programming

[amber fractal]

Honestly I'm a long way off from that point, I'm still on CPU as I still need to make sure this whole project works at all.

[silent cloak]

just do small conceptual prototypes

see how it runs tests on the GPU

[amber fractal]

Not at that stage yet, I need it to function in itself first

[mighty thorn]

How dare Elon make it difficult for me to scrape the data I should rightly be able to steal without permission

[silent cloak]

ha take that

no data for u

[mighty thorn]

When I get it working, it’ll be even sweeter

[maiden geyser]

@grok is this true

[silent cloak]

Gork

imprison this creature

[amber fractal]

I already have an idea on GPU preformance of the current iteration, but I'm not liking having to manually loop in the training portion. I hope the rewrite with everything I'm adding can allow me to add everything I need to without that ugly looping.

[mighty thorn]

Using the official api looks bleak

[amber fractal]

forward pass just works, backwards pass has some issues.

[rigid snow]

inb4 everyone solves the api problem with implementing x402 or something

[silent cloak]

https://tenor.com/view/ai-slop-tung-tung-sahur-tung-tung-tung-sahur-tung-tung-tung-tung-sahur-now-gif-16546899190744295000

[rigid snow]

[rigid snow]

honestly a pretty cool idea

do we hate ai so much we don't mind crypto

or nah

[mighty thorn]

I don’t like it cause it makes it harder for me to commit mass theft

[mighty thorn]

I’d assume most people here hate crypto more

But general public?

[rigid snow]

[mighty thorn]

I’d believe it

Ai hater tunnel vision is crazy

[amber fractal]

Honestly I don't consider cypto the worst, I just think of if as a game market.

[opaque sigil]

why not both

[rigid snow]

nowadays the only true reason to hate crypto is the culture which is very understandable, it's good tech otherwise

[amber fractal]

There is enough hate for everyone's bad choices

[rigid snow]

and culture here is irrelevant x402 is a way to programmatically pay for requests not engage with cryptobros

[amber fractal]

This is where I stand, family member does it and enjoys it so I'm biased to be netural

[amber fractal]

My brain just clocked in on what that means, yeah that makes sense

[silent cloak]

they are both shit and consume a ton of hardware

[amber fractal]

took me too long to realize that is the error code for payment required

[rigid snow]

there is a reason i said nowadays

no chains people actually use for payments are proof-of-work

[rigid snow]

i'm talking about this

[rigid snow]

not that wasteful anymore

[quick condor]

I mean Im looking at an L4 to additionally augment my db server but I havent seen anyone else mention this line

[amber fractal]

-# Every other person has flamed the line for not making sense compared to other lines

[quick condor]

I mean the A2 has worked flawlessly for its task, fits the hardware and isnt jank so IDK

[amber fractal]

I just see it as cloud GPU and nothing more, it feels like it was built to be a vGPU considering how Tesla P40 has the same ram and bandwidth and was built for that task.

[quick condor]

Ive worked with P40s they are a headache

[amber fractal]

You pay with time not cash

at least in this market

[quick condor]

Im not running primary inference on a db server lol

[amber fractal]

All I'm saying is that it is cheap for mortals, you pay for that cheapness by being out of nvidia driver support and setup/usage time.

I personally was aware of that when I made the choice to buy them

[quick condor]

I was originally running just an A4000 and a gtx1080 (pascal). The 1080 made things brutally slow due to no tensor cores

[amber fractal]

That is a very imbalanced system, plus the common knowledge that fitting in one gpu is always preferred to having multiple cards.
I don't even do AI training and I know that due to how often it is talked about both from users here and marketing.

[mighty thorn]

This is true

Less true if you have data center hardware with high speed interconnection

But mostly true

[amber fractal]

I mean it still exists, but not much you can do as your limited by interconnect speed either by system or networking at supermassive scales.

[sage crag]

[quick condor]

Which is why Im thinking about selling ram and getting a rtx pro 6000

[mighty thorn]

[amber fractal]

there was a post in here forever ago saying that an h100 cluster was always at half usage due to just all of the interconnects

[mighty thorn]

Bro does not have 9tb of ddr5 to sell for that gpu

[quick condor]

thats mostly due to poor optimization and the fact that most DCs heavily throttle h100s to keep their temps down

I have 1.5tb of ddr4 32gb server modules

[mighty thorn]

That is not 10k worth of memory brother

[amber fractal]

-# Every day I hope that my NN project works so I never have to deal with the markup on tensor core cards

[mighty thorn]

What exactly are you doing that is gonna solve vram?

[quick condor]

lol, have you seen prices? I could get that taking it to one of my vendors at work

[amber fractal]

Nothing on that front I just know I won't need tensor cores so I don't need to pay for cards that have a lot of them

Something nice would be to deal with bandwidth, but speed issues will always exist due to interconnects.

At the same time, I am the type to say skill issue if I can't train on my hardware so

I don't have the economy nor desperate enough to cloud compute

[quick condor]

I mean thats like me saying "I can quant 123b models" on my hardware. I can but Im not going to do cpu quanting

[amber fractal]

You clearly picked too large of a model then

If a 7b equivalent is all I can do, then I'll make it the best damn 7b as I can

if I have to wait days for it to work, then I better hope I have some good cooling set up to wait

[olive sable]

[mighty thorn]

[olive sable]

I have eaten enough dessert to bankrupt a small country

I have gained 10kg

[opaque sigil]

i hate vscode, updated to 1.107 and now explicitly setting the editor background to #111111 sets it to #181818
how does this even happen

[quick condor]

I mean it depends on use... I can run 70b but dont care for llama and its significantly slower. For the primary model I use a 24B, for DB augment a 8B, and for embedding a 0.1b model

[olive sable]

Merry christmas

[amber fractal]

All of my cluster is 4 P40 cards on very bad amounts of bandwidth, if the model size for my usecase isn't able to run well enough then either I compromise or work my ass off to make something that will run well.
I can't rely on any tensor performance, so I'm working on something that doesn't need them at all and can just be ran on the normal cores at fine speeds.

[quick condor]

You might be able to cap out with 24B models on individual cards with good quantization but anything beyond that really has to be VLLM and you are going to tank in t/s

[amber fractal]

That I call a skill issue for compromising in the first place, that just means Transformers as an architecture is not fit for my needs. That's why I'm working on not using Transformers at all.

It isn't easy and it may not even work, but I'll try my damn hardest to explore the options I have.

[obsidian mantle]

what about ddr6

i heard from several sources that it was better to wait for ddr6 in september

[amber fractal]

ddr6 won't help me if I'm still on gddr5 memory

the 3060m is not real

[opaque sigil]

true you should wait the 2-3 years until ddr6 ram exists

[amber fractal]

dislike it's existance so much that I'm running vrchat in VR (half quest 2 res) using the framework 13 (12th gen i7) out of spite

runs at 40fps, good enough for me

[nocturne olive]

What you sillying here?

