programming

[sage crag]

it is a vegetable and a fruit

[olive sable]

sure, i guess im both

[sage crag]

✅

bananas arent actually fruit though

[olive sable]

[fiery anchor]

fruit flies like a banana

[olive sable]

Yes, a banana is both a fruit and, specifically, a botanical berry

a berry is a fruit no?

[sage crag]

nop

ever seen

a fruit that looks like a banana

excluding bananas (which arent fruit)?

[olive sable]

uhhh

[leaden crest]

starfish

[sage crag]

thats what i thought

[leaden crest]

get destroyed

[sage crag]

starfish is a fruit

ohisee

[leaden crest]

also plantain

[sage crag]

planes are not fruit

[olive sable]

all berries are fruit, but not all fruit are berries

[sage crag]

strawberries arent berries

[olive sable]

strawberries arent berries, but they're fruit

[olive sable]

yellow fingerlimes

[leaden crest]

what the actual https://www.thetakeout.com/1591578/banana-passionfruit-banned-explained/

[olive sable]

banana passionfruit

[leaden crest]

what

this is a thing

[olive sable]

"illegal"

banana mangos

[tender river]

wrr

[tender river]

wrrr

[olive sable]

wrrrrrr

[tender river]

@real sierra congratulations on your marriage

[olive sable]

[leaden crest]

(/s on the kiwi)

[real sierra]

i am not married

[olive sable]

@real sierra spill the tea

[obsidian mantle]

why is there a kiwi inside of banana

[tender river]

divorced so fast

[real sierra]

oh i forgot to answer really the topic moved on too fast

yeah not a lot has changed

[olive sable]

you said so already

[leaden crest]

you said the same thing

[olive sable]

shiro has alzheimers

my condolences

[real sierra]

well yes im going to expand on it but i need to reiterate the context for the rest of the paragraph to make sense

[leaden crest]

he can never remember when his birthday is

[real sierra]

[leaden crest]

so we have to remind him constantly

it’s exhausting but we must do the duty to remind our beloved shiro

[real sierra]

there's less telephone to play to get things done

since i can just do them myself

[olive sable]

[sage crag]

shiro got married?

[real sierra]

NO

[sage crag]

when was the wedding

[olive sable]

lmao

[real sierra]

[sage crag]

o

wrrrr

[olive sable]

awawawawa

[sage crag]

ye

[leaden crest]

redston made it official

[sage crag]

shiro do you remember your vowels (not vows)

[tender river]

shiro my server offer is still available its just an overloaded single-board computer but it should at least somewhat suffice

[sage crag]

[leaden crest]

pics or it didn’t happen ❌
pics then it did happen ✅

[real sierra]

a e i o u with y in a quantum superposition

[olive sable]

i can host a server too btw

its not much, but its a pc

[real sierra]

[sage crag]

not false ❌
true ✅

[real sierra]

you're gonna laugh at my lack of linux knowledge

[leaden crest]

why cannnnnntttt you seeeeeee my viewwwwww

[real sierra]

and probably backseat me

but ill consider it once i have time to develop and test fumoquest again

[olive sable]

if you need help you knwo where the programming channel is

[tender river]

[olive sable]

mister admin

[tender river]

i would never

backseat

anyone

[sage crag]

shiro what does sudo rm -rf / --no-preserve-root do

[leaden crest]

where

[olive sable]

i lost it

shiro help

[warped narwhal]

I forgor

[leaden crest]

bakers can yall help me find it

[real sierra]

dont call me admin for the love of god it has the worst image associated with it

[olive sable]

[leaden crest]

happy admirthday

[olive sable]

shiradmino

[real sierra]

i did not spend all these years taking good care of my image as shiro just for an admin label to be stamped on my thick skull

[warped narwhal]

step-admin?

:3

[sage crag]

[olive sable]

no

[sage crag]

no

[real sierra]

that's the worst thing i've read in the last 6 months

[trim valve]

community peace keeper

[warped narwhal]

you're welcome

[olive sable]

[sage crag]

did you know that shiro is actually an anagram of admin

also if you spell shiro backwards its written as admin

[leaden crest]

s h i r o
s h o i r
a h o i n
a d m i n
checks out

[sage crag]

shoir is pronounced like choir btw

[olive sable]

shiro
remove the s
remove the h
add the a
remove the r
add the d
remove the o
add the m
add the n

[tender river]

did you know

[real sierra]

ok idk if id go that far to the other end of the reputation spectrum either

[tender river]

if you convert shiro to leetspeek

[real sierra]

i definitely cause my share of problems

[tender river]

it turns into shiro birthday

[glad path]

you've pinged vedal like 11 times

[sage crag]

5h170

[olive sable]

https://tenor.com/view/stare-compute-think-complicated-gif-10481635

[real sierra]

those arent even the ones i mean

[leaden crest]

mods ban shiro

[glad path]

still

[sage crag]

shwire

[olive sable]

that does not help

[sage crag]

o

[real sierra]

im talking about back in february 2023 when i got sad over a dispute with someone and left the server and like half of genchat followed me out

[sage crag]

sam dent

[tender river]

what

what

[olive sable]

the problem with englsih is that its pronounced however it is, and i dont know

[tender river]

is choir pronounced chwire

[leaden crest]

what

[sage crag]

no

like kwire

[leaden crest]

wait then how long has the shiro been in here for

[vague sonnet]

@tender river why did you untoast yourself

[real sierra]

jan 1 2023

[tender river]

[sage crag]

not like chore

[glad path]

yeah

more or less

[leaden crest]

ancient

[sage crag]

like kwire

[glad path]

kwire yeah

[leaden crest]

no wonder you have so many birthdays

[tender river]

felt like it

[olive sable]

is choir pronounced like the ch in chair, or the ch in charisma

[vague sonnet]

[glad path]

charisma

[real sierra]

character

[vague sonnet]

kwayear

[glad path]

kwyre

[maiden geyser]

just like queue
also the word is french so

[olive sable]

so shoir is like kwire?

what the hell

[real sierra]

sometimes my brain pronounces queue as kwayway

[vague sonnet]

I could pronounce that as "k-why-reh"

[leaden crest]

I say cue-eee

[warped narwhal]

I say q

[tender river]

queue pronunciation is so convenient

when i have a

queue variable

[warped narwhal]

queue = que = q

[tender river]

i name it q

[leaden crest]

the ch in choir is pronounced like the ch in chappy chirthday chiro

[sage crag]

i say kew or Q.U.E.U.E or kewww

[vague sonnet]

ueue

[real sierra]

🕊️ coo

[silent cloak]

https://tenor.com/view/spawn-hell-hurl-through-hell-fall-to-hell-fire-gif-17250068

[sage crag]

coo, coo

[olive sable]

Redston is here

[dusty jungle]

there was a reason i didn’t post that here

[leaden crest]

https://tenor.com/view/teto-kasane-teto-jamie-paige-birdbrain-teto-dance-gif-5360684036479908174

[sage crag]

happy birthday redston

[vague sonnet]

holy shit is that redstone from minecraft???

