinity years before

[olive sable]

apple - helps make usb-c and uses it on ipads and macbooks for years
also apple - lightning on iphone

[fast pagoda]

tim apple's got a master plan

the plan: money

[olive sable]

tim apple's master plan is talking about ai

[fast pagoda]

but never making any

[olive sable]

and then not having any ai

[fast pagoda]

just buying it from googl

[olive sable]

yes

[fast pagoda]

and then call it apple intelligence triumphantly

[cobalt sigil]

apple intelligence (powered by Gemini)

btw you prob won't believe it but Gemini is finally accessable in where I live without using a vpn

[fast pagoda]

apple's patented buy competitiveness from your competitor and pretend you did anything

[fast pagoda]

what's normally available like before

curious if you had like z.ai and chinese services before

which z is

but i realized there was a lot to name lol

[olive sable]

[cobalt sigil]

grok and perplexity

what else

the Microsoft ai

they can be used

oh copilot

[fast pagoda]

i found a screenshot from bing ai yesterday

[cobalt sigil]

sorry Microsoft

[fast pagoda]

i forgot it was bing first

[cobalt sigil]

deepseek?

I enjoy using grok more

and gemini

[fast pagoda]

deepseek, z, qwen,

[cobalt sigil]

never heard of those 2

[cobalt sigil]

china have a lot of ai service that I can't even name

[fast pagoda]

https://tenor.com/view/take-it-away-ern-ernie-harry-potter-knight-bus-gif-18140725

[cobalt sigil]

but yeah I only uses grok and Gemini the most

what's weird is that if you have a workspace account then you can still use Gemini without vpn before

[fast pagoda]

local servers

[cobalt sigil]

no

don't think so

it's a very weird relationship

[fast pagoda]

workspace, you configure the region specifically

so i imagine it's something to do with that sort of control

[cobalt sigil]

not quite sure, but definitely not hosted by ourselves

[fast pagoda]

i meant more like

local (google controlled)

[cobalt sigil]

(idk)

but hey, at least now everyone can use it

[fast pagoda]

ERNIE 5?

(i have no clue how ernie is but i like its name)

ernie is baidu's

[sick owl]

Ernie has historically been a bit crap and subject to domestic controversy

[cobalt sigil]

ohh 文心一言

[sick owl]

But hey I wouldn't write off a new model release just cause the past ones weren't the best, could always turn it around someday

[cobalt sigil]

tbh I've never heard of the English name :P

[fast pagoda]

Enhanced Representation through kNowledge IntEgration

lmao

[cobalt sigil]

just call it baidu ai

tbh baidu the search engine itself is kinda

mid

[fast pagoda]

qwen is pronounced qianwen

since that's what it's short for

[cobalt sigil]

千問

the name itself sounds familiar but no idea what company is it under

[fast pagoda]

alibaba

[cobalt sigil]

ohh