[amber fractal]

Just a discussion

I can't really silly when I'm not at home

[nocturne olive]

Ver sil

Meanwhile I guess I'm playing HL2 for the 3rd time but this time to learn some cool speedrun stuff

[amber fractal]

[nocturne olive]

See this funny gap? I jumped over without a car

[hearty notch]

damn nostalgic fkn screenshot to see

[nocturne olive]

HL2 holds up insanely well even today

And it doesn't run like garbage with whatever DLSS framegen garbage

This is what games should be like

• runs on potato ✅
• looks visually good ✅
• gameplay works ✅

[amber fractal]

another point to stylization is better than realistic

[rigid snow]

i'm sorry but if you think hl2 looks good by todays standards you might want to go get your eyes checked

or play some aaa 3d games released later than 2008

[rigid snow]

also what

hl is realistic

[amber fractal]

I'd consider the screenshot as stylized in this era of more detail than reality era. It's a bad take but it is my opinion
-# My eyes are already checked and I do need glasses

[rigid snow]

that's, like, the whole point for hl2 in particular

and probably hl as a whole

[jagged turtle]

👍 thanks microsoft

[amber fractal]

I'm not denying that, I'm just thinking if a recently released game did something similar it'd be considered stylized imo.

It wouldn't be good, but it would be a choice the developer made

[opaque sigil]

Idk to me this looks like every other game that tries to be realistic

Albeit doing a shitty job in comparison because 20yo game

[amber fractal]

agree to disagree

[opaque sigil]

Fair enough

[olive sable]

😭 dont set the bar too high

I cant do much better than hl2

[rigid snow]

you can

that game is made for ancient hardware

pbr not a thing yet

blah blah

[olive sable]

We're doing npr

Anime shaders

[rigid snow]

then why are you comparing to fucking half-life sam

[olive sable]

In terms of physics and detailing

[rigid snow]

"looks"

[olive sable]

Im not getting anywhere close to the current anime games like genshin

[rigid snow]

hl2 is an awesome game

but it just objectively doesn't look good

by todays standards

[olive sable]

Hmmm

[rigid snow]

like visually

[olive sable]

Id say it looks fine

They do a fuckton with the limited processing power they're using

In terms of min-maxing its great even

[rigid snow]

it would NEVER release today as is visually, no matter by a studio large or small

[olive sable]

If we're comparing it to pc then sure it looks bad. But this could run on a phone easily now

[rigid snow]

no matter if their goal is to optimize or minmax or whatever

[olive sable]

Different context, but looks on par for phone games

[rigid snow]

honestly not even

[olive sable]

Eh

[rigid snow]

materials are shit shaders are shit

[olive sable]

I remember pubg looking about that quality

That was on a galaxy tab 2 tho actually. The graphics prob got upgraded by now

[rigid snow]

oh mobile

i was about to call you insane

i remember pubg mobile looking way worse than hl2

[olive sable]

I think the most jarring part of hl2 is the low poly models. The physics are decent, the lighting is good, there's AO.

[rigid snow]

it's consistent ye

[olive sable]

[rigid snow]

2016-18 mobile hardware could not run hl2 graphics

[olive sable]

hl2 shouldnt be the end goal for modern graphics at all, but its a decent checkpoint

[rigid snow]

it shouldn't be a checkpoint because the tech is so much different, you will just not have hl2 graphics using modern practices

[olive sable]

I mean

Hl2 isnt too different i feel

We just use more polygons and higher texel density textures in modern games as standart

Does hl2 use opengl?

Nah direct x???

[rigid snow]

dx6 or some shit

yeah

[olive sable]

I see

[rigid snow]

but source games also have a software rendering setting

oh and opengl too why did i forget

[olive sable]

Idk how dx (9???) Works so i cant fully say its the exact same

But i dont think texture mapping and gpu shaders changed too much

[rigid snow]

with the engine upgrades ye it would be 9 actually

[warped narwhal]

We gotta go back to the goldsrc days when games were really good

[rigid snow]

latest version officially ported to source

[olive sable]

Was gpu still fixed pipeline then?

Thats prob the biggest difference if it was

[warped narwhal]

Dx9 has some shaders depending on the tier, but everything prior is fixed function

[olive sable]

[rigid snow]

yeah so there’s your answer

[silent cloak]

ABH my beloved

[nocturne olive]

It's so peak truly

I need AFH though so I can see where I'm going at hyper speed

[patent shard]

My phone is pulling apart 😔

[tender river]

me miss this

look good wanna eat

[jagged turtle]

pseudo-css during hype train

[mighty thorn]

Do you come from the land down under?

[fiery dew]

Excuse me im new here what this chat for

[quick condor]

It's whisper

Vedal didn't set up a proper ram buffer it seems

[jagged turtle]

?

[rigid snow]

yeah llm output is whisper bru

tooootally it

[quick condor]

Whisper (or faster-whisper) is pretty much the only sane VTT program that can be used for something like Neuro, especially when dealing with something like discord. Discord individually streams everyone's audio. With 10+ people joining the call it triggered a hard catch.

Input not output

[rigid snow]

cool how does this relate

[quick condor]

The error occurred when everyone joined the call. The config wasn't set up for 10+ people in a call

Just explaining the picture

[rigid snow]

neuro does not see errors unless function calling but even that’s besides the point

[rigid snow]

your explanation makes absolutely no sense

what does whisper have to do with what she says

i mean, okay sure it can output some bullshit that influences her to say stuff

but why would the model say anything about the amount of people in a call?

[quick condor]

Because whisper doesn't handle stream overload cleanly. It drops valid data and halucinates

[rigid snow]

what the fuck are you on about

whisper is not aware of multiple speakers, it’s not a thing it was trained for

why would it say anything about the number of speakers

why css

nothing you said has any relation to the picture

an error is not model output

an error triggers the someone tell vedal

which it did

but that’s not what the pic says

the actual explanation is probably what i just mentioned happened, vedal said something about it and that prompted her to output the pseudo css

[quick condor]

Each time an error isn't caught properly when dealing with a multi model pipeline it cascades down. Rewatched it and he doesn't say anything that would have triggered it so it's likely

1. Discord api notifying names of people rapidly joining
2. VTT stream buffering related.
3. No VTT buffering system in place and max beams was default.
4. No catch to handle multiple output streams from whisper (very unlikely)

[past hull]

print(‘Hello World!’)