[tender river]

not like q

[real sierra]

https://tenor.com/view/jod-gif-22535623

[sage crag]

q is kew

same

identical

[silent cloak]

yeah thats what i had in mind lmao

i love that gif

[tender river]

english

confusing

[silent cloak]

thats why we speak programming languages

[olive sable]

the person that made gif pronounces it like jif.
so luckily they're also going to hell

[silent cloak]

correct they are

[maiden geyser]

jeople jho jronounce jif jas jif jeing jent jirectly jo jell jafter jeeting Jod jat jhe jearly jates jand jalling jim Jod

[olive sable]

what

[real sierra]

its inevitable

[silent cloak]

its the same with people who pronounce vedal like medal instead of veedull (not as science intended)

[sage crag]

did you wear a dress or a suit or

[tender river]

i would pronounce kew as きゅう and q as roughly k-you

[olive sable]

im not active in shirocord, and i dont think im even in redstoncord

i should invade

[tender river]

@sage crag

[real sierra]

suit

[silent cloak]

im active in cordcord

[dusty jungle]

[sage crag]

so you did get married

[leaden crest]

wh are you pinging

[sage crag]

[real sierra]

no i was just a guest

people wanted to dance with me

[silent cloak]

thats cool u had 2 parties

a wedding and a birthday party

[sage crag]

naturally you get invited to your own wedding

[warped narwhal]

would it really be invited?

[real sierra]

i

[warped narwhal]

I would say your presence would jsut be assumed

[sage crag]

if you werent invited you couldnt go

[real sierra]

it was my sister's wedding

but she knew it was also my birthday

[silent cloak]

my mom was late to my birth

[olive sable]

understrandable, its shiro afterall

[real sierra]

and she went on stage with the mic and told everyone it was my birthday

[warped narwhal]

...what?

[sage crag]

[real sierra]

entire event's attendees sang happy birthday to me

[silent cloak]

yeah she missed the whole thing

[real sierra]

🥀

[tender river]

birthday curse follows shiro in real life

[sage crag]

we gather here today to celebrate the... birthday of my sibling, 1 verse please

[real sierra]

it was like when the waiters sing you happy birthday at a restaurant but worse

[warped narwhal]

it's your birthday? why didn't you tell us?

we could've sung for you as well

[sage crag]

27th march

[olive sable]

so did you get bron in the car on the way there or what?

[real sierra]

standing there in the middle of a crowd of like 100+ people

[leaden crest]

it’s jonna follow shiro to heaven and jod will sing for him too

[silent cloak]

na i just had to be born without her

[olive sable]

ohhh

[silent cloak]

it wasnt a problem thankfully

[olive sable]

you're synthethic

[warped narwhal]

ah, you were summoned from the aether

[olive sable]

testbuiskindje

[silent cloak]

the doctors built a herobrine totem and spawned me in

[olive sable]

[warped narwhal]

damn, where did they get the 8 cubic metres of gold to do that?

[olive sable]

[silent cloak]

at the end of a rainbow

[leaden crest]

wait a minute

[sage crag]

imagine if the y2k problem was a feature of numbers and not of the storage of numbers and people born after the event had a birth certificate with a negative birthday

[warped narwhal]

we have the opportunity of a lifetime

[trim valve]

[tender river]

shiro you have to go to the neuro live

[leaden crest]

@real sierra how would you like to break a world record

[tender river]

[olive sable]

[silent cloak]

easily break that record

[real sierra]

[warped narwhal]

[olive sable]

we must

for it is our duty

[tender river]

put shiro to guiness world records

[silent cloak]

https://tenor.com/view/ksi-floating-stairs-sidemen-gif-20767306

shiro after gaining the power of 100,000 happy birthdays

[sage crag]

ohisee

[olive sable]

shiro is like the dragonball power sharing thing

https://tenor.com/view/send-energy-sending-healing-dragon-ball-spirit-bomb-gif-16443313

[leaden crest]

https://tenor.com/view/accept-the-cake-green-avm-red-avm-green-red-gif-17837322140711312998

[warped narwhal]

the first second of that clip gives off a very different vibe from the rest of it

[silent cloak]

lmao

[leaden crest]

same

[silent cloak]

https://tenor.com/view/rigby-freaky-cat-cat-meme-tongue-out-gif-1492882876567450372

[vague sonnet]

nah he's just clenching his cheeks really hard to focus his power to fly

[silent cloak]

its his keybind

[umbral wigeon]

I wanna transfer my brain to robot body

Elon musk please make it true in future

[olive sable]

soma ahh story

[silent cloak]

sorry champ that would be a copy

u would have to stream ur brain instead

[olive sable]

its a copy, unless you

[vague sonnet]

of all people to ask you choose the twitter man? shivers

[warped narwhal]

what if you pulled a "ship of theseus"?

[real sierra]

still a copy

[olive sable]

hmmm

[real sierra]

your brain is a storage medium and it encodes info in a funky way

if you create a new storage medium and replicate the data, it's a copy

[silent cloak]

we also dont completely understand our brains

[olive sable]

hmmm

[warped narwhal]

I mean, we duplicated the brain of a fruit fly and put it in a simulation, surely it's not much harder to do a human one

[olive sable]

what if you put the brain in the server and replace everything besides the brain?

not a copy technically

but barely human

[real sierra]

"you" is not well-defined to begin with

it's not a copy of your mind if you do that

[silent cloak]

if its only the signals that make us up then u would need to magically freeze ur signals and move them into a mechanical brain and resume them

[vague sonnet]

brains are inefficient anyway
they have a lot of bloated features related to keeping the vessel working

[real sierra]

so if your conscious thought is what you consider "you" then yeah

thats not a copy

[umbral wigeon]

Before technology could even do that, I'm gonna die of old age

[real sierra]

i think that's probably the most realistic definition since almost every other body part is replaced on an individual basis already

[leaden crest]

we’re back to the ai sentience convo @sick owl

[real sierra]

people have artificial arms, legs, organs

just not all at once usually

[dusty jungle]

the second season of steins gate is about this

[olive sable]

from an outside perspective a non-biological storage medium would still be a copy

[silent cloak]

based reference

[olive sable]

TRUE

[real sierra]

ik i have amadeus on my phone

[silent cloak]

i was just listening to a steins;gate song

[olive sable]

steins;gate 0, the GOAT

[silent cloak]

stop ratting my computer

[silent cloak]

fun fact

[dusty jungle]

[silent cloak]

amadeus is what got me into LLM development

[olive sable]

honeslty

i want an amadeus

[vague sonnet]

it being biological doesn't change anything
like if you rewrite a new brain with yours' data

[silent cloak]

i was using machine learning for anticheats and other purposes

but that show made me take it to text when the show came out

[olive sable]

i need to hire quack to make me an ARC. i can do the actual graphics for a 2D anime girl myself

[leaden crest]

here’s a link to remove any spyware from your computer

just click upvote and dont ask questions

https://colornames.org/color/cd2edb

[warped narwhal]

if they could remove the part that makes me need to sleep that would be great. I want 8 more hours of slaving away at a desk job please

[umbral wigeon]

I have hope in elon musk

[silent cloak]

i got into LLM research around 2018-2019 if i recall

[warped narwhal]

https://tenor.com/view/ip-grabber-ip-grabber-thanos-gif-21846609

[vague sonnet]

what's your fav c++ library to use for ML?

[sick owl]

My opinion of AI sentience is that we'll never know for sure if/when we reach it

[olive sable]

the problem is we still dont have the digital version thats 100.0000% the same as a biological brain

[fiery anchor]

the brain is an organ like every other. imo the conciousness is only a temporary state of neurons firing and the paths they take, the brain being a tool for modifications of those paths, same as the rest of the body

[silent cloak]

originally didnt use one because my previous uses never needed it (LVQ networks etc)

i think I used tensorflow later

but i remember hating it

[sick owl]

I take back this statement, apparently the definition of sentience is the ability to perceive and feel things

We can prove both of those in humans, and current LLMs definitively have 50% of that

[leaden crest]

am I real if I can feel

[silent cloak]

AI sentience wont happen with current tech

[warped narwhal]

I think therefor I am

[vague sonnet]

oh yeah? give me a few months

[umbral wigeon]

Where did Consciousness come from?

[dusty jungle]

Nah, we’ll know because it’ll have to be specifically made sentient. An llm will never achieve it but an ai designed from the ground up to be sentient isn’t impossible, although probably won’t happen in our lifetime

[silent cloak]

sure thing big dog

[umbral wigeon]

Even google don't know

[olive sable]

from what i personally believe, it is def possible to make a digital brain if you just go deep enough into how the human brain works and replicate each system. but its probably not feasible

[sick owl]

See I would have said that but the topic gets very complicated

https://www.anthropic.com/research/emotion-concepts-function

Its not unthinkable that we manage it by accident

At some point

[dusty jungle]

It’s because it’s trained on human datasets

[leaden crest]

we have to understand how the human brain works first

[dusty jungle]

its told to act sad so it does

[sick owl]

We already accidentally created models with an incredibly complicated understanding of emotions and functional, albeit probably not phenomenological, emotion circuits

[silent cloak]

thats why they talk about taking over the world alot when they are told they are AI

[real sierra]

my opinion of ai sentience is that people want to believe one or the other and will shift the goalposts to agree with that viewpoint because sentience is not well defined and has no value in an academic setting

[silent cloak]

alot of the data about AI they are trained on happens to be about them being evil lmao

[fiery anchor]

wasn't there an article somewhere where thry replicated a fly brain, put that into an virtual reality and it started trying to clean istself...

[silent cloak]

yeah

[olive sable]

i think the only feasable way to make digital brain is to cut some corners and not simualte each individual neuron in excruciating detail

[vague sonnet]

yeah, LLMs are not the way
text generators cannot be made to actually think/feel under the hood

[sick owl]

@shell jasper You may have a horse in this particular race

[silent cloak]

people forget that LLMs are closer to a auto complete than to a human brain

infact beyond terms 99% of what makes them up is unrelated

[dusty jungle]

Llms are just predictive text

[leaden crest]

the term artificial intelligence is probably misleading in itself

[silent cloak]

yup

NLG

[vague sonnet]

that's what I'm saying
it's so sad to see so many people get fooled by that tech

[silent cloak]

a term i dont hear anymore

[real sierra]

ai used to mean npc

now it means tuned function approximator

[silent cloak]

current AI is far more A than I

[real sierra]

🥀

[leaden crest]

did yall see this

https://youtu.be/WAUnmQt2Z7Y?si=D43p2Sv7QEch1xDX

[sick owl]

Correct

[vague sonnet]

it's not, just heavily misinterpreted
it's artificial because it mimicks real intelligence, without being it

[dusty jungle]

now instead of ai we just have me

[sick owl]

We have a full digital scan of a fly brain that we can use for simulation

[olive sable]

llm is def not the way to ai sentience.
but just like how animals learn to be afraid from snakes by watching their parents be, i think that learning from behavior could be successful

[sick owl]

And by jolting that brain we have been able to recreate fly behaviour in a sim

[silent cloak]

LLMs were more innovative for data science than neural science

[olive sable]

ye

[sick owl]

Prediction of the kind they do is incredibly complicated though, I think people undersell what that actually entails a lot of the time

[vague sonnet]

and yet ironically they produce a lot of inaccurate data

[olive sable]

observational conditioning

[sick owl]

I don't think current models are sentient but I don't think the fact they're predictors is as deflationary as many frame it

[silent cloak]

they are predictors though

thats their function

[sick owl]

They have functional circuits that mimic emotion for the express purpose of predicting emotional response in text

Is it not incredible that we already came that far on accident

[dusty jungle]

oh ik it’s complicated, i just also know it’s nowhere near sentient. It’s just a really complicated math equation

[trim valve]

nah current ai is more agi then i, haven't you read the news

[silent cloak]

its based solely on the data given from text though

text can be associated with emotion

keep in mind too LLMs are single shot and have no concept of time or neural plasticity

[sick owl]

Given that this is the case I think its arrogant of us to handwave it away as "not intelligence" or to assume that at some point down the line we could end up with something sentient without specifically aiming for that goal

[silent cloak]

for sure just not with this tech

[olive sable]

sentience requires self-awareness, so instead of the only input being text, i believe that feedback from the physical world could be a step in the right direction.
instead of LLMs they would be learning from scratch with normal neural networks.
thats what i believe at least

[silent cloak]

its not really arrogance as much as is it just understanding how it operates

[sick owl]

On the former point, you can have intelligence without sentience and the complexity of prediction carried out by current leading models I'd argue is truly intelligent

[silent cloak]

it can pull data but thats it

it doesnt ever truly "think"

[sick owl]

The mistake people make is going with an anthrophomorphic definition of intelligence

[silent cloak]

again its closer to a plinko machine than a brain

[real sierra]

does anyone?

[silent cloak]

https://tenor.com/view/plinko-stake-gif-25294696

[fiery anchor]

for me, llms are more a muscle that a brain or a soul, and therefore unable to achieve sentience, or intelligence.