[jagged turtle]

"yeah that's expected, you're not designed to handle that many people" idk man

[quick condor]

Whisper can handle multiple streams at the same time. It's actually more efficient when doing so but you have to maintain the Metadata (eg discord provides Metadata about the user ID attached to the stream. You have to maintain which stream and corresponding output that is going into whisper to properly reinject the speaker name to the model

[jagged turtle]

sure seems related to me

[rigid snow]

that’s cool and all but once again

[quick condor]

Yes it related but it fundamentally doesn't make sense from a backend perspective

[rigid snow]

yeah whatever gn

this is fucking stupid

[jagged turtle]

what backend perspective do I need to understand this lmfao, you can pretty damn well tell she is doing a joke

[jagged turtle]

agreed

[quick condor]

audio_buffer: deque = field(default_factory=lambda: deque(maxlen=1000))
Found it, Not saying its the exact problem but its the best logical guess. If hit with that many streams, whisper will just start dropping packets and that can lead to a garbled output.

[visual pawn]

was this before or after Vedal said that Neuro probably couldn't handle so many people in the call?

[quick condor]

right after

[visual pawn]

if it was after, it could be her just pulling out a relevant line rather than an actual error

I think either way, this depends entirely on how vedal set up the software stack, so we can't know for sure; he could've had whisper or equivalent send error output down the line or not

[quick condor]

I dont think it was intentionally whisper but it could have been the custom wrapping code. Generally there is zero reason to touch that specific line so if it was overloaded it would drop packets. As for if whisper was used... there isnt really a better gpu accelerated vtt

[visual pawn]

but you can't discount the possibility that it was Neuro-sama riffing off of what Vedal said right?

[quick condor]

It's possible but doesn't really fit unless she is pulling a memory which still doesn't fit. It's no different than the day before when evil output a dots_ocr special token

[rigid snow]

if you still think that the error somehow triggered this we know when she gets errors directly into her context because she loves to talk about them. like with twitch polls, something about client id when content too long. this (her complaining about the error) only happens with function calling. wrapping code failing triggers someone tell vedal there’s a problem with my ai. she doesn’t see the error. she only sees she said the phrase.

[mighty thorn]

(yes i know this is overfitting, it's intentional for this test)

[rigid snow]

she never talks about those in detail because she doesn’t see them

[quick condor]

They also aren't supposed to output special tokens, even if the try.

[rigid snow]

and this is just completely unrelated fym no different

ok? whisper??

[quick condor]

Dots_ocr is pre model. It's special stolen should be caught

[rigid snow]

this is actual ragebait holy

[jagged turtle]

"it's possible" have you considered that what is most recently inputted to the model might be the thing that influences the model's next output

[quick condor]

You got me I had no idea how context caching or dynamic prompting works

[rigid snow]

1. she does not see harness errors
2. whisper not aware of N of speakers therefore cannot output “N speakers” into neuros context

[jagged turtle]

^ + if there was an error thrown it is caught overall by vedal's inference system (probably, could also be the thing that connects the STT to neuro) and then surfaced back up in the form of someone tell vedal

[quick condor]

It did, 10 seconds earlier

[jagged turtle]

yeah and so why would that be the cause of Neuro spitting out pseudo-css

[rigid snow]

here’s what that looks like in neuros ctx:
Neuro: Someone tell Vedal there is a problem with my AI.

which is why she can say that completely unprompted and then be like “haha jk”

[visual pawn]

it doesn't have to be pulling from a memory (I assume you mean RAG?); making a reference to an error with too many people directly fits with responding to someone saying that she shouldn't be able to handle that many people

it's something I'd expect an LLM to say without any other context from anything else

well, I guess the context is basically "I am an AI, an AI can output errors, I'm being told that I can't handle so many people, I can say something that sounds like an error about there being too many people"

[quick condor]

More like this

2025-12-24 23:33:46,252 - INFO - LLM REQUEST - 2025-12-24T23:33:46.252960
2025-12-24 23:33:46,253 - INFO - ================================================================================
2025-12-24 23:33:46,253 - INFO - Model: XXX
2025-12-24 23:33:46,253 - INFO - Temperature: XXX, Max Tokens: XXX
2025-12-24 23:33:46,253 - INFO - ----------------------------------------
2025-12-24 23:33:46,253 - INFO - SYSTEM PROMPT:
2025-12-24 23:33:46,253 - INFO - XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX

2025-12-24 23:33:46,253 - INFO - ----------------------------------------
2025-12-24 23:33:46,253 - INFO - MEMORY CONTEXT: Yes
2025-12-24 23:33:46,253 - INFO -  
2025-12-24 23:33:46,253 - INFO -   
2025-12-24 23:33:46,253 - INFO -   - XXX 
2025-12-24 23:33:46,253 - INFO -   - XXX 
2025-12-24 23:33:46,253 - INFO -   - XXX 
2025-12-24 23:33:46,253 - INFO -   - XXX
2025-12-24 23:33:46,253 - INFO - ----------------------------------------
2025-12-24 23:33:46,253 - INFO - USER MESSAGE (add_to_history=True):
2025-12-24 23:33:46,253 - INFO - XXX
2025-12-24 23:33:46,253 - INFO - ----------------------------------------
2025-12-24 23:33:46,253 - INFO - FULL MESSAGES ARRAY (3 messages):
2025-12-24 23:33:46,253 - INFO -   [0] SYSTEM:
2025-12-24 23:33:46,253 - INFO -       XXX
2025-12-24 23:33:46,253 - INFO -       XXX
2025-12-24 23:33:46,253 - INFO -       XXX
2025-12-24 23:33:46,253 - INFO -       XXX
2025-12-24 23:33:46,253 - INFO -       XXX
2025-12-24 23:33:46,253 - INFO -       
2025-12-24 23:33:46,253 - INFO -       
2025-12-24 23:33:46,253 - INFO -       
2025-12-24 23:33:46,253 - INFO -       
2025-12-24 23:33:46,253 - INFO -       
2025-12-24 23:33:46,253 - INFO -       
2025-12-24 23:33:46,254 - INFO -   [1] SYSTEM:
2025-12-24 23:33:46,254 - INFO -       [Relevant memories from past conversations:]
2025-12-24 23:33:46,254 - INFO -       [XXX]
2025-12-24 23:33:46,254 - INFO -       
2025-12-24 23:33:46,254 - INFO -       - XXX
2025-12-24 23:33:46,254 - INFO -       - XXX
2025-12-24 23:33:46,254 - INFO -       - XXX
2025-12-24 23:33:46,254 - INFO -       - XXX
2025-12-24 23:33:46,254 - INFO -       - XXX
2025-12-24 23:33:46,254 - INFO -       - XXX
2025-12-24 23:33:46,254 - INFO -       
2025-12-24 23:33:46,254 - INFO -       [Current conversation:]
2025-12-24 23:33:46,254 - INFO -   [2] USER:
2025-12-24 23:33:46,254 - INFO -       XXXX
2025-12-24 23:33:46,254 - INFO - ----------------------------------------
2025-12-24 23:33:47,487 - INFO - ASSISTANT RESPONSE:
2025-12-24 23:33:47,487 - INFO - XXXXXXXXXX```

[rigid snow]

??????????