[silent cloak]

its plinko with set parameters on the points

[sick owl]

If we go by the definition LLMs absolutely are artificial intelligence

In context learning is a thing after all

[umbral wigeon]

Aren't we a bunch of parameters too

[silent cloak]

100% not intelligent then

they learn once for model creation

[real sierra]

really i think the gap is on the biology side

[silent cloak]

they have 0 neural plasticity

they cant ever learn again

[olive sable]

idk how yall think, but isn't it ussualy that you kinda talk to yourself in your mind going through the stuff you're doing?

[real sierra]

it could very well be that our brains work very similarly to current AI approaches and we just dont know it

[silent cloak]

i 100% disagree

we have decay rates for our signals

LLMs dont have a temporal variable

[real sierra]

similarly is a load-bearing word

there are obvious differences

[silent cloak]

we are more "Event driven"

[fast pagoda]

They learn in context it's just ephemeral

[real sierra]

not the least of which is the continuous nature of human thought

[silent cloak]

while an LLM is closer to just pouring out a glass of water through a filter

[real sierra]

idk who you're arguing with i agree

[silent cloak]

not really its pretty much just reprompting with context

their weights dont change

the context just adds... well more context

[umbral wigeon]

So llm is like fixed weights

[silent cloak]

correct

[sick owl]

Read the message I placed below the image

In context learning is a thing, its not permanent but it fits the definition

[silent cloak]

yeah i still wouldnt call it intelligence though

[fast pagoda]

The way they affect the remaining output within that session does change though, since the weights are only part of the equation

[silent cloak]

machine learning != AI

[fiery anchor]

percieved randomness and change may be an incomplete understanding of the algorithms

[vague sonnet]

doesn't have to be physical world and you can pretrain it to a degree
but yeah, text shouldn't be the input, human languages are too abstract and imprecise and if that's the only thing you give it it will still learn to be a text generator

[silent cloak]

i mean the randomness is half caused by floating point issues

[olive sable]

[fast pagoda]

Other half on purpose lol

[silent cloak]

yeah u often get random seeds

[vague sonnet]

easiest way to check is just run the exact same prompt twice (with reset contexts)
if inputs are different then there's a temperature parameter to randomise it

[fiery anchor]

for all we know, we could be ai living in a simulated world. Only because the possibility seems unfathomable, it is not impossible.

[sick owl]

By definition it is intelligence

[olive sable]

having it be fully stuck to the same weights is also a problem for true consciousness. so at the very least variable weights are needed

[sick owl]

You can argue it doesn't fit your subjective view of intelligence, but objectively the term "artificial intelligence" isn't incorrect as some frame it

[silent cloak]

its being trained technically through an external tool

a tool that isnt shipped during inference

[jovial remnant]

that is just incorrect by definition, all machine learning is AI, but not all what is under term AI is machine learning

[silent cloak]

AI has become a clusterfuck of a term that means something different to everyone

[sick owl]

Again, in context learning is a thing

https://arxiv.org/pdf/2310.15916

See the above

[vague sonnet]

simply feed back the outputs as inputs and have the model self-train

[sick owl]

Francois Chollet also wrote a piece on this

[olive sable]

eh

[silent cloak]

yeah but its not "learning"

the models remain fixed

its all context

no matter what u say to an LLM there are things that wont change

[olive sable]

im not too knowledgeable on this, but i feel like learning from its own outputs wouldnt be ideal

[silent cloak]

when ur having a conversation with a model its just taking the chat history and feeding it back through again

[sick owl]

Not entirely accurate

https://arxiv.org/abs/2212.07677

Just because its dependent on the existence of that context doesn't mean its not learning

[silent cloak]

seems like a dated paper but ill give it a read

[sick owl]

You say dated I say foundational

[vague sonnet]

you need to add a special ingredient to make it work
but reading its own outputs will give it continuity between inferences and an increased "sense of self" since it will eventually learn how its outputs interact with the rest of inputs

[olive sable]

hmmm

i wouldnt know

[silent cloak]

honestly alot of this seem very dated

and unsupported beyond small tests

[olive sable]

if i do ai stuff in the future ill get back on this

[silent cloak]

like sofmax attention implementing GD requires multi-head correction

there is 0 information backing it up on actual production / deployed models either

there are a few other explanations for their results though

[sick owl]

You know what, that's valid

It is a toy setup in that paper

[silent cloak]

seen some papers saying it could also be implicit Bayesian inference

[ebon basin]

well it not just the content even the structure can change by some "neuron" being dead this is what Gemma make use of

[silent cloak]

ICL involves more than just gradient-based learning too

[leaden crest]

https://tenor.com/view/neurosama-neuro-sama-gif-11499204033432984688

[faint sandal]

https://tenor.com/view/ari-aria-sunna-chinatsu-zzz-gif-14672163086010667151

[ebon basin]

Still working on my chess engine . Just trying to find a better MC I also try alpha-beta bruning I add tons of feature since I don't know which settings I let copilot write genetic algorithm let it now run for 100 generations

[sick owl]

Okay, leaving behind the gradient descent idea since that does appear dated

Would you not qualify induction heads as a learning mechanism?

[silent cloak]

not at all honestly its a single-pass retrieval so nothing is refined

they dont generalize beyond whats literally present in the context either

[sick owl]

The definition of learning is "the acquisition of knowledge or skills through study, experience, or being taught"

[silent cloak]

its n-gram copying though

[sick owl]

Knowledge can be acquired through experience without being permanent

[silent cloak]

which is predicting

[sick owl]

Does it not still fit the definition though?

If you give a model updated data beyond its learning cutoff it can update assumptions based on what you taught it for as long as that context cache exists

Its temporary, but it is acquiring new knowledge in a functional sense

[ebon basin]

So the discussion is about how a transformer lerns ?

[jovial remnant]

how would it even make sense for the model to be able to generalize to what is not in the context? What do you mean by that?

[ebon basin]

I watched a video about how a team analyzed how transformer hallucinate. It turns out the same "neurons" which pick stuff up which is rarely repeated doing the output for it . When you turn them off = the stuff which is in a low amount in the trainign data the LLm can't answer them anymore.

[silent cloak]

if you solely has induction heads they would not be able to do anything nearly as good as a full model

[vague sonnet]

you're making it sounds like there are "correct nodes" and "incorrect nodes"
which is plain wrong

[silent cloak]

they arent designed for reasoning as thats not the point of the tech

[sick owl]

Well naturally, you could make the same argument about any given slice of a brain

[silent cloak]

the fuzzy matching they do helps but they struggle MASSIVELY for anything beyond non-local patterns

they cant learn from such things

they are relient on token level patterns too

things that need semantic understanding they are very bad at

[ebon basin]

Not really it more like LLM needs to dig into nodes which contains rare information which contains a lot of nosie . It seems the information is more sorted by intensity then by catrogory which make sense

[silent cloak]

not to mention all the repetition issues they can cause sometimes

[jovial remnant]

what do you think is the point of the tech then?

[sick owl]

Also, it was my understanding that subsequent interpretability work has shown induction heads do generalise to out of distribution patterns in a true sense, such as made up words for instance

If it was just n-gram copying it wouldn't be able to learn abstract patterns and unique made up tokens surely

And this is just one component of the much grander mechanism responsible for ICL

[vague sonnet]

generating text

xdd

it is straight up a text completion algorithm

which based on training data picks the next most probable word

[jovial remnant]

well for one thing transformer architecture isn't bound to text, it can represent any abstract modality

if you think of it this way then it allows to not think of the kinda reductionist framework of it being "just text completion algorithm"

[olive sable]

the thing with artificial brain. if humans do figure out how to do it,t hen they will also be able to make smarter than human brains
so then id be jobless

[vague sonnet]

if the tokens take form of text
then yeah it is a text completion algorithm

which is the case with LLMs

[silent cloak]

yeah i think thats from the fuzzy / semantic induction heads that match with similarity

[ebon basin]

Yeah and the gradient descend works also really straight forwards the problem is for researcher this is not enough they want to know how the structures internal works . Basically how the words get link to each other which honestly I also don't understand what there isn't to understand

[jovial remnant]

that's kinda like saying every piece of software is just logic completion algorithm, and while true, there is a ton of stuff you are just missing in this kind of explanation

[vague sonnet]

yeah software is just a sequence of instructions

[silent cloak]

i guess in the end it depends on what u call "learning"

[jovial remnant]

yeah, but there is more to the topic of software than just "sequence of instructions"

[vague sonnet]

learning implies it's permanent

[silent cloak]

a nearest neighbor classifier generalizes to inputs via a fixed matching rule and people will call that learning

[silent cloak]

yeah by that definition then no

[vague sonnet]

if it's an added instruction added to the context that will get forgotten when you run out of space in the context
then it's not learning, it's just a bigger input

[silent cloak]

u can see very easily where it starts to break down as the context gets longer and complicated

[jovial remnant]

idk about that definition, I'm certain I learned quite a bit of stuff that I no longer remember

[silent cloak]

not to mention grokking

[vague sonnet]

you're just being ignorant at this point

[jovial remnant]

explain how

[silent cloak]

a big issue too honestly is that these heads are very local with their recognition

you will have a tough time trying to "train" a model during inference

[ebon basin]

This is like wanting to get a prove liek this :"lan Turing proved in 1936 that no general algorithm exists to determine whether an arbitrary computer program will halt or run forever. This result, known as the Halting Problem, establishes that the problem is undecidable for Turing-complete systems. " The problem is each NN can be totally different because of this developing a neural network is more like cooking or growing a plant then programming itself . The same way you can't make a function for all function . You can't generlize even transfoerms

[sick owl]

No it doesn't

The definition of learning doesn't make any such qualification

[sick owl]

I think it would be more accurate for you to argue its not parameter learning rather than it not being learning outright

[silent cloak]

i think its just closer to recognition than anything

[jovial remnant]

and then at that point the argument is pointless, cause of course stuff learned in context with current models is not permanent if just left as is

[silent cloak]

the whole purpose of these models is pattern recognition

[sick owl]

Precisely

Well actually

I wouldn't say its pointless

[vague sonnet]

my bad, didn't know english isn't your first language
or second language it appears

[sick owl]

We're just arguing about different things

[sick owl]

How needlessly rude

You're wrong, by the by

[vague sonnet]

LLMs don't acquire new knowledge
they are given it as input

[ebon basin]

Moment didn't deepseek add recently permanent memory to transformers ?

[silent cloak]

yeah i dont think that definition works in ur favor tbh

they cant really develop any of that beyond simple local recognitions and small short terms adaptions

[vague sonnet]

he's just trying to stretch definitions to match his narrative

[sick owl]

Just because the basic model weights aren't altered doesn't mean its not acquiring that knowledge in a functional sense

[jovial remnant]

I mean yeah conversation isn't actually pointless, just more or less saying that the argument solves itself

[sick owl]

You're being needlessly narrow with your definitions here to suit your own point moon

Not to mention oddly standoffish

[silent cloak]

the context in general to me feels like a patchwork solution to a problem rather than actual learning

[ebon basin]

Dose definition are like trying quantum locality pudding on the wall if you know what I mean

[silent cloak]

because remember the context is just feeding back its own chat history

its nothing "special" about the tech and model itself

[sick owl]

Its not learning in the human sense true, but I'd argue it is a sort of learning even if it only occurs in brief flashes

[silent cloak]

i think adapting is a better compromise

[sick owl]

Limited and brief, but still within the bounds of what you'd conventionally consider learning

[silent cloak]

it can shift to fit the context at times but it will struggle to take that information far

like if u tried to teach a model a whole language by feeding it data during inference u would run into a ton of issues

[sick owl]

I'd define that as limited learning, but we are basically arguing semantics by that point since we mean a similar thing

[silent cloak]

thats half of the AI debate tbh

[sick owl]

And you're right about this, I was just arguing that it qualifies as learning, not that it qualifies as complete robust learning in the sense of humans

[ebon basin]

Fair I personal when I try to make non technical people to understand how NN learnsI say it is like a sandform the NN tried to fit

[vague sonnet]

you can save conversation data with semantic chunking and retrieve it by matching tokens
which is still just extending the prompt, but somewhat solves the issue of completely forgetting everything
no one uses that because any sort of permanency is undesirable for corporations that want to sell the technology as a service, having to adjust model weights or store more data is extra costs

[silent cloak]

i used to build models from scratch during the researching eras

ive moved on since with how terrible the field has become though i still research in branching fields at times

(just not transformer models kek)

[silent cloak]

thats because its a text generator

to make it a "chat bot" you need context/history

so they slapped it on

and called it a day

[ebon basin]

I think what I remember were the engram stuff which is kinda on top https://www.youtube.com/watch?v=xUlX6jvwVfM&t=251s

[mighty thorn]

Don’t even get me started

On Deepseek engrams

[ebon basin]

I have time ^^

[mighty thorn]

I fucking love Deepseek engrams

I have plans

Deep, complex plans

I’m gonna use them harder than Deepseek did

[vague sonnet]

which is completely fine
but the issue is how they set the expectations
it outputs in a human language form and it does take your context into consideration so it's not full random
but beause of that a lot of people are suddenly thinking it's capable of real understanding or thoughts
while it was trained with massive amounts of dirty data
and that marketing made investors buy into it
which made the companies go heavier on it
so we're now stuck in a weird timeline where people are investing crazy money into such inefficient and inconsistent tech

[ebon basin]

I also need my short time memory to make sentences so it is fine

[mighty thorn]

Don’t pay them too much mind, SNN is at least 5 years from passing the Turing test