[quick condor]

thats what is actually sent to the model

[rigid snow]

oh my god there are words i want to use ill get permad for to describe what you think of vedal apparently

[mighty thorn]

DPO

[quick condor]

hes not writing a model from scratch. I dont think anything negative regarding him. Most of this isnt that complex

[rigid snow]

tell me WHY would bro bloat 50% of his context with the least token efficient timestamp format imaginable

appended to EACH entry

[quick condor]

Thats just me being lazy with a log

[rigid snow]

that’s just you saying shit for the sake of saying it

nothing works like this

unless this is supposed to be a log in which case that’s not what the model sees and isn’t relevant to what i said

[sage crag]

tortura

[amber fractal]

I'd take ms offsets compared to that

[visual pawn]

I'm not sure how relevant this is to whether the model sees the context of an error from the other end of a pipeline

[amber fractal]

and that I really consider stupid

[visual pawn]

and also frankly, the twins have claimed things about themselves that can't possibly be how they actually work, so there's precedence

[mighty thorn]

There is intense ragebaiting occurring rn

[quick condor]

dots_ocr is obscure enough and fit the context well enough that that at least wasnt something they are smart enough to do

[jagged turtle]

dude I mentally tapped out a while ago why do I still care atp

[rigid snow]

i have nothing better to do

[quick condor]

same

[visual pawn]

wait, when did this become about outputting a dots_ocr token?

[jagged turtle]

since like 500 or so minutes ago idefk anymore

[visual pawn]

isn't that a pretty different claim from them reporting their own error?

[jagged turtle]

ask the person who brought it up lmao

[rigid snow]

we are being baited deal with it

[jagged turtle]

i dont see why ocr has anything to do with audio processing lmao

let alone this

[rigid snow]

i’m aware i am i just like arguing

[rigid snow]

they do drop vlm related tokens sometimes that is true

would have been huge if it was in any way relevant

based on this is it reasonable to make an assumption that their base models have an image modality?

[visual pawn]

I'd have to see how they dropped it but could it be forwarded from a separate VLM?

[quick condor]

<s>[SYSTEM_PROMPT]XXX
<THINKING>XXX</THINKING>
<MEMORY_QUERIES>
XXX
XXX
</MEMORY_QUERIES>
<UTTERANCE>XXX</UTTERANCE>
<READY>XXX</READY>
CRITICAL RULES:
1. XXX
2. XXX
3. XXX
4. XXX
5. XXX
UTTERANCE GUIDELINES:
- XXX
- XXX
- XXX
- XXX 
What memories should be searched? Output the tags.
Context: Already have: - XXX[/INST]
23:54:38-510932 INFO     Output generated in 5.54 seconds (38.05 tokens/s,                          100 tokens, context 423, seed 867017211)           
23:54:38-536714 INFO     STOPPING_STRINGS=                                  
['</s>', '[/INST]']
23:54:42-824727 INFO     Output generated in 4.28 seconds (21.97 tokens/s,  
                         94 tokens, context 481, seed 549546900)            
23:54:42-850435 INFO     STOPPING_STRINGS=                                  
['</s>', '[/INST]']
23:54:42-858950 INFO     PROMPT=                                            
<s>[SYSTEM_PROMPT]XXX
[Current conversation:][/SYSTEM_PROMPT][INST]XXX[/INST]Hello. My day has been quiet. It's been nice to have some time to settle into things. How about yours?</s>[INST]XXX[/INST]
23:54:44-847353 INFO     Output generated in 1.99 seconds (16.61 tokens/s,  
                         33 tokens, context 724, seed 1525060346)   ```
Better

[visual pawn]

I'm still failing to see how this is relevant to anything

[rigid snow]

that’s still stuff in context that’s irrelevant to the model

which no one does i hope

[visual pawn]

what do you mean, the model has to know the seed right? /s

[quick condor]

you said "thats all she sees in relation to the error but you dont know that

[rigid snow]

yes the fuck i do and i explained why

i have watched thousands of hours of neuro

[visual pawn]

how is this demonstrating that she can see the error?

[rigid snow]

she never ever talks about harness errors

ne ver

only function calling ones

[quick condor]

I mean it took about 15 minutes of watching her alone with chat to determine the model family, more hours doesnt mean you know everything

[visual pawn]

you're 100% confident about the model family?

[quick condor]

Im about 90% confident based on refusal method

[rigid snow]

i thought that was a wild statement too ye lol

[quick condor]

they have traits and it depends on where in reasoning the base guards are injected

[rigid snow]

[visual pawn]

I'll actually humour it, which do you think it is?

[sage crag]

🎤

[rigid snow]

atorin emote

[sage crag]

⚰️

[visual pawn]

going with the mainstream bet of LLaMa or a wildcard?

[rigid snow]

i know the model family

it’s a language model guys

[visual pawn]

maybe it's a vision language model

maybe it's a vision language action model

[quick condor]

she soft refuses after reasoning and attempts to change topic. I can say its not a gpt, claude or deepseek model.
95% chance its not mistral (refusals just drop)
Most likely a llama3.3 or less likely Qwen

[rigid snow]

noooo

[visual pawn]

maybe it's a vision language action memes minecraft model

well, not GPT or deepseek were already safe bets

just based on timeline and API requirement

[quick condor]

it was down to mistral or llama before with low prob qwen but the refusal method ruled out mistral

[visual pawn]

frankly I don't disagree with the guess; it's just I wouldn't bet a ton of money on it

[sage crag]

ragebaiting whole chat

[rigid snow]

deepseeks models were still too shit when neuro already was a thing. gpt only recently had an open weights model that’s not gpt2 which is just not it. anthropic has never had an open weights model

[visual pawn]

there's a difference between I figure it's probably Llama 3.3 or Qwen versus I absolutely know what the model family is

[rigid snow]

qwen is relatively new too

[jagged turtle]

I assume vedal has not upgraded the llm behind neuro since he started?

[visual pawn]

yeah timelines usually lead people to LlaMa

[visual pawn]

for sure he has compared to the first iterations

[rigid snow]

oh he FOR SURE has that’s a wild thing to assume

[visual pawn]

the first streams were like

GPT-J or even GPT-Neo level

[quick condor]

Hes 100% finetuned, in theory he could but its risky

[visual pawn]

the chicken nugget era

[visual pawn]

a lot of effort for little gain imo

[quick condor]

as long as he has the original training data and the training configs he can upgrade. Its a slow process and I agree generally not worth it

[trim valve]

neuro is a Markov chain

[rigid snow]

this is consensual we already talked about this

[sage crag]

[visual pawn]

that said, I think it's possible he hasn't upgraded in quite a while

I remember there being one or two huge jumps over the years

[jagged turtle]

yeah, idt my brain thought well for that assumption

[quick condor]

First-person self boundaries, action focused prohibition, and firm redirection instead of moralization -> execution gated
Qwen is too literal and mistral too blunt

[jagged turtle]

how did we get to this point

[quick condor]

given the quantity of training data he gets hes free to use stupid low training rates

[visual pawn]

well, Discord released in 2015...

[sage crag]

popopopopopopo

[rigid snow]

given the amount of money he gets he’s even probably free to pretrain

[sage crag]

e nub

[quick condor]

pretrain is risky

[sage crag]

electronic nub

🌩️

[rigid snow]

not electric

[sage crag]

[jagged turtle]

?

[sage crag]

nubs

[visual pawn]

"first the big bang happened" joke

[quick condor]

from what I've gathered that doesn't seem like something he'd do with the possible downsides. Small improvements and sub models that run in the background for improving memory recall are safer to mess with as long as they dont touch the original transcripts

[sage crag]

[jagged turtle]

ah

[sage crag]

nnnnnub

wrrr

nub mcmeal

[quick condor]

for the record, I also think his evasiveness to neuro's question line recently is also not as simple of an answer as chat seems to think. Embedding a firm no in her memories (or a firm yes) isnt a good idea with the current state of tech and erasing it isn't super simple if he does because he would both have to edit the memories (questionable) and make sure it doesnt leak back in from chat.

For now, "no answer" is safe

[rigid snow]

ok now i’m lost too

sam would probably combust if he reads this

[quick condor]

The other thing that I dont think people seem to understand is the base model realistically doesnt matter. Its actually a very small segment of the stack

[rigid snow]

yeah idk about that

[quick condor]

There would be some changes for sure if he dropped an off the shelf model in but the memory storage recall and context embedding, tool call features, and system prompt generation methods are all custom. She would be off with a different model but in a "something is wrong" kinda way.

The model is what helps with the fine touches that connects everything. It wouldnt survive without the stack

but her model has 100% had better/more tuning than evil's... IDK how to explain but evil's feels rough.

It would be interesting to see what would happen if he mixed up the models though. I cant see that ever happening in a way we would see but I can dream

[visual pawn]

evil and neuro? maybe for april fool's

I'm sure I've heard that idea floated around

maybe from him back when he was saying they were identical

[quick condor]

It's interesting but I'm sure especially this deep it would feel wrong.

Like I'd like to see it as an outside observer but I don't think I would do it if it was my call.

[young plover]

He mentioned on Ellie's stream once that transferring the personality is a complicated process, but naturally gave 0 technical clues.

[quick condor]

I mean from a technical standpoint it shouldn't be. But it's more of a complex subject when it comes to the more meta side of the code. If they don't keep the memories of it happening then it's kinda pointless, but if they do who keeps the memory.

Would neuro's model with evils past memorys go to neuro's memory afterwards or would evil remember it? Or would they both remember both?

I'm assuming that there is three general components: personality, model, and memory. The question isn't so much how to technically do it and more what to swap

With no coupling there is 8 permutations total, 2 core and 6 permutations and it gets messy quickly. I don't think the UK has enough alcohol for that

[tender river]

simd

wwwwwwwwrrrrrrrrrrrrrrrr

[sage crag]

[tender river]

[rigid snow]

smug instruction mug data

[tender river]

= simd

true!

cups

parallel

[rigid snow]

[sage crag]

[sage crag]

mugs instruction mug data

(mimd)

[tender river]

single instruction no data

my brain

[sage crag]

sin'd

[tender river]

[sage crag]

no christmas presents

[tender river]

[sage crag]

i got

socks

[tender river]

russian christmas celebrated in january purely religious holiday

we have

new year

[sage crag]

i also got

a book by alice roberts

[tender river]

anthropology

me like

evelyn reed

[sage crag]

:ALERT:

[rigid snow]

uuh

[tender river]

:ALERT:

emoji id start with 123456

fancy

[rigid snow]

!vedalUUH

[sage crag]

[rigid snow]

grilled cheese with freshly ground black pepper

[olive sable]

goodmorning

[olive sable]

i got an external ssd nvme enclosure, and a hdmi cable for christmass

exactly what i bought for myself and then handed off to my family so they could give it to me

how do i move my linux to a new ssd

[sage crag]

copy paste

only home folder

@tender river ee

[olive sable]

only home folder cuz i need to install the os itself again cuz proper drivers and everything?

[sage crag]

yes

[olive sable]

[sage crag]

this is one of the reasons people like nixos

its trivially easy to reinstall the system part if you use it

[olive sable]

fair enough

i think config files are jsut too much work for me tho

[sage crag]

nixos not config file

[tender river]

technically you can just move the files and only reinstall the bootloader easier just install from scratch

[sage crag]

it program

[olive sable]

i change my cachyos config maybe once in 3 months at most

[sage crag]

but the problem is

when you want to do exactly what youre doing

you have to redo everything from memory

[olive sable]

my memory is my worst part of my brain

[sage crag]

https://nixos.wiki/wiki/NixOS_Installation_Guide
read this and see if it interests you

[tender river]

wrong wiki

http://wiki.nixos.org/wiki/NixOS_Installation_Guide

[sage crag]

not my fault SEO failure

[tender river]

Your ISP (and/or VPN provider) has been banned from nixos.wiki due to ongoing attacks originating from their IP space.

To regain access to the wiki, you can:

• Use another internet service provider.
• Contact your ISP and urge them to take action that will result in a stop of the attacks.

[sage crag]

i used that first guide when installing

[sage crag]

[olive sable]

that jsut explains me how i need to instasll it tho

i odn even know if i want to install it

[sage crag]

https://github.com/Misterio77/nix-starter-configs
take a look at this then

my nix config looks nothing like these

[tender river]

not a bad template i guess

[sage crag]

mine is much simpler

[tender river]

i ended up adding similar things independently

[rigid snow]

ok regret

heartburn speedrun

[tender river]

but it

doesnt showcase nixos at all

[sage crag]

[hallow flume]

yo everyone, is there gonna be a game jam for this subathon?

[tender river]

❤️‍🔥

[rigid snow]

heart burn

[sage crag]

❤️‍🔥

[tender river]

toasted heart

yum

[olive sable]

no. but we're planning on maybe doing an unnoficial one.

#1336984748882399242

[sage crag]

esl debuff

[hallow flume]

I see thank ya!

[sage crag]

why is shiro shanta and not cheero

[olive sable]

?