Are just mad that their slightly different field wasn’t the one that suddenly become worth 400b

[vague sonnet]

nah man I have to gatekeep my tech on purpose

we're all cooked if it gets out

[mighty thorn]

Ughhh shur up shakes but

No way you AND quack invented agi while nvidia google facebook and Deepseek couldn’t

This is just mythos marketing but for SNN

[vague sonnet]

corporations have different goals

[mighty thorn]

No offense to you

[minor crag]

I'm failing to keep Track of what this argument is actually about

[olive sable]

same

[vague sonnet]

it's mostly about nothing

[mighty thorn]

SNNbros thinks ANNslop is slop while saying their SNNslop is better but not even having any SNNslop to show the ANNbros

“Oh yeah my laser point is wayyyy brighter than that.. what, no I can’t show you, it’s way too bright for the likes of you”

[olive sable]

ok but like

i feel like that's just still in the research phase

[minor crag]

SNNs are cool but currently very hard to scale and ANNs are currently the best available tech that can be scaled

[warped narwhal]

what are SNNs? synthetic I'm guessing?

[olive sable]

spiking neural network

[mighty thorn]

Spiking

[minor crag]

Spiking neural nets

[mighty thorn]

Ever so slightly closer to real brain stuff

[sick owl]

I'll save myself the 🥉

[mighty thorn]

But still basically not even close

[olive sable]

it is better at least

[minor crag]

Idk how to easily explain them

[mighty thorn]

And all of their groundbreaking results that would shatter society as we know it are always suspiciously secret and never demonstrated

Basically a pyramid scheme but for gatekeeping instead of money

[vague sonnet]

I have some sentient AIs in your local network already

they are everywhere

[minor crag]

They are cool but are very inefficient on modern hardware

[vague sonnet]

not as inefficient as LLMs tbh

[minor crag]

Ok Mr pseudoscience

[vague sonnet]

lol

[minor crag]

Try and run a 14B SNN on a £800 gpu

Yea they are more efficient but the hardware to use them doesn't really exist

But I can run a 14B LLM on my £800 GPU since the hardware does exist

[sick owl]

He's joking, its not especially funny but that's not a serious statement

[minor crag]

Yea I get that

[vague sonnet]

you thinking it's something that needs explaining says a lot lol

[mighty thorn]

SNNs use 100% less power since they aren’t real

[opaque sigil]

if it can't be efficiently implemented on gpus it's basically doa

[vague sonnet]

bet, double that number even

[opaque sigil]

[ebon basin]

Ah I done that too

[sick owl]

That response could go either way, this is the internet after all

[olive sable]

so far

[minor crag]

You have fun running a 28B SNN on a GPU that can't run a 1B SNN

[sick owl]

And again with the personal jabs, you've got a real attitude problem

[minor crag]

Unless you simplify the spikes into some sort of integer and simplify the math used... That kinda sounds like ANNs

[mighty thorn]

Yeah

SNNbros hate this one simple hack

“Okay but can you actually run it on a computer”

[minor crag]

Then you could use matrix multiplication to optimise that math even further

Well now I stopped the argument with the logic of a teenager I'm here to talk about why I hate CPP

Wait

I forgot why I don't like CPP

Uhhh... CPP is old and smelly

[mighty thorn]

Chinese People Party

[minor crag]

Wha

No?

[opaque sigil]

https://www.youtube.com/watch?v=7fGB-hjc2Gc dw i gotcha

[minor crag]

Thank you

Anyways I've gotta rewrite lots of CPP code and I hate it

[vague sonnet]

oh that says a lot
def not gonna take your takes seriously

[warped narwhal]

I actually watched the whole thing a couple weeks ago, and god dammit they are 100%

[olive sable]

i agree that c++ is bad, but i dnt think it deserves to be called the worst of all time

[vague sonnet]

it's the best

[mighty thorn]

[olive sable]

it is def not the best, as a c++ user its kinda ass sometimes

[minor crag]

Have fun with your imaginary 28B SNN you saw on an ai generated YouTube short

I'm gonna be working on my realistic 14B dense transformer

[opaque sigil]

c++26 is going to fix all the issues don't worry

[mighty thorn]

Guys

Please

Your scaring the kids

[olive sable]

the kids shouldnt be here, this is #programming

[vague sonnet]

the experience scales with the skill
at the point where you're not throwing logs at your own legs you can really appreciate all the features it has and the freedom it gives you

[mighty thorn]

Yeah move this to #debating-and-calling-each-other-names

[vague sonnet]

you would know a lot about being dense

last one I swear

[olive sable]

no

[leaden crest]

to be fair my old middle school gc had people spamming [filtered]

[olive sable]

being able to use the language does not make the language good

[minor crag]

To be fair that one is not bad

[leaden crest]

in here there’s at least a shiro to protect me

[mighty thorn]

Sometimes

[leaden crest]

no all the time watch

I cast happy birthday @real sierra

[vague sonnet]

more like
the disadvantages of c++ come from it being hard to use right
but once you reach the skill threshold it's amazing

[olive sable]

no

the disadvantages of c++ come from it being bad

[vague sonnet]

compiler also does a lot so you can easily write nice readable code without performance loss

[mighty thorn]

https://tenor.com/view/basedspongebob-based-spinchslob-epicbob-epic-gif-22636716

[vague sonnet]

what are you, a rust fanboy?

[minor crag]

Rust is the best language according to developers :3

[vague sonnet]

according to rust developers*

[leaden crest]

nvm you were right

[sick owl]

Oh god not this debate

Take cover chat

[olive sable]

i despise rust, but mainly due to the people

[vague sonnet]

oh so you're not one of those
then what's your problem with c++? genuinely

[olive sable]

i use c++ as my main language. and im fine with admitting it has its flaws

i dont see why you cant

[vague sonnet]

I don't run into issues when using it idk
only really others' code but nothing to do about it at work, not realistic to rewrite millions of lines of code myself

[minor crag]

Well according to developers who have worked with rust (you can't like something you haven't used)

And developers who have worked with multiple languages usually prefer rust over languages like C and CPP

[olive sable]

just because you dont run into issues doesnt mean the language in itself doesnt have issues

[mighty thorn]

I like rust

It’s what allowed me to calculate the minanumber for Cerber

Even tho I still took almost a week straight

[vague sonnet]

well if my workflows aren't affected by said issues
then I guess I am him or something

[mighty thorn]

And 100 gb of ram

[opaque sigil]

defeated by a style guide

[olive sable]

ye

[opaque sigil]

now that i think about it it's been well over a year since i've written any rust

maybe it's time to gaslight myself into thinking i should change it again

cutile-rs

[olive sable]

how c++ handles having multiple files is absolutely atrocious and you cant convince me otherwise

[young plover]

[vague sonnet]

realise something is bad and stop using it -> time passes -> forget -> repeat

[olive sable]

bro

https://tenor.com/view/jordan-stop-it-gif-15830202965985496707

[vague sonnet]

yeah kinda
the disadvantage of its freedom

it's way easier to fuck yourself up

but you can basically always do something the best way

if you know how

[opaque sigil]

i'll still pick rust over cpp for things that need to only run on a cpu if i have the choice

[olive sable]

the fact that its easy to shoot yourself in the foot, jsut shows the language is far from being the best one

[vague sonnet]

ok I remember a c++ issue

[sick owl]

Why are we having this debate when we all know HTML is the best programming language

[vague sonnet]

v-tables

[opaque sigil]

idk i'm bored

[olive sable]

the casting in c++ is absolutely terrible

i hate static_cast

fuck that

[opaque sigil]

what?

[minor crag]

What about CSS?

[sick owl]

It sucks and is bad

[vague sonnet]

is it?
you should rarely ever cast
and when you do need it it works

[minor crag]

Oh :(

[opaque sigil]

are you just too lazy to type static_cast or what's the problem

[minor crag]

I'm gonna go rewrite my llama CPP fork in HTML give me a second

[leaden crest]

.md

[vague sonnet]

I will take that any day over the program not doing what I told it to
like c# not having an always clear answer if you will pass something by value or reference

[minor crag]

Can we all agree C# is bad?

[rigid snow]

??

no

[young plover]

Really only need static_cast in the case of narrow-casting primitives. In which case it's still bad because narrow-casting should be checked and there's no good built-in solution for that.

[vague sonnet]

it's not bad-bad like python
but I kinda hate using it

using c# always makes me miss c++

[minor crag]

Python is good for many uses

[young plover]

• Python sucks but is used for many things

[vague sonnet]

I'm a python hater I'm not even going to bother

[leaden crest]

python itself is great but I hate dependencies because anything useful you want to do with it is dependent on them

[minor crag]

Yes

Wait

The more I read that sentence the less sence it made

[leaden crest]

does that help

[vague sonnet]

the one good thing about python is that it's easy to include libraries and there are a lot of them that make things quick to do
but everything else about it I hate

it's like instant soup
it tastes like shit but it's quick and edible enough

[rigid snow]

praising python for its ecosystem

[leaden crest]

but so many of the libraries are dependent on each other having certain versions

[rigid snow]

literally the worst part

[minor crag]

I think I understand what you mean and I think I agree

[vague sonnet]

I know
its biggest upside is still a flaw

but it is undeniably convenient to write a quick script

[olive sable]

the problem is probably on my side, but what the hell are all these random casts even for?
to make my vulkan shaders i have to do reinterpret_cast<const uint32_t*>(code.data()); what even is all that?? why do we need to do that?

[minor crag]

The python language is good but the ecosystem and such are not good

[opaque sigil]

sounds like a shitty api

[olive sable]

its vulkan

[opaque sigil]

more than anything

vulkan-hpp?

[minor crag]

Also I've been trying to get a screenshot where toast, toast, and toast are typing at the same time and its very difficult

[vague sonnet]

sorry I will type more

[minor crag]

That's the main reason I am still here

[opaque sigil]

[vague sonnet]

I think a few people screenshotted that before

[leaden crest]

I use this for the joke it has 5 toasts but not concurrently

wrong embed wrong embed

[minor crag]

I GOT IT!!!

[opaque sigil]

[minor crag]

That took like 30 minutes of waiting

[leaden crest]

fishing for toasts

[olive sable]

idk, the linter wont tell me where its from, might be from volk. but i dont think i changed that code when switching to volk

[opaque sigil]

the triangle is in plain vulkan

[olive sable]

nah its not

volk.h and volk.c come with the files

[opaque sigil]

i meant the plain c api

[olive sable]

actually, i should check if volk still makes it faster or if it makes it slower now

[olive sable]

i dont get what you mean

[opaque sigil]

vulkan is a c library, vulkan-hpp is a thin c++ wrapper around it to make life easier

[olive sable]

oh

ye its plain vulkan

[vague sonnet]

ok so what the fuck is code.data() and why does it not already output the correct format

[opaque sigil]

pointer to 4 byte array if i had to guess

[olive sable]

VkShaderModule FrameManager::createShaderModule(VkDevice &device, const std::vector<char>& code)
{
    VkShaderModuleCreateInfo createInfo{};
    createInfo.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO;
    createInfo.codeSize = code.size();
    createInfo.pCode = reinterpret_cast<const uint32_t*>(code.data());

    VkShaderModule shaderModule;
    vkCreateShaderModule(device, &createInfo, nullptr, &shaderModule);

    return shaderModule;
}

[opaque sigil]

oh right it casts to u32*

wait

is code the shader string?

[olive sable]

ye

the spirv

[opaque sigil]

so already compiled

[olive sable]

ye

[leaden crest]

VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO 💔
when the coding practices are too good

[olive sable]

i didnt make that

Contrary to popular belief, i, sam, did not invent vulkan

[leaden crest]

my world is shattered

[olive sable]

[leaden crest]

next you’re gonna say the earth is round

[young plover]

reinterpret_cast is for taking a pointer and treating the address as a pointer to a different type. It's generally only well-defined to cast a pointer to a trivially-copyable type to a byte pointer or the other way around as long as the pointer is correctly aligned. Casting to reinterpret_cast<const uint32_t*> is weird. uint32_t array is probably just the most convenient way for them to process SPIR-V.

[minor crag]

Its a Velociraptor‭ and nobody can change my mind

[vague sonnet]

you should be able to static cast pointer to pointer no?

[olive sable]

i can try, but its been a while since i dusted off the old triangle

🔺

cathy, the triangle does not make me trans.

[opaque sigil]

each instruction is 32 bits apparently so i guess that's why the u32*

[minor crag]

But 🏳️‍⚧️

Also I never said you is trans I just said 🏳️‍⚧️

[opaque sigil]

there might be some alignment requirements who knows

[olive sable]

im not saying im not trans, im saying im not trans due to the triangle

[leaden crest]

the triangle is transphobic?

[minor crag]

Was it due to the 🏳️‍⚧️ I sent?

[young plover]

Only if it's a polymorphic relative, or void*. The void* part is goofy because you can just static_cast to void* then a different type to bypass the normal pointer static_cast rules.

[vague sonnet]

oh ok

[young plover]

C-casts (T*) are an abomination because they can be either static_cast or reinterpret_cast depending on the context.

[mighty thorn]

Most are

[olive sable]

that im trans, that the triangle does not make me trans, or the fact im even mentioning this in the first place?

[leaden crest]

toast you’re gonna get cancelled at some point explaining yourself

[young plover]

and const_cast, that's the biggest problem

[opaque sigil]

yeeeeeeeah....