[sage crag]

[tender river]

it heart burn

smh

[rigid snow]

why is it called heartburn actually

that is stupid

[sage crag]

unofficial

[olive sable]

uh

i dont see where im worng

[rigid snow]

it not in heart

[sage crag]

unnoficial
unofficial

[rigid snow]

or near heart

[sage crag]

sam van "dyslexia" maele

[olive sable]

[sage crag]

[tender river]

medieval doctors

[olive sable]

konii grammar police lg crt tv

[rigid snow]

upper torso = heart

[olive sable]

the time cas come to burn your heart

[sage crag]

why is a cold called a cold when you ove in the hot temperature

[olive sable]

[rigid snow]

you feel cold

[sage crag]

i dont i just get sneeze

[rigid snow]

when you have cold

[olive sable]

i thought its a cold cuz you get it from being in the cold

[rigid snow]

or that

[sage crag]

flu

[rigid snow]

🕊️

it flu

[sage crag]

no

it dove

[rigid snow]

LMAO

[sage crag]

[rigid snow]

it flu but then it dove

[sage crag]

historians will argue

[rigid snow]

it alternate history

i made up

[sage crag]

[tender river]

follow wikipedia source

Given the location of the pain
and the concept of cardiac dominance in medicine, it was assumed that the origin
of the discomfort was the heart and not the esophagus
useless

[olive sable]

i dutch we say verkoudheid, which is basicly (prefix that signifies change change)coldness. an di think thats beautifull

[sage crag]

seem to go back to idea of four humours

maybe that where the name cold go back too as well

[olive sable]

Before germ theory, people thought prolonged exposure to cold, wet conditions directly caused colds, leading to the name.

i was right

[sage crag]

phlegm was associated with an excess of cold

[rigid snow]

nerd nerd nerd

[sage crag]

which means symptom of cold(ill) were associated with cold(temp)

[tender river]

[sage crag]

i studied history of medicine

[olive sable]

konii has the top 2 requirements of nerd, but what about the obsession requirement?

[sage crag]

[olive sable]

why?

[rigid snow]

too much stuff obsessed with > illusion of lack of obsession

[tender river]

programming

[olive sable]

damn

you're right

[sage crag]

autismu

[sage crag]

fun

[olive sable]

what are "overlays on exports"?

for nix

[sage crag]

overlays are thing you dont have to think about, and ignore

[rigid snow]

why is sam learning nix

[sage crag]

[sage crag]

<-

[tender river]

wtf do they mean by exports

[olive sable]

im gonne install linux on my new 990pro, and they're trying to indoctrenate me

[sage crag]

i dont know

it confused me

[olive sable]

imm abe honest here. i have no fucking clue

[ivory plinth]

[trim valve]

[sage crag]

che

[olive sable]

hi bred

[trim valve]

hi

[olive sable]

as a child, were you dogh?

[rigid snow]

nixos for unemployed hypernerd spend 20 hours a week write config file to make trivial thing work

just install nix if need nix

[sage crag]

esl debuff

[rigid snow]

dogh

[trim valve]

[olive sable]

i purpusefully removed the u in dough cuz bred is bread without the a

[trim valve]

I think I made bread once or twice from scratch

so surely that counts

[sage crag]

make it again

[trim valve]

hm

[olive sable]

i want to give it a chance.
indoctrenation is fun

[trim valve]

might be fun

[trim valve]

ime nix is nice

other than tailscale

grr

[olive sable]

what is a tail scale?

[rigid snow]

nix is fine

[sage crag]

what point of installing nix if you dont nixos

[rigid snow]

development environment

[trim valve]

VPN that connects your devices together

[sage crag]

@tender river ples confer

[sage crag]

bad

[olive sable]

i dont think i need that

[trim valve]

I use it a fair bit

[rigid snow]

would you rather use docker? because that’s what happens in actual corpa

docker container as dev env

i wish this was a joke

[sage crag]

nevermind

[olive sable]

how good is nix with gaming performance?

[sage crag]

same as all other linux

though for me it was better than cachyos

[olive sable]

uh

there are differences

[trim valve]

you can always swap out to another kernel iirc

[sage crag]

only between bleeding edge and stable kernels really

[olive sable]

what is a flake.nix?

[sage crag]

flakes is newer alternative to nix channels

[rigid snow]

erm just watch a tutorial maybe

or something

if that exist

[sage crag]

basically makes it easier to version control

[olive sable]

uh

[tender river]

uuuh
nix dev env
home-manager
ye i've got nothing

[olive sable]

i odnt think i care much about version control

[tender river]

you can think of flakes as packages

[sage crag]

it makes it more likely for your system to work perfectly on reinstall

flakes are less messy in general

[olive sable]

why wouldnt it tho? and why would i even reinstall?

[tender river]

you are doing it right now

[sage crag]

[olive sable]

nuh uh

[sage crag]

ye

[olive sable]

or well, i guess i am

[sage crag]

nix config = can copy to other system have same setup

flake make it easier to copy without random change

[olive sable]

[sage crag]

flake have lockfile

[olive sable]

but i could jsut do the same to my cachyos config tho

[sage crag]

you cant

[tender river]

os is not just whats in /etc its also what you have installed etc

and stuff in /etc can randomly change without you even noticing

[jagged turtle]

Why are these not aligned properly

[sage crag]

also if you copy etc from one system to another its pretty much guaranteed to not work

[jagged turtle]

I can't unsee this now

[ivory plinth]

[olive sable]

how cumbersome is needing to put stuff in the config to use something?

[tender river]

same effort for me

[sage crag]

not particularly

[trim valve]

it depends on the package

[sage crag]

if theres a nixpkg its easy

if theres not, you have to write one

[trim valve]

but only because there's less magic involved

[ivory plinth]

https://media.tenor.com/RRpwGVODEIAAAAAM/me-boy-white-shirt-boy-ne.gif

[sage crag]

nixpkgs is huge though

[trim valve]

😔

[olive sable]

can i use hyprland with nix?

[trim valve]

still no openfoam on nixpkgs

[sage crag]

yes, easily

[trim valve]

though in my friends experience it can get messy

[ivory plinth]

i need custom build thing
me need make nixpkg if want to use nixos

[tender river]

surely maintainers wont leave their packages unmaintained for years

[olive sable]

uh

[tender river]

this my last commit

[sage crag]

https://wiki.nixos.org/wiki/Hyprland

[sage crag]

need to send email complaint to ms pavyluk

[olive sable]

nvidia hardware acceleration?