[leaden crest]

ahhh no I actually called him toast

[minor crag]

Are you trans because I said 🏳️‍⚧️

[olive sable]

no

[minor crag]

If so I must exploit this power

[minor crag]

:(

[vague sonnet]

crazy how a single reaction emote can make a conversation weird

[leaden crest]

the toasts are getting to me

[olive sable]

[glad path]

why is it 2 hours

oh

[opaque sigil]

why wouldn't it be

[glad path]

another rust cultist

[leaden crest]

if the triangle makes him not trans, we must take away his triangle to make him trans

[fickle rain]

Just mprotect() to make sure the variable is for sure writable even if it's in executable readonly section

[minor crag]

Maybe if we just double the 🏳️‍⚧️ then it will out power the 🔺

[glad path]

wheres' the triangle leaderbaord again

[olive sable]

i never said i wasnt trans, but that has nothing to do with the triangle. the triangle is ungendered

cuz its 3 vertices

[leaden crest]

I gotchu

[leaden crest]

^

[minor crag]

That's what the 🔺 wants you to think

[olive sable]

[minor crag]

I think the power of 🏳️‍⚧️ has beat the power of 🔺

Wait

[leaden crest]

WHOA

[olive sable]

[leaden crest]

WHOA WHOA WHOA

[minor crag]

I pressed the wrong thingy

[leaden crest]

we”re not giving sam a castration (yet)

how is that not filtered

[olive sable]

crazy how we're humanizing a shader running at 126000 fps now

[nocturne olive]

[leaden crest]

question is why was the🔺next to the 🍆

[vague sonnet]

[olive sable]

\🔺 and \🍆,the classic combo

[minor crag]

My frequently used emojis happened to have 🔺 next to 🍆

[leaden crest]

[olive sable]

how is 🔺 in your frequently used emotes?

[leaden crest]

^

[leaden crest]

^

[olive sable]

[vague sonnet]

needing to censor emotes is crazy work

[minor crag]

I censored the server icons of the servers I am in

[olive sable]

no, the frequently used bar gives the stuff you use the most at the top, triangle shouldnt be 2nd after only using it a couple times today

[leaden crest]

do those servers require the use of 🍆 more often

[minor crag]

A lot of things require the use of 🍆

[leaden crest]

[minor crag]

Yes logic :3

[olive sable]

so

why the 🔺 ?

[minor crag]

Wha?

Sentence no English?

[olive sable]

why does the triangle appear as the 2nd most frequently used thing in your emotes bar?

that cant be only from today

[minor crag]

It is?

[olive sable]

hmmm

alright, no further questions

[minor crag]

I don't use the small colourful shapes often so using something like 5 times gets it into my frequently used

[stark needle]

??????

[fickle rain]

Vibe modding...

[olive sable]

joke cuz they mixed them up

[stark needle]

Classic 🍆 🔺 Moment

[olive sable]

[stark needle]

Wait this looks sus

[olive sable]

that it sure does

[silent cloak]

Egg plant

[olive sable]

im pretty sure the chat history from today will eventually be used against me somehow

https://tenor.com/view/better-call-saul-james-morgan-mcgill-slippin-jimmy-bob-odenkirk-mugshot-gif-18613260

[silent cloak]

https://tenor.com/view/life-imprisonment-cat-imprisoned-for-gif-26724402

[stark needle]

[olive sable]

i odnt think i even said anything bad

i might have said i was trans tho

[stark needle]

Wym "might have said"😭 😭 do u not remember 😭

[olive sable]

uhm

lets see

[olive sable]

here

[rigid snow]

scatter

[olive sable]

hi lmao

[rigid snow]

hello

[olive sable]

[rigid snow]

the meds fucked my sleeping schedule up aaaaahghg

[stark needle]

Wait so are u trans or not I'm confused😭 😭 😭 do u need new pronouns now or what did i miss

[olive sable]

idk

[stark needle]

[olive sable]

[rigid snow]

[vague sonnet]

simply never be famous enough for that to matter

[olive sable]

im not planning on being an npc my whole life

[vague sonnet]

npc - neuro power clipper

[rigid snow]

not a bad digital footprint

[olive sable]

so far

[rigid snow]

well keep it that way

[olive sable]

[rigid snow]

can you not control it

[olive sable]

mostly ye

[vague sonnet]

I can't wait for the things my haters find after I make it big

there has to be a goldmine somewhere

[minor crag]

About 1 hour ago

[olive sable]

i dont think i have anything to worry about, besides the old fortnite vc

[vague sonnet]

not even a fraction of my full power

I'm actually especially tame here

[olive sable]

not something to be proud of 😭

[minor crag]

If you are that 28b SSN AGI AI thingy are not a faction of your power I am scared

[olive sable]

[vague sonnet]

AGI is just a weekend throwaway project

[minor crag]

Sure

[vague sonnet]

you wouldn't get it

[minor crag]

*head pats*

Why does discord not show the *

[olive sable]

because ** makes it ittalic

[vague sonnet]

is discord italian-phobic?

[olive sable]

and ** ** makes it bold

[minor crag]

Ohhh mark down

Yeeee

Uhhh

How about * head pats *

Yay fixed

[olive sable]

and *** *** makes it bold itallic

you can jsut use backslash to cancel the effect

like i did

[minor crag]

Huh

[olive sable]

\*headpats\*

[minor crag]

OHHHHH

[olive sable]

*headpats*

[minor crag]

*head pats*

[vague sonnet]

\

[minor crag]

JREUDJDNS

I is saving to my clipboard for future use

[rigid snow]

bbbb

[olive sable]

bbbb

[rigid snow]

create aeronautics out

[olive sable]

so far im better at hydrodynamics

[rigid snow]

what about kinematics

[olive sable]

i know some of it

mainly the stuff i learned in school, and the blender rigging system

[rigid snow]

fk ik

[olive sable]

the hardest thing about ik is not even the math, its the collisions

cuz you can calcualte the position where the knee should be if the ground is at y height

but it the knee is then in a wall you're fucked

or you choose to ignore it

at the cost of it looking bad

[vague sonnet]

for a 3 point setup it's not even that bad tbh

since you can rotate

if there's bending there is a circular path on which the middle point can move

[olive sable]

you can rotate it a bit, but ideally your knees arent behind you

[vague sonnet]

yeah you clamp it

[olive sable]

ye

limited range

[vague sonnet]

not the right limits

it's upper leg that applies here

[olive sable]

they both apply

im going up from the feet

but you can go down from the hip

[vague sonnet]

nah like
you can check the knee collision after doing IK

[olive sable]

if you want to keep the body in place at least, you can do stuff in different ways

the issue is that if it collides and you have to check if a new position will work, then you might be wasting a lot of cpu for 20 different position checks

so its better to check as early as possible