[sage crag]

yes

[tender river]

** **s/yl/ly/

[sage crag]

oops

[tender river]

ukrainian surname

[sage crag]

first name easier to remember

commutator of naanbread

[tender river]

bread os

nix breads

[sage crag]

i would dislike having my name public

cant contribute to nixos

[rigid snow]

ye why are we discussing legal names

[sage crag]

we arent

[trim valve]

illegal names

[sage crag]

only public names different

[olive sable]

my real name is sam but you can call me sam

🔺

[sage crag]

are you sure

[olive sable]

no

[rigid snow]

my name probably not secret but no reason to mention

[olive sable]

you specifically konii

you can call me 🔺

[sage crag]

what is your legal full name, phone number, social security number, bank card details

[olive sable]

uh

[sage crag]

[olive sable]

idk

i know the first 2

[sage crag]

oh ok

nevermind then

[olive sable]

you already have my name and i odnt care if you have my phone number

[sage crag]

i dont want it

[olive sable]

ye i thought so

[sage crag]

if you send it i will sign you up for 20 newsletters

[olive sable]

[rigid snow]

sms newsletter

[olive sable]

daily verse form the bible

single nixos configs in your area

[ivory plinth]

john john, 7355808, 01101110, 01101111

[olive sable]

mr hohn john, it looks like you were born on 10 november in the year 110

[ivory plinth]

[sage crag]

[olive sable]

what if i thing i want isnt on nixos?

[rigid snow]

refer to this

[olive sable]

oh

[tender river]

you can alternatively

use distrobox or steam-run

steam-run creates a fhs-compatible environment for running programs

[sage crag]

basically

steam-run aprogram

it work

[tender river]

and distrobox = just another distro in a box

[olive sable]

so uhm, what is fhs

[rigid snow]

you can alternatively

not run nixos

[tender river]

guix

[rigid snow]

no

[sage crag]

filesystem hierarchy standard

[tender river]

[sage crag]

leaf not expand acronym because expanded version not any more understandable

[tender river]

fhs thing that say

program in /usr/bin

[olive sable]

i see

[tender river]

nixos say i dont care about fhs program in /nix/store

program compiled for other distro care about fhs

steam-run pretend nixos have fhs can run game and other stuff

[sick owl]

HOLY

https://www.cnbc.com/2025/12/24/nvidia-buying-ai-chip-startup-groq-for-about-20-billion-biggest-deal.html

Guess the competition spooked them

[rigid snow]

we already talk and more news since then

[sage crag]

why they call groq

[sick owl]

Been around since before Musks nonsense

[olive sable]

according to google, nixos is slower than cachyos

[tender river]

grok favorite word of silicon valley vc types

[sick owl]

CachyOS is the gold standard in linux gaming imo

[sage crag]

you believe everything you read?

[rigid snow]

it’s mainly a poaching thing nvidia just wanted the ceo groq stays independent

new ceo

[olive sable]

everything you have told me so far is also things i read.

[rigid snow]

and tech of course

[sage crag]

touche

[sick owl]

Is it not gonna be like windsurf where the company is left as a husk with its top talent bought off?

[rigid snow]

maybe?

[sage crag]

the answer is no, nixos is not meaningfully slower than cachyos

and in fact, no rolling release linux distro is

[sick owl]

Losing their tech moats gotta be bad news too

[olive sable]

[sage crag]

in particular, my frames are better on nixos than they ever were on cachyos

because i have more control over the garbage that runs

[olive sable]

does rebuilding take long?

[tender river]

around 10-20s

[olive sable]

oh thats nothing

[tender river]

i'd say it's pretty bad but still perfectly usable

[olive sable]

is bluetooth still as weird as its on arch?

[tender river]

https://wiki.nixos.org/wiki/Bluetooth

[olive sable]

oh

its literally bluetoothctl again

[tender river]

or, if you read carefully

[warped narwhal]

I've never understood how people have had problems with Bluetooth on Linux

All I had to do was install bluez, start the service and then everything connected fine

[tender river]

it list gui/tui option before explaining the cli

[olive sable]

oooh

blueman?

[olive sable]

if i use my headphones on windows, they become completely unpaired in cachyos for me
bluetoothctl goes "ive never seen these headphones in my entire life"

which is bad and sucks

[tender river]

ye that solvable https://wiki.archlinux.org/title/Bluetooth#Dual_boot_pairing

[silent ether]

https://x.com/lundukejournal/status/2003293402528428209

Translation: You have until 2030 to switch to Linux

The Lunduke Journal (@LundukeJournal)

Microsoft has set a goal to “eliminate every line of C and C++ from Microsoft by 2030.”

What are they going to try to replace that C & C++ code with?

You guessed it. Rust.

And they’re going to use AI to do the “Rust re-write” at an insane speed.

“Our strategy is to combine

X · Dec 23, 2025, 2:35 AM

[tender river]

lunduke

[silent ether]

Good decision by Nvidia, Groq is way better at inference than their current GPUs

[olive sable]

rust, sure its a decent language. but that they use ai to do it
surely rust is a safe language that the ai can make no mistakes in

[silent ether]

Yeah the biggest issue is that they’re planning to rewrite it in Rust using AI

[olive sable]

does nixos not replace the old verison witht he new version?

like, if i update blender would i suddendly have 2 blenders?

cuz i think that would eat a lot of storage

sure its a 2tb drive, but i dont want to use all my storage

[sick owl]

The linux kernel is also integrating Rust now though

[tender river]

theres garbage collection that you can run manually or on a timer

and theres hardlinking which deduplicates files so new version reuses files of the old one

[sick owl]

Oh

Oh the AI bit

Yeah that aint ideal for a kernel

[silent ether]

Yeah Microsoft is basically trying to paste their entire codebase into ChatGPT and ask it to rewrite in Rust lol

[sick owl]

Oh also merry christmas nerds

[silent ether]

Merry Christmas!

[tender river]

at the end ye nix not perfect for storage constrained setups, my router has 8gb which is barely enough for having two nixos versions side by side

[rough bloom]

well no but actually yes
you have "generations" of your system, each of which is a version of your system that you built with nixos-rebuild
but everything that's shared with other generations is only stored once, it's not like your entire system actually gets copied
and once you no longer need the old generation you can garbage collect it to free up the storage

[tender river]

mhm it uses symlinks

[rigid snow]

if i speak

bye bye server

[olive sable]

ok so

whtr im hearing is

nixos is as fast, and as configurable as cachyos

but you get the joy of banging your head against a config file even more?

[obsidian mantle]

Do they have 1 engineer left or what. Why do all this shit

Whats wrong with microsoft

[rigid snow]

it’s very unbelievable and unlikely they’re gonna do it

[obsidian mantle]

Ah

[tender river]

me use same config for months, but ye reason nixos makes you spend more time on complicated config is because you know it wont disappear after a reinstall

[rigid snow]

every line? like bro this kernel is 30 years of cpp code

yeah i’m sure we can just riir and it’s gonna work

exactly the same

[tender river]

its an r&d position not a "we're doing it right now" one

[stray dragon]

i wouldn't even put it past them at this point, every company seems to be racing to make things as awful for the user as possible

[tender river]

or at least i don't expect microsoft to be that badly mismanaged

[olive sable]

but wouldn't you need to rebuild everytime you install a new app?
sure i havent installed many new apps recently, but i feel like the frequency of new apps is like exponential decay where you install a lot more at the start

[tender river]

exactly, i dont really need to install anything in my daily life

if i want to try out a program i can run it in nix-shell

no need to add it to my system/home-manager config

[olive sable]

hmmm

[tender river]

but if i do

its a one line change

same effort as running paru -S ...

[rigid snow]

[olive sable]

and i can still use my hyprland configs for stuff like waybar too?

[rigid snow]

time to change battery

[hard raptor]

meow

[tender river]

might have to port it a bit since nixos likes structured config should be mostly transferrable tho

[olive sable]

meow

so i can use this thing still. except for that screenshot stuff?

[tender river]

read this https://wiki.nixos.org/wiki/Hyprland

[jagged turtle]

168% usage

[olive sable]

just means it got charged

[rigid snow]

ye

[tender river]

tl;dr: yes, if you dont use home-manager, it will just use the same config files as it does on other distros

[trim valve]

one dude experimenting btw

not to mention lunduke slop

[olive sable]

i might give this nixos stuff a chance i think

i will keep this cachy install on this sata ssd tho just to be sure

[sage crag]

[olive sable]

you have succesfully indoctrenated me

cangrotas

[sage crag]

mission complete

[olive sable]

what the fuck did i type here

[rigid snow]

sounds fun

[olive sable]

single user or multi user?

[trim valve]

huh

hasn't the default been multi user for ages

[olive sable]

idk

[rough bloom]

you're not installing Nix, you want NixOS
you don't get a choice I think, it's always multi-user

so yes, multi-user

[olive sable]

oh

[rough bloom]

single-user is only for systems that can't do multi-user for some reason AFAIK

[olive sable]

i had to scroll down

https://nixos.org/download/ has the package manager as the first thing you see for some reason

[tender river]

ye

it same thing

but not take your family hostage

[olive sable]

uh

sure i guess

[ivory plinth]

[sage crag]

[silent cloak]

Merry Christmas

[olive sable]

Merry chrsitmass

[hearty notch]

MERRY CHRISTMAS 🎄

[silent cloak]

https://tenor.com/view/sleigh-my-name-walter-white-breaking-bad-jolly-jollyposting-gif-4288259068272680801

[olive sable]

the payload has been activated

imma eat first tho

[trim valve]

[sage crag]

graphical install

[olive sable]

well you see

[rigid snow]

jraphical

[olive sable]

i want to see what im doing

[trim valve]

is that sneakily a ventoy usb

[rigid snow]

giraffeical

[trim valve]

or did you just put the iso on a usb stick

[olive sable]

a what?

[ivory plinth]

geographical

[sage crag]

ventoy is a multiboot usb program

[olive sable]

oh

it shoudlnt be

[sage crag]

it allows you to have multiple isos on the same usb

[olive sable]

i have multiboot thing somewhere else on my pc

[sage crag]

you cant just put an iso on a usb

[rigid snow]

kni what happen you used to sound less like nerd

[sage crag]

you need to flash the usb

[olive sable]

what

but i odnt want to lose my usb files

[sage crag]

me listen to nerd audioboook

make me feel smart

[rigid snow]

make you look dork

[sage crag]

wrr

[olive sable]

true

[sage crag]

you dork

[olive sable]

ngl

[sage crag]

[olive sable]

sorry konii

[sage crag]

me normal sound normal

[olive sable]

it feels like talking to someone who maxxed out their intelligence but has brain damage in the language part of the brain.
and you even grammer check me a lot

[sage crag]

[rigid snow]

ok this is rude kinda

[olive sable]

oh

sorry if it came across as rude

[sage crag]

perpetuate harmful stereotype

[rigid snow]

my bad g you cool

[sage crag]

me sleepy

[olive sable]

its noon

[rigid snow]

omg same i haven;'t

slept since when i said i slept for 6 or 7

[sage crag]

carr

carrrr

sleepy

yrhreh

[olive sable]

ok now you're jsut doing it on purpose

[sage crag]

key spam

[rigid snow]

sound like something bird would say

are you raven tv hybrid

[sage crag]

me

sleepy tv

[rigid snow]

hybird

[sage crag]

feel motion sick

bwrr

[olive sable]

[rigid snow]

are you living in a self driving rv

[sage crag]

bye me close eyes now

[olive sable]

[rigid snow]

good afternoon?

[olive sable]

it should be 12:10 there rn

so i guess

depends on what afternoon means where you live

[rigid snow]

You set a personal best for Deaths Without Firing a Shot (6). Get your full stats for the match here

[olive sable]

[rigid snow]

[olive sable]

hmm

i havent been too tolerant this last week

need to fix that

[nocturne olive]

Wa

[olive sable]

wa

[nocturne olive]

You're switching to NixOS?

[umbral thorn]

[nocturne olive]

Elvy spotted

[sage crag]

wrr

[still garden]

Merry christmas programming chat

[olive sable]

one of the christmas edits of all time

[opaque sigil]

Another one down the rabbit hole

[onyx scarab]

Merry Christmas to all our resident femboys🎄

[rigid snow]

the ending card got me

powerpoint ahh transition

[sage crag]

bwwrrr

bwrr

[obsidian mantle]

what happened konii

[midnight sigil]

what I want for christmas

[silent cloak]

Got a new 3D printer

Now I can work on more projects

[midnight sigil]

I wish I have a 3D printer

I would print neuro lul

[uneven pulsar]

i'd print more ram

[midnight sigil]

bro copium'd himself

@sage crag

[hearty notch]

Hey guys i want to wish you all a merry christmas. Discord is full of copypasted christmas messages, that people don't even read and just forward to other servers. I don't like that. I like to write what i deeply wish and what comes from my heart. Our friendships, from the most deep ones to the most virtual ones are very important and will never be represented by a simple message copied from elsewhere. This being said I would like to thank all of you. You are the best hockey team i've ever played with. A big hug from me, fluffy rabbit.

[stiff trout]

Question so how often does Evil have a stroke?

https://www.youtube.com/watch?v=vix7cJ8FVr0

not sure how asking evil to say OK helps

[hearty notch]

not often and the strat was to dilute context

[nocturne olive]

It was a bug

The bug no longer exists

[stiff trout]

Sorry but mind explaining dilute context?

vedal went over her context limit?

[hearty notch]

it was a one off thing and i dont konw if the strat actually became relevant but the idea is if you induce output that doesnt have the corruption often enough it lowers the probability it recurs

[stiff trout]

oh if vedal breaks the loop long enough they get back on track?

[hearty notch]

something like that but it never happened really before or after that

arguablylike neuro long ago repeating random things over and over and evil repeating random things over and over are like